Wnt-regulated dynamics of positional information in zebrafish somitogenesis

Segmentation of the anteroposterior axis is a widespread feature of animal body plans. Interpretation of positional information from signaling gradients (Aulehla and Pourquié, 2010; Chen et al., 2012; Porcher and Dostatni, 2010; Rogers and Schier, 2011) appears key in this process despite different modes of segmentation in different phyla. In the syncytial blastoderm of the fruit fly Drosophila, molecular signaling gradients supply positional information to initiate a hierarchical genetic cascade that simultaneously subdivides a body of fixed size into segments of similar length (Chen et al., 2012; Porcher and Dostatni, 2010). By contrast, vertebrate segmentation, known as somitogenesis, occurs sequentially and rhythmically at the posterior of a continually elongating body axis where multicellular blocks, known as somites, are periodically segregated from the anterior of the presomitic mesoderm (PSM) (Pourquié, 2011). Somitic precursor cells continuously join the posterior end of the PSM during elongation, creating an anteriorward cell flow through the PSM. Signaling gradients in the PSM are believed to supply positional information that regulates segment length (Aulehla and Pourquié, 2010), but how positional information is set in a continuous flow of cells remains unclear.

The rhythmicity of somitogenesis is governed by a segmentation clock consisting of cellular oscillators (Harima et al., 2013; Oates et al., 2012). Kinematic waves of oscillating gene expression sweep anteriorly across the PSM, arriving in the anterior as each new segment forms (Aulehla et al., 2008; Delaune et al., 2012; Masamizu et al., 2006; Palmeirim et al., 1997). These waves are believed to arise from the gradual slowing of oscillations in individual cells as they approach the anterior PSM (Morelli et al., 2009). This tissue-level pattern of oscillating gene expression is thought to repeat as each new somite forms, and the time interval of the repeat is termed the period of the segmentation clock.

The anteroposterior (A/P) length of a segment is proposed to be set when oscillating cells interact with a wavefront of cellular change that sweeps posteriorly down the axis (Cooke and Zeeman, 1976). In this scenario, which is termed the clock and wavefront mechanism, the wavefront converts the temporal information of the oscillating cells into a permanent segmental pattern (Oates et al., 2012). The clock and wavefront mechanism is a generic proposal about the dynamics of the system and, regardless of the molecular details, at steady state the segment length is fixed by the period of the segmentation clock and the wavefront velocity (Gomez et al., 2008; Morelli et al., 2009; Oates et al., 2012). Events associated with the wavefront include oscillation arrest and determination of segment boundary position; morphological somite formation is downstream of these rhythmic patterning events (Dahmann et al., 2011). It is therefore expected that segment length and the segmentation clock period are closely related to the morphologically observable somite length and somitogenesis period, respectively.

The currently favored hypothesis for how wavefront velocity is regulated involves gradients of Wnt, fibroblast growth factor (Fgf) and retinoic acid (RA) spanning the PSM (reviewed by Aulehla and Pourquié, 2010). Wnt and Fgf ligands are transcribed in the regressing tailbud and are thought to build gradients by gradual decay of ligand mRNA and/or protein in the cell flow across the PSM (Aulehla et al., 2003; Dubrulle and Pourquié, 2004). The contribution of potential ligand diffusion (Yu et al., 2009) to the shape of the gradients, or to the downstream gradients of signal reception, has not been examined. It remains difficult to interpret the effects of perturbations to these signaling gradients in regulating wavefront velocity because Wnt, Fgf and RA also play important roles in embryonic elongation and posterior body development (Wilson et al., 2009). Experimental uncoupling of gradient dynamics from posterior development and elongation of the embryo would simplify interpretation. Furthermore, to interpret changes in segment length resulting from signaling gradient perturbation, wavefront velocity and clock period must be simultaneously assessed. To date, these criteria have not been met in any system and the specific roles of Wnt, Fgf or RA in supplying positional information in somitogenesis remain unclear (Oates et al., 2012). Although the wavefront may be influenced by each of these signals, here we focus on the role of Wnt signaling in regulating wavefront velocity.

Current understanding of the role of Wnt in regulating the wavefront comes from experiments in mouse and chick embryos. A qualitative gradient of Wnt signaling activity has been reported across the mouse PSM, with highest levels in the posterior (Aulehla et al., 2008). Ubiquitously active Wnt signaling in the mouse PSM dramatically changes the position of oscillator arrest, creating a longer PSM (Aulehla et al., 2008; Dunty et al., 2008). Furthermore, implantation of Wnt3-overexpressing cell clusters in chick PSM shifts the locally forming somite boundary anteriorly (Aulehla et al., 2003). Consistent with these observations, it was proposed that posterior PSM cells are maintained in an undetermined and oscillating state by high levels of Wnt signaling, and that the wavefront is triggered directly in anterior PSM below a concentration threshold in a Wnt activity gradient spanning the PSM (Aulehla and Herrmann, 2004; Aulehla et al., 2003). However, in the absence of quantitative analysis of the signaling gradient and the timing of its downstream targets during segmentation, it is still an open question as to when and where Wnt signaling supplies positional information to the flow of cells in the PSM.

Here, using time-controlled manipulation of Wnt signaling and time-lapse imaging of live zebrafish embryos, we define conditions under which wavefront velocity can be tuned without perturbing embryonic elongation or somitogenesis period, resulting in segments of abnormal length. Measurements of the distribution and dynamics of the Wnt signaling gradient, as well as the timing of downstream events, reveal that the wavefront is not triggered directly in the anterior PSM by a Wnt activity threshold. Instead, our results support a model in which Wnt-regulated positional information in the posterior PSM is relayed and refined by downstream targets and is transported by cell flow to the anterior PSM.

A role for Wnt signaling in zebrafish segmentation is not yet established, partly owing to evolutionarily conserved early roles in A/P axis specification (Shimizu et al., 2005; Thorpe et al., 2005) and later axial elongation (Martin and Kimelman, 2009; Martin and Kimelman, 2012). To avoid these roles of Wnt, we explored time-controlled inhibition or activation of the Wnt pathway during somitogenesis. We used heat shock-inducible transgenes expressing either Axin1, an intracellular inhibitor of the pathway (Kagermeier-Schenk et al., 2011); Dkk1, an extracellular inhibitor (Stoick-Cooper et al., 2007); or Wnt8, a ligand (Weidinger et al., 2005). Transgenic embryos and their wild-type (WT) siblings were briefly heat shocked during trunk somitogenesis (1- to 9-somite stage), then elongation and somite length were measured from brightfield time-lapse movies of multiple embryos developing simultaneously. We measured the elongation velocity e relative to a reference point in the already formed somites, along the line of the axis (Fig. 1A,B, white line).

Inhibition of Wnt signaling by axin1 overexpression dramatically slowed embryonic elongation, thus preventing specific analysis of segmentation (supplementary material Fig. S1A-C′, Movie 1). Consistent with previous work, overexpressing Dkk1 at the 1-somite stage also led to posterior truncation (Martin and Kimelman, 2012); however, elongation was not slowed until the twelfth to thirteenth somite post-heat shock (phs) (Fig. 1A-C′; supplementary material Table S1, six out of seven experiments), leaving the interval up to 13 somites phs open for segmentation analysis. Wnt8 induction did not affect elongation.

We observed significant changes in somite length following overexpression of Dkk1 or Wnt8 (Fig. 1A′,B′,D; supplementary material Fig. S1D-G, Movies 2, 3) compared with WT siblings, starting ∼8 somites phs (Fig. 1D; supplementary material Fig. S1F,G). Inhibition of Wnt signaling by Dkk1 overexpression led to the systematic formation of up to six longer somites starting eight somites phs [Fig. 1D; 21±7% longer (mean ± s.d.), seven independent experiments, n=50]. These effects were consistent for heat shocks delivered at 1- to 9-somite stages. Activation of Wnt signaling by Wnt8 overexpression had the opposite effect: formation of up to three smaller somites (supplementary material Fig. S1F,G; ninth somite 14±6% smaller, four independent experiments, n=29). Although the eighth and ninth somites phs were smaller in all experiments, this effect was not fully penetrant.

In all transgenic and WT embryos, the sixth somite formed phs was significantly shorter and sometimes followed by a longer somite (Fig. 1D; supplementary material Fig. S1F). This effect was due only to the heat shock, and proved a useful temporal mark to compare experiments. As a long-term effect, shorter somites formed after Dkk1 overexpression starting at ∼17 somites phs (Fig. 1D), possibly as a consequence of the system’s recovery. However, because of associated effects on tail outgrowth at these later stages interpretation is complicated. We conclude that time-controlled modulation of the canonical Wnt pathway by Dkk1 or Wnt8 induction during somitogenesis modifies somite length, starting eight somites phs, without affecting embryonic elongation.

To test whether this effect on somite length results from an earlier defect in segmental patterning, we measured the distance between mespb gene expression stripes in the anterior PSM. mespb is a marker and determinant of the rostral compartment of the presumptive somites of zebrafish (Sawada et al., 2000) and other vertebrate species (Buchberger et al., 2000; Saga et al., 1997). In zebrafish, the distance between the posterior two stripes is one of the earliest markers of segment length [presumptive somite -I, where somite I is the most recently formed somite (Pourquié and Tam, 2001)] (Fig. 1E,F; supplementary material Fig. S1H,I). At 3.5 hours (∼7 somite cycles) phs, the length of the mespb interstripe was 32±24% larger after Wnt inhibition and 14±9% smaller after Wnt activation (Fig. 1E-G; supplementary material Fig. S1H-J). These values are qualitatively consistent with the effects on somite length and show that segment length in the anterior PSM had already been altered ∼6 somites after heat shock modulation of Wnt signaling (Fig. 1G; supplementary material Fig. S1J). In conclusion, these results are the first evidence that Wnt regulates segmentation in zebrafish.

The change in segment length that we observed could have resulted from a modification in the clock period T and/or the wavefront velocity v. From time-lapse movies, we measured the period of somitogenesis T_s by calculating the slope of the linear increase in somite number versus time (Fig. 2A,A′) (Schröter et al., 2008). We use this period as a proxy for segmentation clock period. Because effects on somite length using Wnt8 overexpression were not fully penetrant and generally restricted to two somites, we only analyzed Dkk1 overexpression. We delivered heat shocks in separate experiments between the 1- and 9-somite stages and measured T_s during the formation of the longer somites, which always occurred 8-13 somites phs (Fig. 2A, dashed lines). We observed no significant increase in somitogenesis period (six out of seven experiments; Fig. 2A′; supplementary material Table S1). This suggests that the increase in somite length is not a consequence of an altered clock period.

We next investigated wavefront velocity as the other potential cause of change in segment length. Because the molecular events associated with the underlying wavefront are not visible in brightfield images, we first measured the progress of a morphological wavefront of somite boundary formation. The morphological wavefront velocity v_m (which anteriorly reduces PSM length) and the elongation velocity e (which posteriorly increases PSM length) combine to set PSM length L (Fig. 2B). If both velocities are equal, PSM length stays constant. However, this is not the case over longer developmental times, when PSM length changes in mouse, chick, snake and fish embryos (Gomez et al., 2008; Schroter and Oates, 2010).

The change in PSM length dL/dt measured from time-lapse movies was significantly increased after Wnt inhibition in six out of seven experiments (Fig. 2C,C′; supplementary material Table S1). To estimate the morphological wavefront velocity v_m, we subtracted dL/dt from the values of elongation measured above. This showed that v_m was increased by 17±10% (Fig. 2D; seven independent experiments, n=50; supplementary material Table S2). The increase in v_m was greater for early- to mid-trunk somitogenesis (30±13%) than for late trunk somitogenesis (14±9%). We then used our measurements of somite length S and somitogenesis period T_s for an independent estimate of v_m using v_m=S/T_s (supplementary material Table S2). With this method, we found a consistent increase of 21±14% after Wnt inhibition. Combined, these results reveal for the first time a specific alteration to morphological wavefront velocity in response to an alteration in Wnt signaling.

If our measurements at the morphological level reflect a change in the underlying wavefront velocity, we would expect to see a corresponding positional change in segment determination and oscillator arrest in the anterior PSM. We therefore examined segment polarity and oscillating gene expression by in situ hybridization in embryos fixed during specification of the larger segments.

We used the position of the posteriormost mespb stripe in the PSM as a marker of the determination front, as previously (Herrgen et al., 2010; Schroter and Oates, 2010). At 3.5 hours (∼7 somites) phs, the distance between the mespb stripe and the posterior end of the notochord was shorter in hsdkk1 transgenics than in WT littermates (91±9%, n=18; Fig. 3A-E). Time series analysis revealed that the effect started at 3 hours (∼6 somites) phs (Fig. 3E). This suggests that the position of the determination front is shifted posteriorly in the PSM by a reduction in Wnt signaling.

We next asked whether the position of oscillation arrest is also shifted posteriorly in response to Dkk1 induction. We examined expression of the cyclic genes her1, her7 and dlc. All are still expressed 3.5 hours (∼7 somites) after heat shock induction of Dkk1 (Fig. 3F-I; supplementary material Fig. S2A-D; data not shown). We observed different phase patterns among Dkk1-expressing embryos within the same clutch, indicating that cyclic genes were oscillating. We estimated the length of the oscillating domain using the distance from the posterior end of the notochord to the anterior margin of the anteriormost cyclic expression stripe of her7. The position of cyclic gene expression stripes in the PSM is thought to change continuously through the segmentation clock cycle. To accurately estimate changes, it is important to compare gene expression stripes in different embryos at similar cycle phases. We subdivided the embryos into two phase-matched groups: those with three stripes (Fig. 3F,H) and those with two stripes (Fig. 3G,I). The distance to the anteriormost stripe was shorter after heat shock in hsdkk1 embryos than in WT littermates for both the two- and three-stripe categories (supplementary material Fig. S2E), indicating that the length of the oscillating domain is reduced. Indeed, this effect was strong enough to be observed when embryos were pooled regardless of phase. A decrease in oscillating domain length was also systematically observed in several experiments using dlc and her1 (supplementary material Fig. S2F; data not shown). A time series from pooled embryos with two or three stripes showed that the arrest front begins to shift ∼3 hours (∼6 somites) phs (Fig. 3J). Thus, although the exact location of oscillator arrest relative to the anteriormost her7 stripe is not known, this result suggests that the position of oscillator arrest in the PSM shifts posteriorly after Wnt inhibition.

Together, these results demonstrate that Dkk1 overexpression controls the position of molecular markers of the wavefront in the PSM. From the consistent morphological and molecular changes observed in our assay, we infer that Wnt signaling has a specific role in regulating wavefront velocity in zebrafish.

How does Wnt activity instruct the position of the wavefront within the PSM? Previous studies have suggested that the levels of Wnt ligand or downstream effects are graded in mouse and zebrafish (Aulehla and Pourquié, 2010; Aulehla et al., 2003; Aulehla et al., 2008; Row and Kimelman, 2009; Thorpe et al., 2005), but quantitative analysis of the distribution and timing of Wnt activity is still lacking. The intracellular response to canonical Wnt binding its receptor includes translocation of β-catenin to the nucleus (Chien et al., 2009), and thus β-catenin concentration in the nucleus indicates canonical Wnt signaling activity. We examined the spatial distribution of Wnt signaling activity in the PSM using fluorescence immunohistochemistry against β-catenin. As a constituent of the cytoskeleton, β-catenin also gives a cortical signal in all embryonic cells (supplementary material Fig. S3B). We therefore developed an image processing procedure to estimate nuclear β-catenin concentration along the A/P axis (supplementary material Fig. S3).

We examined normally developing embryos during mid-trunk somitogenesis and observed a gradient of nuclear β-catenin fluorescence intensity, with highest levels in the tailbud and posterior PSM and lower levels more anteriorly (Fig. 4A,C,D). Importantly, a basal level was reached at ∼50% of PSM length as measured from the tail tip, and the intensity profile was flat in the anterior half of the PSM (Fig. 4D). This distribution of nuclear β-catenin is consistent with the posteriorly restricted expression of both wnt3a and wnt8 genes and known Wnt target genes in the zebrafish tailbud, such as ntl or axin2 (supplementary material Fig. S4) (Jho et al., 2002; Kelly et al., 1995; Lekven et al., 2001; Martin and Kimelman, 2012; Shimizu et al., 2005; Stulberg et al., 2012). By contrast, the nuclear β-catenin fluorescence intensity is not graded in the anterior PSM, where oscillators arrest and markers of segmental determination such as mespb are expressed (Fig. 4B,D).

We next measured the response of the nuclear β-catenin gradient to time-controlled perturbation of Wnt signaling in our assay. The maximum levels of nuclear β-catenin in the tailbud were reduced to ∼60% between 30 minutes and 1 hour after the dkk1GFP heat shock when compared with control embryos, and the position where signaling activity reached basal levels was shifted posteriorly (Fig. 4E-K). Thus, inhibition of Wnt in our assay was already effective 1 hour (∼2 somites) phs. This corresponds to a four to five somite delay between when Wnt signaling is reduced in the posterior and when wavefront markers are affected in the anterior PSM.

We conclude that the observed posterior distribution of the nuclear β-catenin gradient in the WT zebrafish embryo and the timing of its response to perturbations in Wnt signaling during somitogenesis are inconsistent with the regulation of the wavefront by Wnt signaling occurring in the anterior PSM.

Our results suggest that the Wnt signaling gradient might be instructing a cellular decision in posterior PSM that is relayed or transported by intermediate players into anterior PSM. The expression or activity of such candidates should be affected after Wnt perturbation, but before observed changes to the anterior wavefront markers, and have a connection to wavefront activity.

The Fgf pathway is proposed to be a regulator of the wavefront in mouse, chick and zebrafish, and in mouse was reported to be regulated by Wnt signaling (Aulehla et al., 2003; Aulehla et al., 2008; Dubrulle et al., 2001; Dunty et al., 2008; Naiche et al., 2011; Sawada et al., 2001; Wahl et al., 2007). To investigate whether Wnt inhibition affects the wavefront by reducing Fgf signaling in our assay in zebrafish, we first asked whether Dkk1 overexpression alters mRNA expression of Fgf pathway members. In situ hybridization with NBT/BCIP color development is not quantitative for mRNA level, but short color development times can be used to detect relative changes and spatial gradients (Dubrulle et al., 2001; Dubrulle and Pourquié, 2004; Picker et al., 2009). We did not observe a reduction in staining intensity of fgf3, fgf8 or fgf receptor 1 (fgfr1) riboprobes in the PSM by in situ hybridization during the timecourse of our assay (Fig. 5A-D; supplementary material Fig. S5A-F). Nor did we find reduced expression levels in the PSM of two Fgf target genes: pea3 and erm (also known as etv4 and etv5b) (Fig. 5E-H; supplementary material Fig. S5G-J). We did, however, observe an increase in erm riboprobe staining intensity in the posterior tailbud (Fig. 5G,H; supplementary material Fig. S5I,J) and a decrease in fgf8 riboprobe staining intensity in the mid-hindbrain boundary (supplementary material Fig. S5E,F), implying that color development had a suitable dynamic range in these experiments. Although small changes in mRNA levels in the PSM cannot be ruled out, these data suggest that reduction of Wnt signaling in this assay does not cause major alterations in the gene expression of Fgf pathway components.

To directly and quantitatively examine the timecourse of Fgf intracellular signaling activity in our assay, we visualized doubly phosphorylated Erk (dpErk) by fluorescence immunohistochemistry. Downstream of Fgf receptor activation, Erk kinase becomes rapidly phosphorylated, and dpErk levels have been used to monitor Fgf signaling in zebrafish (Sawada et al., 2001; Stulberg et al., 2012). In WT embryos at ten somites, the dpErk signal was distributed in a striking bi-modal profile (Fig. 5I,J). Highest levels were observed at ∼25% of PSM length, corresponding to the posterior end of the notochord; posteriorly, a graded signal decreased to about half-maximum at the tip of the tailbud, and anteriorly the signal was at background levels at ∼60% of PSM length. More anteriorly, an increase in dpErk signal was seen; this is associated with myogenesis in the forming somites (Groves et al., 2005).

We assessed the effect of Dkk1 overexpression on dpErk levels by comparing both the maximum intensity of the dpErk signal and the PSM location at which the anterior dpErk gradient dropped to 50% relative intensity (logEC50) at intervals phs. We observed no difference in either measure before 3.5 hours phs (Fig. 5K,M,N), by which time changes to segment determination and cyclic gene expression had already occurred in the anterior PSM (Fig. 3). At 3.5 hours phs, we detected a significant decrease in the maximum intensity of dpErk signal and logEC50 of the gradient (Fig. 5L-N). Although these data do not exclude some contribution, together they argue against a major role of Fgf signaling in mediating the effects of canonical Wnt signaling on the wavefront in our assay.

We next examined the expression timecourse of the transcription factor genes spadetail (spt; also known as tbx16) and mesogenin 1 (msgn1) in the PSM following Dkk1 overexpression. Msgn1 is a Wnt target gene in mouse (Wittler et al., 2007) and spt is downstream of Wnt signaling in zebrafish (Szeto and Kimelman, 2004; Thorpe et al., 2005). Msgn1, together with Spt, induces tbx6 (also known as fused somites or tbx24) expression (Fior et al., 2012; Yabe and Takada, 2012), which is a well-established regulator of segmental determination (Holley et al., 2000; Nikaido et al., 2002; Oates et al., 2005; van Eeden et al., 1996).

Whole-mount in situ hybridization revealed that the spt expression domain decreases after Dkk1 overexpression (Fig. 6A-D). To determine the timing of this change, we plotted the intensity profiles for several embryos and measured the length of the expression domain (Fig. 6E-G). This revealed a posterior shift in spt expression starting between 2 and 2.5 hours (four to five somites) phs, ∼1.5 hours after the effect on nuclear β-catenin and 1 hour before the effect on the wavefront (Fig. 6A-G). Similarly, inhibition of Wnt signaling led to downregulation of msgn1 expression (Fig. 6H-M) between 2 and 2.5 hours (four to five somites) phs (Fig. 6N). Thus, msgn1 expression is also downstream of changes to the Wnt activity gradient and upstream of changes to the wavefront during segmentation. We conclude that the Msgn1 and Spt transcription factors are plausible candidates to relay posterior Wnt-mediated positional information to the anterior PSM.

Finally, we asked whether cell flow in the PSM could contribute to the spatiotemporal changes following heat shock (Fig. 7). The timecourse of changes we have observed after Wnt perturbation delineates a successive chain of events across the PSM with a total duration of ∼4 hours (∼8 somites). From the timing (4 hours) and the distance from the start of a linear Wnt activity gradient to the anterior end of the PSM at ∼14 somites (∼300 μm), we estimate the velocity with which the changes propagate across the PSM following a reduction in Wnt signaling activity as ∼1.25 μm/minute. Assuming a constant and linear cell flow in the region of interest (Morelli et al., 2009), we approximated the magnitude of cell flow velocity in the PSM using the elongation velocity measured from time-lapse microscopy (supplementary material Table S1); this value is ∼1.0 μm/minute, in good agreement with the velocity of information propagation. Thus, a simple transport process of cell flow in the PSM could account for the observed velocity of information propagation. Additional, faster processes, such as signaling relays or diffusion, need not be considered. In other words, a cell located in the Wnt activity gradient in the posterior PSM at the time of heat shock would be swept with the cell flow into the anterior PSM with the correct timing to participate directly in the observed alterations to segment length.

Canonical Wnt signaling is well known for its role in the formation and later development and elongation of the posterior body structures, including the PSM (Greco et al., 1996; Martin and Kimelman, 2009; Martin and Kimelman, 2012). The difficulty in dissociating these Wnt functions from those in PSM patterning has blocked specific analysis of Wnt in segmentation in zebrafish. In mouse, evidence for a specific role first came from a Wnt3a hypomorph in which defects in posterior development appeared after segmentation had already started, leaving a time window for analysis (Aulehla et al., 2003). Here we temporally dissociate effects on segmentation from effects on posterior development and elongation using a time-controlled mild perturbation of the Wnt pathway. This perturbation is a distinctive test of the clock and wavefront mechanism (Cooke and Zeeman, 1976). Previous work indicated that segment length can be altered by changing the clock period while leaving wavefront velocity unchanged (Harima et al., 2013; Herrgen et al., 2010; Schroter and Oates, 2010). Here, we achieve the same outcome by altering wavefront velocity while leaving clock period unchanged. This apparent independence of clock period and wavefront velocity is difficult to reconcile with any model of somitogenesis in which the wavefront arises from coupling between oscillators and in which the period is strictly dependent on the wavefront velocity (Murray et al., 2011).

In chick and mouse there is some evidence for Wnt regulation of the segmentation clock period (Gibb et al., 2009). Although our results do not rule out a function of Wnt signaling in regulating the segmentation clock period in zebrafish, they are consistent with a conserved role for Wnt in regulating wavefront velocity throughout vertebrates.

How does the Wnt signaling gradient regulate the dynamics of wavefront position? The simplest model proposed that local levels of Wnt activity in the anterior PSM directly define the wavefront position: PSM cells arrest oscillations and define their position in the future somite once they reach a given threshold in the anterior PSM in a gradient of Wnt activity that spans the PSM (Aulehla and Herrmann, 2004; Aulehla et al., 2003). By contrast, we conclude that Wnt regulates wavefront velocity in the posterior PSM, and this is temporally and spatially separated from the expression of segment determination markers and the arrest of oscillations. Our work therefore raises important questions about the mechanisms downstream of the Wnt activity gradient that transport and refine positional information.

We show that spt and msgn1 are potential candidates for mediating the transfer of information from posterior Wnt signaling events towards the anterior in zebrafish. This list is almost certainly not exhaustive, and our studies do not address the mechanism by which these factors act upon the clock. Nevertheless, when combined with our estimates of the velocity of information propagation and the velocity of cell flow across the PSM, our findings lead us to a new model for how Wnt regulates the wavefront. In this scenario, Wnt provides positional information in the posterior that is transported by cell flow to determine segmental pattern and arrest cellular oscillations in the anterior, with a delay of several somite cycles. In other words, the falling level of Wnt starts an intracellular timer, or countdown, in posterior PSM cells that regulates when these cells will eventually switch off PSM behavior.

Strong evidence has accumulated that Wnt activity is required for the expression of transcriptional activators Tbx6 and Msgn1 in the mouse PSM, which are in turn required for expression in the anterior PSM of transcription factors vital for segment determination, such as Mesp2 and Ripply (Chalamalasetty et al., 2011; Dunty et al., 2008; Galceran et al., 2004; Hofmann et al., 2004; Wittler et al., 2007; Yasuhiko et al., 2006). However, this work did not examine where and when in the PSM Wnt activity is required, nor did it consider the consequences of cell flow through the tissue for wavefront regulation. Recent work in zebrafish has revealed striking roles for spt and msgn1 in the differentiation and movement of mesoderm progenitor cells from the tailbud into the PSM, thereby contributing to the flow of cells through the tissue (Fior et al., 2012; Yabe and Takada, 2012). Following msgn1 morpholino knockdown, Fior and colleagues reported the formation of smaller somites, whereas we observe larger somites in our assay. This apparent discrepancy might arise from the different mechanisms involved in the long-term versus short-term effects of msgn1 reduction and/or the level of reduction. To explain the shorter somites, the authors invoke a prolonged reduction of cell flow out of an enlarged tailbud (Fior et al., 2012), albeit without measurement of elongation or somitogenesis period or reference to wavefront velocity. This link between altered cell flow and somite length warrants further investigation. These studies also examined the role of spt and msgn1 in the maturation of paraxial mesoderm, showing a cooperative activation of tbx6 expression (Fior et al., 2012; Yabe and Takada, 2012). Tbx6 expression is a marker of mature PSM and is also an important transcriptional regulator of segment polarity (Nikaido et al., 2002). However, in these studies wavefront regulation was not considered. Our proposal explicitly joins the threads of a Wnt-initiated transcriptional cascade and cell flow through the PSM in the regulation of wavefront velocity.

This cell flow model for the regulation of wavefront velocity in zebrafish is consistent with scenarios explaining the distribution of posterior morphogen gradients along the mouse and chick embryonic axes by an mRNA or protein decay in a cell flow - the so-called ‘gradient by inheritance’ mechanism (Aulehla and Pourquié, 2010). Originally proposed as a mechanism to describe the graded distribution of Wingless signal away from the parasegment boundary in the Drosophila blastoderm (Pfeiffer et al., 2000), three examples have been described in vertebrates: the graded distribution of Cdx transcription factor as an input to Hox gene boundaries (Gaunt et al., 2003); a putative gradient of Wnt3a protein as an input into the direct concentration-threshold model of wavefront regulation (Aulehla et al., 2003); and the gradient of Fgf8 mRNA and protein, also involved in wavefront regulation (Dubrulle and Pourquié, 2004). The model suggested by our findings extends these scenarios because the information that is transported is not a single, decaying molecular species, but a regulatory cascade of components. This cascade downstream of Wnt signaling sets the wavefront position as it is transported with the cell flow. Segment length is determined by the velocity of this wavefront, which depends on the velocity of the cell flow as well as on the time scales of production, interaction and decay of the components in the downstream regulatory cascade.

Surgical manipulations in chick show that segments form rhythmically and sequentially even when the PSM is separated from the formed somites and tailbud (Palmeirim et al., 1998; Palmeirim et al., 1997), suggesting that the PSM has an autonomous schedule of wavefront progression. Furthermore, in grafts of entire chick PSM that are inverted along the A/P axis, the sequence of segment formation follows the pre-graft orientation of the PSM piece (Christ et al., 1974; Menkes and Sandor, 1969; Menkes et al., 1968; Palmeirim et al., 1998). Although the timing and lengths of the resulting segments were not measured in these experiments, this suggests that the schedule, once established, is relatively refractory to new input. The cell flow-based models above describe a mechanism consistent with these classical explant and inversion experiments. In the future, theoretical frameworks in which advection plays a prominent role will be essential for quantitative understanding of gradient formation and propagation of information in the PSM (Chisholm et al., 2010; Ibañes et al., 2006; Morelli et al., 2012).

Cross-talk with other signaling pathways may influence the positional information that PSM cells read from a given single gradient. Indeed, it is widely believed that cells are likely to integrate inputs from several gradients to determine wavefront position (Aulehla and Pourquié, 2010). Previous work in several species suggests that a gradient of Fgf signaling from the posterior and an opposing gradient of RA from the anterior regulate the wavefront (Dubrulle et al., 2001; Dubrulle and Pourquié, 2004; Moreno et al., 2008; Moreno and Kintner, 2004; Naiche et al., 2011; Sawada et al., 2001; Wahl et al., 2007). In mouse, strong gain and loss of β-catenin function in the PSM through conditional alleles causes an increase and decrease, respectively, in Fgf pathway gene expression (Aulehla et al., 2008; Dunty et al., 2008). In zebrafish, Fgf signaling appears to have a strong effect on wavefront positioning (Sawada et al., 2001), although the Fgf pathway does not appear to be a major downstream target of Wnt (Shimizu et al., 2005; Thorpe et al., 2005). Consistent with this, our results suggest that, for mild and time-controlled perturbation of Wnt signaling, the effects on the wavefront are not primarily dependent on alterations in Fgf signaling.

Using the distribution of dpErk, we have quantitated the Fgf signaling gradient in the PSM for the first time. In the tailbud, the Wnt and Fgf signaling gradients are oriented in different directions; whereas the Wnt gradient is flat or decreases monotonically towards the anterior, the Fgf signaling gradient rises from the posterior of the tailbud to a maximum level adjacent to the posterior end of the notochord. From this point, the orientation of the gradients is remarkably parallel, with Wnt and Fgf signaling reaching a minimum ∼50-60% of the distance along the PSM. Therefore, neither gradient would be expected to trigger events via a concentration threshold in the anterior half of the PSM.

From the delayed response of Fgf signaling levels to Wnt reduction observed in our experiments, it seems likely that Fgf signaling is not simply a response to Wnt signaling in the zebrafish PSM. Rather, it is possible that cells receive parallel inputs from Wnt and Fgf signaling as they flow thorough the PSM. Wnt and RA have been proposed to regulate each other indirectly through Ntl in the tailbud (Martin and Kimelman, 2010; Martin and Kimelman, 2012), but the effects of RA on the wavefront in zebrafish remain unknown. Further studies will be required to address the integration of Wnt, Fgf and RA signaling in zebrafish segmentation and to identify the mechanisms that influence the precision of their positional information. Ultimately, the wavefront defines segment boundaries in cooperation with the clock, and this interaction will also influence the precision and interpretation of positional information. The approach we describe should help address how cells integrate these signals to define their position in the PSM.

In considering any pattern-forming system, mechanisms mediating the propagation of information can be weighed by the ratio of diffusion to advection of signals. The case of vertebrate segmentation is striking in that gradients of potentially diffusive signaling molecules act on a continuous flow of cellular oscillators that also transport the signals. It appears, at least for Wnt signaling, that the advective process of cell flow is dominant over diffusive processes in establishing and transporting positional information. In light of these properties, the segmentation clock is an important paradigm for developmental systems that use morphogen gradients. Our approach, in which we dissociate the dynamics of positional information from cell flow and segmentation clock period, might prove useful in investigating how positional information is provided, transported and refined to specify the future segment boundaries.

hsaxin1::YFP (Kagermeier-Schenk et al., 2011), hsdkk1::GFP (Stoick-Cooper et al., 2007) or hsWnt8::GFP (Weidinger et al., 2005) transgenic zebrafish were outcrossed with wild-type (WT) AB. Transgenic and WT siblings were simultaneously heat shocked for 30 minutes at 37°C. Somite number was recorded before heat shock (time 0). Multiple-embryo time-lapse microscopy was according to Herrgen et al. (Herrgen et al., 2009).

Elongation velocity e was estimated from the distance of the tail tip to the posterior boundary of the fourth already-formed somite at the start of the movie. Instantaneous somite length S and somitogenesis period T_s were measured as described (Schröter et al., 2008). The rate of PSM shortening dL/dt was obtained from the distance between the last formed boundary and tail tip along the axis, each time a new somite formed. For e, T_s and dL/dt, slopes were calculated for each embryo and significance tested using Student’s t-test, two-sided, unequal variance. For S, significance was determined by two-way ANOVA test with Bonferroni correction.

To control for technical variation, control and transgenic embryos were processed in parallel, using identical procedures and reagents in a 24-well plate. Probe details are available in supplementary material Table S3. For immunohistochemistry, embryos were fixed in 2% (β-catenin) or 4% (dpErk) paraformaldehyde in 1×PBS at 4°C, washed in PBST (PBS with 0.1% Tween 20) and dehydrated overnight in methanol at -20°C. After progressive rehydration in PBST they were deyolked (β-catenin) and blocked in 10% sheep serum, 10 mg/ml BSA in PBS with 0.5% Triton X-100 (β-catenin) or 0.1% Tween 20 (dpErk). Anti-β-catenin antibody (Sigma, clone 15B8) was diluted 1/300 and goat anti-mouse IgG-coupled Alexa594 (Molecular Probes) and Hoechst 34222 were diluted 1/1000. Mouse monoclonal anti-dpErk [which recognizes Erk1/2 (Mapk3/1) in their double-phosphorylated state; gift of Ben Shilo, Weizmann Institute, Israel] was diluted 1/2000.

Distances between expression pattern and/or anatomical landmarks and expression profiles were measured on flat-mounted embryos. myoD (myod1), axin2, ntl, spt and msgn1 expression profiles were generated using a segmented line of 40 pixel width on 8-bit RGB, color-inverted images using ‘Analyze->Plot profile’ in Fiji (Schindelin et al., 2012). To determine antibody staining intensity, the PSM was imaged with a 25× objective using confocal microscopy (Zeiss LSM 510) in several parts that were stitched together (Preibisch et al., 2009). For β-catenin, color channels were split, a nuclear mask was obtained with Cell Profiler software from the Hoechst signal and used to retrieve average intensity from the β-catenin signal. A corresponding position for each nucleus was determined using a custom Fiji plugin, which is available on request. Nuclear β-catenin intensity values were averaged for nuclei along the mediolateral axis with bins of 1.5% of PSM length (10 μm), using an R script. Intensity profiles from dpErk-immunostained embryos were obtained from grayscale images using a line of 50 pixel width, using ‘Analyze->Plot profile’ in Fiji. To reliably compare treatment effects, we averaged over multiple embryos. To determine the mid-point of graded distributions, we fitted a logarithmic response curve, for convenience, and calculated logEC50. To determine maximum intensity, we calculated dynamic range by subtracting the minimum from the maximum value for each embryo and normalizing to the average value of heat shocked WT embryos. Significance was assessed by multiple t-test.

We thank G. Weidinger and B. Kagermeier-Schenk for reagents and sharing preliminary results; M. Esner for R coding; L. Rohde, A. Webb, K. Uriu, G. Valentin, R. Desai, J. Briscoe and D. Kimelman for comments on the manuscript; S. Pralow, K. Straube and C. Fröb for technical support; and the fish and light microscopy facilities of the MPI-CBG.