Dynamics of an incoherent feedforward loop drive ERK-dependent pattern formation in the early Drosophila embryo

Cells in a developing embryo must obtain information about their location to enact appropriate morphogenetic and fate decisions. In many developmental contexts this positional information takes the form of stripes of gene expression that subdivide a coordinate axis into distinct regions. Stripe formation requires precise spatial coordination of activating and inhibiting inputs to delineate boundaries. This complex signal processing has spurred decades of study of both natural and engineered stripe-forming biological systems (Schaerli et al., 2014; Stanojevic et al., 1991; Tabor et al., 2009). However, in many cases, how a stripe is produced by simple asymmetric inputs is still incompletely understood.

One of the simplest and best-studied developmental stripes is that of brachyenteron (byn) expression in the posterior region of the pre-gastrulation Drosophila embryo. Byn is an essential gut regulator and homolog of vertebrate brachyury (Kispert et al., 1994; Kusch and Reuter, 1999; McFann et al., 2022; Singer et al., 1996). It is primarily controlled by just a single input: activation of the Torso receptor tyrosine kinase at the embryo poles, which produces a gradient of active, doubly phosphorylated ERK (dpERK), known as the terminal pattern (Casanova and Struhl, 1989; Lu et al., 1993; Sprenger and Nüsslein-Volhard, 1992). ERK then inactivates the transcriptional repressor Capicua (Cic), leading to the expression of two essential transcription factors, tailless (tll) and huckebein (hkb) (Fig. 1A) (Bronner and Jackle, 1991; Coppey et al., 2008; Jiménez et al., 2000; Keenan et al., 2020; Patel et al., 2021; Pignoni et al., 1990; Strecker et al., 1986; Weigel et al., 1990). tll is sensitive even to low doses of ERK and so is expressed broadly, whereas hkb requires a high dose and is expressed at only the most posterior positions (Ghiglione et al., 1999; Greenwood and Struhl, 1997). Expression of byn is positively regulated by Tll and repressed by Hkb, so byn forms a stripe at low doses of ERK signal (Kispert et al., 1994).

Although the basic principles underlying byn stripe formation are well understood, some experimental results cannot be explained by the canonical model of steady-state regulation by tll and hkb. First, the pattern of byn expression is dynamic: it is initially expressed as a cap covering the entire posterior domain, only resolving into a stripe in late nuclear cycle (NC) 14 (Fig. 1B) (Kispert et al., 1994). Recently, Keenan et al. (2022) showed that these dynamics derive from changes in byn transcription: there is a domain of transient byn transcription at the pole and a domain of sustained byn transcription at interior positions. The source of these byn dynamics is not known, nor is it clear how byn might be expressed in the same posterior domain as its negative regulator hkb. Second, recent optogenetic studies demonstrated that replacing the terminal gradient by a strong all-or-none light input was sufficient to fully rescue embryogenesis (Johnson et al., 2020). This all-or-none input was observed to eliminate spatial differences in tll and hkb expression, such that the two genes were expressed in fully overlapping domains. How can a uniform input correctly pattern byn?

The architecture of the byn circuit provides a clue as to what may be missing from our understanding of byn regulation. The byn circuit has the general structure of an incoherent feedforward loop: a single input (dpERK) takes two paths (tll and hkb) with opposite signs to regulate a single output (Fig. 1A) (Alon, 2007). Incoherent feedforward loops can generate both steady-state stripes (Schaerli et al., 2014) and dynamic pulses (Mangan and Alon, 2003), but the differences in the delay time through each path is a crucial parameter for the incoherent feedforward loop's function. Whether the tll and hkb branches might exhibit different delays in the byn circuit is unknown and poses a conceptual challenge, given that the same ERK-dependent transcriptional regulator (Cic) mediates signal transfer to both target genes.

Here, we set out to examine dynamic transmission through the gene network regulating byn expression by combining optogenetic stimulation with endogenously-tagged transcriptional reporters at multiple network nodes. Optogenetics has proven to be a powerful approach to interrogate developmental circuits; the precision of light patterning enables the experimentalist to systematically vary the spatial and temporal parameters of an input while monitoring outputs ranging from gene expression to morphogenesis (Guglielmi et al., 2015; Huang et al., 2017; McFann et al., 2021). In particular, measuring responses after acute changes in stimulus can provide information about the timescale with which signals are transmitted through a developmental gene network (Keenan et al., 2020; Patel et al., 2021; Singh et al., 2022). We find that low- and high-amplitude optogenetic ERK stimuli are dynamically interpreted to generate either transient or sustained byn transcription. These dynamics are also reflected in the relative timing of tll and hkb transcription, with tll transcription occurring rapidly (<5 min) but hkb transcription exhibiting a dose-sensitive delay of up to 1 h post-illumination. Our data are consistent with a dose-dependent shift from transient to sustained byn expression dynamics that depends on the relative timing of Tll and Hkb protein accumulation. Finally, we show that a uniform input can also produce a byn stripe in space at a site near, but not overlapping with, the input stimulus, revealing a role for protein diffusion and mixing in the spatial interpretation of the terminal pattern.

Previous studies have shown that posterior byn expression is dynamic and forms two different domains (Keenan et al., 2022; Kispert et al., 1994). At the posterior pole, byn expression occurs in a transient, 15-min pulse at the beginning of NC14, whereas at more interior positions, byn is sustained for 45 min throughout NC14 (Fig. 1B; Fig. S1). We first sought to test whether optogenetic inputs to ERK signaling are sufficient to reproduce these byn dynamics without crosstalk from the endogenous terminal pattern. We generated Drosophila embryos that expressed the OptoSOS system for Ras/ERK pathway activation (Johnson et al., 2017; Toettcher et al., 2013) and a fluorescent MCP protein for visualizing transcription of MS2-tagged transcripts (Garcia et al., 2013). These OptoSOS/MCP embryos also contained MS2 stem loops in the endogenous byn 5′ untranslated region (UTR) (byn-MS2) (Keenan et al., 2022). To avoid overlap with the endogenous terminal system, we stimulated a region on the ventral surface of the embryo that is sufficiently far from the pole to lack byn, tll or hkb transcription when embryos are grown in the dark, but where tll, hkb and byn expression could be readily induced by optogenetic stimulation (Fig. S2A,B). The light input was limited to the right half of the embryo (Fig. 1C), providing an internally-controlled, unilluminated region at the same anterior-posterior and dorsoventral position. Blue light was delivered beginning in NC10, coinciding with the start of endogenous terminal signaling, and continuing through to NC14, a duration of roughly 90 min (Coppey et al., 2008).

We varied the intensity of the blue light dose to identify intensities that would produce byn responses similar to those endogenous responses observed by Keenan et al. (2022). Under a high-intensity blue light dose (‘high light’), byn was expressed in a transient pulse, with MCP-labeled transcriptional foci detected for only 10-15 min after the start of NC14 (Fig. 1D, top; Movie 1). This input was a similar intensity as that used previously to rescue the terminal pattern (Johnson et al., 2020). In contrast, an 8-fold lower low light dose (‘low light’) produced prolonged byn transcription that was maintained in NC14 for 25 min or more (Fig. 1D, middle; Movie 1). Although our ectopic stimuli were able to recapitulate the differences in byn transcription duration observed in wild-type embryos, we note that the prolonged ectopic bursts under low light were somewhat shorter in duration than those of the endogenous byn stripe, where bursting persisted for 45 min (Fig. S1C). In the dark, there were no byn bursts (Fig. 1D, bottom).

To quantify light-induced byn responses, we measured the number of bursting nuclei from a region of fixed size (Fig. S2C,D), and we defined a nucleus as being in a ‘bursting state’ for all time points between the first and last appearance of an MCP focus within a nuclear cycle (Fig. 1E, inset). Quantification from multiple embryos confirmed that low-intensity light stimulation drove byn bursts over a significantly longer duration than high light (Fig. 1E) and that the bursting state under low light persisted significantly later into NC14 (Fig. 1F). Together, these data confirm that two different levels of light-induced ERK activation are filtered into distinct byn dynamics, producing transient and sustained transcriptional responses that resemble byn endogenous dynamics at the posterior pole (Fig. 1G).

We observed some additional effects of high and low light stimuli on byn expression beyond the regulation of transcription duration. In some embryos under low light, the total domain of byn expression was smaller and limited to the posterior portion of the illuminated region, likely arising from differences in transcriptional sensitivity to ERK inputs along the anterior-posterior axis due to the expression of other gap genes (Fig. S2C,D) (Kim et al., 2013). We restricted our analysis to the sensitive domain under all light conditions, where both stimuli induced byn expression in 60-80% of nuclei (Fig. S2A), a response that matched endogenous byn transcription at the posterior in early NC14 (Fig. S1B). We did not detect a difference between the intensity of bursts under high and low light (Fig. S3A), suggesting that burst duration, not amplitude, is the primary dose-dependent parameter.

We hypothesized that the differences in byn transcription in each light condition would be reflected in the expression of tll and hkb, the well-characterized positive and negative regulators of byn (Fig. 2A). We applied the same high and low light stimuli to OptoSOS/MCP embryos, the endogenous tll or hkb genes of which contained MS2 stem loops in the 5′ UTR (tll-MS2 and hkb-MS2) (Keenan et al., 2022). These endogenously-tagged MS2 reporters differ from the enhancer sequences used in previous studies (Johnson et al., 2020; Keenan et al., 2020) and are expected to better reflect the full regulatory environment of each gene. Both high and low stimuli induced tll transcription in NC14, consistent with the standard model whereby tll transcription is triggered even by weak ERK stimuli (Fig. 2B; Fig. S2D) (Greenwood and Struhl, 1997). We also found that both high and low light could induce hkb transcription, with four out of seven embryos under low light exhibiting high hkb expression in NC14 (Fig. 2B; Fig. S2D). This result was unexpected because byn is sustained in NC14 under low light conditions (Fig. 1D) and hkb is a known repressor of byn expression. Moreover, hkb is thought to only be induced by strong ERK stimuli. We observed byn and hkb expression in the majority of nuclei under low light, making it unlikely that byn and hkb transcription exclusively occur in different sets of nuclei. However, it is impossible to rule out this possibility with our single-locus imaging approach. The transcription of both tll and hkb declined after the first 15 min of NC14 (Fig. S3B,D), consistent with a loss of input sensitivity of both genes that is also observed for their endogenous expression patterns at the posterior pole (Keenan et al., 2022). Although hkb expression persisted slightly longer than tll expression, neither gene showed a significant difference in the bursting duration in NC14 between high and low light (Fig. S3C,E).

Although high and low light stimuli drove similar tll and hkb transcription in NC14, the expression of these two genes was quite different in earlier nuclear cycles (NC11-13) (Movie 2). For tll, MS2 transcriptional foci appeared rapidly in NC11 and most nuclei maintained bursting states during NC11-14 (Fig. 2B,C, teal). Rapid induction of tll also occurred under low light, although there was a small reduction in the percentage of bursting nuclei, an increase in the variability between embryos, and a decrease in the size of the expression domain (Fig. 2C; Fig. S2D). In contrast, expression of hkb was delayed. Under high light, hkb-MS2 transcriptional foci were largely absent in NC11, but the percentage of nuclei in a bursting state increased in NC12 and remained high in NC13 and 14 (Fig. 2B,C, magenta). The delay between illumination and hkb transcription was much longer under low light, where hkb transcription was low or absent until NC14 (Fig. 2B,C, magenta). These results reveal that tll and hkb exhibit different temporal expression dynamics in response to ERK activation, with hkb transcription delayed relative to tll. They also identify hkb as a factor with transcriptional dynamics that change as a function of ERK input strength. These results differ from our previous observations using enhancer-based MS2 reporters (Johnson et al., 2020; Keenan et al., 2020), where we generally observed much weaker transcription in fewer nuclei and were unable to detect a difference in timing between tll and hkb. These differences may point to the importance of performing measurements at endogenous transcriptional loci, particularly when assessing gene expression dynamics.

We next set out to relate ERK inputs to the transcriptional delay in tll and hkb gene expression for individual embryos. OptoSOS expression can vary between embryos due to variability in Gal4/UAS-based control, so even a single light dose may produce different degrees of pathway stimulation. To account for this variability, we reasoned that embryo-specific inputs can be quantified by measuring the light-induced increase in SSPB-tagRFP-SOS^cat membrane fluorescence. By measuring the timepoint at which at least 50% of nuclei entered the bursting state as a function of the SSPB-tagRFP-SOS^cat membrane recruitment of each embryo, we were thus able to quantify the delays in tll and hkb expression over a continuum of ERK activity levels (Fig. 2D). We found that the onset of tll transcription was rapid in nearly all embryos, reaching 50% bursting within 8 min of the start of illumination, and exhibited weak scaling with the level of OptoSOS recruitment (Fig. 2D, teal). Onset of tll was most variable in embryos with the lowest OptoSOS recruitment, suggesting that the low light stimulation of these embryos was near the minimum threshold for tll induction. In contrast, the onset of hkb expression occurred later than tll at all signal amplitudes and scaled strongly with ERK signal amplitude, ranging from 20 min to 60 min depending on membrane OptoSOS levels (Fig. 2D, magenta). In the three embryos with the lowest ERK activity, hkb was never expressed. Taken together, these data demonstrate that nuclei of early Drosophila embryos convert the amplitude of ERK signaling into differences in the relative timing of tll and hkb transcription.

How might the relative delay in tll and hkb transcription affect their appearance at the protein level? To address this question we endogenously tagged tll and hkb with the LlamaTag system to rapidly visualize protein-level accumulation of these transcription factors (Bothma et al., 2018) (Fig. S4A). At the posterior pole of wild-type embryos, Tll and Hkb-LlamaTags form the expected nested domains (Fig. S4B,C). In embryos expressing the LlamaTag system and OptoSOS, we applied our high light stimulus from NC10 onward (Fig. S4D), a condition under which tll expression precedes hkb by ∼10 min and a pulse of byn expression is observed in early NC14. We found that, although nuclear Tll protein was detected both 10 min (early NC14) and 25 min (late NC14) into NC14, nuclear Hkb protein was only detected in late NC14 (Fig. S4D-F). These data confirm that the relative delay between tll and hkb transcription is also reflected at the protein level for both factors. Further, they show that there is a large increase in Hkb protein in mid-NC14 that occurs around the same time as repression of byn begins.

Together, these results suggest a conceptual model for how ERK signals of different amplitudes produce either transient or sustained dynamics of byn expression (Fig. 2E). When ERK activity is high, tll expression precedes hkb expression by a short but important window, leading to the pulse of Tll-mediated byn expression in early NC14 before the accumulation of Hkb repressor protein. When ERK is lower, tll expression is still rapidly initiated but hkb expression is substantially delayed (or, at sufficiently low ERK doses, altogether absent). The resulting Hkb protein levels do not rise until too late in NC14, leading to sustained byn expression. At the core of this model is the observation that differences in ERK dose are converted into temporal differences in hkb transcription onset.

Our conceptual model holds that tll and hkb are induced by ERK after different time delays, producing distinct byn dynamics. This model has so far been tested only under a simple stimulus regime: a step-change in light delivered at NC10 and maintained until gastrulation. A strength of the optogenetic approach is that stimuli can be varied over time, either by switching between light and dark conditions or by varying the developmental time at which the stimulus is applied. Importantly, such dynamic experiments are possible because transcriptional repression of Cic targets (i.e. tll and hkb) is restored within minutes of turning off an optogenetic input to the ERK pathway (Keenan et al., 2020; Patel et al., 2021).

We first applied a pulse of light to interrogate the consequences of the delay between tll and hkb expression. If a high light stimulus is delivered as a pulse that is sufficiently long to express tll but removed before sufficient hkb expression is initiated, it should be possible to express the Tll activator without Hkb-mediated repression, thereby shifting byn expression from transient to sustained. Importantly, this prediction differs from the canonical model, where a high ERK input would give rise to both tll and hkb expression, leading to byn repression. We applied high light to OptoSOS/MCP embryos as a pulse from NC10-13, a duration of roughly 45 min (Fig. 3A, left), and compared byn-MS2 expression with our previous experiment in which we applied a continuous stimulus beginning in NC10 (Fig. 1). This stimulus regime resulted in detectable Tll but not Hkb protein in NC14 (Fig. S4D-F, ‘pulse’ condition). We found that this pulsed versus continuous input had a profound effect on byn expression (Fig. 3A, right). In pulse-stimulated embryos, byn expression in NC14 was sustained, persisting for over 25 min in nearly all illuminated nuclei, as predicted by the absence of Hkb repressor (Fig. 3A,B; Movie 3). In contrast, continuous high light led to a brief pulse of byn expression during the first 15 min of NC14 (Fig. 3A, top). We note that the pulse of high light induced sustained byn in a higher proportion of nuclei than continuous low light (Fig. 1F), suggesting that that pulse input surpasses low constant illumination by driving higher Tll expression, lower Hkb expression, or both. In summary, two ERK-activating stimuli of the same amplitude but different durations produce distinct dynamics of byn expression, consistent with a necessary role for NC14 hkb expression in repressing byn as well as the sufficiency of NC10-13 tll expression in activating byn. This result is also consistent with our previous observations concerning the role of Hkb in repressing snail during ventral furrow formation, where we found that sustained illumination over a time period that includes late NC13/early NC14 was most effective in suppressing ventral furrow invagination (McFann et al., 2021).

One key feature of our conceptual model is a time delay before hkb expression that depends on ERK stimulus intensity. However, an alternative model could hold that the timing of hkb transcription is primarily controlled by developmentally regulated changes in its sensitivity to ERK. For example, hkb transcription may exhibit different sensitivities at different nuclear cycles, switching from being resistant to low-dose ERK stimulation during NC11 to being more sensitive during NC13-14. We reasoned that optogenetics could also be used to distinguish these models. By applying the same light input in different nuclear cycles and measuring the timing of tll and hkb expression, it is possible to separate a stimulus-response delay from a developmental timer.

We illuminated OptoSOS/MCP embryos using the high light condition at two different time points – an ‘early’ stimulus beginning in NC13 or a ‘late’ stimulus beginning at the start of NC14 (Fig. 3C, left). We then assessed transcription of both tll and hkb using endogenous MS2 reporters. For tll, we observed a robust, immediate pulse of transcription in NC14, regardless of whether the light began in NC13 or 14 (Fig. 3C,D, left; Movie 4). Even light stimulation in early NC14 produced an extremely rapid tll response, with MS2 foci detectable within just 4 min of illumination (Fig. 3C, center panel) in agreement with previous findings (Keenan et al., 2020). Taken together with our observation of transcription in NC11 for stimuli delivered in NC10 (Fig. 2), these data suggest that tll remains highly sensitive to ERK activity from NC10 to early NC14. For hkb, we observed that bright illumination in NC13 was sufficient to drive a pulse of transcription in NC14, but it was not sufficient when the illumination began in NC14 (Fig. 3C,D, right; Movie 4). In sum, tll is always induced immediately following ERK activation, whereas hkb exhibits delayed transcription, regardless of the developmental time at which the stimulus is applied. Taken together with our observation of a dose-dependent delay in hkb expression, these results support a model where the dynamics of the tll/hkb/byn system are intrinsic to this incoherent feedforward circuit, and not externally imposed by developmental time. More broadly, the experiments described above demonstrate the power of optogenetic manipulations to interrogate candidate models of developmental signal interpretation.

We have thus far investigated the temporal dynamics of byn expression, but previous work also suggests that its spatial pattern may be under complex regulation. In mutant embryos lacking the endogenous terminal pattern, an all-or-none OptoSOS input at the poles is sufficient for normal embryogenesis even though it eliminates both quantitative differences in Ras activity and spatial differences between tll and hkb expression (Johnson et al., 2020). OptoSOS embryos were also viable when all-or-none light inputs were applied on top of the endogenous gradient (Johnson et al., 2017). These results pose a conundrum because uniform high ERK activity and a complete overlap between tll and hkb expression would be naively predicted to eliminate a stripe of sustained byn expression. Having shown that byn dynamics might also be regulated by the temporal patterns of tll and hkb transcription, we next performed a closer examination of the circuit dynamics near the spatial boundary of an all-or-none OptoSOS stimulus (Fig. 4A).

To quantify tll, hkb and byn expression near the illumination boundary, we simply extended our analysis from previous experiments into the unilluminated half of each embryo (Fig. 4B; Fig. S5A,B). Our viewing window comprised both illuminated and unilluminated regions at the same anterior-posterior position and was symmetric across the ventral axis, thus enabling comparisons between these regions without introducing anterior-posterior or dorsoventral bias. Remarkably, when we considered the illuminated and unilluminated regions together under high light, we observed the dynamic formation of a byn stripe in response to the all-or-none input (Fig. 4C; Movie 5). Initially, byn-MS2 foci were detected throughout both the illuminated and unilluminated region, up to 50 μm away from the illumination boundary. After ∼15 min, byn foci were lost in the illuminated region but persisted in the unilluminated region (Fig. 4C,D). This non-local effect only occurred under high light; under low light there was no byn expression in the unilluminated region (Fig. S5E-G). In sum, a bright all-or-none light pattern produced a byn stripe in the nearby unilluminated region with similar spatial dynamics to the wild-type pattern but rotated 90° to match the illumination boundary.

This observation of sustained byn expression in the unilluminated region suggested that this region may also exhibit rapid tll expression but delayed or absent hkb expression. To test this prediction, we analyzed the unilluminated regions of tll-MS2 and hkb-MS2 embryos subjected to the high light condition. This analysis revealed that tll was transcribed beginning in NC11, whereas hkb transcription was again delayed until NC14 (Fig. 4E). The transcription of both genes then declined after the first 15 min of NC14 (Fig. S5C,D). These results suggest that for tll, hkb and byn, the unilluminated region of embryos subjected to a high light stimulus is functionally equivalent to the illuminated region under low light conditions (Fig. 4F).

How does a uniform optogenetic input to the Ras/ERK pathway act at a distance to induce tll, hkb and byn transcription ∼50 μm outside of the illuminated region? We envisioned two possibilities. First, light scattering or imprecision of the optogenetic input might lead to blurring of the all-or-none pattern to accidentally form a gradient. Second, active cellular components might diffuse within the syncytial embryo so that an initially sharp stimulus pattern would produce a graded response. We took two strategies to discriminate between these possibilities. First, we quantified the spatial responses at multiple nodes including our light input, OptoSOS membrane recruitment, an ERK activity biosensor (Moreno et al., 2019) and Tll/Hkb protein distributions (Fig. S6A-D). Overall, we found that our light stimulus and membrane OptoSOS recruitment formed relatively precise boundaries, whereas ERK activity and Tll/Hkb expression were graded and extended ∼50 μm away from the illumination boundary. Byn expression was most sustained in the region where Tll protein was present, but Hkb was low or absent (Fig. S6E). Second, we tested for intracellular component mixing using a photoconvertible tdEOS-tubulin construct (Lu et al., 2013), the green-to-red photoconversion of which we could control using the same optical path. Our illumination setup was able to produce an initially-sharp pattern of photoconverted tdEOS that subsequently underwent substantial blurring over a time period of 10 min; no blurring occurred when the same embryos were photoconverted after cellularization (Fig. S6F,G). Taken together, these data suggest that an all-or-none Ras stimulus is converted into a graded transcriptional response at least in part through the diffusion or transport of activated intracellular components.

As a final test of our model, we asked whether the same uniform OptoSOS input could rescue patterning of the byn stripe at the endogenous position in an embryo lacking the endogenous terminal pattern. Embryos mutant for torso-like (tsl), an activator of the Trunk ligand, lack terminal ERK signaling and do not express byn (Fig. 5A, left plot) (Savant-Bhonsale and Montell, 1993). Although we recently reported that an all-or-none optogenetic stimulus at the poles was sufficient to rescue subsequent development (Johnson et al., 2020), we did not examine byn expression under these conditions.

We generated OptoSOS/MCP embryos from mothers that were also homozygous for the tsl loss-of-function mutation (Fig. 5A, right schematic). Unlike in previous experiments, where we restricted illumination to the ventral surface of the embryo away from the poles, here we applied our light stimulus directly at the posterior pole beginning in NC10 and visualized byn-MS2 transcription in both the illuminated and neighboring unilluminated regions. At early time points in NC14 (Fig. 5B, 5 min) we observed byn expression in both the illuminated and unilluminated regions. However, by 10 min into NC14, byn resolved into a stripe in the unilluminated region that persisted for at least 20 min (Fig. 5B,C). There was an apparent difference in the early timing of byn expression between the two regions (Fig. 5C; 0-5 min), but close examination revealed that this difference was due to a pole-to-center wave of mitosis (Deneke et al., 2016), staggering the start of NC14 and initiation of transcription. Overall, these data demonstrate that byn stripe formation in response to a simple all-or-none stimulus is a robust property of its regulatory circuit that can occur both at endogenous and ectopic positions in the embryo.

Here, we have dissected the regulation of the byn stripe by combining precise optogenetic inputs in space and time with live biosensors of target gene expression. Using ectopic activation of Ras on the ventral side of wild-type embryos, we defined high- and low-amplitude OptoSOS inputs that induce distinct byn transcriptional dynamics – a pulse of expression in early NC14 versus more sustained expression – that match its endogenous responses in the posterior terminus and stripe-forming region. We then used these conditions to characterize the tll and hkb inputs that explain these byn dynamics in space and time.

Our approach yielded novel insights about both the temporal and spatial interpretation of ERK inputs to pattern the byn stripe. First, differences in signal amplitude are interpreted through the timing of tll and hkb expression (Fig. 5D, left). The onset of tll expression is always rapid, occurring as quickly as 4 min after signaling onset, whereas there is a dose-dependent delay in the onset of hkb expression. This delay in hkb expression is a function of Ras/ERK input amplitude, not of developmental time. These data are consistent with previous observations in OptoSOS embryos that hkb RNA only accumulates to high levels in response to blue light inputs over 30 min (Johnson and Toettcher, 2019). They also broaden our conception of the thresholds for tll and hkb expression: tll and hkb can be induced by inputs of the same amplitude, but hkb requires that the signal persist for a longer time. If the amplitude is low enough, the signal must persist longer than the developmental window allows, and hkb is never expressed. Thus, cumulative dose of ERK input (amplitude integrated over time) appears to be the relevant feature sensed by the circuit, as has been proposed for the terminal pattern as well as other systems (Gillies et al., 2017; Johnson and Toettcher, 2019). Integration of signal over time has similarly been shown to be important for interpretation of several morphogen pathways including Hedgehog, Wnt, Nodal and BMP (Alexandre et al., 2014; Dessaud et al., 2007; Tozer et al., 2013; van Boxtel et al., 2015). The byn circuit then processes this input through the relative timing of tll and hkb, rather than simply their presence, to determine local byn expression.

This more nuanced understanding of byn regulation resolves a conundrum of the endogenous pattern: how can the transient pulse of expression of byn in the high-ERK, Hkb-positive domain be reconciled with the presence of its inhibitor? Here, we show that at the high light levels which produce a comparable pulse of byn transcription, hkb transcription is delayed relative to tll and this delay is also evident in the accumulation of their protein products. Thus, there is a temporal window in which only the positive regulator is present, allowing for a pulse of byn expression, before accumulation of the repressor. The sequential appearance of Tll and Hkb was hypothesized during the initial characterizations of posterior patterning (Casanova, 1990) but has only now been directly shown. It is interesting to note that Tll has been characterized as a transcriptional repressor, implying that there is an intermediate node between tll and byn (Morán and Jiménez, 2006). However, the identity of this node and how it affects the timing of byn activation and repression remain unknown.

Our improved understanding of byn regulation also explains how a byn stripe can form in conditions where tll and hkb transcription have the same spatial domain, as in the rescue experiment of Johnson et al. (2020). Our current study revisits these results with improved tools, in particular endogenously tagged transcriptional reporters of tll and hkb that are able to clearly resolve differences in transcriptional dynamics that were obscured by enhancer-based reporter constructs. We find that stimulus conditions that support sustained byn can also support hkb expression in NC14, but under these conditions hkb expression is largely absent from earlier nuclear cycles. The co-expression of sustained byn with hkb under low light differs from the wild-type pattern, where the byn stripe forms in a region only expressing tll. Presumably the endogenous ERK gradient induces tll expression at even lower activity levels than our optogenetic inputs. We note that the shortened bursting duration of sustained byn at the ectopic position (∼25 min) compared with the endogenous stripe (∼45 min) suggests that the late-appearing hkb under low light does ultimately repress byn in late NC14. It is also possible that the network dynamics we report here provide robustness to the byn circuit, allowing it to produce different outputs for even a narrow range of input strengths.

Our study reveals that the tll-hkb-byn circuit can be classified as an incoherent feedforward loop with rapid activation and delayed repression, a circuit with well-characterized pulse-generation and stripe-forming properties (Fig. 5D, left) (Alon, 2007; Schaerli et al., 2014). A unique feature of this circuit however is that the delay in hkb expression is dose-dependent, meaning that differences in signal amplitude are converted to differences in hkb dynamics and thus different byn responses (i.e. transient if hkb onset is fast, sustained if hkb onset is slow). Interestingly, similar dose-dependent delays in transcriptional onset were recently shown for Dorsal and BMP signaling targets (Carmon et al., 2021; Hoppe et al., 2020; Keller et al., 2020). What is the mechanism underlying this delay in hkb onset? The dose-dependence of tll and hkb has been a longstanding open question even without the complexity of temporal dynamics (Furriols and Casanova, 2003; Smits and Shvartsman, 2020). ERK signaling activates transcription of both tll and hkb through relief of the same repressor, Cic, and it is unclear why these genes would require different doses of ERK signaling. Our experiments rule out a few possible explanations. Developmental time does not appear to be crucial, given that the delay in hkb transcription is observed regardless of when light is applied and both the tll and hkb loci are known to be accessible early (Blythe and Wieschaus, 2016). We can also rule out interactions with other components of the anterior-posterior patterning machinery given that we are able to produce an ectopic byn stripe rotated 90° from its endogenous counterpart. One intriguing possibility, supported by previous ChIP-seq results, is that Cic leaves the enhancers of hkb more slowly than those of tll (Fig. S7) (Keenan et al., 2020). It is also possible that signaling-dependent chromatin changes are involved (Hannon et al., 2017; Semprich et al., 2022). These models will be tested in future studies.

The second major finding is that the boundary of a uniform OptoSOS input is blurred in space downstream of Ras to produce two domains from a single input – a transient byn domain within the high-ERK illuminated region and a sustained domain in the low-ERK unilluminated region (Fig. 5D, right). These non-local effects of a local Ras input are most likely mediated by diffusion of active intracellular components, a well-established contributor to developmental patterning in the syncytial Drosophila embryo (Driever and Nüsslein-Volhard, 1988; Gregor et al., 2005). It remains unknown whether the endogenous terminal dpERK gradient is produced from a similar gradient of active Torso receptors, or is due to the combination of a discrete domain of Torso activity at the poles and cytoplasmic diffusion of downstream components (Casanova and Struhl, 1989; Coppey et al., 2008). If the latter model is correct, the developmental rescue by an all-or-none OptoSOS input may not be an example of a simple input replacing the function of a complex one, but rather a good approximation of endogenous activation in the terminal system. A number of systems once thought to depend strictly on input concentration have similarly been shown to depend on an unexpectedly simple form of the input (Economou et al., 2022; Zhu et al., 2022).

Several limitations of our optogenetic system reveal opportunities for future investigation. We have performed these experiments at an ectopic position in the embryo where position-specific gene expression may influence ERK interpretation differently than at the poles. For example, the gap gene knirps has been shown to repress tll in the center of the embryo (Kim et al., 2013), and we observed that the total domain of tll and byn expression was smaller under low light. Because of these positional differences in ERK sensitivity, it is not possible to make absolute comparisons about input and output strengths with the endogenous terminal pattern. In the future it will be interesting to investigate this circuit in embryos lacking other sources of positional information, preventing localized gap gene expression (Hannon et al., 2017; Singh et al., 2022). Also, it is possible that our methods left some transcriptional bursts undetected, and we cannot distinguish whether our upper bound of ∼75% transcriptionally active nuclei represents true transcriptional heterogeneity or an experimental limit of detection. These limitations could be overcome by future studies using techniques that simultaneously label the target DNA locus and measure transcription in live embryos (Alamos et al., 2023), or advances in high-quality volumetric imaging and machine-learning approaches (Lammers et al., 2020). Finally, many questions remain about the precise temporal relationships between ERK activation, gene transcription and protein accumulation. What is the relative influence of tll and hkb transcripts produced by early versus late nuclear cycles, and what is the delay between RNA production and protein accumulation? Combining transcriptional and protein reporters in the same embryo with mathematical models will allow us to address these questions (Birnie et al., 2023; McFann et al., 2021; Wilson et al., 2017).

Altogether, this work provides a blueprint for dissecting a developmental circuit with optogenetic tools to reveal new insights about network architecture. We have manipulated amplitude, duration, timing and spatial pattern of the signal to understand the contributions of each factor to signal interpretation. This framework will be an effective strategy for dissecting other developmental circuits in the future.

Drosophila melanogaster stocks used in this paper include the maternal tubulin Gal4 driver 67; 15 [P{matα-GAL-VP16}67; P{matα-GAL-VP16}15, Bloomington Drosophila Stock Center (BDSC), #80361], UAS-OptoSOS (Johnson et al., 2017), UAS-OptoSOS; MCP-mCherry (McFann et al., 2021), endogenous MS2-tagged tll-24xMS2, hkb-24x MS2 and byn-24xMS2 (Keenan et al., 2022), 67; 15 tsl⁴/TM3 Sb and MCP-mCherry; UAS-OptoSOS tsl⁴/TM3 Sb (Johnson et al., 2020), tub-miniCic-NeonGreen (a gift from Romain Levayer, Institut Pasteur, Paris, France), vasa-mCherry (a gift of Hernan Garcia, University of California, Berkeley, USA), tll-mCherryLlamaTag and hkb-mCherryLlamaTag (this paper), and UAS-alphaTub84B.tdEOS (BDSC, #51313). For details of the generation of Tll and Hkb LlamaTag flies please see the supplementary Materials and Methods, and see Table S1 for primer sequences for cloning. Flies were cultured using standard methods at 25°C.

For all imaging experiments, a cage of parental flies was placed in the dark at room temperature for at least 2 days before collection with standard apple juice plates and yeast paste. For wild-type experiments, female 67/OptoSOS; 15/MCP-mCherry virgins were mated with homozygous MS2 males. For tsl mutant experiments, female 67/MCP-mCherry; 15 tsl⁴/OptoSOS tsl⁴ virgins were mated with homozygous byn-MS2 males. Embryos were mounted between a semi-permeable membrane and a coverslip in a 3:1 mix of halocarbon 700/27 oil. Wild-type embryos were mounted with their ventral side facing the coverslip and tsl embryos were mounted with their lateral side facing the coverslip.

All imaging was carried out at room temperature at 40× on a Nikon Eclipse Ti microscope with a Yokogawa CSU-X1 spinning disk and an iXon DU897 EMCCD camera. Optogenetic stimulation was performed with a Mightex Polygon digital micromirror device using an X-Cite LED 447-nm blue LED. Embryos with similar red fluorescence levels were selected for imaging with the goal of choosing embryos expressing similar levels of OptoSOS. For wild-type embryos, a stimulation region was drawn at a position slightly posterior to the center of the embryo encompassing only the right half of the embryo, with the ventral midline as the illumination boundary. For tsl embryos, a stimulation region was drawn encompassing roughly the most posterior 60 μm of the embryo. Unless otherwise noted, blue light stimulation began in NC10, when nuclei could first be observed at the surface. A 0.6 s pulse of blue light was applied to the embryo every 30 s. Light amplitude was varied by changing the LED power over an 8-fold range. Images were acquired every 30 s with 1 μm z-slices over a range of 22-30 μm. At high light intensities, SSPB-tagRFP-SOS^cat rapidly relocalized to the membrane. Any embryos without visible SOS^cat localization were excluded. At low light intensities, this membrane enrichment was only barely detectable.

Images were processed using ImageJ. All images containing MS2 were background subtracted using rolling ball subtraction with radius 50 pixels. A maximum projection was then made with z-slices containing the nuclei. Images were analyzed manually, and each nuclear cycle was analyzed independently. For each nucleus, the position at the start of each nuclear cycle was identified. Although the mitotic wave staggers entry to the nuclear cycle, the start of a nuclear cycle was globally defined as the time when reformed nuclei were first visible in the most posterior positions.

The nucleus was then annotated with the time of the first and last visible MS2 focus, defining the ‘bursting duration’. The nucleus was defined to be in a ‘bursting state’ at any timepoint within this bursting duration, even if an MS2 focus was not detected. The percentage of nuclei in a bursting state during each nuclear cycle or at a given timepoint within a nuclear cycle was calculated out of the total number of nuclei. To determine the ‘time to 50% nuclei in bursting state’ for each embryo, we identified the first nuclear cycle in which at least 50% of nuclei were in a bursting state and at least 50% of nuclei remained in a bursting state for all subsequent nuclear cycles. We then identified the timepoint (measured in minutes since illumination began) at which 50% of nuclei had entered the bursting state within that nuclear cycle.

We observed some heterogeneity within the viewing region. At low light intensities, tll and byn varied in how far in the anterior direction their transcription extended in the illuminated region. To account for this variation, we quantified a box of defined size (50×45 μm) positioned at the most anterior extent of transcription in NC14 when comparing high and low light (see Fig. S2C,D). Similarly, when comparing transcription under high light in the illuminated and unilluminated region, we found that tll, hkb and byn all varied in how far from the illumination boundary their transcription extended. Therefore, we similarly defined a 30×75 μm box that shifted in the unilluminated region to a position at the farthest extent of transcription away from the illumination boundary in NC14 (see Fig. S5A,B). Any nuclei within this quantification box were included in quantifications.

To measure membrane enrichment of OptoSOS, we segmented out the nuclei from the background subtracted and maximum projected images, leaving only the cytoplasm/membrane portions. In NC14, when this analysis was carried out, there was very little visible cytoplasm in these planes so we included any non-nuclear region in the membrane analysis. To measure OptoSOS recruitment as a proxy for ERK signaling (Fig. 2D), we measured the difference in mean membrane intensity in the illuminated region and an unilluminated region at the same anterior-posterior position. To measure the profile of OptoSOS recruitment relative to the illumination boundary (Fig. S6B), we measured the median intensity at each pixel position at increasing distances from the boundary and normalized it to the highest intensity measurement in each embryo.

Nuclei were segmented using a model fine-tuned from the Cellpose ‘cyto2’ model, trained using default hyperparameters on manually segmented frames. MS2 bursts were identified through manual segmentation. To produce burst intensity visualizations, the nucleus mask nearest to each identified burst was colored to indicate detection.

All statistical analysis was performed in Prism 9 (GraphPad). Details for statistical tests used can be found in the figure legends. Figure legends also indicate the number of embryos or nuclei analyzed for each condition (n). All graphs show the mean±s.e.m. unless otherwise noted. In all cases, significance was defined as *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001. n.s. indicates no significance.

We thank Romain Levayer for the kind gift of miniCic-NeonGreen flies and Hernan Garcia for the kind gift of vasa-mcherry flies. We thank Eric Wieschaus and all members of the Toettcher and Shvartsman labs for helpful discussions. Stocks obtained from the Bloomington Drosophila Stock Center (NIH P40OD018537) were used in this study.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201818.reviewer-comments.pdf.