Dynamic modulation of enhancer responsiveness by core promoter elements in living Drosophila embryos

Abstract Regulatory interactions between enhancers and core promoters are fundamental for the temporal and spatial specificity of gene expression in development. The central role of core promoters is to initiate productive transcription in response to enhancer's activation cues. However, it has not been systematically assessed how individual core promoter elements affect the induction of transcriptional bursting by enhancers. Here, we provide evidence that each core promoter element differentially modulates functional parameters of transcriptional bursting in developing Drosophila embryos. Quantitative live imaging analysis revealed that the timing and the continuity of burst induction are common regulatory steps on which core promoter elements impact. We further show that the upstream TATA also affects the burst amplitude. On the other hand, Inr, MTE and DPE mainly contribute to the regulation of the burst frequency. Genome editing analysis of the pair-rule gene fushi tarazu revealed that the endogenous TATA and DPE are both essential for its correct expression and function during the establishment of body segments in early embryos. We suggest that core promoter elements serve as a key regulatory module in converting enhancer activity into transcription dynamics during animal development.


INTRODUCTION
Communication between enhancers and core promoters is critical for the temporal and spatial specificity of gene expression. Enhancers are distal regulatory elements that contain a cluster of binding sites for sequence-specific transcription factors and co-activators, which usually span several hundreds to thousands base pairs (bp) in length. Recent quantitative imaging studies reported that enhancers are responsible for driving transcriptional bursting from their target genes (e.g. [1][2][3]. However, little is known about the role of core promoters in the process. Core promoters are short segments of DNA, which typically range −40 to +40 relative to the +1 transcription start site (TSS). They serve as a docking site for general transcription factors and RNA polymerase II (Pol II) for the assembly of the pre-initiation complex. Importantly, core promoters alone are not sufficient to drive a high level of transcription, but they typically require enhancer's activation cues to facilitate the assembly of active transcription machineries. Thus, core promoters act as a 'gateway to transcription', converting enhancer activity into gene expression (reviewed in 4).
Core promoters consist of several sequence motifs located at fixed positions relative to the TSS. The TATA box (TATA) (5) is typically located 25-30 bp upstream of the TSS and directly interacts with TATA-binding protein (TBP), a subunit of the transcription factor IID (TFIID) complex. Initiator (Inr), motif 10 element (MTE) and downstream promoter element (DPE) are sequence motifs downstream of TATA (6)(7)(8) and serve as an additional docking site for TFIID via direct interaction with TBP-associated factors (TAFs) (9)(10)(11)(12)(13). In addition, core promoters often contain binding-sites for sequence-specific DNA binding proteins such as Zelda and GAGA factor (GAF) in Drosophila (14)(15)(16). The precise composition of core promoter elements varies substantially among Pol IItranscribed genes (17)(18)(19), which is thought to play a key role in determining the level of gene expression by changing the responsiveness to enhancers (20)(21)(22)(23). We previously reported that swapping of entire core promoter sequence can largely alter bursting activities (2). However, it remains to be determined how individual core promoter elements modulate temporal dynamics of gene transcription in living multicellular organisms.
Importantly, quantitative measurement of enhancer responsiveness has been technically challenging with traditional bulk approaches because transcriptional output is controlled not only by core promoter sequences, but also by combinatory effects of surrounding regulatory landscapes such as enhancer strength and chromosome topology (reviewed in 24). Moreover, there is no universal core promoter architecture, making it difficult to compare expression profiles of different genes. Here, we developed a liveimaging system that permits unambiguous comparison of the roles of individual elements using a standardized, optimized synthetic core promoter placed under the control of a fixed enhancer. We have systematically analyzed a series of newly produced Drosophila MS2 strains that contain a variety of core promoter modifications at different positions. Quantitative image analysis revealed that each core promoter element differentially modulates the functional parameters of transcriptional bursting in developing embryos. Our data suggest that the timing and the continuity of burst induction are common regulatory steps on which core promoter elements impact. We also show that core promoters use the upstream TATA to increase the amplitude of transcriptional bursting upon activation by distal enhancers. On the other hand, Inr, MTE and DPE mainly contribute to the regulation of the bursting frequency. In addition, promoter-proximal Zelda sites were found to facilitate burst induction at the beginning of new cell cycle, adding another layer of complexity to this process. To further dissect the function of core promoter elements in the context of endogenous genome, we combined genome editing and sitedirected transgenesis. Using the pair-rule gene fushi tarazu (ftz), we show that both TATA and DPE mutations dramatically change transcription dynamics throughout its expression domains, resulting in disrupted spatial patterning of gene expression and misregulation of the downstream segment polarity gene engrailed (en). We therefore suggest that core promoter elements serve as a key regulatory module in converting enhancer activity into transcription dynamics during animal development.

Plasmids
Plasmid construction is detailed in the supplementary materials.

Site-specific transgenesis by phiC31 system
All reporter plasmids were integrated into a unique landing site on the third chromosome using VK00033 strain (26). PhiC31 was maternally provided using vas-phiC31 strain (27). Microinjection was performed as previously described (28). In brief, 0-1 h embryos were collected and dechorionated with bleach. Aligned embryos were dried with silica gel for ∼7 min and covered with FL-100-1000CS silicone oil (Shin-Etsu Silicone). Subsequently, microinjection was performed using FemtoJet (Eppendorf) and DM IL LED inverted microscope (Leica) equipped with M-152 Micromanipulator (Narishige). Injection mixture typically contains ∼500 ng/l plasmid DNA, 5 mM KCl, 0.1 mM phosphate buffer, pH 6.8. mini-white marker was used for screening.

cDNA synthesis
Total RNA was extracted from 40 adults of Oregon-R using TRIzol reagent (Thermo Fisher) followed by chloroform purification and isopropanol precipitation. Three g of total RNA was subjected to reverse transcription using PrimeScript 1st strand cDNA Synthesis Kit (Takara).

Fluorescence in situ hybridization
Embryos were dechorionated and fixed in fixation buffer (1 ml of 5× PBS, 4 ml of 37% formaldehyde and 5 ml of Heptane) for ∼25 min at room temperature. Vitelline membrane was then removed by shaking embryos in a biphasic mixture of heptane and methanol for ∼1 min. Antisense RNA probes labeled with digoxigenin and biotin were used. Hybridization was performed at 55 • C for overnight in hybridization buffer (50% formamide, 5× SSC, 50 g/ml Heparin, 100 g/ml salmon sperm DNA, 0.1% Tween-20). Subsequently, embryos were washed with hybridization buffer at 55 • C and incubated with Western Blocking Reagent (Roche) at room temperature for 1 h. Then, embryos were incubated with sheep anti-digoxigenin (Roche) and mouse anti-biotin primary antibodies (Invitrogen) at 4 • C for overnight, followed by incubation with Alexa Fluor 555 donkey anti-sheep (Invitrogen) and Alexa Flour 488 goat anti-mouse (Invitrogen) fluorescent secondary antibodies at room temperature for 1 h. DNA was stained with DAPI, and embryos were mounted in ProLong Gold Antifade Mountant (Thermo Fisher). Imaging was performed on a Zeiss LSM 900 confocal microscope. Plan-Apochromat 20×/0.8 N.A. objective was used. Images were captured in 16-bit. Maximum projections were obtained for all z-sections, and resulting images were shown. Brightness of images was linearly adjusted using Fiji (https://fiji.sc).

MS2 live-imaging
Virgin females of nanos >MCP-GFP, His2Av-mRFP/CyO (25) were mated with males carrying the MS2 allele. The re-sulting embryos were dechorionated and mounted between a polyethylene membrane (Ube Film) and a coverslip (18 mm × 18 mm), and embedded in FL-100-450CS (Shin-Etsu Silicone). Embryos were imaged using a Zeiss LSM 800 (Figures 1 and 4 S3, S6, S8 and S10, a stack of 20 images separated by 0.63 m was acquired at each time point, and the final time resolution is 12.8 s/frame. Images were captured in 16-bit. Images were typically taken from the end of nc13 to the onset of gastrulation at nc14. During imaging, data acquisition was occasionally stopped for a few seconds to correct zposition, and data were concatenated afterwards. For each cross, three biological replicates were taken. The same laser power and microscope setting were used for each set of experiments.

Image analysis
All the image processing methods and analysis were implemented in MATLAB (R2019a, R2019b MathWorks).

Segmentation of nuclei
For each time point, maximum projection was obtained for all z-sections per image. His2Av-mRFP was used to segment nuclei. 512 × 512 maximum projection images were initially cropped into 430 × 430 (Figures 1, 4, 7) or 300 × 430 (Figures 2, 3, 5, Supplementary Figures S1, S3, S8 and S10) to remove nuclei at the edge, and used for subsequent analysis. In the analysis of rhoNEE reporters (Supplementary Figures S5 and S6), 512 × 512 maximum projection images were initially cropped into 300 × 500 to remove nuclei outside of the expression domain. For nuclei segmentation, we used two different methods. In the first method, His2Av images were pre-processed with Gaussian filtering, top-hap filtering, and adaptive histogram equalization, in order to enhance the signal-to-noise contrast. Processed images were converted into binary images using a threshold value obtained from Otsu's method. Nuclei were then watershedded to further separate and distinguish from neighboring nuclei. Subsequently, binary images were manually corrected using Fiji (https://fiji.sc). In the second method, His2Av images were first blurred with Gaussian filter to generate smooth images. Pixels expressing intensity higher than 5% of the global maxima in the histogram of His2Av channel were removed. Processed images were converted into binary images using a custom threshold-adaptative segmentation algorithm. Threshold values were determined at each time frame by taking account of (i) the histogram distribution of His2Av channel and (ii) the number and the size of resulting connected components. Boundaries of components were then modified to locate MS2 transcription dots inside of nearest nuclei. In brief, pixels with intensity twice larger than mean intensity of MS2 channel were considered as transcription dots, and new binary images were cre- ated for each time frame. The Euclidean distances between the centroid of binarized transcription dot and all boundaries of segmented nuclei were calculated. Boundary of the nucleus with the smallest Euclidean distance was modified in order to capture transcription dot within a nucleus. Centroids of connected components in nuclei segmentation channel were used to compute the Voronoi cells of the image. Resulting binary images were manually corrected by using Fiji (https://fiji.sc).

Tracking of nuclei
Nuclei tracking was done by finding the object with minimal movement across the frames of interest. For each nu-cleus in a given frame, the Euclidean distances between the centroids of the nucleus in the current time frame and the nuclei in the next or previous time frame were determined. The nucleus with the minimum Euclidean distance was considered as the same lineage.

Recording of MS2 signals
3D raw images with all z-sections of MCP-GFP channel were used to record MS2 fluorescence signals. Using segmented regions from max projected images of His2Av-mRFP channel, fluorescence intensities within each nucleus were extracted. 3D fluorescence values were assigned to the nearest segmented regions of projected images. Signals of MS2 transcription dots were determined by calculating an integral of fluorescence intensities around the brightest pixel within each nucleus using a 2D Gaussian fitting method as described below. (i) The xyz position of transcription site was determined as a brightest pixel in each nucleus. (ii) A 2D Gaussian fitting was performed in a 11 × 11 pixels region with a single z-plane centering the transcription site to estimate a fluorescent dot intensity and local background. Fitting was performed with the following formula where α is the local background intensity, I 0 is the amplitude of the peak fluorescence intensity, x 0 and y 0 are the center of the peak, σ x and σ y are the spreads of the fluorescent dot. When the size of estimated fluorescent dot was larger than a fitted region of 11 × 11 pixels due to a lowquality fitting, whole time-point data of corresponding nucleus was excluded from further analysis. (iii) The intensity of MS2 transcription dot was calculated as 2πσ x σ y I 0 from fitting parameters as an estimated integral value after subtracting the local background (31). Subsequently, minimum MS2 intensities were determined for individual trajectories and subtracted to make the baseline zero.

Detection of transcriptional bursting
A transcriptional burst was defined as a local change in fluorescence intensity. First, signal trajectories were smoothed by averaging within a window of 5 timeframes. When a nucleus had above-threshold transcription activity, burst was considered to be started. Burst was considered to be ended when the intensity dropped below 55% of the local peak value of each burst. Location of defined burst was then  moved 2 timeframes afterwards to better capture the center of individual bursting event. When the burst duration is less than 5 timeframes, it was considered as a false-positive derived from detection noise. It was confirmed that this filtering method does not affect quantification of the burst duration (Supplementary Figure S4E and F). When signal trace exhibited continuous decreasing at the beginning of burst detection, it was also not considered as a burst. Same method and threshold value were used for each set of experiments.

Description of bursting properties
From each trajectory, number of bursts, amplitude and duration of each burst, and total integrated signal (output) produced by each nucleus were measured. To determine amplitude, the peak value during the burst was measured using trajectories after smoothing by averaging within a window of 5 timeframes. The duration was determined by measuring the length of each burst. Total output was measured by taking the area under the raw trajectory. The amplitude and duration for each nucleus were determined by taking average of all analyzed bursts in a single nucleus. The rate of burst induction was determined as a frequency divided by a length of time after the initial burst for each active nucleus.

Fraction of instantaneously and cumulative active nuclei
For each time frame, nuclei with MS2 intensity above the threshold were considered as active. Threshold was deter-  mined for each group of nuclei by calculating the 10% of the maximum MS2 intensity across all trajectories.

Mean MS2 intensity per actively transcribing nucleus
MS2 intensities in instantaneously active nuclei at each time point were averaged.

Autocorrelation analysis of MS2 signal
Autocorrelation values were used to estimate a periodicity of each MS2 trajectory. First, MS2 trajectories were smoothed as described in the previous section. To extract periodicity, the first derivative of smoothed MS2 signal was calculated to exclude general trends of signal changes during nc14. The autocorrelation value was calculated with 16.8 s timestep lags according to the time resolution of live-imaging data (16.8 s/frame). All autocorrelation values were normalized to 1 at the smallest lag time. Random data was generated from time-shuffled MS2 trajectories of the WT reporter by using MATLAB rand function.

Computational reconstitution of ftz expression
Newly synthesized ftz mRNAs were considered to be linearly degraded with a half-life of 7 min according to previous measurements in early embryos (32). Amount of ftz mRNA remained to be undegraded by the end of analysis was estimated using live-imaging data in Figure 7. Using segmentation mask, individual nuclei were false-colored with the pixel intensity proportional to the level of intact ftz mRNA in a given nucleus. Resulting image was then colored and layered over the maximum projected image of His2Av-mRFP.

Inr, MTE and DPE modulate bursting frequency
To quantitatively visualize how each core promoter element regulates enhancer responsiveness in living Drosophila embryos, we employed the MS2/MCP live-imaging method (33,34). We constructed a reporter system in which the yellow gene was placed under the control of the Drosophila synthetic core promoter (DSCP) (35), a modified even-skipped (eve) core promoter containing TATA, Inr, MTE and DPE at optimal positions relative to the TSS ( Figure (36,37). It has been previously shown that the sna shadow enhancer, but not the proximal primary enhancer, is essential for the mesoderm invagination during gastrulation (36,37). Thus, the use of the sna shadow enhancer provides a nice model for visualizing transcriptional regulation by developmentally essential enhancers. In this synthetic locus, sna shadow enhancer is located ∼6.5 kb away from the TSS, which is similar to the enhancer-promoter distance found at the endogenous sna locus (∼7 kb). Transgenes were integrated into a specific genomic landing site via phiC31-mediated transgenesis (26,38). Importantly, expression of the reporter gene was almost completely abolished upon deletion of enhancer sequence from the synthetic locus (Supplementary Figure S1B-E), indicating that the induction of transcription from DSCP is mediated by a linked enhancer. First, we examined the core promoter elements Inr, MTE and DPE by introducing mutations that compromise their direct interaction with TFIID ( Figure 1C) (6,7,9,13,39). Notably, the mutation at the +18 to +22 positions eliminates MTE-dependent transcription without affecting DPE function (7,13), allowing us to examine the individual contributions of MTE and DPE in the same core promoter context. Transcription activity was monitored from the entry into nuclear cycle 14 (nc14), when the major wave of zygotic genome activation starts to take place. To unambiguously compare the activities of different reporter genes at the same dorsal-ventral (DV) position, we focused on the ventralmost nuclei, as defined by the location of ventral furrow formation at the onset of gastrulation ( Supplementary Figure S2A). MS2 signal intensity and background fluorescence were determined by performing a 2D Gaussian fit at the z-plane corresponding to the highest intensity value for each time point (Supplementary Figure S2B and C). Quantitative image analysis revealed that the sna shadow enhancer induces fewer number of transcriptional bursts from the MTE mutant (mMTE) than the WT ( Figure 1D, Supplementary Movie S1). In comparison, even less frequent bursts were observed when the Inr mutant (mInr) and the DPE mutant (mDPE) were linked to the enhancer ( Figure  1D, Supplementary Movie S1). This trend was clearly seen when MS2 trajectories in all analyzed nuclei were visualized as a heatmap ( Figure 1E). These data suggest that each of these elements individually contributes to the regulation of transcriptional bursting, with Inr/DPE playing a major role and MTE playing a more supplemental role. Similar reduction of burst induction was also seen when enhancerpromoter communication was attenuated by the insertion of gypsy insulator (40) (Supplementary Figure S3), supporting the idea that inefficient TFIID recruitment to the core promoter region reduces responsiveness to enhancers. We then analyzed individual bursting events and quantified their functional parameters, including amplitude, duration, and frequency (Supplementary Figure S2D). As seen in individual MS2 trajectories ( Figure 1D), the frequency of transcriptional bursting and the total output were largely diminished when each of these elements was mutated (Figure 1F and G, Supplementary Figure S4A). On the other hand, the amplitude and the duration remained to be less affected ( Figure 1F, Supplementary Figure S4E and F), although Inr and MTE mutations caused moderate yet significant reduction in these parameters ( Figure 1F). It should be also noted that DPE mutation increased variance of the duration and the amplitude within a population ( Figure  1F), implicating that the direct interaction between TFIID and DPE helps to ensure production of homogeneous transcriptional bursts. Importantly, all core promoter variants exhibited significant delay in the onset of the first round of transcription ( Figure 1H), and substantial fraction of nuclei remained inactive during the analysis even though the sna shadow enhancer itself is in an active state as evidenced by the profile of the WT reporter (Supplementary Figure S4B). We next determined the rate of burst induction as a number of bursts divided by a time length after the onset of first burst for each active nucleus. There was an overall reduction in the efficiency of producing subsequent bursts after the first round of burst in all core promoter variants (Figure 1I). These data suggest that reduced bursting frequency is attributed to (i) the initial delay of first burst and (ii) the inefficient induction of subsequent bursts in response to activation cues from distal enhancers. Essentially same results were observed when the sna shadow enhancer was replaced with another key developmental enhancer, rhomboid neuroectoderm element (rhoNEE) (41) (Supplementary Figure  S5). Furthermore, reduction of the bursting frequency was also seen when DPE was mutated from the minimal rho core promoter linked to its cognate rhoNEE enhancer placed in a same distance as in the endogenous locus (Supplementary Figure S6). Overall, our data suggest that Inr, MTE and DPE support rapid and consecutive induction of transcriptional bursting, thereby facilitating the production of a high level of nascent transcripts during early development. Lastly, to obtain kinetic information underlying burst regulation, we examined the distribution of ON and OFF duration of individual events (Supplementary Figure S7). We found that OFF duration between two consecutive bursts nicely fits to a single exponential distribution, suggesting that burst initiation is driven by a single rate-limiting step. On the other hand, ON duration of individual bursts did not fit well to an exponential distribution, implicating that multi-step kinetics are involved to turn the promoter off.

TATA modulates bursting frequency and amplitude
We next examined the role of the upstream TATA, which we mutated according to previous characterization in order to abrogate its function ( Figure 2A) (7,21). Live-imaging analysis revealed that the TATA mutation causes more profound changes in overall transcription activity comparing to other modifications ( Figure 2B-E, Supplementary Figure S4C, Supplementary Movie S2). Importantly, the sna shadow enhancer could only induce bursts with substantially lower amplitude from the TATA mutant (mTATA) ( Figure 2D), suggesting that TATA mutation largely diminishes the number of Pol II released per burst. These results are consistent with previous studies showing that TATA affects the burst size and gene expression noise in yeast and mammalian systems (3,(42)(43)(44)(45)(46)(47)(48). Burst duration was largely unaffected by the loss of TATA ( Figure 2D, Supplementary Figure S4F), indicating that the size of each burst is mainly determined by the amplitude, but not the duration, of individual activation events. There also was a clear reduction in the frequency of transcriptional bursting when TATA was mutated ( Figure 2E). As seen for other modifications (Figure 1H and I), reduced frequency appears to be attributed to delayed and inefficient burst induction ( Figure 2F and G, Supplementary Figure S4D). Strong TATA-dependency was also observed when DSCP was placed under the control of the rhoNEE (Supplementary Figure S5) or the minimal sna and rho core promoters were linked to their cognate en-  Figures S6 and S8). We therefore suggest that core promoters use the upstream TATA to ensure rapid and consecutive induction of strong bursts upon activation by enhancers.
We then tested if the conversion of a natural TATA-less core promoter into a TATA-containing core promoter has an opposite effect. It has been previously shown that the core promoter regions of the Drosophila homeotic (Hox) genes are typically depleted of TATA (49). Indeed, the core promoter of the Hox gene labial (lab) lacks an optimal TATA motif ( Figure 3A and B). When the lab core promoter was placed under the control of the sna shadow enhancer, only weak and infrequent bursts were produced ( Figure 3C and D). We then introduced two nucleotides substitution at the −28 and −30 positions to covert 'TCT-GAAA' to an optimal TATA motif, 'TATAAAA' ( Figure  3B). Intriguingly, this minimal modification dramatically increased overall activities including the amplitude and the total output of transcriptional bursting ( Figure 3C-F, Supplementary Movie S3). In addition, the timing of first burst ( Figure 3G, Supplementary Figure S9A) and the continuity of subsequent bursts ( Figure 3H) were also augmented, resulting in a higher frequency of transcriptional bursting ( Figure 3F). These results support the idea that the upstream TATA helps to facilitate consecutive induction of strong transcriptional bursting. Burst duration was only moderately impacted by this modification (Figure 3E), implicating that TATA does not largely alter the stability of active transcription machineries at the core promoter region. We next examined if enhancers regulating the expression of natural TATA-less genes also exhibit strong TATA-dependency. To this end, a well-characterized IAB5 enhancer that regulates the expression of the TATA-less Hox gene Abdominal-B (Abd-B) was linked to the mTATA DSCP (50) (Supplementary Figure S10A). Similar to the sna shadow and rhoNEE enhancers (Figure 2, Supplementary Figure S5), there was a clear reduction in the amplitude and the frequency of transcriptional bursting (Supplementary Figure S10B-G), suggesting that IAB5 enhancer can also utilize TATA to facilitate induction of strong bursts. On the other hand, DPE mutation was found to preferentially impact the bursting frequency (Supplementary Figure  S10B-G). These data support the idea that TATA and DPE can differentially modulate bursting profiles under various genomic configurations.

Promoter-proximal Zelda sites facilitate burst induction
In Drosophila, core promoters often contain binding sites for sequence-specific DNA binding proteins, Zelda and GAF (15,51), that are thought to recruit chromatin remodeling complexes such as NURF to increase chromatin accessibility of regulatory regions (52)(53)(54). However, it remains unclear how they modulate core promoter functions because previous studies have mainly focused on their roles at enhancer regions (e.g. 30,53,55,56). DSCP contains a GAFbinding site (GAGA) from the eve core promoter (Supplementary Figure S1A) (57), but lacks any known Zelda binding motifs. To examine how promoter-proximal GAGA and Zelda sites influence transcription dynamics, we either mutagenized GAGA site or added Zelda sites ( Figure 4A). We found that the GAGA mutant (mGAGA) exhibits only a moderate reduction in the bursting amplitude and duration ( Figure 4B-D). On the other hand, addition of Zelda sites led to rapid induction of strong bursts ( Figure 4B and C, Supplementary Figure S9B) and an overall increase in transcription activities ( Figure 4D and E, Supplementary Movie S4). Especially, the timing of initial burst was found to be dramatically accelerated in the presence of Zelda sites (Figure 4F, Supplementary Figure S9B), suggesting that Zelda helps to increase the enhancer responsiveness by opening up the promoter chromatin to allow subsequent TFIID recruitment during the onset of nc14. On the other hand, addition of Zelda sites only moderately increased the continuity of burst induction after the first round of transcription ( Figure 4G). Intriguingly, we noticed that periodic patterns of burst induction are present in the heatmap profile of the Zelda reporter gene ( Figure 4C). To examine if the addition of Zelda sites actually facilitates to drive periodic transcription, we detected repeating patterns in MS2 trajectories by performing autocorrelation analysis. Although the unmodified WT reporter also drives periodic bursts, our data showed that the Zelda reporter gene exhibits even stronger periodicity ( Figure 4H, Supplementary Figure S11). At this point, molecular mechanism underlying periodic bursts is unclear, but these data suggest that it typically takes ∼5-6 min for core promoters to be ready for responding to next activation cue from distal enhancers once Pol II is released for transcriptional elongation in this system.

Zelda can compensate weak activities of TATA-less core promoters
It is known that key segmentation genes such as hairy and paired lack TATA but contain promoter-proximal Zelda sites in Drosophila (58). To dissect the role of Zelda in the context of TATA-less core promoters, Zelda sites were added to the mTATA DSCP reporter ( Figure 5A). This modification led to partial increase in the amplitude and the duration ( Figure 5B-D, Supplementary Movie S5). In addition, this reporter gene exhibited substantial increase in the frequency of transcriptional bursting ( Figure 5E). Intriguingly, the onset of first burst from the Zelda/mTATA reporter became even earlier than the unmodified WT reporter ( Figure 5F), suggesting that Zelda can facilitate the recruitment of TFIID to the core promoter region even in the absence of TATA. The rate of burst induction also became ∼2-fold higher upon addition of Zelda sites ( Figure  5G). When the same modification was introduced to the WT DSCP, the rate of burst induction was only moderately increased ( Figure 4G), implicating that the burst induction rate was already saturated when all the other elements were present. Overall, our data suggest that the promoter opening by Zelda can compensate low enhancer responsiveness of TATA-less core promoters to facilitate rapid and consecutive induction of transcriptional bursting.

Core promoter modification at the endogenous ftz locus
To explore how core promoter elements impact enhancerpromoter communication at the endogenous locus, we next focused on one of the best studied developmental patterning genes, fushi tarazu (ftz) (59)(60)(61)(62)(63). ftz is expressed in seven transverse stripes spanning across the anterior-posterior (AP) axis and is regulated by multiple enhancers located 5 and 3 of the transcription unit (64)(65)(66)(67)(68). The ftz core promoter contains TATA, Inr and DPE elements ( Figure  6A) (49). Reporter assays in Drosophila S2 cultured cells have suggested that DPE, but not TATA, is specifically required for the activation of the ftz core promoter by the homeodomain-containing transcription factor Caudal (49). In this model, mutation of DPE is expected to most severely affect ftz expression in stripe 5 and 6 because they are regulated by Caudal-dependent enhancers (69). To test this idea, we developed a genome engineering approach for visualizing impacts of core promoter modification at the endogenous locus. First, the entire ftz transcription unit was replaced with attP site via CRISPR/Cas9-mediated genome editing (Supplementary Figure S12A). Subsequently, a fulllength ftz transcription unit containing the modified ftz core promoter and 24× MS2 sequence was integrated into the attP site via phiC31-mediated transgenesis ( Figure 6B, Supplementary Figure S12A). Fluorescent in situ hybridization assay revealed that ftz expression is diminished equally across all seven stripes upon DPE mutation ( Figure 6C), indicating that the ftz core promoter uses DPE for responding not only to Caudal-dependent stripe 5 and 6 enhancers but also to all the other stripe enhancers. Moreover, ftz expression was almost completely lost upon TATA mutation ( Figure 6C). As a consequence of irregular ftz patterning, expression of the downstream segment polarity gene engrailed (en) was specifically lost from even-numbered stripes in mTATA and mDPE embryos ( Figure 6D). Cuticle preparation analysis further revealed that the misexpression of ftz and en leads to loss of even-numbered body segments in developing embryos ( Figure 6E and F). We therefore suggest that both TATA and DPE are required for the correct expression and the function of ftz during Drosophila embryogenesis.
To elucidate the molecular basis underlying these phenotypes, we carried out live-imaging analysis of individual ftz-MS2 complementation alleles. We focused on anterior stripe 1/2 and posterior stripe 5/6 ( Figure 7A, Supplementary Movie S6) because they are known to be regulated by different enhancers located 5´and 3´of the gene (Supplementary Figure S12A) (64,68). While both the mTATA and mDPE ftz alleles failed to restore en expression and the formation of body segments ( Figure 6D and F), their MS2 profiles differ dramatically. The mDPE allele produced delayed discontinuous but strong bursting activity ( Figure 7B and C), whereas the mTATA allele further led to an overall diminishment of MS2 intensity in all analyzed stripes ( Figure  7B and C). As a consequence, the mTATA allele exhibited a ∼81-89% reduction in the total RNA production, while the mDPE allele showed a more modest reduction ( Figure  7D). We then characterized the functional parameters of MS2 trajectories. As we previously reported (25), individual bursting events were hard to be discerned due to the continuity of bursting activities ( Figure 7B), suggesting that the endogenous genomic configurations are highly optimized for efficient production of transcriptional bursting within a short period of time. To minimize ambiguity in defining individual bursting events, we quantified the instantaneous fraction of active nuclei ( Figure 7E) and the mean MS2 in-tensity in actively transcribing nuclei ( Figure 7F). We found that mutation of TATA and DPE both diminish the fraction of active nuclei ( Figure 7E) and also delay the onset of transcription (Supplementary Figure S12B and C). Importantly, the level of MS2 intensity was more severely reduced in the mTATA allele ( Figure 7F), suggesting that the TATA mutation reduces the number of Pol II entering into productive elongation even when core promoter in an active state. These profiles were similar to those of mDPE and mTATA DSCP linked to the sna shadow enhancer at the synthetic locus (Supplementary Figure S12D and E). We therefore suggest that DPE and TATA differentially regulate the enhancer responsiveness by changing the efficiency and the strength of burst induction. Lastly, to determine how misregulation of bursting activities affects the stripe formation, we computationally reconstituted the spatial patterning of ftz by calculating the total RNA production at each nucleus and the ftz mRNA turnover rate in early embryos (half-life: 7 min) (32). Although the mDPE allele exhibited somewhat stronger MS2 activities than the mTATA ( Figure 7B-F), both resulted in highly sporadic stripe patterns (Supplementary Figure S13). Thus, we concluded that the endogenous configuration of ftz core promoter is essential for the uniform expression at the stripe regions to ensure proper control of downstream genes such as en during embryo segmentation.

Molecular mechanism of burst regulation by core promoters
In this study, we provided evidence that each core promoter element differentially modulates the responsiveness to enhancers in early Drosophila embryos. Our data suggest that Inr, MTE and DPE mainly affect the frequency of transcriptional bursting by changing the timing of first burst and the continuity of subsequent bursts (Figure 1). On the other hand, TATA mutation dramatically diminished overall transcription activity including the bursting amplitude (Figures 2 and 3). Our genome engineering approach further revealed that the endogenous ftz core promoter requires both TATA and DPE to initiate rapid and productive transcription upon activation by stripe enhancers (Figures 6 and 7). We also show that each element can exert its function independently of enhancer-promoter distances or differential combinations of enhancer-promoter pairs (Supplementary Figures S5, S6, S8 and S10). Importantly, recent cryo-EM studies have revealed that TFIID recognizes the core promoter in a stepwise fashion (11,12,70), with initial contact with Inr, MTE and DPE through TAF1/7 and TAF2 subunits, followed by dynamic structural rearrangement that facilitates recognition of the upstream TATA by TBP. We speculate that the rate of initial TFIID loading onto the downstream core promoter region mainly affects the timing and the frequency, and the subsequent TBP loading onto the upstream TATA helps to further increase the bursting amplitude. Supporting this view, a recent biochemical study suggested that downstream core promoter interactions of TFIID increase the efficiency of transcription reinitiation in yeast (71). It was also reported that TATA and Inr can differentially modulate the burst size and the frequency during PD1 expression in primary B-cells (72),  T1  T3  A2  A4 A6   A8  T1  T3  T2  A8  A7  A6  A5  A4  A3  A2  A1   T1  T3  A2  A4 A6   A8  T1  T3 A2  A4 A6   A8  T1  T3  T2  A8  A7  A6  A5  A4  A3 A2 A1 although this study focused on their function as a negative regulator of transcription. More recently, it was reported that the loss of TATA or downstream elements alters entire conformation of TFIID by changing molecular spacing between TBP and TAFs (73,74). They further showed that core promoters undergo two distinct PIC assembly pathways depending on the initial conformation of the promoter-bound TFIID. Thus, dynamic structural rearrangements of TFIID during the assembly of PIC can also contribute to differential bursting profiles seen in this study.
As a limitation of current study, it is important to note that we engineered core promoter elements by introducing mutations that alternate entire sequences within individual motifs. Future functional studies are needed to fully elucidate the roles of core promoters by analyzing intermediates between the most optimized and dysregulated motifs in the context of various core promoter sequences. We believe that our study will serve as a critical starting point toward understanding of how temporal dynamics of gene expression is encoded in the eukaryotic genome.  Total RNA produced (AU)   quantile and the open circle indicates the median. Whiskers extend to the most extreme, non-outlier data points. A total of 234, 236 and 227 nuclei at stripe 1, 248, 241 and 242 nuclei at stripe 2, 243, 238 and 240 nuclei at stripe 5, and 232, 239 and 240 nuclei at stripe 6 were analyzed from three independent embryos for the ftz-MS2 with WT, mTATA and mDPE core promoter, respectively. Median values relative to the ftz-MS2 with WT core promoter are shown at the bottom. The P-values of two-sided Wilcoxon rank sum test are shown at the top. AU; arbitrary unit. (E) Instantaneous fraction of actively transcribing nuclei at each expression domain. A total of 234, 236 and 227 nuclei at stripe 1, 248, 241 and 242 nuclei at stripe 2, 243, 238 and 240 nuclei at stripe 5, and 232, 239 and 240 nuclei at stripe 6 were analyzed from three independent embryos for the ftz-MS2 with WT, mTATA and mDPE core promoter, respectively. (F) Mean MS2 intensity per actively transcribing nucleus at each expression domain. A total of 234, 236 and 227 nuclei at stripe 1, 248, 241 and 242 nuclei at stripe 2, 243, 238 and 240 nuclei at stripe 5, and 232, 239 and 240 nuclei at stripe 6 were analyzed from three independent embryos for the ftz-MS2 with WT, mTATA and mDPE core promoter, respectively. Shades represent the standard deviation of the mean across active nuclei at a given time.

The role of promoter-proximal Zelda sites in burst regulation
During early development of Drosophila embryos, a zincfinger DNA binding protein Zelda plays an essential role in zygotic genome activation (75). At the enhancer regions, it is well established that Zelda exerts its pioneering activity by lowering nucleosome barriers to assist subsequent recruitment of transcription factors and coactivators (53,55). Importantly, our data showed that the addition of Zelda sites at the core promoter region augments both the amplitude and the frequency of transcriptional bursting ( Figures  4 and 5). In analogy to Zelda's function at the enhancer regions, it is conceivable that the promoter-opening by Zelda contributes to the initial TFIID recruitment and the subsequent TBP loading at the core promoter region. It has been previously shown that Zelda is pre-loaded onto a thousand of promoter regions prior to zygotic genome activation in early Drosophila embryos (15). Thus, it appears that promoter-proximal Zelda sites increase the responsiveness to enhancers by making core promoters poised for activation even before distal enhancers start to drive transcription. We suggest that multi-layered mechanisms of Zelda and other elements help to diversify the enhancer responsiveness of core promoters across the genome because there are substantial variations in the composition of core promoter elements among Pol II-transcribed genes. For example, it is estimated that only a small fraction of protein coding genes contain TATA, both in human and Drosophila (17)(18)(19). Importantly, our data showed that the TATA-less core promoter of the Hox gene lab is naturally depleted of TATA to limit the level of total RNA production by reducing the amplitude of transcriptional bursting ( Figure 3). As suboptimal transcription factor binding sites are important for tissue-specific gene activation by enhancers (76,77), suboptimal core promoter architectures might play a key role in ensuring the spatial and temporal specificity of gene expression during animal development. Alternatively, it can be possible that a subset of enhancers is capable of driving strong bursts even when TATA or other elements are not present as suggested by previous enhancer trapping assay in Drosophila (21). In this regard, our data suggests that the promoter-proximal Zelda sites can help to compensate weak activities of TATA-less core promoters in the Drosophila genome ( Figure 5).

Drosophila TATA regulates minute-scale bursting activities
Regulation of transcriptional bursting by core promoter elements seems to be a common mechanism conserved across species. Analysis of actin family genes in Dictyostelium cells also reported that swapping of entire promoter sequence can alter profiles of transcriptional bursting (78). Previous studies showed that TATA mutation reduces the burst size and gene expression noise in yeast and mammals (3,(42)(43)(44)(45)(46)(47)(48). Consistently, our data also showed that TATA largely affects the amplitude of transcriptional bursting (Figures 2 and 3). However, there seems to be a clear difference in a time-scale of bursting activities that TATA impacts between species. In mouse and human cells, transcriptional bursts are often separated by refractory period of several hours (e.g. [79][80][81][82]. Importantly, live-imaging analysis of HIV-1 reporter gene suggested that the mutation of TATA specifically affects hour-scale, but not minutescale, transcriptional bursting in HeLa cells (47). Recent single-cell RNA-seq study also suggested that TATA affects hour-scale bursting activities in mouse and human cultured cells (3). In contrast, TATA clearly influenced minute-scale bursting profiles in our Drosophila system (Figures 2 and  3). Thus, it appears that Drosophila embryos use TATA in a different way from mammalian systems to regulate bursting activities at a finer time-scale, which may help to control dynamic ON/OFF patterns of gene expression during rapid processes of early embryogenesis.