The formation of reiterated somites along the vertebrate body axis is controlled by the segmentation clock, a molecular oscillator expressed within presomitic mesoderm (PSM) cells. Although PSM cells oscillate autonomously, they coordinate with neighboring cells to generate a sweeping wave of cyclic gene expression through the PSM that has a periodicity equal to that of somite formation. The velocity of each wave slows as it moves anteriorly through the PSM, although the dynamics of clock slowing have not been well characterized. Here, we investigate segmentation clock dynamics in the anterior PSM in developing zebrafish embryos using an in vivo clock reporter, her1:her1-venus. The her1:her1-venus reporter has single-cell resolution, allowing us to follow segmentation clock oscillations in individual cells in real-time. By retrospectively tracking oscillations of future somite boundary cells, we find that clock reporter signal increases in anterior PSM cells and that the periodicity of reporter oscillations slows to about ∼1.5 times the periodicity in posterior PSM cells. This gradual slowing of the clock in the anterior PSM creates peaks of clock expression that are separated at a two-segment periodicity both spatially and temporally, a phenomenon we observe in single cells and in tissue-wide analyses. These results differ from previous predictions that clock oscillations stop or are stabilized in the anterior PSM. Instead, PSM cells oscillate until they incorporate into somites. Our findings suggest that the segmentation clock may signal somite formation using a phase gradient with a two-somite periodicity.
In vertebrates, the process of somitogenesis generates mesodermal blocks of tissue – called somites – from the undifferentiated presomitic mesoderm (PSM). Somites flank the notochord and eventually give rise to structures such as axial muscles and vertebrae. Somites are formed sequentially with regularity in size, periodicity and number within each species. Segmentation is resilient and adjusts for changes in PSM size (Lauschke et al., 2013) and temperature (Schröter et al., 2008). To account for the regularity of somite formation, Cooke and Zeeman (1976) proposed two hypothetical interacting mechanisms behind somitogenesis: a ‘clock’ and a ‘wavefront’. They hypothesized that cells in the PSM would oscillate between a permissive and restrictive phase (clock), periodically forming somites based on their interaction with a positional signal (wavefront), the latter progressing across the embryo at the same rate as tailbud elongation. PSM cells in the permissive phase of the clock would respond to the wavefront by pinching off and rapidly transitioning from undifferentiated PSM cells into segmented somites. Their model informed the search for factors that molecularly define the clock and wavefront, and forms the basis of most somitogenesis models today (Lewis et al., 2009; Aulehla and Pourquié, 2010; Pourquié, 2011; Oates et al., 2012).
Through experimental and theoretical data across a wide breadth of organisms, the periodicity and timing of somite formation is now thought to be controlled by a molecular network known as the segmentation clock (Pourquié, 2011; Oates et al., 2012; Palmeirim et al., 1997; Holley et al., 2000; Jouve et al., 2000; Henry et al., 2002; Oates and Ho, 2002; Bessho et al., 2003). Clock genes oscillate in individual PSM cells and some of them are essential for proper somite formation. PSM cell oscillations are locally coordinated, with the Notch pathway necessary for synchronizing expression in neighboring cells (Pourquié, 2011; Oates et al., 2012). The synchronous oscillations of clock cycling generate a wave of gene expression that is propagated as a narrowing stripe from the posterior to the anterior PSM and is kinematic, produced by local, coordinated expression in individual cells rather than the bulk transport or transduction of molecules or cells. The periodicity of oscillatory expression of cyclic genes in vivo in the posterior PSM matches the rate of somite generation, with each wave of clock expression reaching the PSM-somite boundary as a new somite is formed (Oates et al., 2012; Aulehla et al., 2008; Masamizu et al., 2006; Takashima et al., 2011; Delaune et al., 2012), although recent work shows that oscillation and segmentation periodicities are offset during tail segmentation (Soroldoni et al., 2014).
A family of core clock components has been shown to oscillate within the vertebrate PSM: the Hairy/Enhancer-of-split (Hes) transcriptional repressors (Palmeirim et al., 1997; Jouve et al., 2000; Bessho et al., 2001). The cycling of these hairy-related genes was first deduced in fixed embryos by in situ hybridization. Fixing the halves of each embryo at different times or exposing each half to different temperatures revealed stripes of gene expression with a period matching that of somite formation repressors (Palmeirim et al., 1997; Jouve et al., 2000; Jiang et al., 2000). In zebrafish, hairy/enhancer-of-split-related (Her) genes, her1 and her7, have been shown to oscillate in the PSM and are necessary for proper somite formation (Pourquié, 2011; Oates et al., 2012; Henry et al., 2002; Oates and Ho, 2002). High-resolution in situ hybridization for Her genes show that clock expression sweeps in a posterior-to-anterior direction: in each stripe of expression, transcripts in the more posterior edge of the stripe are localized in the cytosol, whereas cells in the anterior have transcripts localized in nuclear puncta (Jülich et al., 2005; Mara et al., 2007; Giudicelli et al., 2007). Recently, we directly observed waves of zebrafish cyclic gene expression using a single-cell resolution clock reporter, her1:her1-venus, in which the 8.6-kb her1 regulatory region drives expression of a transcript encoding a Her1-Venus fusion protein, flanked by her1 5′- and 3′-UTRs to facilitate rapid transcript turnover (Delaune et al., 2012). Using this reporter, we described the behavior of oscillating cells at a local level, confirming that the Notch pathway synchronizes neighboring cells and revealing that daughter cells oscillate synchronously after mitosis. The her1:her1-venus reporter is thus a powerful tool to explore how segmentation clock signal is translated to pattern each forming somite.
In both fixed and live embryos, the cyclic gene expression wave slows as it approaches the anterior PSM, a feature not described in the original clock and wavefront model (Cooke and Zeeman, 1976; Delaune et al., 2012). As more data were gathered about the slowing of clock gene oscillations, models have been proposed to explain how this slowing influences segment formation. The function of clock slowing is still unclear; recent models suggested the clock freezes as it interacts with a theoretical ‘arrest front’, effectively stopping and stabilizing clock expression (Oates et al., 2012; Giudicelli et al., 2007; Morelli et al., 2009; Herrgen et al., 2010). The observed clock slowing would then account for the continuous transition from a finite period in the tail bud to an ‘infinite period’ at the front. Some calculations based on clock oscillations in mice suggest that clock expression does not stabilize in the anterior PSM, but may only slow to a 1.5-segment periodicity (Niwa et al., 2011). Importantly, transition from oscillating systems to somites does not require a priori a diverging period (François and Siggia, 2012). Understanding how the clock behaves in the anterior PSM may have important implications in understanding somite patterning and will also inform somitogenesis models, ensuring that models accurately reflect phenomena observed in the developing embryo. For example, the dynamics of clock slowing could help to determine where somite boundaries and/or somite polarity are patterned.
Here, we follow oscillating PSM cells in vivo to investigate the slowing of the clock relative to somite boundary formation. We focus on cells that eventually form each somite boundary, using the her1:her1-venus reporter to examine clock oscillation patterns in future boundary cells through developmental time. We find that the clock initially oscillates in the posterior PSM with a periodicity that matches the rate of somitogenesis, gradually slows as PSM cells become more anterior, and increases in amplitude during the final two oscillations. The clock slowing in anterior PSM cells creates a phase distribution where cells at a one-somite distance are actually in opposite phases of clock expression. Importantly, we do not find evidence for an arrest front that would cause clock expression to drastically increase in period as it stops or stabilizes in the anterior PSM, ceasing oscillations. Based on these results, we propose an updated interpretation of how the segmentation clock patterns somites.
To understand the dynamics of clock slowing and the relationship of slowing to somite formation, we followed oscillations in cells progressing anteriorly in the PSM. We tracked cells over time in the transgenic line her1:her1-venus, a single-cell resolution clock reporter (Delaune et al., 2012), focusing on cells that will eventually constitute each somite boundary. Zebrafish embryos were injected at one-cell stage with h2b-cerulean and lyn-mcherry mRNA to mark nuclei and membranes, respectively (Delaune et al., 2012; Megason, 2009). The PSM was imaged beginning at the 10-12 somite stage for 4-6 h and PSM cells were tracked through time using a semi-automated cell tracking software (Delaune et al., 2012). As in our previous study (Delaune et al., 2012), we imaged embryos at 23°C to lengthen the somitogenesis period (Schröter et al., 2008) and allow for more time to image embryos. Each peak of expression was correlated to the approximate anterior-posterior global position of the cell in the PSM and its position relative to the most recently formed somite. We used conventions in the field (Pourquié and Tam, 2001), with the newest fully-formed somite named S1 and the second-newest formed somite named S2. The region constituting the forming somite is named S0, while regions forming the two subsequent future somites are named S-1 and S-2, respectively (Fig. 1A). To identify the oscillations of each cell relative to its final oscillation, we propose a nomenclature that is similar to that of somitogenesis: the final oscillation peak of a given cell is named ‘P0’, while the second to last is named ‘P-1’, third to last ‘P-2’, and so forth. As previously noted by us and others, we observed that reporter oscillations are faster in posterior PSM cells than in anterior ones, and that reporter expression levels in anterior PSM cells are larger in amplitude than in posterior PSM cells (Fig. 1B,C) (Ay et al., 2014; Lauschke et al., 2013; Delaune et al., 2012). We quantified the periodicity of oscillations by measuring the time between each peak of fluorescence and normalizing it to the somitogenesis period. All peaks of oscillations were identified by both raw fluorescence measurements and calculated sinusoidal waves using a validated smoothing heuristic (Delaune et al., 2012). As expected, cells in the posterior PSM oscillate with a periodicity that approximately matches somitogenesis (Fig. 1D, ‘P-5’ to ‘P-3’) (Delaune et al., 2012; Oates et al., 2012; Holley et al., 2000; Giudicelli et al., 2007). The periodicity increases in a linear fashion as cells progress anteriorly, with a 50.7% longer period (s.e.m.=1.2%) between P0 and P-1 (Fig. 1D) compared with the periodicity of somitogenesis. We do not see evidence of period divergence to infinity, contrary to suggestions in previous models (Oates et al., 2012; Giudicelli et al., 2007; Morelli et al., 2009; Herrgen et al., 2010). We also quantified the increase in amplitude of clock expression in the last two oscillations of anterior PSM cells (Fig. 1E). Clock expression in the posterior PSM has little or no change in amplitude when compared with the previous oscillation (Fig. 1E, ‘P-5’ to ‘P-3’). However, we noted a 58.9±3.6% increase in fluorescence in comparing the second-to-last oscillation with the third-to-last oscillation (‘P-2’ versus ‘P-1’), and an additional 70.1±4.6% increase when comparing the last oscillation with the second-to-last (Fig. 1E, ‘P-1’ versus ‘P0’). The gradual signal increase and slowing of the clock was observed in all PSM cells (four embryos, 243 cells), regardless of the final position of the cell within a formed somite.
With cells in the posterior PSM oscillating with the same periodicity as somitogenesis, one wave of clock expression is propagated through the PSM for each somite formed. As the clock slows, these waves condense into narrowing peaks of expression. Current models of somitogenesis routinely position peaks of clock signal at one-somite length intervals in the anterior PSM, where period eventually diverges towards infinity at the PSM-somite boundary (Oates et al., 2012; Giudicelli et al., 2007; Morelli et al., 2009; Herrgen et al., 2010). To examine the dynamics of clock slowing in the anterior PSM, we followed reporter expression over time in cells that eventually form either side of somite boundaries (Fig. 2A,A′). We found that oscillations in anterior boundary cells of the forming somite (cells circled in red, Fig. 2A′) are nearly synchronous to cells in the adjacent posterior boundary of the previously formed somite (circled in blue, Fig. 2A′,B). A prominent distinction between these two neighboring populations is that cells in the anterior S0 will oscillate one more time compared with the cells incorporated into posterior S1, which cease to oscillate. The last oscillation of S1 posterior boundary cells appears similar to the adjacent S0 anterior boundary cells, ceasing oscillations without becoming fixed in a persistently ‘on’ or ‘off’ state. We observed the same pattern in boundary cells that formed the previously formed somite (circled as green and orange in Fig. 2A′): synchronous oscillations, with cells that form the future anterior S1 boundary oscillating one more time after boundary cells that form the future posterior S2 boundary have stopped oscillating (Fig. 2D). Mitosis in the PSM produces daughters that oscillate in tight synchrony (Delaune et al., 2012); if the characteristic ‘extra’ cycle is a robust feature of neighboring cells that join adjacent somites, we expected to observe that pattern in highly synchronized daughters, but only if they eventually contribute to opposite sides of a somite boundary. In a rare example where cell division occurred early enough to allow tracking and the daughters separated enough to join adjacent somites, that is exactly the pattern observed (supplementary material Fig. S1). Finally, our cell-tracking analyses also reveal that cells that incorporate into a posterior somite boundary cease oscillations prior to cells that incorporate into the anterior boundary of the same somite [Fig. 2B-E, compare anterior S1 cells (green) with posterior S1 cells (blue); supplementary material Fig. S2], contrary to what is predicted by current models. Thus, within the same presumptive somites, clock oscillations stop from a posterior-to-anterior direction, without any ‘freezing’ of the phase or period divergence (Fig. 2B-E; supplementary material Fig. S2). This is in contrast to clock and wavefront models where the clock is expected to stop from an anterior-to-posterior direction following continuous wavefront dynamics (Oates et al., 2012; Giudicelli et al., 2007; Morelli et al., 2009; Herrgen et al., 2010).
As mentioned above, previous work suggested that oscillations of cells one-somite length away should be synchronized in the anterior PSM (Oates et al., 2012; Giudicelli et al., 2007; Morelli et al., 2009; Herrgen et al., 2010). As expected, we find that posterior PSM cells oscillate in synchrony, even when at a one-somite distance (supplementary material Fig. S3). However, in the anterior PSM, cells that will form the anterior border of S0 have opposite levels of expression compared with cells that will form the anterior-most border of S1 (Fig. 2C): i.e. when presumptive anterior border S0 cells have peak expression levels, presumptive anterior border S1 cells are at the expression trough, and vice versa. This distinct anti-phase relationship is observed when comparing any two groups of anterior PSM cells separated by a one-somite length, including cells that incorporate into the posterior-most border (Fig. 2E) or center of formed somites (not shown). These oscillation relationships among PSM cells were consistent at every forming somite boundary we examined (15 boundaries across five embryos): cells at a one-somite distance initially oscillate in synchrony in the posterior PSM, but as the clock slows in the anterior PSM, they shift into anti-phase so that adjacent waves of cyclic expression are not synchronized in the anterior PSM.
To compare globally anterior PSM oscillation phase across multiple boundaries and embryos, we quantified the synchrony of these anterior PSM cells, using methods described previously (Delaune et al., 2012). Briefly, we used a smoothing heuristic to estimate the oscillations and clock phase of each cell at each timepoint; phase calculations were then used to quantify the phase difference between any two cells at any given timepoint as long as both cells were still oscillating (Delaune et al., 2012). Tracked cells were indexed based on their final position within a developing somite, either at the anterior or posterior border. The phases of cells constituting each boundary were compared with each other in a combinatorial fashion, calculating phase differences between cells in each group at every timepoint. Cells in the same compartment exhibited a high level of synchrony with very little phase difference between cells at any given time point (35,740 comparisons across four embryos, Fig. 2F). As expected, cells on either side of a somite boundary are slightly less synchronized than cells on the same side of a somite boundary (17,353 comparisons across four embryos, Fig. 2G). Across a larger distance, clock synchrony in the anterior PSM continues to decrease (15,003 comparisons across four embryos, Fig. 2H), with maximum phase difference at one-somite length away (25,326 comparisons across four embryos, Fig. 2I). These data validate previous observations that cells in close proximity are synchronized (Delaune et al., 2012), and we show here that a somite boundary does not create an exception to that finding. Instead, a gradual phase gradient is distributed along the length of the anterior PSM, with adjacent cells oscillating in phase and cells at a one-somite distance in anti-phase.
To gain a better sense of how clock oscillations behave throughout the entire PSM, we performed a tissue-level analysis of clock behavior, retrospectively examining populations of cells based on their final position in the developing embryo. By broadly analyzing the waves of clock expression across space and time, we searched for repeating patterns of clock oscillations that match the repeated formation of somites. Using automated scripts (see Materials and Methods), z-stacks of images at each timepoint were processed to detect the axis of the embryo based on injected nuclear and membrane labels. We divided the somite and PSM tissue into sectors at the end of each timelapse – with each future somite delineated into thirds (Fig. 3A) – and then retrospectively measured the total combined fluorescence within cells of each sector at each timepoint. The clock expression pattern in each presumptive somite recapitulated our single-cell observations: in the anterior PSM, sectors of tissue separated by a one-somite distance were in anti-phase with each other and cells that were separated at a distance of two somites were in phase with each other (two embryos shown, Fig. 3B,C).
An embryo-wide profile of clock expression dynamics generated by measuring reporter fluorescence across the entire AP axis (without individual cell tracking) resembles the spatial readout of mRNA levels using in situ hybridization. However, our system measures spatial expression profiles at multiple time points over the course of development in a single embryo rather than at a single timepoint in a fixed embryo, and provides a correlation across the presomitic mesoderm between clock dynamics and eventual somitic fate. Tissue-wide fluorescence was quantified by totaling reporter signal within each digital sector of PSM. Based on morphological furrowing of somite boundaries, we examined fluorescence reporter levels in the PSM at each moment a somite was forming (Fig. 3D-E‴). At each timepoint, we observed peaks of expression in the forming somite (S0) and at a two-somite distance (S-2), with minimal clock expression at the S-1 position (Fig. 3F,G). This pattern recapitulates the two stripes of expression seen in fixed embryos, with an alternating gap of gene expression between the newly forming S0 and the subsequent S-2. As the cells in S0 form a somite, cells that were previously in the S-1 position are now in S0. The pattern repeats, with cells in the new S0 and S-2 peaking in expression, while S-1 cells (halfway between these two peaks) are at a minimum level of expression. The only difference between these two patterns is that peaks are shifted one-somite length posteriorly as each somite forms. Cells that end up in adjacent somites have opposite levels of clock expression in the anterior PSM, both temporally and spatially. This alternating pattern was recapitulated even when the clock was slowed using a hes6 antisense morpholino, which has been previously shown to create larger somites (Schröter and Oates, 2010) (supplementary material Fig. S4).
Since the discovery of cyclic segmentation clock genes, researchers have explored how oscillations are translated to pattern future somites. Much of the work focuses on clock dynamics in the posterior PSM, where clock periodicity matches that of somite formation (Oates et al., 2012). Because a kinematic wave of clock expression narrows as it sweeps across the PSM, it is clear that the clock slows anteriorly (Giudicelli et al., 2007; Morelli et al., 2009; Ay et al., 2014). The gradual slowing of the clock has received less attention, largely due to the lack of real-time tools to measure changes in expression dynamics. We investigated how the longer clock period in the anterior PSM may influence forming somites.
Using an in vivo clock reporter, we examined clock expression in individual PSM cells as an embryo develops, correlating clock oscillations with morphological somite formation. We observed that segmentation clock periodicity and amplitude increases in the anterior PSM, with each wave of clock expression corresponding to a forming somite boundary. Our measurements of clock slowing match mathematical predictions made for the mouse (Niwa et al., 2011). The increase in amplitude may be due to a longer period, allowing for more protein to accumulate, though changes in expression or degradation rates may also play a role. A recent study has shown that increases in translational time of Her1 protein in the anterior PSM contribute to this slowing (Ay et al., 2014). Although it is not clear what regulates the change in cyclic gene expression dynamics in the anterior PSM, we know that her1 expression in the anterior PSM is controlled by distinct regulatory elements (Gajewski et al., 2003; Brend and Holley, 2009), including a 500 bp upstream region with binding sites for Tbx24, Su(H) and Hairy-related transcriptional regulators (Brend and Holley, 2009). Additionally, signaling gradients, including FGF, Wnt and RA (Aulehla and Pourquié, 2010), could contribute to differential her1 regulation in the anterior versus posterior PSM.
In vivo imaging of clock dynamics confirms that clock reporter expression gradually slows, with periodicity in the anterior PSM becoming almost double that in the posterior PSM. This periodicity creates an interesting phenomenon where at any given time in the anterior PSM, peaks of clock expression are spatially separated at a two-segment periodicity, with a trough of expression between them. After a round of somite formation, there are again two peaks of clock expression in the anterior PSM, shifted one-somite length compared with the previous two peaks, with the cells previously in the trough of expression now at a maximum (Fig. 3). In this way, there is an alternating two-segment periodicity, with each presumptive somite experiencing a peak of clock expression before boundary formation. As discussed later, this separation of clock expression peaks could be important in determining somite boundaries or defining anterior-posterior somite polarity. Clearly, real-time imaging is required to measure clock dynamics directly, and our study has elaborated, extended and refined what was previously inferred from studies in fixed embryos.
The slowing of the clock creates a spatial and temporal phase gradient, with expression marking two future alternating somites, with a gap of expression between them. This in vivo pattern mirrors previous fate mapping and in situ data: the distance between stripes of anterior PSM her1 expression was measured to be up to two somites in length (Holley et al., 2000; Muller et al., 1996). Recent models of clock expression assert that expression peaks are separated by only one-somite length in the anterior-most PSM (Oates et al., 2012; Giudicelli et al., 2007; Morelli et al., 2009; Herrgen et al., 2010). The distance between each domain of segmentation gene expression differs among arthropods, with segmentation genes expressed at a one-segment length in spiders (Damen, 2007) or a two-segment length in insects such as beetles and fruit flies (Damen, 2007; Sarrazin et al., 2012). Our analyses show that peaks of clock expression in zebrafish differ depending upon position relative to the tailbud. Although her1 is expressed as a dynamic kinematic wave, it is intriguing that her1 spatial distribution in the anterior PSM at any given time has a similar alternating segment pattern to that of classic pair-rule genes (Muller et al., 1996; Nüsslein-Volhard and Wieschaus, 1980). The difference between classic pair-rule expression and what we observe, however, is that each future segment will alternate expression in the zebrafish anterior PSM, whereas only every other segment will ever express the classic pair-rule genes during Drosophila segmentation (Nüsslein-Volhard and Wieschaus, 1980).
In the original clock and wavefront model, the clock oscillates between a permissive and restrictive phase, interacting with the wavefront to mark distinct blocks of cells for somite formation (Cooke and Zeeman, 1976). One feature of this model is that somites can only be formed if groups of cells alternate between the permissive and restrictive phases of the clock. Later models incorporated additional features, such as the slowing of the clock. In these models, clock periodicity gradually increases to infinity at the PSM-somite boundary, with peaks of clock expression separated by one-segment length in the anterior PSM (Oates et al., 2012; Giudicelli et al., 2007; Herrgen et al., 2010; Morelli et al., 2009). We find that this does not accurately reflect key features of clock stopping, because such models predict that the clock period continuously diverges to infinity from anterior to posterior, with the anterior half of each presumptive somite ceasing oscillations before the posterior half. Because of the continuous change of period both within one somite and along the antero-posterior axis, we describe such models as ‘wavefront stopping’ models of clock oscillations. These models predict a smooth pattern of stopping in anterior, with a one-somite periodicity, and would be consistent with observations of the half-somite width of the anterior-most stripe of her1 transcripts in a fixed embryo (Holley et al., 2000; Sawada et al., 2000) and on spacing of stripes in the anterior PSM (Giudicelli et al., 2007).
Our real-time imaging directly shows that the her1:her1-Venus clock reporter is not constantly expressed in the anterior half of the next presumptive somite. The half-somite stripe of clock gene expression observed in fixed embryos is likely the last part of the kinematic wave of clock expression sweeping through the PSM (Fig. 4A). Differences in interpretation may also be due to the focus on transcripts in fixed embryos, compared with protein levels in our in vivo experiments (cyclic transcripts and proteins have been shown to be expressed in offset domains) (Bessho et al., 2003; Takashima et al., 2011; Delaune et al., 2012; Giudicelli et al., 2007). We show that clock reporter expression in the anterior PSM is not restricted to any subset of cells; instead, all cells are oscillating, even at the PSM-somite boundary, so that the period does not increase gradually to infinity. Our reporter also revealed that within a future somite, presumptive posterior boundary cells cease oscillating before their anterior boundary counterparts (Fig. 4B), which is contrary to the classical view of a smooth posterior-progressing wavefront. We observe that cells cease oscillating with the same directionality as the waves of clock expression, consistent with the idea that the clock itself plays a role in determining the wavefront (Lauschke et al., 2013). We describe our model as ‘clock wave stopping’. At the whole-tissue level, the clock stops from anterior to posterior, as observed with various real-time clock reporters (Lauschke et al., 2013; Aulehla et al., 2008; Masamizu et al., 2006; Takashima et al., 2011; Delaune et al., 2012; Soroldoni et al., 2014); however, within a given forming somite, clock stopping progresses from posterior to anterior. Thus, cells in each presumptive somite cease oscillating in discrete groups, rather than stopping in an anterior to posterior direction by a sweeping wavefront, as in the ‘wavefront stopping’ model (Fig. 4C). In our ‘clock wave stopping’ model, cells in the anterior PSM continue to oscillate with their neighbors, regardless of future somite position (Fig. 4C), consistent with the idea that synchrony is regulated by interactions between Delta-expressing and Notch-expressing cells (Delaune et al., 2012; Mara et al., 2007; Lewis, 2003; Horikawa et al., 2006; Riedel-Kruse et al., 2007; Özbudak and Lewis, 2008). Our data further suggest that a simple mechanism defining positional information could be at work. For example, as clock expression increases over the last two oscillations, levels may reach a threshold required to permanently repress its own expression and drive differentiation of PSM cells into a somite, which would explain the pattern of clock stopping we describe (Figs 2 and 4).
In a ‘clock wave stopping’ model, there is no requirement that the period of oscillations diverges to infinity to give rise to a smooth periodic one-somite pattern. Instead, the two-segment periodicity we observe might be crucial to sharply define antero-posterior polarity. With oscillation peaks at a two-somite length, cells in future adjacent somites are experiencing a maximal difference in clock expression, so that there is a clear polarity established within and between somites. Within one future somite, clock expression could then be, for example, discretized into anterior-posterior fates using a downstream bi-stable system (Meinhardt, 1982; François et al., 2007). This two-somite periodicity is preserved in the anterior PSM through the slowing of the clock, even though a wave of clock expression is generated in the posterior PSM every time one somite forms. Wavefront stopping models contend that peaks of expression in the anterior PSM are spaced at a one-somite periodicity, but we do not observe this in our study. What we observe instead – opposite phase at consecutive boundaries in the anterior PSM – reflects clock slowing and creates an alternating two-segment periodicity. The spatial two-segment periodicity not only makes it easier to distinguish consecutive boundaries due to their opposite phases, but has also been well described in other systems, such as Drosophila, the beetle Tribolium and mice (Niwa et al., 2011; Damen, 2007; Sarrazin et al., 2012). The correlation of gene oscillations and specific morphological landmarks may be a broad theme in development; for example, periodic expression of a cyclic reporter in Arabidopsis marks the future position of lateral root formation (Moreno-Risueno et al., 2010).
Heat-shock experiments and wavefront manipulations in zebrafish have suggested that boundary determination may occur as early as S-5 (Roy et al., 1999; Sawada et al., 2001; Akiyama et al., 2014). Experimental and modeling data have demonstrated that the sweeping wave of segmentation clock oscillations may not even be necessary for proper somite formation (Hester et al., 2010; Soza-Ried et al., 2014). Although initiation of somite patterning may occur in the posterior PSM, additional patterning mechanisms also operate later. For example, we have shown that a PSM cell dividing in S-3 can generate daughters that end up on opposite sides of a somite boundary, with appropriate oscillation profiles based on their final location (supplementary material Fig. S1). Conversely, local cell rearrangements can separate cells that were originally neighbors and are similarly fated; cell tracking work has shown that some PSM cells apparently switch sides right before boundaries form (Henry et al., 2000). Similarly, time-lapse analyses of neural tube patterning show that cell-type specification occurs early, but is spatially irregular; local rearrangements then sort like cells together (Xiong et al., 2013). Signals in the posterior PSM clearly have an important role in determining boundary formation, but our observations suggest that the final boundary decision is refined in the anterior PSM. Manipulating oscillations in different PSM regions will resolve whether the oscillatory changes we observe influence somite fate, or are an output of the fate decision. We note that our model is based only on observations of Her1 oscillatory expression and do not include other segmentation clock components. Future experiments in the zebrafish segmentation clock system will clarify the role of other PSM genes in distinguishing and separating PSM cells into discrete somites.
As we continue to investigate somitogenesis, more sophisticated analytical tools will expand our understanding of the underlying mechanisms. Our findings have revealed complex behaviors of the segmentation clock and how the clock may play a more central role in segmenting boundaries. The slowing of the clock in the anterior PSM is precise, generating separation of phase both temporally and spatially. Real-time reporters are important to capture dynamics not otherwise observable, and will continue to enhance our understanding of the segmentation clock.
MATERIALS AND METHODS
Adult fish strains were kept at 28.5°C on a 14 h-light/10 h-dark cycle. Embryos were obtained by natural crosses or in vitro fertilization, and staged via established criteria (Kimmel et al., 1995). The Tg(her1:her-Venus)bk15 transgenic line has been described previously (Delaune et al., 2012) and consists of her1 gene cyclic regulatory elements driving expression of a Her1-Venus fusion protein. All imaging experiments were performed in embryos heterozygous for the transgene, obtained by crossing fish homozygous for the transgene to the wild-type AB strain. Animal experiments were performed in accordance with institutional and national guidelines, and regulations and were approved by the UC Berkeley and Ohio State University Animal Care and Use Committees.
Time-lapse images were generated using methods modified from Delaune et al. (2012). Embryos were imaged on a LSM 780 with 32-channel GaAsP detector on the AxioExaminer microscope (Carl Zeiss) with the W Plan-Apochromat 20×/1.0 NA objective (Carl Zeiss), using Zeiss Zen 2010 software. Owing to low level fluorescence, we have found that the numerical aperture of the objective is a crucial component. Confocal sections were taken every 2 µm, with a stack of 30-35 slices taken approximately every 4 min. Images were converted to 8-bit tagged image file format before processing.
3D cell contour detection and tracking
Semi-automated contour detection MATLAB scripts were used to segment individual cells, as described previously (Delaune et al., 2012). As in our previous work, automated tracking of all individual cells was manually corrected and validated before further analyses.
Phase smoothing and phase calculation
A sine wave for each cell was generated using a smoothing heuristic, as described previously (Delaune et al., 2012). The phase at each timepoint was then extrapolated based on the smoothed sine wave. The extrapolated phase of a cell at a particular timepoint was compared with the phase of other cells at the same timepoint, and the absolute values for these phase differences were plotted onto histograms.
Tissue-wide partitioning and fluorescence binning
Cell positions and contours in the developing embryos were detected based on nuclear and membrane fluorescence within each z-stack. Image stacks were treated with low-pass filters to remove background and noise (Delaune et al., 2012). A binary image was constructed to detect embryo contour and to compute a skeleton corresponding to the center axis. We then defined a curved coordinate along this axis, and computed the position of each cell along this coordinate. This reference axis was then partitioned into equal parts, with the extended segment boundaries corresponding to bins. Fluorescence within each slice was measured as an average compared with slice size, and stored based on tissue position and timepoint. Embryo movement was accounted for by anchoring tissue position to a known reference cell as a landmark.
We thank the UC Berkeley and Ohio State Zebrafish Facilities staff for excellent zebrafish care. We thank Holly Aaron, Samuel Coleman, the UC Berkeley Molecular Imaging Center and the OSU Neuroscience Imaging Core for confocal access and advice, and Susan Cole for reviewing the manuscript. N.P.S. thanks Craig Miller, David Weisblat, and Patricia Zambryski for their invaluable input and feedback, and Paul Wang for excellent technical assistance.
N.P.S. performed timelapse imaging and cell tracking experiments, which were conceived and designed by all authors. MATLAB-based computer analyses and modeling were performed by N.P.S. and P.F. N.P.S., E.A.D. and S.L.A. provided reagents and materials. All authors contributed intellectually and discussed the data and manuscript. N.P.S., P.F. and S.L.A. wrote the manuscript and all authors participated in the editing process.
The work was funded by the Association Française contre les Myopathies (E.A.D.), a Marie Curie Outgoing International Fellowship (E.A.D.) and an National Institutes of Health (NIH) grant [1-R01-GM061952 to S.L.A.], supplemented by the American Recovery and Reinvestment Act. P.F. is supported by the Natural Science and Engineering Research Council of Canada, Discovery Grant program, Fonds de recherche du Québec – Nature et technologies, and partially supported by the Human Frontier Science Program and a Simons Foundation Investigator Award in the Mathematical Modeling of Living Systems. N.P.S. was supported by an NIH Training Grant [GM007127]. Deposited in PMC for release after 12 months.
The authors declare no competing or financial interests.