Dynamics of Spaetzle morphogen shuttling in the Drosophila embryo shapes gastrulation patterning

The crude onset and subsequent refinement of spatial information shape the future body pattern of embryos. Morphogens, key instructive elements in this context, are secreted signaling molecules that induce cells to adopt different fates depending on the morphogen concentration they experience. Establishing a morphogen gradient over a field of naïve cells patterns the cell layer into distinct domains of gene expression (Green and Sharpe, 2015; Wolpert, 1971). A common strategy for generating a morphogen gradient is to produce the morphogen in a restricted group of cells, giving rise to its graded distribution in the surrounding cells (Lecuit et al., 1996; Nellen et al., 1996). In this scenario, the morphogen-producing cells are typically distinct from the responding cells.

An alternative strategy of morphogen distribution is applicable to situations in which the morphogen is broadly expressed, and the gradient is generated within the field of expressing cells. This scenario is applicable to early embryos, in which broad transcriptional domains have been established, but these domains are not sufficiently restricted to provide for a local morphogen source. In such situations, restricting morphogen signaling to a narrow domain becomes a challenge, as diffusion tends to spread, rather than restrict, ligand distribution.

Studies in several systems identified the ‘shuttling’ mechanism as a robust solution to this challenge (Shilo et al., 2013). Here, a morphogen gradient is established not merely by its diffusion away from the production source, but through an effective translocation of the morphogen into the center of the field. This translocation is purely diffusion driven, and is mediated by a distally produced inhibitor. The resulting gradient is sharp and robust, displaying limited sensitivity to gene dosages or reaction rate constants. Shuttling provides robustness by concentrating the morphogen into a restricted domain, which allows excess levels to be stored in regions of maximal signaling, without modifying the resulting cell fates. Morphogen gradients established by the shuttling mechanism include the bone morphogenetic protein (BMP) gradient in the early embryos of Drosophila, other insects and sea urchins (Eldar et al., 2002; Lapraz et al., 2009; Shimmi et al., 2005; van der Zee et al., 2006; Wotton et al., 2017). Shuttling is also used for forming the BMP gradient in Cnidaria (Genikhovich et al., 2015) and in the Xenopus embryo, where it has acquired additional features that allow for scaling of the gradient with embryo size (Ben-Zvi et al., 2014, 2008).

Compelling evidence for shuttling was provided by comparing mutant phenotypes with the predictions made by computational models (Ben-Zvi et al., 2008; Eldar et al., 2002; Haskel-Ittah et al., 2012). It was also demonstrated that ligand produced ectopically in one part of the embryo can be translocated to its normal activity zone, defined by the expression domain of the shuttling molecule (Reversade and De Robertis, 2005; Wang and Ferguson, 2005). Experimentally, these data were obtained through the analysis of fixed embryos. Yet, the essence of the gradient-forming mechanisms resides in their dynamics. What is the time frame during which the gradient is established? How fast is gradient formation relative to its readout, i.e. does the resulting gene expression take place after the gradient was formed and refined by shuttling or does gene expression respond dynamically to the continuous sharpening of the gradient? Furthermore, considering the dynamic environment of the early embryo – where the nuclei undergo multiple division cycles during which their environment is perturbed – is the gradient stably formed, or is it subject to subsequent cycles of refinement? Insight into these questions requires monitoring the dynamic distribution of the morphogen within living single embryos.

The ability to observe the dynamics of morphogen gradient formation is challenging because the ligands typically function at low levels. Furthermore, the morphogen may be present not only in its active form, but also in a non-processed, inactive form, or bound to an inhibitor. Finally, the morphogen is present in both extra- and intracellular locations. Most studies therefore follow the patterning processes by quantifying the intracellular outcome of morphogen signaling as a proxy for active morphogen distribution, using antibodies against the activated (e.g. phosphorylated) states of signaling pathways triggered by morphogens (Dorfman and Shilo, 2001; Gabay et al., 1997; Tanimoto et al., 2000). Such approaches, however, cannot be used to follow live embryos, as they rely on immunostaining of fixed samples. An alternative is to follow the transcriptional outcomes of morphogen signaling, but this analysis is already quite removed from the original morphogen gradient itself, and is compounded by additional regulatory inputs controlling the expression of the target genes.

Patterning the dorso-ventral (D-V) axis of the Drosophila embryo provides a powerful system in which to analyze the dynamics of morphogen gradient formation (Lynch and Roth, 2011). The early Drosophila embryo is a syncytium, i.e. a collection of nuclei that occupy a common cytoplasm enclosed by the embryonic plasma membrane. Within this syncytium, the nuclei undergo 13 rapid and synchronous divisions without any change in embryo size or shape. During the final four division cycles, the nuclei form a monolayer just beneath the plasma membrane (Foe and Alberts, 1983; Schejter and Wieschaus, 1993). The signaling protein Spaetzle (Spz) is secreted from the syncytium into the surrounding peri-vitelline fluid, positioned between the plasma and vitelline membranes, and a gradient of active Spz forms within it, along the radial D-V axis. The processed Spz morphogen binds the transmembrane Toll receptor positioned on the plasma membrane (DeLotto and DeLotto, 1998; Morisato and Anderson, 1994; Schneider et al., 1994; Weber et al., 2003), and the resulting signaling pathway triggers translocation of the transcription factor Dorsal into the syncytial nuclei. The processed form of the Spz ligand therefore functions as the morphogen at this stage.

We previously showed that graded active Spz distribution is established by a shuttling mechanism. In this case, shuttling is implemented in a self-organized manner through interplay between the active ligand and its pro-domain, which can accommodate diverse structures. Cleavage of the covalent link between the Spz pro-domain and the ligand retains the association between the two moieties but allows binding to Toll. Upon binding, the pro-domain is released. Occasionally the active ligand may also dissociate. Re-association between the ligand and the pro-domain generates an inactive complex that translocates the active ligand to the ventral midline (Fig. 1A,B) (Haskel-Ittah et al., 2012; Shilo et al., 2013). Shuttling results in a sharp ventral gradient of active Spz (Fig. 1C), and provides robustness to a variety of perturbations in the level of pathway components.

To visualize the dynamics of gradient formation, we considered the nuclear localization of Dorsal, a direct output of Spz signaling. Entry of Dorsal into the nuclei can be followed in single live embryos carrying a Dorsal-GFP fused protein (DeLotto et al., 2007). We use light-sheet fluorescence microscopy (LSFM) for live imaging of Dorsal-GFP nuclear localization during the final nuclear division cycles of the syncytial Drosophila embryo. The resulting dynamics show the two signatures of ligand shuttling within the broad ventral domain: a transient increase in signaling in the lateral regions, which is then reduced in order to preferentially increase signaling at the ventral midline, and the resolution of two lateral peaks to a single central (midline) peak.

We find that ligand shuttling is an ongoing process, which repeats itself following each nuclear division. Further, nuclear re-entry of Dorsal at the beginning of nuclear cycle 14 refines the spatial pattern of gradient readout, as closely positioned nuclei begin to express key zygotic target genes at slightly different times. This converts the morphogen gradient information to a graded cytoplasmic distribution of zygotic transcripts, which subsequently impinge on the pattern of gastrulation. The secreted protein WntD, an inhibitor of Toll signaling, buffers the D-V patterning gradient against fluctuations through an integral negative feedback (Rahimi et al., 2016). We now find that in wntD mutant embryos, the Dorsal peak becomes flattened, and leads to an abnormal increase in the number of cells simultaneously undergoing the initial step of gastrulation, highlighting the significance of properly shaped gradient formation. Thus, diffusion-based ligand shuttling, coupled with a dynamic readout, establishes a refined spatial pattern within the environment of early embryos.

The shuttling mechanism makes a number of counter-intuitive predictions regarding the dynamics of pattern formation. In particular, it predicts that signaling at positions that are at an intermediate distance from the center of the field will initially increase, as the ubiquitously produced ligand begins to accumulate, but will subsequently be reduced, because ligand is continuously being shuttled to the center of the field. This non-monotonic behavior is a defining property of the shuttling mechanism that concentrates ligand, but is absent from other diffusion-based mechanisms that establish a graded pattern. In a certain parameter range, shuttling also predicts transient formation of a double-peak pattern, as a consequence of the gradual ventral translocation of ligand from the two lateral edges. This property is also absent from naïve gradient-forming mechanisms that display temporally monotonic behavior throughout the tissue. Uncovering such features is possible only by monitoring the dynamics of gradient formation in live embryos.

Using LSFM, we followed individual embryos carrying a Dorsal-GFP fusion protein expressed under the endogenous dl promoter (Fig. 1D, Movie 1). Consistent with previous reports (DeLotto et al., 2007; Kanodia et al., 2009), we observed a D-V gradient of nuclear Dorsal-GFP already at nuclear cycle (NC) 12. This gradient was further refined and elaborated during the next two cycles. To enable quantitative analysis of Dorsal-GFP nuclear dynamics, we used an area-preserving transformation to project the 3D images onto a 2D sheet. We restricted our analysis to a region surrounding the antero-posterior (A-P) midline, where distortion due to 2D projection is negligible (Fig. 1E, Movies 2 and 3) (Heemskerk and Streichan, 2015). Next, we automatically segmented the nuclei and averaged the nuclear Dorsal-GFP signal over nuclei occupying a similar D-V axis position (Materials and Methods; Fig. S1).

Our measurements defined the quantitative, spatio-temporal dynamics of Dorsal-GFP at a 1-2 min time resolution (Fig. 1E,F; Materials and Methods). These dynamics are likely dictated by the extracellular active Spz gradient. However, inferring the profile of this extracellular gradient from Dorsal-GFP dynamics is confounded by the fact that Dorsal nuclear accumulation is established anew at every nuclear cycle, because Dorsal exits the nucleus at mitosis upon nuclear envelope breakdown. We therefore needed a framework to suitably infer properties of the extracellular active Spz gradient, and distinguish between models of gradient formation.

Toll signaling, at each given position along the D-V gradient, defines the rate by which Dorsal is translocated to the nucleus. Thus, at the beginning of each division cycle, following re-establishment of the nuclear envelope and before significant Dorsal nuclear accumulation is observed, nuclear Dorsal levels increase at a rate that is proportional to the level of nearby Toll signaling. Conversely, at later time points, nuclear Dorsal levels approach a steady state that is proportional to the extracellular Toll signaling itself. We therefore plotted the dynamics of both parameters, nuclear Dorsal levels and its temporal derivative, during the course of NC 14. Notably, we observed that these qualitative dynamics differ, depending on the spatial position of nuclei along the D-V axis. In the ventral-most regions they increased monotonically. In contrast, in lateral domains nuclear Dorsal displayed an overshoot, initially increasing but then starting to decrease (Fig. 2A-D, Fig. S2). Clearly, such a decrease in nuclear Dorsal is only possible if Toll signaling at this position decreases as well. Therefore, the data suggests that the external Spz gradient continues to evolve through the early part of NC 14, showing a distinct position-dependent, non-monotonic temporal signature.

To more rigorously infer dynamic properties of the external gradient from the measured pattern of nuclear Dorsal, we used computer simulations, modeling Dorsal nuclear entry while assuming different temporal patterns of Toll signaling (Fig. 2E-L). Specifically, we searched for a qualitative signature that distinguishes between three scenarios: (1) Toll signaling that is monotonically increasing in time, as expected in naïve gradient-forming models; (2) temporally constant Toll signaling; and (3) Toll signaling formed by a shuttling mechanism, giving rise to non-monotonic behavior in lateral regions. Our simulations showed that these scenarios are best distinguished by comparing the derivative of changes in nuclear Dorsal [d(Dl)/dt] with the levels of nuclear Dorsal. In the first two cases – constant or monotonically increasing Toll activity – the relationship between these two parameters was invariably linear or concave (Fig. 2E-L). In contrast, in a regime of shuttling-based dynamics, a convex relation was obtained, with a pronounced negative temporal derivative at the lateral regions (Fig. 2M-P).

The measured data are not consistent with dynamics that are defined by a monotonically increasing or a constant Toll signaling. Rather, it shows a clear signature of shuttling-like dynamics. Our simulations, which include the full shuttling model that establishes the active Spz gradient combined with Dorsal nuclear transport (see supplementary Materials and Methods), confirmed that this model is fully capable of recapitulating the experimentally observed dynamics, including the overshoot at the lateral regions.

An additional notable property of Dorsal nuclear entry dynamics was the initial formation, at every nuclear cycle, of two ventro-lateral signaling peaks, that eventually converge to a single ventral peak (Figs 1F and 2A, Fig. S2). Thus, by 10-15 min into NC 14, when the robust expression of zygotic Dorsal-target genes is induced, the initial two-peak gradient had refined to a single sharp peak. The initial two-peak pattern provides another unique signature of the shuttling dynamics. This is expected when parameters are such that the mean distance traveled by the shuttling complex before it is cleaved is much smaller than the distance to the ventral-most site. In this case, ligand will initially accumulate at lateral regions, followed by gradual ventral translocation (see supplementary Materials and Methods).

Although a sharp peak was observed at the end of NC 12 and NC 13, at the onset of the following cycle the pattern of nuclear Dorsal exhibited a double peak that converges over time to a sharp peak. The reappearance of the double peak at every nuclear cycle may reflect a process of extracellular ligand mixing in the peri-vitelline fluid, possibly caused by reorganization of the cortical actin-based cytoskeleton and deformation of the plasma membrane associated with the nuclear divisions (di Pietro and Bellaïche, 2018; Zhang et al., 2018).

Following Toll activation, the nuclear entry of Dorsal is facilitated by degradation of the Cactus (Cact) protein, which normally retains Dl in the cytoplasm (Belvin and Anderson, 1996). cact mutant embryos exhibit a general elevation in nuclear Dorsal levels. However, on top of this elevated background, peak ventral signaling is still observed due to Toll-mediated phosphorylation of Dorsal, which promotes its nuclear entry (Rahimi et al., 2016).

Previous work has suggested that the ventrally biased degradation of Cact may generate a cytoplasmic flux of Dorsal towards the ventral midline (Carrell et al., 2017). To determine whether the shuttling hallmarks we observed stem from extracellular Spz shuttling or intracellular Dorsal shuttling by Cact, we examined the dynamic distribution of nuclear Dorsal in cact knockdown embryos (Fig. 3A,B). If Cact-mediated intracellular shuttling is a significant driving force of the dynamics we observed, we anticipate elimination of shuttling signatures in this mutant background. This, however, was not the case: a double peak that converges to a single peak was also observed in cact knockdown embryos, and in fact appears to be even more prominent than that detected in wild-type embryos (Fig. 3C,D, Fig. S3). Note that, due to the expected global elevation in Toll signaling, it was not possible to reliably follow the feature of overshoot in lateral nuclei.

In conclusion, the dynamic behavior of Dorsal-GFP supports a continuous process of extracellular Spz shuttling, displaying two of its defining signatures: non-monotonic dynamics of nuclear Dorsal entry in lateral regions, and the transient formation of a double-peaked gradient.

Dorsal nuclear localization dynamics can be used for refined analysis of mutant phenotypes. We applied this approach to study WntD, an inhibitor of Toll signaling which provides a negative feedback that buffers the D-V patterning gradient against fluctuations (Rahimi et al., 2016). wntD, itself a target of the Toll pathway, is transcribed locally at the posterior terminus of the embryo, and the secreted protein diffuses within the peri-vitelline fluid to attenuate Toll signaling (Fig. 4A) (Helman et al., 2012; Rahimi et al., 2016).

Upon binding to the Frizzled 4 (Fz4) receptor, WntD is anchored to the membrane. Genetic epistasis experiments are consistent with WntD association with the extracellular domain of Toll, compromising Toll's capacity to bind Spz (Rahimi et al., 2016) (Fig. 4B). However, the consequences of this inhibition on gradient formation remained unclear. One possibility is that WntD uniformly decreases signaling in all regions. Alternatively, with the shuttling mechanism in mind, our simulations suggested that the binding of WntD to the extracellular domain of Toll promotes ligand shuttling. More specifically, within the model, binding of WntD to the Toll receptor decreases the number of available receptors, thus decreasing signaling preferentially in lateral regions and sharpening the gradient. Reducing the number of free receptors further compromises the ability of Spz to bind Toll, thereby increasing the probability that it will bind the free shuttling molecule, leading to more effective shuttling. Uniform distribution of moderate WntD levels would therefore result in redistribution of the ligand to more ventral regions, impacting not only the strength of Toll signaling, but also sharpness of its gradient (see Fig. 4C,F for simulation results, and supplementary Materials and Methods for model details).

Using live imaging in a wntD mutant background, we directly tested the impact of WntD on shuttling. The initial signature of a double peak was more prominent and persistent, consistent with higher signaling levels in the absence of WntD. Identification of shuttling in this background indicates that the core mechanisms of shuttling are still operating in the absence of WntD. However, in contrast to wild-type embryos, in which gradual sharpening of the gradient takes place, in a wntD-mutant background the final gradient was relatively wide and displayed a flattened, single peak (Fig. 4C-J, Fig. S4), consistent with the predicted role of WntD in promoting the ventral translocation of the Spz ligand (Fig. 4K,L).

To follow the morphological consequences of a wider peak of nuclear Dorsal distribution, we monitored wntD mutant embryos for an extended period of NC 14, observing the processes of gastrulation and ventral furrow formation. We defined the edges of the furrowing domain by marking the two lateral-most nuclei that alter their angle with respect to the embryo circumference upon gastrulation. Working backwards to an earlier phase of NC 14, when the nuclei are still in a monolayer, we can accurately count the number of nuclei between these edges. In contrast to gastrulating wild-type embryos, in which the initial invagination was observed in approximately ten cells, in wntD mutants a broader front of up to 15 cells invaginated at the same time (Fig. 5). Thus, the shape of the Dorsal activation gradient is essential for normal patterning and gastrulation. When the final gradient peak is not sharp, a larger cohort of ventral cells takes part in furrow formation.

wntD mutants display perturbed Dl dynamics already at NC 13 (Fig. 4H, Fig. S5) suggesting that WntD normally exerts its gradient-modulating effects at this early stage. This implies that zygotic expression of wntD, its translation, secretion to the peri-vitelline fluid and diffusion of the protein, have commenced by then. To verify that this is possible, we applied our LSFM visualization setup as a sensitive assay for defining the onset and temporal dynamics of wntD expression.

wntD expression is activated by nuclear Dorsal, and is restricted to the posterior region, where Torso signaling relieves Capicua (Cic) repression of wntD (Helman et al., 2012). We generated a wntD::MS2 reporter, utilizing wntD genomic upstream regulatory sequences (Fig. 6A). The early expression profile of this reporter at the posterior part of the embryo mimics the known pattern of wntD expression (Rahimi et al., 2016). Importantly, all eight embryos examined expressed wntD, implying that excess of Toll signaling, which triggers expression of the WntD ‘buffer’, is a common consequence of D-V gradient signaling. When monitoring the dynamics of wntD expression, the number of nuclei transcribing wntD decreased continuously between NCs 12 and 14 (Fig. 6B-F). These results suggest that the WntD protein already exerts its attenuating effect on Toll signaling at an early time. Notably, the diminishing wntD expression in NCs 13-14 is attributed to a reduction in Toll signaling rather than in Torso signaling, as constitutive activation of Toll leads to stable wntD expression throughout NC 14, indicating the persistence of Torso signaling at this phase (Rahimi et al., 2016). The marked reduction in active transcription of wntD at NCs 13-14 is in contrast to the persistence of wntD transcripts (Rahimi et al., 2016), suggesting that the transcripts produced at earlier cycles were not yet fully degraded.

The gradient of Dl nuclear localization defines three major domains of zygotic gene expression along the D-V axis. Within each of these regions, several target genes are uniformly expressed. The mesoderm is defined by highest levels of nuclear Dorsal and uniform expression of the zygotic target genes twist (twi) and snail (sna) (Rusch and Levine, 1996). There is, however, an additional requirement of a graded zygotic response within the mesoderm: a gradient of apical myosin II recruitment, peaking at the ventral midline, is essential for the ordered apical cell constriction that drives ventral furrow formation (Heer et al., 2017).

Can graded activation of the Toll pathway define not only the broad zones of zygotic gene expression, but also the graded expression pattern of cardinal target genes? We reasoned that this may be the case because, if zygotic expression of Dorsal-target genes depends not only on the final, steady-state level of nuclear Dorsal, but also on the dynamic profile of its accumulation, the signaling output could be further sharpened. A dynamic phase of Dorsal nuclear entry takes place during the initial 20 min of NC 14 (Figs 1F and 2B), the nuclear cycle associated with a major onset of zygotic gene expression. A consequence of these dynamics is that ventral-most nuclei will reach the threshold for expression of a given zygotic gene earlier than more lateral ones. These ventral nuclei will begin to express the gene earlier, therefore expressing it for a longer period than more lateral nuclei. Transcript accumulation should therefore be ventrally biased, if these transcripts are sufficiently stable.

A zygotic target gene that may lead to graded myosin II distribution is T48, because the T48 protein recruits RhoGEF2 to the apical membrane, triggering the accumulation and contractile activity of an apical actomyosin network (Kolsch et al., 2007). We therefore examined whether the dynamics of Dorsal nuclear entry play a role in generating a graded transcriptional response within this region, by monitoring expression of the T48 target gene. Previous analysis of an MS2 reporter for live T48 transcription demonstrated that the ventral-most nuclei initiate transcription 15-20 min earlier than the lateral ones (Lim et al., 2017). As T48 expression requires not only Dorsal but also the product of the Dorsal-target gene twi (Urbansky et al., 2016), it will be triggered by Dorsal nuclear entry only at NC 14, and not at earlier cycles.

To examine the consequence of the graded onset of T48 transcription on mRNA accumulation, we carried out quantitative single-molecule fluorescence in situ hybridization (FISH) using T48 probes. The signal obtained comprised two components: prominent puncta in the nuclei representing active transcription of the gene, and sparse weaker spots in the cytoplasm marking accumulated individual mRNA molecules. We focused on embryos that demonstrated ongoing transcription of T48 in all ventral nuclei (Fig. 7A), allowing us to monitor the level of active transcription in each nucleus at the time of fixation, as well as the quantity of transcripts that accumulated in the cytoplasm since the onset of transcription. The number of transcribing nuclei in ventral and lateral positions within the expression domain was similar, and the transcription intensity in the different rows appeared comparable (Fig. 7B,C). Most nuclei exhibited two foci, indicating that both chromosomes were transcribing. These features indicate that at the level of individual nuclei, once T48 transcription is initiated, it progresses at a constant rate independently of the position of the nucleus along the D-V axis.

We next quantified the levels of cytoplasmic T48 mRNA, using TransQuant software (Bahar Halpern et al., 2015). A clear D-V gradient of cytoplasmic mRNA levels was observed across several cell rows, peaking at the ventral midline (Fig. 7B,D). This result indicates that the T48 mRNA is sufficiently stable during the temporal window of early NC 14, such that the time of onset of its transcription along the D-V axis, governed by the dynamics of Dorsal nuclear entry, correlates with the level of mRNA that accumulates in the adjacent cytoplasm. Dorsal nuclear entry dynamics thus appear to be a crucial factor regulating graded T48 levels along the D-V axis.

Because nuclear Dorsal displayed a flat peak in wntD mutant embryos, we predicted that these embryos would display an altered distribution of T48 transcripts. Indeed, single-molecule FISH in wntD mutant embryos at the stage when all ventral nuclei express the T48 gene (Fig. 7E, Fig. S6A-D) revealed a significantly shallower gradient of accumulating T48 transcripts in the cytoplasm (Fig. 7F-H, Fig. S6E-H). Based on the role of T48 in triggering apical constriction in the invaginating mesoderm (Kolsch et al., 2007), the uniform accumulation of T48 transcripts could contribute to the broader invaginating front that we observed in wntD mutants (Fig. 5). Thus, WntD activity is important not only for restricting the zone of sna and twist expression, but also for maintaining a sharper peak of Dl nuclear entry, limiting the number of invaginating cells at the onset of gastrulation.

The milieu of the early Drosophila embryo poses extreme challenges for the generation and maintenance of extracellular morphogen gradients. Most notably, the peri-vitelline fluid surrounding the embryo facilitates rapid diffusion of molecules (Stein et al., 1991). Analysis of the early morphogen gradients operating in this environment, including ventral Spz/Toll activation and the subsequent BMP gradient patterning the dorsal aspect, should consider this highly dynamic environment. In the case of the Toll pathway, the active Spz ligand is generated by proteolytic processing within the extra-embryonic peri-vitelline fluid in a broad ventral region, defined by the activation domain of the Easter (Ea) protease (Cho et al., 2012). Diffusion-based shuttling can continuously sharpen the Spz activation gradient, within this broad ventral domain. While a variety of experiments and computational analyses indicated the utilization of a ‘self-organized shuttling’ mechanism that is driven by the Spz pro-domain (Haskel-Ittah et al., 2012), it was imperative to infer the actual dynamics of the process.

We were able to infer the dynamics of the extracellular Spz gradient by following the kinetics of Dorsal-GFP nuclear accumulation in individual live embryos, during the final syncytial nuclear division cycles and the early phase of NC 14. We identified clear hallmarks of ligand shuttling, most notably the initial non-monotonic dynamics (overshoot) predicted in lateral regions, and the presence of two lateral peaks which converge to a central ventral peak. This convergence takes place within a time frame of minutes, and repeats at every nuclear cycle. Because new protein molecules of the extracellular components are continually translated, the ongoing activity of the shuttling process is vital. Therefore, shuttling is important not only for generating the gradient, but also for maintaining it, in the face of rapid diffusion and mixing within the peri-vitelline fluid. Importantly, by ∼10-15 min into NC 14, when the robust induction of transcription of the cardinal zygotic Dorsal-target genes twi and sna initiates, the nuclear gradient of Dorsal is sharp and a single ventral peak is resolved.

Having described the dynamics of Dorsal nuclear entry and gradient formation, we were in a position to use our experimental approach in order to examine regulatory processes affecting Toll signaling. The Wnt family ligand WntD provides an essential buffering system to variations in Toll signaling between embryos (Rahimi et al., 2016). WntD is likely to exert its inhibitory effect by associating with the extracellular domain of Toll (Rahimi et al., 2016), thereby reducing the number of Toll receptors that are available for binding Spz. As shuttling relies on competition between ligand binding to the shuttling molecule versus the receptor, this mode of inhibition implies that the effect of WntD would sharpen the shape of the global gradient. The observed dynamics of Dorsal-GFP in wntD mutant embryos reported here indeed confirm this prediction. Because WntD impinges on the extracellular properties of the Toll receptor, the active ligand is deposited in more ventral regions, where the concentration of the pro-domain is lower. Thus, WntD does not simply reduce the overall profile of Toll activation, but actually re-directs the ligand from the lateral regions to the ventral domain, as shown by our simulations. We have previously shown that shuttling-dependent accumulation of excess ligand at the center of the field is an effective mechanism to buffer noise. Because activation in this region is already maximal, the excess ligand will not alter the resulting cell fates (Barkai and Shilo, 2009; Shilo and Barkai, 2017).

WntD is produced locally, and the rapid secretion and diffusion of the protein in the peri-vitelline space generates a uniform attenuation of Toll signaling throughout the embryo surface. The rapid timing of processes in the early embryo and the short duration of interphases between nuclear divisions raise the question of whether it is actually possible to produce sufficient levels of WntD that will drive the morphogen distribution to its desired steady-state profile. When monitoring wntD transcription directly utilizing the MS2 system, we saw that most, if not all, embryos express wntD, indicating that Toll signaling overshoots in most embryos. Furthermore, within single embryos the number of nuclei expressing wntD was reduced between NCs 12 and 13, and completely terminated by NC 14, implying that WntD impinges on the Toll gradient and its own expression by this time. The intron-less arrangement of the wntD gene and the rapid secretion of the protein, which does not require post-translational modifications (Herr et al., 2012), may facilitate the process.

The ventral cohort of zygotic target genes, including twi and sna, is induced by the Toll activation gradient, and the threshold for their induction corresponds to ∼50% of maximal Dorsal nuclear localization (Kanodia et al., 2009; Liberman et al., 2009). Within the ventral domain, nuclei exhibit a similar level of sna transcription (Bothma et al., 2015; Lagha et al., 2013). Robust transcription of these genes ensues at NC 14, after the Dorsal gradient is stabilized and a distinct activation peak generated.

Are there zygotic target genes that respond to the dynamics of Dorsal nuclear targeting, before it stabilizes, within the first 15 min of NC 14? This appears to be the case for T48, which encodes a transmembrane protein that facilitates recruitment of RhoGEF2, and ultimately Rho1 and actomyosin, to mediate apical constriction of invaginating mesodermal cells (Kolsch et al., 2007). Graded distribution of myosin II was shown to be crucial for proper invagination of these cells to form the ventral furrow (Heer et al., 2017).

We provide evidence that the graded distribution of T48 mRNA results from the dynamics of Dorsal nuclear re-entry at NC 14. The ventral-most cells reach the threshold of T48 induction earlier than more lateral cells, and hence will express the gene longer (Lim et al., 2017). Integration of the length of expression along the D-V axis could then lead to a gradient of cytoplasmic T48 mRNA levels, if the T48 transcripts are sufficiently stable. This example represents a unique case, in which graded morphogen activation instructs the generation of a gradient of target gene expression. The strict dependence on the timing of transcription initiation provides a mechanism to generate differences in transcript levels between adjacent nuclei along the D-V axis. This mechanism is distinct from a more trivial scenario, in which at steady-state the activity of the morphogen signaling pathway within each nucleus will be linearly translated to the rate of target gene expression.

In conclusion, this work has utilized live imaging of the outcome of Toll pathway activation, to identify and characterize the hallmarks of ligand shuttling (Fig. 8). We find that this process is rapid and takes place continuously throughout the final nuclear division cycles, to generate and maintain a sharp activation gradient in the diffusible environment of the peri-vitelline fluid. WntD impinges on Spz shuttling, and is responsible not only for buffering variability between embryos, but also for generating a sharp activation peak. This peak may be utilized to induce a graded expression of zygotic target genes that are essential for executing processes that drive gastrulation. Thus, diffusion-based ligand shuttling, coupled with a dynamic readout, establishes a refined pattern within the dynamic environment of early embryos.

For the Dorsal-GFP experiments we used Sco/CyO; dl-GFP/TM3, Sb flies as the wild-type genotype (DeLotto et al., 2007). The wntD null allele wntD^CR1 served as the wntD mutant background. It was generated by CRISPR-induced excision of the entire wntD gene, using the following oligonucleotide pairs to generate the guide RNAs: CTTCGAAACCACCTGTAGCTAAAAC, AAACGTTTTAGCTACAGGTGGTTTC; and CTTCGAAGTCCTGTCTGCGTAGCAC, AAACGTGCTACGCAGACAGGACTTC.

cact^GL00627 (Bloomington Drosophila Stock Center stock 37484) was used for cact RNAi-based knockdown, and induced maternally by nos-Gal4.

To generate wntD::MS2-Fhos-RH, the 1162 bp region upstream of the wntD transcription start site was synthesized and followed by 24 repeats of the MS2 sequence, and placed within the 5′UTR. The sequence was inserted into a pAttB vector with NotI and KpnI sites. The Fhos-RH sequence was further ligated into the NotI site to generate wntD::MS2-Fhos-RH. Virgin females expressing both MCP::GFP and His:RFP were crossed with males of the reporter line wntD::MS2-Fhos-RH to collect embryos for imaging.

Embryos were imaged using a Light Sheet Z.1 microscope (Zeiss) equipped with two sCMOS cameras PCO-Edge, 10× excitation objectives and Light Sheet Z.1 detection optics 20×/1.0 (water immersion). The embryos were collected, dechorionated, and up to four embryos were sequentially mounted perpendicularly into a glass capillary (Brand) in a 1% low melting agarose solution (Roth). Imaging was performed using dual-side illumination, zoom ×0.8. GFP excitation: 488 nm, emission/detection - BP 505-545; RFP excitation: 561 nm, emission/detection: BP 575-615 nm.

Stellaris RNA FISH probe sets for the T48 gene (5′ UTR and coding, not including the 3′ UTR), were designed by Stellaris Probe Designer and purchased from LGC Biosearch Technologies. Three hours after egg laying (AEL) wt embryos were fixed for 25 min in 4% formaldehyde, washed in methanol and kept at −20°C. The following day, embryos were washed in methanol and then in ethanol, rocked in 90% xylene, 10% ethanol for 1 h, post-fixed in 4% paraformaldehyde/PBS/0.05 M EGTA for 25 min, washed three times (10 min each wash), then incubated for 6 min with Proteinase K and post-fixed again. Embryos were transferred gradually to 10% FA (Formamide Deionized, Thermo Fisher Scientific, AM9342) in 2× SSC+10 μg/ml ssDNA preheated to 37°C and prehybridized for 30 min at 37°C. Hybridization buffer included 10% FA, 10% dextran, 2 mg/ml bovine serum albumin (Thermo Fisher, AM2616), RVC (Ribonucleoside Vanadyl Complex, NEB, S1402) and ssDNA+ tRNA in 2× SSC, containing the probe set (1 ng/μl) (Trcek et al., 2017). Hybridization was carried out overnight at 37°C. Next morning the embryos were shaken gently and incubated for another 30 min. Embryos were washed twice for 30 min each wash at 37°C with 10% FA in 2× SSC+10 μg/ml ssDNA and gradually transferred to PBS-0.5% Tween and mounted with Vectashield+DAPI Mounting Medium (Vector Laboratories). Fluorescence was visualized with a Nikon Eclipse Ti2 microscope, and analyzed by the TransQuant script as previously published (Bahar Halpern and Itzkovitz, 2016). TS intensities were measured using ImageJ.

To enable quantitative analysis of the nuclear Dl gradient, we projected the 3D scans of the embryo from the light-sheet microscope, into a 2D flat surface, for every time point imaged. This was possible, because all the nuclei are arranged on the surface of the embryo, the shape of which resembles an ellipsoid. This ellipsoid can be projected onto a 2D surface, which contains all the nuclei and therefore the entire nuclear Dorsal gradient. To this end, we used an area-preserving transformation with minimal distortion far from the anterior and posterior poles, implemented by the IMSANE tissue cartography tool (Heemskerk and Streichan, 2015). IMSANE was used with the following specifications: Planar Illastik surface detector and cylinder chart type. Surface detection was performed at the last time point for each embryo, and the detected surface was then used to project all earlier time points. Because the embryo is, to a good approximation, a cylinder apart from the anterior and posterior poles, embryo circumference was defined as the largest circumference of the ellipsoid fitted to the embryo surface by IMSANE.

The nuclei were detected separately for each time point, using the following segmentation method. Automated local thresholding of the image was carried out in order to create a binary mask. This was done in ImageJ using the Bernsen algorithm with a contrast threshold of 15. The resulting binary mask underwent further refinement to segment the nuclei using MATLAB image analysis filters as follows: (1) all connected objects in the mask large enough to be nuclei (over 50 pixels in size) were located and classified into three size groups (small, 50-150 pixels; medium, 150-600 pixels; large, 600 pixels and over), (2) each size group underwent erosion using ‘imerode’ and then dilation using ‘imdilate’ with appropriate filter sizes for each group, and (3) objects belonging to the ‘large’ group underwent another round of erosion and dilation with same filter size as in the first round.

The resulting objects in the binary mask were filtered by size to exclude objects too small (under 50 pixels) or too large (over 3000 pixels) to be a single nucleus. Nuclei locations were detected by overlaying the mask on the original image.

Gradient measurement was performed by first manually discarding all time points between the nuclear cycles. Then, for every detected nucleus, for all remaining time points, the value of nuclear Dorsal was calculated as the mean intensity inside the nucleus (located by the above segmentation method). The location of the A-P midline was manually selected for each embryo. A rectangular area around the A-P midline was then defined. The width of the area along the A-P axis was 15% of the entire A-P length and it spanned the entire D-V axis. For NC 14, this definition corresponds to approximately eight to ten columns of nuclei closest to the A-P midline. Only the nuclei inside this area were taken into account for gradient measurement; monitoring several columns of nuclei along the A-P axis gave rise to averaging of Dorsal nuclear intensity along this small window. The spatial axis for the gradient was defined as a relative axis, x/L, indicating the location on the D-V axis, x, divided by embryo circumference, L. In order to assign a location on this relative D-V axis for each nucleus, the location of the D-V midline was manually selected and defined as x/L=0. This resulted in the raw intensity function, Dl_raw(x/L, t) measuring nuclei intensity along the relative D-V axis, over time. This function, Dl_raw(x/L, t), was then smoothened in space for each time point separately using the MATLAB ‘smooth’ function with a smoothing coefficient of 0.23. The smoothened data was then fitted by a smoothing spline using the MATLAB ‘fit’ function and evaluated on 1000 linearly spaced x/L locations. The resulting function, Dl_smooth(x/L, t), was plotted in the main text figures.

For the calculation of Dorsal nuclear intensity over time at a specific location, x̃/L, we used Dl_smooth(x/L, t) at that location: Dl_smooth(x̃/L, t). Background subtracted values were calculated separately for each nuclear cycle by subtracting the minimal intensity observed in a nucleus for that nuclear cycle. Dl_smooth(x̃/L, t) was then smoothened in time using the MATLAB ‘smooth’ function with the LOESS method and a smoothing coefficient of 0.5. It was then fitted with a smoothing spline, resulting in the function Dl_x̃/L^fitted(t). The temporal derivative of nuclear Dorsal, at a specific location, x̃/L, was calculated by applying a third-order finite differences formula to Dl_x̃/L^fitted(t) and then smoothing using ‘smooth’ with a smoothing coefficient of 0.6 and fitting a smoothing spline.

The peak sharpness measure for an embryo was calculated based on the values of Dl_smooth(x/L, t), around the ventral-most location. For each time point, t̃, Dl_smooth(x̃/L, t) was normalized by dividing by its maximal value at t̃. Peak sharpness was then calculated as the standard deviation divided by the mean, in percentage, of values close to the peak, i.e. within the range (x/L)∈[−0.1, 0.1]. This measure captures how different from each other are values close to the peak.

We are grateful to Y. Addadi and O. Golani for help in acquisition and analysis of LSFM images, to S. Itzkovitz for advice on single-molecule FISH, and S. Streichan for support in 3D image processing. Imaging using the light-sheet microscope was made possible thanks to the de Picciotto-Lesser Cell Observatory in memory of Wolf and Ruth Lesser. We thank the members of the Shilo and Barkai labs for fruitful discussions.