Abstract
Information from developmental signaling pathways must be accurately decoded to generate transcriptional outcomes. In the case of Notch, the intracellular domain (NICD) transduces the signal directly to the nucleus. How enhancers decipher NICD in the real time of developmental decisions is not known. Using the MS2/MCP system to visualize nascent transcripts in single cells in Drosophila embryos we reveal how two target enhancers read Notch activity to produce synchronized and sustained profiles of transcription. By manipulating the levels of NICD and altering specific motifs within the enhancers we uncover two key principles. First, increased NICD levels alter transcription by increasing duration rather than frequency of transcriptional bursts. Second, priming of enhancers by tissue-specific transcription factors is required for NICD to confer synchronized and sustained activity; in their absence, transcription is stochastic and bursty. The dynamic response of an individual enhancer to NICD thus differs depending on the cellular context.
Introduction
Genes respond to external and internal cues through the action in the nucleus of transcription factors and effectors of signalling pathways. Regulatory regions that surround genes, termed enhancers, integrate the information from these inputs to produce an appropriate transcriptional output. During development some of these decisions can occur in a matter of minutes, but usually the outcomes are measured many hours later. Rarely have transcription dynamics been analyzed in vivo in the real-time of the developmental signalling pathways, so we know little about how recipient enhancers decipher the signals. For example, enhancers could respond in a digital manner, working as simple on off switches, or as analog devices, operating as a rheostat so that levels of the signal can modulate the output (Blackwood et al. 1998; Garcia et al. 2013; Lammers et al. 2018). In either case they must also have the capability to detect and transduce key parameters to the transcription machinery, such as input signal duration and thresholds.
With the advent of precise and quantitative methods to measure transcription, such as single molecule fluorescence in situ hybridization (smFISH) or live imaging, it has become evident that transcription is not a continuous process. Instead, genes that are being actively transcribed undergo bursts of initiation that are often separated by inactive intervals (Chubb et al. 2006; Golding et al. 2005). Bursting is thought to occur because the dynamics of enhancer-promoter activation leads to episodic polymerase release. One consequence of this behaviour is that factors modulating the output levels of transcription can do so by changing either the frequency with which a burst occurs (measured by the gap between bursts) or the size of each burst (measured by changes in burst duration and/or amplitude). Since forced looping of the beta-globin enhancer to its promoter led to an increase in burst frequency (Bartman et al. 2016), it has been proposed that transcription factors activate transcription by modulating enhancer-promoter interactions, and hence bursting frequency; although other studies suggest enhancer-promoter interactions are not the underlying basis of transcriptional bursting (Lim et al. 2018; Chen et al. 2018). Though the molecular origin of bursting remains unknown, bursting frequency rather than burst duration or amplitudes seems to be the major parameter modulated in different species and contexts (So et al. 2011; Senecal et al. 2014; Xu et al. 2015; Desponds et al. 2016; Padovan-Merhar et al. 2015; Lammers et al. 2018; Berrocal et al. 2018). For example, enhancers controlling early patterning genes in Drosophila embryos all produce similar bursting size but have different bursting frequencies, which can be attenuated by the presence of insulators (Fukaya et al. 2016). Similarly, steroids increase the bursting frequency of target enhancers to regulate their activation kinetics (Larson et al. 2013; Fritzsch et al. 2018). However, it remains to be discovered whether all transcription factors alter transcription dynamics in this way and specifically whether it is these or other properties that are modulated by developmental signals to confer appropriate outputs in the in vivo setting of a developing organism.
Transcriptional bursting is thought to make an important contribution to heterogeneity in the output of transcriptional activity between cells (Raj et al. 2008). For example, in cells exposed to estrogen, response times for activation of transcription measured live were highly variable and there was no coherent cycling between active and inactive states (Fritzsch et al. 2018). Stochastic transcriptional behaviour has been found of key importance in many developmental decisions, such as the differentiation of photoreceptors in the Drosophila eye (Wernet et al. 2006), hematopoietic cell differentiation in mouse cells (Chang et al. 2008; Ng et al. 2018) or during neuronal differentiation in the zebrafish retina (Boije et al. 2015). But while an attractive feature for promoting heterogeneity, inherent variability in responses could be extremely disruptive in developmental processes where the coordinated response of many cells is required to pattern specific structures. In some cases this may be circumvented by mechanisms that allow cells to achieve the same average mRNA output and so produce homogeneous patterns of gene expression (Little et al. 2013). For example, cells that express the mesodermal determinant Snail average their transcriptional output over a period of 20 minutes by mRNA diffusion to produce a homogeneous field of cells and a sharp boundary in Drosophila syncytial embryos (Bothma et al. 2018). However it is only in rare circumstances that mRNA diffusion can operate and it is unclear whether other averaging mechanisms would be effective over shorter time intervals. To effectively achieve reproducible patterns, cells must therefore overcome the variability that is inherent in transcriptional bursting and stochastic enhancer activation.
Notch signaling is one highly conserved developmental signaling pathway that is deployed in multiple different contexts. It has the unusual feature that the Notch intracellular domain (NICD) transduces the signal directly to the nucleus, when it is released by a series of proteolytic cleavages precipitated by interactions with the ligands. NICD then stimulates transcription by forming a complex with the DNA binding protein CSL and the co-activator Mastermind (Mam) (Bray 2006). The lack of intermediate signalling steps and amplification makes this a powerful system to investigate how signals are deciphered by responding enhancers. Furthermore, there may be differences in the levels and dynamics of NICD produced by different ligands (Nandagopal et al. 2018). However, although its role as a transcriptional activator is well established, at present we know little about how enhancers respond to NICD in the real time of developmental decisions. For example, do the enhancers operate as simple switches, detecting when NICD crosses a threshold? Or are they sensitive to different levels of NICD, in which case does NICD, like other factors, modulate bursting frequency? Nor do we know what features in the sequence of the responding enhancers confer the output properties, although it has been suggested that enhancers with paired CSL motifs (referred to as SPS motifs)(Bailey et al. 1995; Nam et al. 2007), whose precise spacing could favour NICD-NICD dimerization, have the potential to yield the strongest responses (Nam et al. 2007).
In order to determine how enhancers respond to Notch activity in real time we have used the MS2/MCP system to visualize nascent transcripts in Drosophila embryos. To do so we used two well-characterised Notch responsive enhancers that drive expression in a stripe of mesectoderm (MSE) cells and analyzed the levels of transcription they produced over time at the single cell level. Strikingly all MSE cells initiate transcription within a few minutes of one another, and once active, each nucleus produced a sustained profile of transcription, without distinct bursts. By manipulating the levels of NICD and altering key motifs within the enhancers we uncover two key principles. First, the ability of NICD to confer synchronized and sustained activity in MSE requires that the enhancers are primed by tissue-specific transcription factors. In their absence, MSE enhancers confer stochastic bursty transcription profiles, demonstrating that different response profiles can be generated from a single enhancer according to which other factors are present. Second, changing Notch levels modulates the transcription burst size but not length of the periods between bursts, in contrast to most current examples for enhancer activation. These two key concepts, that we have uncovered by analysing the dynamics of transcription profiles produced by enhancer variants in different signalling conditions, are likely to be of general importance for gene regulation by other signalling pathways in developmental and disease contexts.
Results
Synchronised and sustained enhancer activation in response to Notch
To investigate how Notch signals are read out in real time, we focused on the well-characterized mesectodermal enhancers (MSEs) from the Enhancer of split-Complex (E(spl)-C) (known as m5/m8) and from singleminded (sim) (Morel et al. 2000; Cowden et al. 2002; Zinzen et al. 2006a). These direct expression in two stripes of MSE cells when Notch is activated in response to Delta signals from the presumptive mesoderm (Fig. 1AB) (Morel et al. 2003; De Renzis et al. 2006; Zinzen et al. 2006a). The MSE converge to the midline during gastrulation, ultimately forming CNS midline precursors similar to the vertebrate floorplate. To visualize transcription from these enhancers in real time, they were inserted into MS2 reporter constructs containing the promoter from the gene even-skipped (peve), 24 MS2 loops and the lacZ transcript (Fig. 1A). When these MS2 reporters were combined with MCP-GFP in the same embryos, nascent transcription was marked by the accumulation of MCP-GFP in bright nuclear puncta, where the total fluorescence in each spot is directly proportional to the number of transcribing mRNAs at any timepoint (Fig. 1AB)(Garcia et al. 2013). In this way the levels of transcription can be followed over time at the single cell level by tracking the puncta relative to the nuclei (which were labelled with His2Av-RFP).
By visualizing transcription in real time, we could see that both m5/m8 and sim switched on transcription in all cells along the MSE stripe within a narrow time-window (~ 10 min) in nc14 (Fig. 1CDE). We note that both enhancers also directed earlier, Notch independent transcription, in broad domains in nuclear cycles 10 to 13 (Movie 1. Movie 2.) and in the first few minutes of nuclear cycle 14 in few scattered cells. However, this was followed by a long period (approximately 20 min) of inactivity before the cells in the MSE stripe initiated transcription concurrently. m5/m8 and sim were then active in a sustained manner in all nuclei - few separated bursts of transcription were detected - throughout the remaining period of nc14 as the embryos underwent the first stage of gastrulation (mesoderm invagination) (Fig. S1E). Transcription then ceased after 30-50 minutes, with less synchrony than at the onset (Fig. 1E).
Sustained activity is a feature of m5/m8 and sim and not a general property of Notch responsive enhancers, as a neuroectodermal enhancer from E(spl)m8-bHLH (m8NE, Fig. 1A) exhibited delayed and stochastic activity in the MSE at this stage (Fig. S1AB). Furthermore, even though the profiles produced by m5/m8 and sim were continuous, the amplitude fluctuated, likely reflecting episodic polymerase release. However it is notable that m5/m8 and sim both direct transcription profiles that are highly co-ordinated temporally, with each conferring a prolonged period of activity that is initiated within a short time-window. Indeed, the mean profile of all the MSE cells was almost identical for the two enhancers (Fig. 1F). This is remarkable given that they contain different configurations of binding motifs and implies that the mesectoderm cells undergo a highly synchronized period and level of Notch signaling.
We next tested the consequences from substituting different promoters with the m5/m8 and sim enhancers, to assess the relative contributions of the enhancer and promoter to the response profiles. First, when peve was replaced by a promoter from sim (psimE), both m5/m8 and sim produced lower levels of transcription, but their overall temporal profiles remained similar and the mean levels were the same for the two enhancers (Fig. S1C). Second, we combined m5/m8 with another heterologous promoter, hsp70, and with four promoters from the E(spl)-C genes that could be interacting with m5/m8 in the endogenous locus. Similar to psimE, substituting these promoters also led to changes in the mean levels of transcription without affecting the overall temporal profile or expression pattern (Fig. S1D). Notably, even in combinations where the overall levels were lower, the transcription profiles remained sustained rather than breaking down into discrete bursts (Fig. S1E) consistent with promoters affecting mean levels of activity without modulating bursting frequencies. Of those tested, pm6 produced the lowest mean levels when combined with m5/m8 (Fig. S1D). This is consistent with the fact that E(spl)m6-BFM is not normally expressed in the MSE and argues for an underlying enhancer-promoter compatibility at the sequence level (Fig. S1D)(Zabidi et al. 2014). Nevertheless, the fact that similar temporal profiles were produced with all the promoters confirms that the enhancers are the primary detectors of Notch signaling activity.
To verify that nc14 MSE transcription was indeed Notch-dependent we measured transcription from m5/m8 in embryos where Notch activity was disrupted by mutations. Embryos lacking Neuralized, an E3 ubiquitin ligase required for Delta endocytosis that is critical for Notch signalling (Morel et al. 2003; De Renzis et al. 2006), had no detectable transcription from m5/m8 in the MSE (Fig. 1G). Likewise, m5/m8 activity was severely compromised in embryos carrying mutations in Delta. Because Delta protein is deposited in the egg maternally (Kopczynski et al. 1988), these embryos contained some residual Delta which was sufficient for a few scattered cells in the MSE stripe to initiate transcription (Fig. S1F). However their transcription ceased prematurely, within <20 min (Fig. 1G, S1F). Together these results confirm that the enhancers require Notch signalling for their activity in the MSE, in agreement with previous studies of these regulatory regions (Morel et al. 2000; Zinzen et al. 2006a), and further show that sustained Notch signalling is needed to maintain transcription, arguing that the enhancers are also detecting persistence of NICD.
Coordinated activity of enhancers within each nucleus
Although m5/m8 and sim confer well coordinated temporal profiles of transcriptional activity, there is nevertheless some cell to cell variability in the precise time of their activation. To investigate whether this cell to cell variability was due to the stochastic nature of transcription (intrinsic variability) or whether it indeed reflects changes in signalling from Notch (extrinsic variability) (Elowitz 2002; Raser et al. 2006) we monitored expression from two identical alleles of the MS2 reporters, supplied by the paternal and maternal chromosomes (Fig. 2A). Transcription from these two physically unlinked loci were detected as distinct puncta in each nucleus so that each one could be tracked independently. We found a remarkable synchrony in the onset of transcription from both alleles of a given enhancer (Fig. 2B). More than 80% of the cells initiated transcription from both alleles less than 5 min apart, (Fig. S2C), which contributes to ~ 6-30% of the total variability (Fig. 2D), indicating that most of the temporal variability in transcription onset between cells was due to extrinsic factors. There was less synchrony between the two alleles in the time at which transcription was extinguished (Fig. 2B S2A), but the extent of variability was much lower than that between cells (only contributing to less than 15% of the total variability, Fig. 1D) and it likely occurs because there will be locus to locus variations in the stage of the transcriptional bursting cycle when the signaling levels decline
Although the overall temporal profiles of transcription from the two alleles were similar to one another, in terms of the onset and overall increases or decreases in levels, the fine grained spikes and troughs were not synchronised (Fig. 2A), in agreement with the expectation that transcription from two different loci is largely uncorrelated (Harper et al. 2011; Little et al. 2013; Fritzsch et al. 2018). However, the fluorescent intensities of two alleles at any time point displayed a small but significant positive correlation (R2 ~ 0.35), compared to a null correlation when these pairs are randomly assigned (Fig. S2B). This argues that the enhancers at the two alleles operate independently while being co-ordinated by the same extrinsic signal information, namely the durations and levels of Notch activity. Even when the m5/m8 and sim enhancers were placed in trans in the same cell, there was comparatively little variation in the onset times, compared to the variation in the onset of the enhancers in different cells (Fig. 2CD S2A). These results indicate that m5/m8 and sim are reliably detecting extrinsic information in the form of Notch activity, which is initiated in the mesectoderm cells within a 5-10 minute time-window, so that within a given nucleus their activation is remarkably synchronized.
Enhancers detect signal thresholds and signal context
The m5/m8 and sim enhancers appear to act as “persistence detectors”, driving transcription as long as Notch signal(s) are present. They may therefore be simple switches detecting when a signal crosses a threshold (digital encoding). Alternatively, the enhancers could respond in a dose-sensitive manner to the levels of Notch activity (analog encoding). To distinguish these possibilities, we tested the consequences from additional Notch activity, by supplying ectopic NICD using the stripe 2 regulatory enhancer from the even-skipped gene (eve2-NICD). This confers a tightly regulated ectopic stripe of NICD which is orthogonal to the MSE (Fig. 3A) (Kosman et al. 1997; Cowden et al. 2002) and was sufficient to produce ectopic expression from both m5/m8 and sim driven reporters (Movie 3.Movie 4.).
Whereas expression from m5/m8 and sim was almost identical in wild-type embryos, clear differences in their behaviour were revealed by ectopic NICD. First, transcription from m5/m8 was detected throughout much of the region corresponding to the eve2 stripe whereas ectopic transcription from sim was only seen in 3-4 cell wide region dorsal to the MSE (Fig. 3B), consistent with previous observations (Cowden et al. 2002; Zinzen et al. 2006a). Second, although both enhancers initiated transcription prematurely, because the ectopic NICD was produced from early nc14 (Bothma et al. 2014), the onset of transcription from m5/m8 was significantly earlier than that from sim (Fig. 3DE). Given that both enhancers are exposed to the same temporal pattern of NICD production, this difference in their initiation times implies that the two enhancers have different thresholds of response to NICD, with m5/m8 responding to lower doses and hence being switched-on earlier. Therefore, we hypothesize that m5/m8 and sim respond at the same time in wild-type embryos because the normal ligand-induced signaling leads to a sharp increase in NICD.
We also detected differences in the dynamics of m5/m8 according to the location of the NICD-expressing nucleus along the DV axis. Nuclei closer to the MSE stripe (in the neuroectoderm, NE) exhibited strong activity, with a temporal pattern that resembled that in the MSE (Fig. 3C, bottom). In contrast nuclei in dorsal regions (dorsal ectoderm, DE) underwent resolved bursts of transcriptional activity (Fig. 3C, top). Ectopic NICD also induced ‘bursty’ expression from sim in the mesoderm (ME), but was not capable of turning on m5/m8 in that region (Movie 5.). The positional differences in dynamics suggest that intrinsic cellular conditions, likely the expression levels of specific transcription factors, influence the way that enhancers “read” the presence of NICD. Such factors must therefore have the capability to modulate the dynamics of transcription.
The fact that m5/m8 and sim are switched on at different times in the presence of ectopic NICD suggests that they require different thresholds for their activation. In addition, they only give sustained transcription profiles in a 2-3 cell-wide region overlapping the MSE, whereas elsewhere they generate stochastic and “bursty” transcription, arguing that they must be differently primed in the MSE region.
Notch activity tunes transcription burst size
To further test how Notch responsive enhancers respond to different doses of signal, we introduced a second eve2-NICD transgene. MSE transcription from sim in the presence of 2xeve2-NICD initiated earlier and achieved higher levels than with 1xeve2-NICD (Fig. 4A, left). This is consistent with the hypothesis that the sim enhancer responds to thresholds of NICD concentration, as the cells will reach a given concentration of signal more quickly in the embryos with 2xeve2-NICD. The mean levels of transcription increased in the ME as well as in the MSE regions (Fig. 4A-C), further indicating a dose-sensitive response. In contrast, the levels and onset of MSE transcription from m5/m8 did not significantly change in 2xeve2-NICD embryos (Fig. 4A, right). The output levels of transcription from the m5/m8 enhancer therefore reached a saturation point with the dose produced by 1xeve2-NICD, possibly due to limiting levels of other factors at this stage. This only occurred in the MSE, as the more stochastic activity in the DE remained sensitive to increases in NICD, becoming responsive in a greater proportion of cells and remaining active over longer periods (Fig. S4A).
To distinguish different models for how NICD confers a dose-sensitive response, we took two strategies to analyze its effect on the transcriptional bursting dynamics. Both approaches assume a two state model where the promoter is switched between an OFF and ON state with switching rates Kon and Koff (representative of the probabilities of switching the enhancer on and off respectively) and confers transcription initiation rate r in the ON state (Fig. 4E)(Peccoud et al. 1995; Larson et al. 2009). In the first approach we used the parameters of bursting amplitude, off period between bursts and bursting length as approximations for r, Kon and Koff, respectively (Fig. 4E). In most previous enhancers analyzed in this way, the off period is the most affected, leading to changes in the frequency of bursting (Fukaya et al. 2016; Fritzsch et al. 2018; Lammers et al. 2018). However, when we quantified the effect from different doses of NICD on sim in the ME, a region where individual bursts of transcription could be distinguished, we found that the bursting length consistently increased with higher amounts of NICD whereas the off period between bursts remained constant (Fig. 4DF). This indicates that the main effect of NICD is to keep the enhancer in the ON state for longer - ie. decreasing Koff - rather than increasing the frequency with which it becomes active (i.e. increasing Kon). The bursting amplitude also increased with 1xeve2-NICD but this was not further enhanced by 2xeve2-NICD (Fig. 4DF). Overall therefore, increasing levels of NICD in the ME result in sim producing an increase in transcription burst size (duration x amplitude) rather than an increase in the frequency of bursts. Transcription in other regions and enhancers (m5/m8 DE and m8NE ME) showed similar increase in burst size in response to the dose of NICD (Fig. S4A-C) suggesting this is a general property of these Notch responsive enhancers.
We developed a second approach, based on the noise properties of transcription, to analyze the changes in the dynamics where single bursts of activity could not be defined. To do so, we used a mathematical model of transcription to account for the initiating mRNA molecules (Fig. S3A). Using derivations from the mathematical model and testing them in simulations, we looked for the signatures that would be produced if the mean of initiating mRNAs (equivalent to the mean fluorescence from the MS2 puncta) were increasing due to changes in r, Kon or Koff. This showed that the effects on the Fano factor ratio between the two conditions and on their autocorrelation function (ACF) could be used to correctly predict which of the three parameters could account for the increase in the mean (Fig. S3B, Supplementary Methods). First we tested the modelling approach with the data from the promoter swap experiments. Analyzing the differences in the mean indicated that they are most likely due to increases in r (Fig. S4D), as expected if promoters influence the rate of polymerase release but not the activation of the enhancer per se. When we then applied the model to the data from the transcription profiles produced by different doses of NICD in the ME the results were most compatible with the causal effect being a increase in r or a decrease in Koff (Fig. S4E) depending on which two conditions were compared, i.e. this approach also indicated that NICD elicits an increase in burst size rather than in burst frequency. Thus the two approaches both converged on the model that, above the critical threshold level of NICD, further increases in NICD levels prolong the period that each enhancer remains in the ON state.
Finally, we then used an enhancer - promoter combination that produces higher mean levels (m5/m8-pm5, Fig. S1D) to investigate whether the saturation that occurred with ectopic NICD was due to the peve promoter having achieved a maximal initiation rate. Strikingly, the substitution of pm5 did not result in significantly higher maximal levels than m5/m8-peve in the presence of eve2-NICD (Fig. S4F) although it did in wild-type signaling conditions (Fig. S1D). This result indicates that the saturation of the m5/m8 response that occurs with higher levels of NICD stems from the m5/m8 enhancer rather than the promoter and argues that enhancers reach a maximal “ON” state that they cannot exceed even if more NICD is provided.
Paired CSL motifs augment burst-size not threshold detection
The m5/m8 and sim enhancers both respond to NICD but they initiate transcription at different thresholds. How is this encoded in their DNA sequence? A prominent difference between the two enhancers is that m5/m8 contains a paired CSL motif (so-called SPS motifs), a specific arrangement and spacing of binding motifs that permit dimerization between complexes containing NICD (Nam et al. 2007), whereas sim does not (Fig, S5A). To test their role, we replaced two of the CSL motifs in sim with the SPS motif from m5/m8 and conversely perturbed the SPS in m5/m8 by increasing the spacing between the two CSL motifs (Fig. S5A). As SPS motifs permit co-operative binding between two NICD complexes, we expected that enhancers containing an SPS motif (simSPS and m5/m8) would exhibit earlier onsets of activity than their cognates without (sim and m5/m8insSPS). However this was not the case for either sim and simSPS (Fig. 5AB) or m5/m8 and m5/m8insSPS in either wild type or eve2-NICD embryos (Fig. S5DE). These profiles suggest that the SPS motifs are not responsible for the difference in the threshold levels of NICD required for m5/m8 and sim activation.
Changes to the CSL motifs did however affect the mean levels of activity. simSPS directed higher mean levels of activity compared to sim in both wild type and eve-NICD embryos (Fig. 5A S5B). Conversely, m5/m8insSPS directed lower levels compared to m5/m8 (Fig. S5D). Analysing the traces from sim enhancer in the ME, where cells undergo bursts of transcription, revealed that the SPS site (simSPS) led to larger burst-sizes - i.e. increased the amplitude and the duration - compared to the wild type enhancer without SPS sites (sim) (Fig. 5CD). Conversely, the continuous profile produced by m5/m8 in the MSE was broken into smaller bursts when the SPS was disrupted (Fig. S5FG). The effects on the bursting size are similar to those seen when the dose of NICD was altered, suggesting that enhancers containing SPS sites respond to a given level of NICD more effectively. They do not however appear to affect the amount of NICD required for their initial activation, i.e. the threshold required for the enhancer to be switched on. This implies that the burst-size modulation and response threshold can be uncoupled and potentially could be encoded independently at the DNA level.
Regional factors prime enhancers for fast and sustained activation
Under ectopic NICD conditions, m5/m8 and sim both produce sustained transcription profiles in the region overlapping the MSE and NE, whereas elsewhere they generate stochastic and “bursty” transcription. This suggests that other factors are “priming” the enhancers to respond to NICD. Good candidates are the factors involved in DV patterning at this stage, the bHLH transcription factor Twist (Twi) and/or the Rel protein Dorsal (dl). Indeed, the region where the enhancers generate sustained profiles in response to eve2-NICD coincides with the domain of endogenous Twist and Dorsal gradients (Fig S6B)(Zinzen et al. 2006b). Furthermore, m5/m8 and sim both contain Twist and Dorsal binding motifs (Fig. S6A) and previous studies indicated that Twist is important for activity of sim although it was not thought to contribute to the activity of m5/m8 (Zinzen et al. 2006a).
To test if Twist and Dorsal are responsible for the different dynamics of transcription observed in m5/m8 we mutated Twist and/or Dorsal binding motifs in m5/m8, which normally exhibits strong activity in the MSE and NE and a ‘bursty’ pattern in DE cells in conditions of ectopic Notch activity (Fig. 3B). Strikingly, mutation of either the three Twist motifs in m5/m8 or the two Dorsal motifs produced a delay in the start of transcription in both WT and eve2-NICD embryos. These effects were even more pronounced when both Twist and Dorsal motifs were mutated together (Fig. 6AB), implying that, without Twist or Dorsal, m5/m8 requires a higher threshold of NICD to become active or responds more slowly to the same threshold. The mean transcription levels were also reduced in all cases (Fig. 6A).
Mutating the Twist motifs had two additional effects: the overall proportion of active cells in the MSE was reduced (Fig. 6C) and out of those active, fewer exhibited the sustained profile observed with the wild type enhancers (Fig. 6DE). Instead most cells displayed a ‘bursty’ transcription profile (Fig. 6D), similar to those elicited by NICD in the DE region. Although the mutated Twist motifs led to bursty profiles in wild type embryos, these effects were partially rescued when ectopic NICD was provided (Fig. 6CE, S6C). However, when both Dorsal and Twist motifs were mutated, the proportions of active cells and of cells with a sustained profile were both decreased even in the presence of ectopic NICD (although mutation of Dorsal motifs alone did not produce a significant decrease in either property) (Fig. 6CE, S6C). The results are therefore consistent with a role for Twist and Dorsal in priming the m5/m8 enhancer to produce sustained activity. In their absence the ability of the enhancer to initiate transcription becomes much more stochastic. Consistently, another Notch responsive enhancer that only contains one Twist motif (the neuroectodermal enhancer m8NE, Fig. S6A) also exhibited a delayed onset of activity (Fig. S6D) and gave stochastic bursting patters (Fig. 6E). This suggest that the two MSE enhancers are especially primed to respond in a fast and sustained manner at this stage.
Discussion
Developmental signaling pathways have widespread roles but currently we know relatively little about how the signaling information is decoded to generate the right transcriptional outcomes. We therefore set out to investigate the principles that govern how Notch activity is read by target enhancers in the living animal, using the MS2/MCP system to visualize nascent transcripts in Drosophila embryos and focusing on two enhancers that respond to Notch activity in the MSE. Three striking characteristics emerge. First, the MSE enhancers are sensitive to changes in the levels of NICD, which modulate the transcriptional burst size rather than increasing burst frequency. Second, the activation of both MSE enhancers is highly synchronous. Indeed, within one nucleus the two enhancers become activated within a few minutes of one another. Third, both MSE enhancers confer a sustained response in the wild-type context. This synchronized and persistent activity of the MSE enhancers is in stark contrast to the highly stochastic and bursty profiles that are characteristics of most other enhancers that have been analyzed (Little et al. 2013; Fukaya et al. 2016; Fritzsch et al. 2018) and relies on the MSE enhancers being “primed” by regional transcription factors Twist and Dorsal. We propose that such priming mechanisms are likely to be of general importance for rendering enhancers sensitive to signals so that a rapid and robust transcriptional response is generated.
Priming of enhancers sensitizes the response to NICD
Transcription of most genes in animal cells occurs in bursts interspersed with refractory periods of varying lengths, that are thought to reflect the kinetic interactions of the enhancer and promoter (Bartman et al. 2016). However, the MSE enhancers maintain transcription for 40-60 minutes, without any periods of inactivity. Calculation of the autocorrelation function in the traces from these nuclei suggest very slow transcriptional dynamics (Fig. S4ED)(Desponds et al. 2016), which would be consistent with one long period of activity as opposed to overlapping short bursts. This fits with a model where promoters can exist in a permissive active state, during which many “convoys” of polymerase can be fired without the promoter reverting to a fully inactive condition (Tantale et al. 2016). The rapid successions of initiation events are thought to require Mediator complex (Tantale et al. 2016), which was also found to play a role in the NICD-mediated increase in residence time of CSL complexes (Gomez-Lamarca et al. 2018). We propose therefore that the sustained transcription from m5/m8 and sim reflects a switch into a promoter permissive state, in which general transcription factors like Mediator remain associated with the promoter so long as sufficient NICD is present, allowing repeated re-initiation.
However, the ability to drive fast and sustained activation is not a property of NICD itself. For example, when ectopic NICD was supplied, cells in many regions of the embryo responded asynchronously and underwent only short bursts of activity. Furthermore, variable and less sustained cell-by-cell profiles were generated in the MSE region when the binding motifs for Twist and Dorsal in the m5/m8 enhancer were mutated. The presence of these regional factors therefore appears to sensitize the enhancers to NICD, a process we refer to as enhancer priming. This has two consequences. First, it enables all nuclei to respond rapidly to initiate transcription in a highly coordinated manner once NICD reaches the threshold level. Second, it creates an effective ‘state transition’ so that the presence of NICD can switch the promoter into a permissive condition to produce sustained activity (Fig. 7). We propose a priming mechanism, rather than classic cooperativity, because Twist and Dorsal alone are insufficient to drive any enhancer activity. Furthermore, since the enhancers immediately achieve sustained activity when NICD is produced, it is most likely that Twist and Dorsal are required prior to the recruitment of NICD, although both may continue to play a role independently of priming after transcription is initiated, as suggested by the lower mean levels obtained when only Twist or Dorsal motifs are mutated.
Our explanation that the synchronous activation of the MSE enhancers reflects their requirements for a critical concentration of NICD is borne out by their responses when the levels of NICD are increased. Notably, while sim and m5/m8 exhibited almost identical dynamics in wild-type embryos, they displayed clear differences in the presence of ectopic NICD, suggesting that they detect slightly different thresholds. Indeed, doubling the dose of ectopic NICD further accelerated the onset times of sim in agreement with the model that the enhancers detect NICD levels. Threshold detection does not appear to rely on the arrangement of CSL motifs, as the onset times of m5/m8 or sim were unaffected by changes in the spacing of CSL paired sites. In contrast, mutating Twist or Dorsal binding-motifs in m5/m8 delayed the onset of transcription, arguing that these factors normally sensitize the enhancer to NICD enabling responses at lower thresholds.
We propose that enhancer priming will be widely deployed in contexts where a rapid and consistent transcriptional response to signaling is important, as in the MSE where a stripe of cells with a specific identity is established in a short time-window. In other processes where responses to Notch are more stochastic, as during lateral inhibition, individual enhancers could be preset to confer different transcription dynamics. This appears to be the case for a second enhancer from E(spl)-C (m8NE) which generates a stochastic response in the MSE cells, similar to that seen for the MSE enhancers when Twist and Dorsal sites are mutated. This illustrates that the presence or absence of other factors can toggle an enhancer between conferring a stochastic or deterministic response to signalling.
NICD regulates transcription burst size
Manipulating the levels of NICD revealed that the Notch responsive enhancers act as analog devices that can measure and broadcast variations in levels. Increased NICD levels have a consistent effect on enhancer activity irrespective of the priming state of the enhancer, in all cases leading to an increase in the burst duration. The effects can be most readily quantified in regions where NICD elicits discrete bursts of transcription initiation, such as the dorsal ectoderm for m5/m8 or mesoderm for sim and m8NE. Transcriptional bursting has been formalized as a two-state model where the promoter toggles between on and off states, conferring a transcription initiation rate when on(Peccoud et al. 1995; Larson et al. 2009). Changes in the duration or frequency of the bursts lead to an overall increase in transcription. Most commonly, differences in the activity of enhancers have been attributed to changes in the probability of the enhancer switching on (Kon), which produces different off periods between bursts, leading to changes in burst frequency (Larson et al. 2013; Senecal et al. 2014; Fukaya et al. 2016; Fritzsch et al. 2018; Lammers et al. 2018; Berrocal et al. 2018). We were therefore surprised to find that higher doses of NICD did not increase the burst frequency. Instead they produced bigger bursts, both by increasing the bursting amplitude, equivalent to the rate of transcription initiation, and the bursting length, indicative of the total time the enhancer stays in the on state. Modifications to the CSL motifs also impact on the same parameters. Thus, enhancers with paired motifs (SPS), which favour NICD dimerization (Nam et al. 2007), produced larger transcription bursts than those where the motifs are further apart. This suggests that paired motifs can ‘use’ the NICD present more efficiently. Interestingly, even though m5/m8 and sim contain different arrangements and numbers of CSL motifs they have converged to produce the same mean levels of transcription in wild type embryos.
Two models would be compatible with the observations that effective NICD levels alter the burst size. In the first model, increasing the concentration of NICD when the enhancer is activated would create larger Pol II clusters. This is based on the observation that low complexity activation domains in transcription factors can form local regions of high concentration of transcription factors, so-called “hubs”, which in turn are able to recruit Pol II (Mir et al. 2017; Tsai et al. 2017; Lu et al. 2018). As the lifetime of Pol II clusters appears to correlate with transcriptional output (Cho et al. 2016), the formation of larger Pol II clusters would in turn drive larger bursts. In the second model, NICD would be required to keep the enhancer in the ON state, for example by nucleating recruitment of Mediator and/or stabilizing a loop between enhancer and promoter, which would in turn recruit Pol II in a more stochastic manner. General factors such as Mediator have been shown to coalesce into phase-separated condensates that compartmentalize the transcription apparatus (Cho et al. 2018; Sabari et al. 2018; Boija et al. 2018) and these could form in a NICD dependent manner. Whichever the mechanism, the clusters/ON state must persist in a state that requires NICD yet is compatible with NICD having a short-lived interaction with its target enhancers (Gomez-Lamarca et al. 2018). Furthermore, the fact that the activity of m5/m8 enhancer saturates with one eve2-NICD construct, and can’t be enhanced by providing a more active promoter, suggests that that there is a limit to the size or valency of the clusters that can form.
Although unexpected, the ability to increase burst size appears to be a conserved property of NICD. Live imaging of transcription in response to the Notch homologue, GLP-1, in the C.elegans gonad also shows a change in burst size depending on the signalling levels (Lee et al. 2018). As the capability to modulate burst size is likely to rely on the additional factors recruited, the similarities between the effects in fly and worm argue that a common set of core players will be deployed by NICD to bring about the concentration-dependent bursting properties.
Materials and Methods
Cloning and transgenesis
Generation of MS2 reporter constructs
MS2 loops were placed upstream of a lacZ transcript and both were driven using different combinations of enhancers and promoters. 24 MS2 loops were cloned from pCR4-24XMS2SL-stable (Addgene #31865) into pLacZ2-attB (Bischof et al. 2013) using EcoRI sites. The m5/m8, sim and m8NE enhancers (Zinzen et al. 2006a; Kramatschek et al. 1994) were amplified from genomic DNA and cloned into pattB-MS2-LacZ using HindIII/AgeI sites (primers in Table 1). Subsequently the promoters hsp70, peve, pm5, pm6, pm7, pm8 and psimE were cloned by Gibson Assembly (Gibson 2011) in pattB-m5/m8-MS2-LacZ, pattB-sim-MS2-LacZ and/or pattB-m8NE-MS2-LacZ (primers in Table 1) using the AgeI restriction site and incorporating a EagI site. All mutations introduced in m5/m8 or sim were first introduced by Gibson Assembly in the enhancers contained in pCR4 plasmids and then transferred to pattB-peve-MS2-lacZ using HindIII and AgeI sites.
Su(H), Twi, dl and Sna binding motifs were identified using ClusterDraw2 using the PWM from the Jaspar database for each transcription factor. Motifs with scores higher than 6 and pvalues 0.001 were selected. Primers to create simSPS, m5/m8insSPS, m5/m8Δtwl, m5/m8Δdl and m5/m8Δtwl Δdl are detailed in Table 1.
The following constructs have been generated and inserted by ΦC31 mediated integration (Bischof et al. 2007) into an attP landing site in the second chromosome – attP40, 25C – to avoid positional effects in the comparisons: pattB-m5/m8-peve-MS2-LacZ, pattB-m5/m8-hsp70-MS2-LacZ, pattB-m5/m8-pm5-MS2-LacZ, pattB-m5/m8-pm6-MS2-LacZ, pattB-m5/m8-pm7-MS2-LacZ, pattB-m5/m8-pm8-MS2-LacZ, pattB-m5/m8-psimE-MS2-LacZ, pattB-sim-peve-MS2-LacZ, pattB-sim-psimE-MS2-LacZ, pattB-simSPS-peve-MS2-LacZ, pattB-m5/m8insSPS-peve-MS2-LacZ, pattB-m5/m8Δtm-peve-MS2-LacZ, pattB-m5/m8Δdl-peve-MS2-LacZ and pattB-m5/m8Δtm Δdl-peve-MS2-LacZ.
Expression of ectopic NICD
To generate eve2-NICD the plasmid 22FPE (Kosman et al. 1997), which contains 2 copies of the eve2 enhancer with five high affinity bicoid sites, FRT sites flanking a transcription termination sequence and the eve 3’UTR, was transferred to pGEM-t-easy using EcoRI sites and from there to pattB (Bischof et al. 2013) using a Not I site. The NICD fragment from Notch was excised from an existing pMT-NICD plasmid and inserted in pattB-22FPE through the PmeI site to create the pattB-eve2x2-peve-FRT-STOP-FRT-NICD-eve3’UTR construct (referred to as eve2-NICD). This was inserted into the attP landing site at 51D in the second chromosome. To increase the amount of ectopic NICD produced, the same eve2-NICD construct was also inserted in the attP40 landing site at 25C and recombined with eve2-NICD51D to produce 2xeve2-NICD. Sequences of all generated plasmids are available in a benchling repository.
Fly strains and genetics
To observe the expression pattern and dynamics from m5/m8-peve, sim-peve and the different promoter combinations (Fig. 1, S1) females expressing His2av-RFP and MCP-GFP (BDSC #60340) in the maternal germline were crossed with males expressing the MS2-lacZ reporter constructs.
To test expression from m5/m8-peve in the Dl and neur mutant backgrounds, His2Av-RFP from His2av-RFP ; nos-MCP-GFP (BDSC #60340) was recombined with nos-MCP-GFP in the second chromosome (BDSC #63821) and combined with a deficiency encompasing the Dl gene (Df(3R)DlFX3, (Vassin et al. 1987)) or a neuralized loss of function allele (neur[11], BDSC #2747). m5/m8-peve-MS2-lacZ was also combined with the Dl and neur alleles and mutant embryos were obtained from the cross His2Av-RFP,nos-MCP-GFP ; mut / TTG x m5/m8-peve-MS2-lacZ ; mut / TTG. Homozygous mutant embryos for Dl or neur were selected by the lack of expression from the TTG balancer (TM3-twi-GFP).
To observe transcripion from two MS2 reporters in each cell (Fig. 2, S2) His2Av-RFP (BDSC #23650) was recombined with nos-MCP-GFP (from BDSC #60340) in the third chromosome and combined with m5/m8-peve or sim-peve MS2 reporters. m5/m8-peve x2 embryos and sim-peve x2 embryos were obtained from the stocks m5/m8-peve-MS2-LacZ ; His2Av-RFP,nos-MCP-GFP and sim-peve-MS2-LacZ ; His2Av-RFP,nos-MCP-GFP, respectively; while m5/m8-peve + sim-peve embryos were obtained from crosssing sim-peve-MS2-LacZ ; His2Av-RFP,nos-MCP-GFP females with m5/m8-peve-MS2-LacZ males.
To observe transcription from MS2 reporters in conditions of ectopic Notch activity the FRT-STOP-FRT cassette had to be first removed from the eve2-NICD construct by expression of a flippase in the germline. To do so flies containing ovo-FLP (BDSC #8727), His2Av-RFP and nos-MCP-GFP were crossed with others containing eve2-FRT-STOP-FRT-NICD, His2Av-RFP and nos-MCP-GFP. The offspring of this cross (ovo-FLP/+ ; eve2-FRT-STOP-FRT-NICD/+ ; His2Av-RFP, nos-MCP-GFP) induced FRT removal in the germline and were crossed with the MS2 reporters to obtain embryos expressing ectopic NICD. We note that only half of the embryos present the eve2-NICD chromosome, which could be distinguished by ectopic MS2 activity and an ectopic cell division of all the cells in the eve2 stripe after gastrulation. The other 50% embryos obtained from this cross were used as the wild type controls. This strategy was used to observe transcription from m5/m8-peve, sim-peve, m8NE-peve, m5/m8-pm5, simSPS-peve, m5/m8insSPS-peve, m5/m8Δtw% -peve, m5/m8Δdl-peve and m5/m8Δtwl Δdl-peve. To measure transcription from 2xeve2-NICD (Fig. 4, S4) removal of the FRT-STOP-FRT cassete was induced from the male germline to avoid recombination. To do so, betaTub85D-FLP (BDSC #7196) females were crossed with 2xeve2-NICD males and the male offspring of this cross (betaTub85D-FLP/Y ; 2xeve2-NICD/+), which induces FRT removal in the germline, were crossed with m5/m8-peve-MS2-lacZ ; His2AvRFP, nos-MCP-GFP or sim-peve-MS2-lacZ ; His2AvRFP, nos-MCP-GFP females. As in the previous strategy, only half of the embryos presented the 2xeve2-NICD chromosome and were distinguished by the ectopic activity.
Live imaging
Embryos were dechorionated in bleach and mounted in Voltalef medium (Samaro) between a semi-permeable membrane and a coverslip. The ventral side of the embryo was facing the coverslip in all movies except when looking at transcription in the DE region (Fig. 3B, S4AC), in which they were mounted laterally. Movies were acquired in a Leica SP8 confocal using a 40x apochromatic 1.3 objective and the same settings for MCP-GFP detection: 40mW 488nm argon laser detected with a PMT detector, pinhole airy=4. Other settings were slightly different depending on the experiment. To observe transcription in the whole embryo (Fig. 1) settings were: 3% 561nm laser, 0.75x zoom, 800 × 400 pixels resolution (0.48um/pixel), 19 1um stacks, final temporal resolution of 10 seconds/frame). To observe transcription from 2 MS2 alleles simultaneously (Fig. 2) settings were: 2% 561nm laser, 1.5x zoom, 800 × 400 pixels resolution (0.24um/pixel), 29 1um stacks, final temporal resolution of 15s/frame). In all experiments with ectopic NICD a ~150 × 150um window anterior to the center of the embryo was captured. Settings were: 2% 561nm laser, 2x zoom, 400 × 400 pixels resolution (0.36um/pixel), 29 1um stacks, final temporal resolution of 15s/frame). All images were collected at 400Hz scanning speed in 12 bits.
Image analysis
Movies were analyzed using custom Matlab (Matlab R2018a, Mathworks) scripts (available at GitHub:FryEmbryo3DTrackin; Briefly, the His2Av-RFP signal was used to segment and track the nuclei in 3D. Each 3D stack was first filtered using a median filter, increasing the contrast based on the profile of each frame to account for bleaching and a fourier transform log filter (Garcia et al. 2013). Segmentation was performed by applying a fixed intensity threshold, 3D watershed accounting for anisotropic voxel sizes (Mishchenko 2015) to split merged nuclei and thickening each segmented object. Nuclei were then tracked by finding the nearest object in the previous 2 frames which was closer than 6 um. If no object was found, that nuclei was kept with a new label, and only one new object was allowed to be tracked to an existing one. After tracking, the 3D shape of each nucleus in each frame was used to measure the maximum fluorescence value in the GFP channel, which was used as a proxy of the spot fluorescence. We note than when a spot cannot be detected by eye this method detects only background, but the signal:background ratio is high enough that the subsequent analysis allows to classify confidently when the maximum value is really representing a spot.
In experiments with two MS2 reporters the maximum intensity pixel per nucleus does not allow to separate transcription from the two alleles. To do so, the 3D Gaussian spot detection method from(Garcia et al. 2013) was implemented in the existing tracking, such that each spot was segmented independently and associated with the overlapping nuclei. In this manner only active transcription periods were detected and no further processing of the traces was required.
MS2 data processing
From the previous step we obtained the fluorescent trace of each nuclei over time. Only nuclei tracked for more than 10 frames were kept. First nuclei were classified as inactive or active. To do so the average of all nuclei (active and inactive) was calculated over time and fitted to a straight line. A median filter of 3 was applied to each nuclei over time to smooth the trace and ON periods were considered when fluorescent values were 1.2 times the baseline at each time point. This produced an initial classification of active (nuclei ON for at least 5 frames) and inactive. Using these inactive nuclei, the mean fluorescence from MCP-GFP was fitted again to redefine the baseline and active:inactive nuclei were classified again. Nuclei were then classified as MSE or earlier stages and the MSE ones were kept for further analysis.
The final fluoresncent values for each nuclei were calculated by removing the fitted baseline from the maximum intensity value for each and normalizing for the percentage that the MCP-GFP fluorescence in inactive nuclei decreases over time to account for the loss of fluorescence due to bleaching.
In all movies time into nc14 was considered from the end of the 13th syncythial division. When this was not captured they were synchronized by the gastrulation time.
Each nucleus was classified into the 4 regions (ME, MSE, NE and DE) by drawing rectangular shapes in a single frame and finding which centroids overlapped with each region. In eve2-NICD these regions along the DV axis were defined within the eve2 stripe (~ 6-7 cells wide in all movies). In wild type embryos ME and MSE regions were drawn in the whole field of view (~ 150×150 um anterior half of the embryo).
Definition of bursting properties
Bursts were defined as periods were the median filtered signal was higher than 1.2 times the baseline for at least 5 frames after the initial burst of activity at the beginning of nc14 (the considered period started at 15 min into nc14). These defined the burst duration and the time off between bursts. The amplitude was defined as the mean value within each burst period.
Onsets and ends of transcription were defined as the beginning of the first burst and the end of the last respectively (also starting at 15 min into nc14). In Fig. 2 to be more precise in measuring the onsets and end-points of transcription for both MS2 alleles they were scored manually as the first and last frame a spot is detected and randomly assigned ‘allele 1’ or ‘allele 2’. The total variability was the variance of all onsets or end points, combining both alleles. The extrinsic variability was calculated as the covariance of onsets and ends between alleles 1 and 2. The remaining (total - covariance) corresponds to the intrinsic variability within each cell.
Author contributions
JFS and SJB planned the experiments; JFS conducted the experiments; JFS,NL,HG developed the computational modelling and analysis; JFS, SJB wrote the manuscript; NL,HG edited the manuscript.
Acknowledgments
We thank members of the Bray Lab and of the Notch community for helpful discussions and Bill Harris and Maria J. Gomez-Lamarca for comments on the manuscript. Thanks to the Sanson, Small and St Johnston, labs for providing flies and plasmids and to Kat Millen and the Genetics Fly Facility for injections. This work was supported by a Programme grant from the Medical Research Council to SJB and by a PhD studentship to JFS from the Wellcome Trust (109144/Z/15/Z). HGG was supported by the Burroughs Wellcome Fund Career Award at the Scientific Interface, the Sloan Research Foundation, the Human Frontiers Science Program, the Searle Scholars Program, the Shurl & Kay Curci Foundation, the Hellman Foundation, the NIH Director’s New Innovator Award (DP2 OD024541-01), and an NSF CAREER Award (1652236). NL was supported by NIH Genomics and Computational Biology training grant 5T32HG000047-18. We also want to thank the Physical Biology of the Cell Course at the Marine Biological Laboratory (Woods Hole, MA), where the modelling approached used in this work developed. The authors declare no competing interests