Multi-level regulation of even-skipped stripes by the ubiquitous factor Zelda

The Drosophila zinc-finger protein Zelda (Zld) is a key regulator of hundreds of zygotic genes during the maternal-to-zygotic transition (MZT) (Liang et al., 2008). Zld binds to TAGteam sites (ten Bosch et al., 2006) in enhancers and promoters (Harrison et al., 2011; Nien et al., 2011; Struffi et al., 2011; Ling et al., 2019; McDaniel et al., 2019), locally increasing histone acetylation levels (Li et al., 2014; Schulz et al., 2015; Sun et al., 2015). Zld binding is thought to facilitate the binding of other sequence-specific factors (Foo et al., 2014; Xu et al., 2014; Mir et al., 2018), which then recruit and/or activate transcription by RNA Polymerase II (Dufourt et al., 2018; Yamada et al., 2019).

However, the effects of Zld on individual target genes can vary significantly. In embryos lacking Zld, some target genes are not activated at all, whereas others show reduced expression levels or delayed activation, as expected for a pioneer factor (Liang et al., 2008; Foo et al., 2014; Xu et al., 2014). In addition, the expression patterns of many target genes are spatially shifted in Zld-deficient embryos (Nien et al., 2011; Combs and Eisen, 2017). Some shifts are caused by the disruption of the normal positive effects of Zld on the binding of Bicoid (Bcd) and Dorsal (Dl), two activator proteins expressed in spatial gradients (Foo et al., 2014; Xu et al., 2014), but other mechanisms that shift patterns are unknown.

Here, we analyze the role of Zld on the pair-rule gene even-skipped (eve), which is expressed in a pattern of seven stripes along the anterior-posterior axis of the early embryo (Macdonald et al., 1986; Frasch et al., 1987). This striped pattern is regulated by five modular enhancers, each of which directs the expression of a single stripe or a pair of stripes (Goto et al., 1989; Harding et al., 1989; Fujioka et al., 1999). Each enhancer contains a unique combination of binding sites for broadly distributed activator proteins and localized repressors encoded by the gap genes (Small et al., 1991, 1992; Stanojevic et al., 1991; Arnosti et al., 1996; Struffi et al., 2011). Importantly, the gap genes themselves are patterned by mutual repression mechanisms that create space for the expression of individual eve stripes (Kraut and Levine, 1991; Wu et al., 1998; Clyde et al., 2003), and previous experiments (Nien et al., 2011) showed that removal of Zld expands the gap gene expression domains into regions normally occupied by the eve stripes.

Our results show that Zld is required for the efficient activation of all five eve enhancers when tested as individual reporter genes. We then focus on the Zld regulation of a 3.8 kb genomic region that contains two independent enhancers and the eve basal promoter (Harding et al., 1989). We show that Zld is required for efficient binding of the activator Bcd to this fragment, and we demonstrate that repressor binding is also abrogated in Zld-depleted embryos. Mutations of Zld sites in this larger fragment caused reductions in stripe intensity, indicating that Zld-binding contributes directly to stripe activation. However, reductions caused by Zld site mutations were consistently weaker than those caused by removing Zld activity in trans, suggesting that Zld also plays indirect roles. We show that one of the most important indirect roles of Zld is to spatially limit the initial expression domains of the gap genes, creating repressor-free gaps that permit the activation of the eve stripes. These results suggest that Zld controls the binding of both activators and repressors to the same cis-regulatory elements, and demonstrate how a single transcription factor can mediate direct and indirect activities that are crucial for the establishment of segmental patterns in Drosophila.

Previous studies have shown that eve mRNA expression is delayed for at least one nuclear cycle in embryos lacking Zld (Liang et al., 2008; Nien et al., 2011), and that its striped pattern is severely disrupted (Struffi et al., 2011). Previous chromatin immunoprecipitation sequencing (ChIP-seq) studies (Xu et al., 2014) have shown that Zld is required for the genome-wide binding of the maternal morphogen Bcd to hundreds of target genes, including eve (Fig. 1B; Fig. S1). Because Zld also binds broadly to the eve locus (Fig. 1B; Harrison et al., 2011), these data indicate that Zld-binding may directly recruit Bcd to specific eve enhancers and function as a pioneer in gene regulation.

To examine the roles of Zld in regulating individual stripes, we analyzed reporter genes containing individual enhancers in nuclear cycle (NC) 14 Zld-depleted embryos. Stripes driven by the upstream enhancers (eve2 and eve37; Fig. 1A) were completely abolished in embryos lacking Zld (Fig. S2B,D). Stripes driven by the downstream enhancers (eve1, 4, 5 and 6) were also substantially reduced in embryos lacking Zld. Specifically, a reporter containing the eve46 enhancer showed lower expression levels and a posterior expansion of stripe 6 (Fig. S2F), and a second reporter containing only the eve1 and eve5 enhancers showed a strong reduction of stripe 1 and a complete loss of stripe 5 expression (Fig. S2H).

If Zld functions as a general pioneer protein, removing it might affect the in vivo binding of repressors as well as activators. To test this, we performed ChIP-qPCR on Zld-depleted embryos using antibodies against Hb and Kni, which repress the eve37 enhancer, and Gt and Kr, which repress eve2 (Fig. S3; Arnosti et al., 1996; Struffi et al., 2011). These experiments showed that binding levels of both Hb and Kni to the eve37 enhancer were significantly reduced in embryos completely lacking Zld (Fig. 2C). Similar results were obtained for Gt and Kr on the eve2 enhancer (Fig. 2F).

These results support the hypothesis that Zld is a pioneer protein that binds directly to the eve locus and facilitates the binding of both activators and repressors, which together form stripes of expression in the early embryo. We tested this hypothesis by mutating Zld-binding motifs in a 3.8 kb genomic fragment that contains the eve37 enhancer, the eve2 enhancer and the eve basal promoter (Fig. 1C; Fig. S4). This fragment drives lacZ reporter expression in a broad diffuse domain at the position of eve stripe 2 in wild-type embryos starting in NC13 (Fig. 1D; Small et al., 1992). This early pattern is controlled by sequences that lie between the eve2 enhancer and the basal promoter (Oberstein et al., 2005) and is replaced by a pattern of three enhancer-dependent stripes (2, 3 and 7) in NC14 embryos (Fig. 1F). In Zld-depleted embryos, the early broad pattern is activated in NC13 (Fig. 1E), but it decays rapidly, and no stripes are detected during NC14 (Fig. 1G).

There are eight canonical Zld-binding motifs in the 3.8 kb eve fragment mentioned above (Fig. 1C; Fig. S4). One motif (TAGGTAG) lies within a 500 bp fragment that is sufficient for eve37 expression (Fig. S4; Small et al., 1996). To test whether this motif is required for eve37 activation, we mutated it alone in the 3.8 kb genomic fragment (construct Mut Z 37; Fig. 3D). We then quantitatively analyzed multiple embryos (Figs S5-S7), converting each stripe for each embryo into a Gaussian distribution, and compared each stripe of the mutated construct with the wild type using a two-tailed t-test (see Materials and Methods). The single mutation of the Zld site in the eve37 enhancer significantly reduced the expression of stripes 3 and 7 compared with the wild-type transgene (P<0.002 for stripe 3 and P<0.0002 for stripe 7; Fig. 3E,F, compare with Fig. 3B,C), but did not affect stripe 2 (P>0.6), consistent with previous results obtained by mutating this site in the minimal eve37 enhancer (Struffi et al., 2011).

In contrast, the minimal eve2 enhancer, which is also completely inactivated in Zld-depleted embryos (Fig. S2D), contains no canonical Zld-binding motifs. However, a previous study identified three variants of the motif in this enhancer, and fragments containing these variants bind Zld weakly in vitro (Fig. S4; Struffi et al., 2011). To test whether these variant motifs are important for eve2 activation, we mutated all three in the context of the 3.8 kb reporter gene (construct Mut Z 2; Fig. 3G). Compared with the wild-type transgene (Fig. 3A-C), this manipulation caused a small but significant reduction of stripe 2 (P<0.0002) without affecting the expression of stripe 3 (P>0.30; Fig. 3H,I), suggesting that these variant motifs are important for Zld activity. It also caused a reduction of stripe 7 (P<0.002), consistent with the previous demonstration that the sequences controlling this stripe are not limited to the minimal eve37 enhancer (Harding et al., 1989; Janssens et al., 2006).

Previous experiments and computational models showed that sequences adjacent to the minimal eve2 enhancer can enhance its in vivo activity (Small et al., 1992; Janssens et al., 2006; Crocker and Stern, 2017). These regions contain Zld motifs, with two canonical motifs within 100 bp of the 3′ end of the minimal fragment (Fig. S4). We mutated these two motifs in combination with the three variant motifs contained within the minimal enhancer (construct Mut Z 2+; Fig. 3J), which caused a stronger reduction of reporter gene expression at stripe 2 (P<2×10⁻⁶) and stripe 7 (P<0.0003), without significantly affecting stripe 3 (P>0.3; Fig. 3K,L).

We next mutated all eight canonical Zld motifs in the 3.8 kb eve-lacZ transgene along with the three variant sites in eve2 (construct Mut Z all; Fig. 3M), expecting more dramatic reductions in the expression of all three stripes. This experiment caused stronger reductions of stripes 2 (P<5×10⁻⁹) and 7 (P<2×10⁻⁶), but only a slight reduction of stripe 3 (P>0.06; Fig. 3N,O). The slight reduction of stripe 3 with the Mut Z all construct was unexpected because mutating the single Zld site in the eve37 enhancer alone caused a stronger reduction (Fig. 3E,F).

Importantly, all three stripes were clearly visible in embryos containing the Mut Z all construct, indicating that the mutations in all 11 Zld motifs did not completely abolish the ability of this construct to activate stripes in the early embryo. In addition, the weak stripes directed by the Mut Z all construct also showed refined boundaries, suggesting that the Zld motif mutations tested here do not affect the binding or activity of the repressors that form these boundaries. In direct contrast, removal of Zld in trans reduces the binding of these repressors to the enhancers in vivo (Fig. 2). One explanation for this discrepancy is that our mutations do not completely interfere with the ability of Zld to bind to this large eve fragment. Alternatively, it is possible that the reduced level of binding observed in embryos lacking Zld is still sufficient for eve stripe boundary formation. Finally, it is possible that Zld plays indirect roles in the complex mechanisms that generate striped patterns in the early embryo.

The gap genes that encode boundary-forming gradients for eve stripes 2, 3 and 7 (hb, gt, Kr and kni) are expressed in one or two broad domains in early Drosophila embryos (Fig. 4A,C,F,H). These patterns are refined by strong repressive interactions between hb and kni, and between gt and Kr (Kraut and Levine, 1991; Wu et al., 1998; Clyde et al., 2003; Yu and Small, 2008). In a previous study, Nien et al. (2011) showed that removal of Zld delays the initial activation of the gap genes by one or two nuclear division cycles, and spatially alters their expression patterns. The most dramatic pattern alterations are anterior expansions of posterior kni domain and the central Kr domain (Fig. 4), which are detected as soon as these genes are activated (Nien et al., 2011). These expansions are visible in NC13 and early NC14 embryos but are not maintained in later embryos (Figs S9, S11).

Previous studies have shown that the anterior boundary of the posterior kni domain is established by Hb-mediated repression (Wu et al., 2001; Yu and Small, 2008), whereas the anterior boundary of the central Kr domain is primarily set by Gt (Wu et al., 1998). Other cis-regulatory studies of kni and Kr identified enhancers that control their expression patterns in early embryos. For kni, the posterior domain that expands in Zld-depleted embryos is controlled by an 800 bp enhancer (KD) that contains at least five Hb-binding sites (Rivera-Pomar et al., 1995). To test whether Zld activity is required for Hb-binding to the KD enhancer, we performed ChIP-qPCR assays using an anti-Hb antibody, which showed significantly lower binding levels in Zld-depleted embryos than in wild type (Fig. 4E). For Kr, two separate enhancers (CD1 and CD2) direct expression of the central Kr domain in early embryos (Hoch et al., 1990), and Gt protein binds in vitro to multiple sites in the Kr CD1 and CD2 enhancers (Capovilla et al., 1992). We performed ChIP-PCR experiments with an anti-Gt antibody, which showed strong binding to the Kr CD1 enhancer in wild-type embryos that was significantly reduced in Zld-depleted embryos (Fig. 4J).

If expansions of the gap proteins into embryonic spaces normally occupied by the eve stripes contribute to the loss of stripe expression observed in embryos lacking Zld, then the concomitant removal of gap gene repression activity and Zld might result in the restoration of enhancer activation. We tested this for the eve37 enhancer, which is completely inactive in Zld-depleted embryos in its unmutated form (Fig. 5A-C). Specifically we examined expression patterns driven by eve37 enhancers containing mutations in Hb- and Kni-binding sites (eve37ΔH and eve 37ΔK; Fig. 5D,G). In wild-type embryos, these enhancers direct patterns that expand into regions normally occupied by the repressors (Fig. 5E,H; Struffi et al., 2011). However, when crossed into embryos lacking Zld, both mutant versions of the enhancer showed a restoration of reporter gene activation in these regions (Fig. 5F,I). Intriguingly, the mutated elements are not expressed in the ventral-most region of Zld-depleted embryos, suggesting that another repressor is preventing expression in that region (see Discussion). These results suggest strongly that the patterning defects in the gap repression system contribute significantly to the failure to activate the eve37 enhancer in the absence of Zld.

We performed a similar experiment with the eve2 enhancer, which is also completely inactive in Zld-depleted embryos (Fig. 5J-L). Previous DNAse I footprint studies showed that Gt binds to at least three regions of the eve2 enhancer (Small et al., 1991); deletions of these regions (12-44 bp) caused an anterior expansion of eve2-lacZ reporter gene expression (Small et al., 1992). To more precisely remove Gt function, we searched for Gt-binding motifs in the eve2 enhancer, but failed to find any close matches to the palindromic sequence bound by Gt in the bacterial one-hybrid system (see Materials and Methods; Noyes et al., 2008). However, there are ten shorter sequences that could represent half sites (Fig. S12), and eight of these do not overlap with sites for other known regulators of eve2. We mutated these eight potential half-sites (Fig. 5M; Fig. S12), which generated a diffuse expression pattern in the position normally occupied by Gt, as expected if these are bona fide Gt-binding sites (Fig. 5N). We next crossed the mutated enhancer into Zld-depleted embryos. In contrast to the results obtained with the eve37 enhancer, expression of the eve2ΔG enhancer was only barely detectable under the conditions of the experiment (Fig. 5O). This suggests that the direct effects of Zld on eve2 are stronger than its effects on eve37, although there are several caveats to this explanation (see Discussion).

Previous studies have hypothesized that Zld is a ubiquitous sequence-specific transcription factor (TF) that functions as a key activator of the zygotic genome in the early Drosophila embryo. The presence of Zld-binding peaks at the positions of the eve2 and eve37 enhancer (Bailey et al., 2009; Harrison et al., 2011; Nien et al., 2011), and their complete loss of expression in Zld-depleted embryos (Fig. S2), is consistent with this hypothesis. Our motif mutation experiments suggest that Zld has direct roles in activating the eve enhancers. For eve2, mutating Zld motif variants within the enhancer causes only a small reduction of expression; mutating additional sites outside the element further reduces, but does not abolish, expression. It is unlikely that our experiments have removed all Zld-binding (e.g. there are two canonical Zld motifs (CAGGTAG) in lacZ itself, and numerous variant motifs throughout the 3.8 kb eve region), and mutating more motifs might completely abolish eve2 expression. However, defining all the crucial sites in this large fragment and mutating them without interfering with other important regulatory sequences is not possible given our incomplete understanding of the precise rules of Zld binding and other known and unknown TFs that regulate these enhancers.

For eve37, mutating the single canonical Zld site within the enhancer (Mut Z 37) reduces expression (Fig. 3F), but the reduction of this stripe is not as strong when this site is mutated in combination with the other motifs in the 3.8 kb transgene (Mut Z all). This clearly indicates that the stripe 3 response is less dependent on Zld for activation compared with stripe 2. The discrepancy between the Mut Z 37 and Mut Z all experiments is not easy to explain, but one possibility is that residual Zld binding to motif variants in the multiply mutated construct is co-opted by the eve37 stripe enhancer because it is intrinsically the strongest enhancer present in the construct.

Our data also suggest that Zld indirectly activates the eve37 enhancer by facilitating the Hb-dependent repression of kni, which forms the posterior boundaries of eve stripe 3. In Zld-depleted embryos, this repression is weakened, which may cause the expression of high levels of kni in nuclei that normally activate eve3 (Fig. 6A,C). Interestingly, the high level of kni RNA in nuclei normally destined to express eve3 is transient (Fig. S9), but this short burst appears to be sufficient for preventing any activation of the eve3 enhancer. Perhaps the protein translated from this mRNA is stable enough to effectively repress the enhancer until the activators are no longer available or active. A complementary expansion of Hb into more posterior regions is not observed because Hb is activated by a threshold-dependent mechanism involving Bcd (Driever et al., 1989; Struhl et al., 1989), which is expressed in an anterior gradient that is not altered in Zld-depleted embryos (Xu et al., 2014). When Kni or Hb repressive activity is removed by binding site mutations, the eve37 enhancer is reactivated in Zld-depleted embryos in regions occupied by those repressors. However, the eve37ΔH and eve37ΔK patterns are still repressed in the ventral-most region of Zld-depleted embryos (Fig. 5F,I). The most likely candidate for this repressor is the terminal gap gene tailless (tll), which encodes a dedicated repressor (Haecker et al., 2007) and is normally expressed in two domains near the poles of the embryo (Pignoni et al., 1990). In Zld-depleted embryos the posterior tll domain tilts ventrally and expands toward the middle of the embryo (Nien et al., 2011).

Our data are less clear for eve2, which forms anterior and posterior boundaries by responding to Gt- and Kr-mediated repression (Small et al., 1991). The strong anterior expansion of Kr caused by weakening of Gt repression in Zld-depleted embryos (Fig. 6B,D) would likely contribute to the complete loss of expression seen in those embryos. In theory, we could test whether loss of Kr in cis would reactivate eve2 expression in embryos lacking Zld, but mutations in Kr sites also interfere with the binding of Bcd (Small et al., 1992), which is the primary activator of the eve2 enhancer (Stanojevic et al., 1991; Arnosti et al., 1996). As an alternative, we attempted to specifically mutate Gt-binding activity in the eve2 enhancer, but the mutated enhancer (eve2ΔG) was nearly silent in Zld-depleted embryos (Fig. 5O). This result supports the idea that Zld has a stronger role in directly activating the eve2 enhancer compared with eve37. However, it is also possible that activation of the eve2ΔG enhancer fails because it is repressed by other repressors that shift in Zld-depleted embryos. It was previously shown that at least three separate repression mechanisms prevent eve2 activation in anterior regions of the embryo (Andrioli et al., 2002).

In summary, our results show that a complete understanding of the roles of Zld on eve expression can only be achieved by considering the full network of activators and repressors that bind the eve enhancers (Fig. 6). They also shed light on general mechanisms of Zld-dependent gene regulation. Most importantly, we show that Zld is required for efficient binding of at least one activator (Bcd) and several repressors to enhancers of the gap and pair-rule genes. These results are consistent with its proposed role as a general facilitator of TF-enhancer binding. However, although Zld may contribute to the binding of many factors, our results show that it differentially affects the efficacy of specific interactions. In particular, the loss of Zld activity causes dramatic expansions of the gap genes kni and Kr (Fig. 4), presumably due to reductions of Hb- and Gt-binding activities respectively. In contrast, mutations of the Zld sites in the eve locus reduce stripe expression levels, but do not visibly affect the repressor activities that establish stripe boundaries (Fig. 3).

How these gene-specific activities occur at the molecular level is not clear. Recent studies suggest that timing is crucial for the role of Zld in initiating and maintaining its functions in zygotic genome activation (Nien et al., 2011; McDaniel et al., 2019). One possibility is that the specific interactions that are most affected in our study (cross-repression by the gap genes and activation of eve2) occur within some critical time interval, whereas those that are less affected (activation of eve3 and repression events that set the boundaries of eve2 and eve3) occur outside this interval. The more sensitive events do normally occur slightly before the less sensitive events, consistent with this idea. Alternatively, the removal of Zld might create a situation where the reduction of binding activity causes some repression activities to fail while others remain functional. In this scenario, repressor binding would be reduced to a certain extent, leaving residual binding that is sufficient to set the boundaries of the eve stripes, but not mediate their normal activities in cross-repression. Previous studies have shown that the gap repressors function as gradients with specific thresholds for repression of different target genes (Wu et al., 1998; Clyde et al., 2003; Yu and Small, 2008), consistent with this idea. Finally, Zld is a very large protein (1596 amino acids) (Liang et al., 2008), and thus could interact with a number of cofactors that might affect its specific activities. Future studies of individual Zld protein domains and the effects of Zld on other individual genes and enhancers will be required to understand how this single protein has such large and varied effects on the overall elaboration of the embryo patterning network.

We used two methods to remove or knock down Zld function in trans: the ovoD-FLP-FRT system (Chou et al., 1993) and misexpression of antisense Zld microRNAs (Zld shmiR) (Sun et al., 2015). Specifically, the experiments in Fig. S2 were performed using Zld germline clones (GLCs), the embryo staining experiments in Figs 1-4 were performed using both methods (with indistinguishable results), and the experiments in Fig. 5 were performed using Zld GLCs. For the ovoD-FLP-FRT method, we used standard crosses to generate virgin females of the genotype zld²⁹⁴ FRT 19A/ovo^D FRT 19A, hsFLP122 (Liang et al., 2008), which were mated with yw males or transgenic males carrying wild-type or mutated reporter constructs. For the Zld shmiR method, we used standard crosses to generate females with the genotype COG-GAL4:VP16/+; Gal4-nos.NGT40/+; nos-Gal4-VP16)/UAS-shRNA-Zld (Sun et al., 2015), which were crossed to yw males.

Canonical Zld sites were defined as four specific core motifs (CAGGTA, TAGGTA, CAGGCA and TAGGCA). There are eight such motifs in the 3.8 kb eve fragment examined here (Fig. S4). To identify Gt-binding motifs in the eve2 enhancer, we used the nearly palindromic sequence (T-T-A/G-C-G/A-T-A-A) derived from one-hybrid experiments using the Gt basic region and leucine zipper (Noyes et al., 2008) as a starting point. There are no matches to this sequence in the eve2 enhancer, but there are ten sequences that correspond to potential half sites (TTAC, TTGC, TTAT and TTGT). These putative half motifs and the mutated sequences tested here are shown in Fig. S12. For the ChIP-qPCR experiments in Fig. 2, we used known binding sites from previous studies (Arnosti et al., 1996; Struffi et al., 2011) to design fragments for PCR. For Fig. 4, we used ClusterDraw (Lifanov et al., 2003) to predict the positions of Hb sites in the Kni KD enhancer and Gt sites in the Kr CD1 enhancer. Primer pairs used to amplify genomic fragments are shown in Fig. S3.

Transgenic flies carrying four mutated Hb sites (eve37 ΔHb) or eleven Kni sites (eve37 ΔKni) have been described previously (Struffi et al., 2011). To generate the other mutant transgenes tested here, subfragments containing different combinations of mutations in individual Zld or Gt motifs were obtained as gBlocks Gene Fragments (IDT) and cloned into reporter genes containing the lacZ coding region and the α-tubulin 3′ UTR. Reporter gene constructs were inserted via ɸC31 integrase mediated cassette exchange at chromosomal position 38F1. Embryos (2-4 h after egg laying) were collected and stained by in situ hybridization using digoxigenin-labeled anti-sense RNA probes as previously described (Small, 2000).

For quantifying the effects of different Zld site mutations on the striped pattern directed by the 3.8 kb eve-lacZ transgene, embryos (2-4 h after laying) were stained for lacZ mRNA expression using the colorimetric alkaline phosphatase method developed by Tautz and Pfeifle (1989). In our hands, this method yields patterns with more precisely separated individual stripes compared with fluorescent in situ hybridization methods, and permits the analysis of a large number of individual embryos. However, the method is not perfect because of the non-linearity of the alkaline phosphate activity, which can saturate the color produced by the enzyme-tethered probe. Therefore, lacZ expression patterns were carefully monitored by eye during staining reactions and, after mounting and image acquisition, two major criteria were used to select individual embryos that were subsequently used for quantification.

First, we selected only embryo images that were precisely oriented laterally by co-staining with lacZ and anti-sense snail (sna) probes. sna is expressed in a ventral domain that overlaps with the ventral regions of the eve-lacZ stripes, and only images with the sna domain at the bottom of the image were selected for further analysis. We also used the sna domain to ensure that selected embryos had developed past the crucial stage of eve stripe initiation. The sna expression pattern exhibits fuzzy boundaries when first expressed in NC13, and then sharpens dramatically after the onset of NC14 (Ip et al., 1992), when the eve stripes are forming. Therefore, we rejected embryos that showed fuzzy sna boundaries. We also rejected embryos containing broad lacZ expression domains at the position of eve2. Previous studies have shown that eve2 expression begins with a broad domain in NC13, which refines from the anterior to form the correct boundary in the first 20 min of NC14. Finally, these constructs also drive expression of a spurious stripe in the region anterior to stripe 2 after it forms (Fig. S5). This stripe gets stronger in the second half of NC14, and we rejected embryos containing strong anterior stripes as being too old. For the Zld depletion experiments, we rejected embryos showing blastoderm defects (irregular nucleus packing, nuclear fallout, etc.; Liang et al., 2008), which appear during the second half of NC14. We estimate that these temporal criteria together define a time window of ∼20 min in the first half of NC14.

Second, each correctly oriented embryo image was then subjected to the following analyses to quantify individual stripe responses and avoid saturation issues. A 100-pixel tall region of interest (ROI) was stretched from the anterior to the posterior in a lateral region that did not overlap the ventral sna domain (Fig. S6A). Expression in the ROI was converted using the plot profile function of ImageJ into a .csv file of gray intensity values of vertical pixel lines with a scale from 0-255. The maximum pixel values for all the embryos analyzed here were in the range of 140-150. To normalize for slight length differences between individual embryos, we converted the pixel line profile into a percentage embryo length (EL) scale, where 100% EL and 0% EL represent the anterior and posterior ends, respectively. We then subtracted the lowest lacZ intensity value from every value to set the baseline to zero, and divided each lacZ value by the mean pixel value in the sna region between stripes 3 and 7 for each embryo to normalize lacZ expression levels between different embryos (Fig. S6A). The measure function in ImageJ was used to calculate an average pixel value in the selected sna region. We expected that this normalization would reduce the variance between individual embryos for a single construct, but the average standard deviations did not appreciably change when the normalization step was left out. However, the normalization with sna was valuable for normalizing intensity values between constructs.

For our Gaussian fitting analysis, .csv files containing the un-normalized data for multiple embryos containing the same construct were imported from ImageJ into the NumPy library (1.19.5) of Python (3.9.1), and two windows were chosen to fit Gaussians onto the .csv files, one for stripe 7 alone, and one for both stripes 2 and 3. The std function from NumPy was used to calculate the standard deviation of the data within a window surrounding stripe 7. The curve_fit function from the optimize sub-library of SciPy (1.6.0) was used to fit Gaussian curves to our data from the .csv file within the given windows. The standard deviations found from NumPy were used as guesses in the parameters of the curve_fit function to increase the accuracies of our fits. A local minimum was found in the stripe 2+3 window and used as a boundary between the stripes (Fig. S6B), which permitted the independent calculation of standard deviations for each stripe, and the function plot from the matplotlib (3.3.3) library was imported to visualize the results of curve_fit.

Detailed protocols for embryo collection, ChIP, qPCR and data processing have been previously described (Xu et al., 2014). Briefly, chromatin was prepared from two biological replicates of wild-type embryos and embryos from Zld GLCs (1-3 h after egg laying) and each experiment was performed three times. ChIP-grade antibodies against Gt, Kr and Kni were generous gifts from Michael Eisen (Department of Molecular and Cellular Biology, UC Berkeley, CA, USA). The anti-Hb antibody (Kosman et al., 1998) was purified using protein A column chromatography. Negative controls without antibodies were run in parallel, and used to normalize binding levels, which were calculated as fold enrichment over amplification levels of the input DNA before immunoprecipitation. Primer sequences used to amplify fragments from the eve2, eve37, KniKD and KrCD1 enhancers after immunoprecipitation are shown in Fig. S3.

We thank Christine Rushlow for fly stocks used to generate Zld mutant GLCs and Zld knockdown shmiRs and for sharing reagents, Miki Fujioka for transgenic lines containing eve1-, eve46-, and eve-5-lacZ transgenes, Michael Eisen for ChIP-quality anti-Gt, anti-Kr and anti-Kni antibodies and Daniel Tranchina for guidance in the statistical analysis for quantifying individual striped patterns in the early embryo. We acknowledge excellent technical assistance from NYU's Fly Facility, the Imaging Facility at NYU's Center for Developmental Genetics, and the Genomics Core Facility at NYU's Center for Genomics and Systems Biology. We thank Michael Erb for providing reagents and lab space to T.R.B. for transgene construction. We are grateful to Christine Rushlow and Claude Desplan for stimulating discussions, and all members of the Datta, Onal and Small labs for support.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201860.reviewer-comments.pdf.