Zygotic pioneer factor activity of Odd-paired/Zic is necessary for establishing the Drosophila Segmentation Network

Because regulatory networks of transcription factors drive embryonic patterning, it is possible that chromatin accessibility states impact how networks interact with information encoded in DNA. To determine the interplay between chromatin states and regulatory network function, we performed ATAC seq on Drosophila embryos over the period spanning the establishment of the segmentation network, from zygotic genome activation to gastrulation. Chromatin accessibility states are dynamic over this period, and establishment of the segmentation network requires maturation of the ground chromatin state. Elimination of all maternal patterning information allows identification of patterning-dependent and -independent dynamic chromatin regions. A significant proportion of patterning-dependent accessibility stems from pioneer activity of the pair-rule factor Odd-paired (opa). While opa is necessary to drive late opening of segmentation network cis-regulatory elements, competence for opa to pioneer is regulated over time. These results indicate that dynamic systems for chromatin regulation directly impact the interpretation of embryonic patterning information.

But embryonic chromatin states themselves are dynamic [4][5][6][7][8][9]. The mechanisms that drive 56 developmental progression can also trigger remodeling of chromatin accessibility patterns on both 57 large and small scales, thereby changing over time what genetic information is available to gene 58 regulatory systems. While in many cases we have near comprehensive understanding of both the 59 genetic components of certain developmental networks and the critical CRMs whereby these 60 components interact, much less is known about how chromatin accessibility states constrain 61 network function and how mechanisms for controlling chromatin accessibility are systematically 62 woven into the developmental program. 63 In the case of Drosophila melanogaster, decades of investigation into the mechanisms of 64 development have exhaustively identified the critical patterning cues and transcription factors that drive early cell fate specification and differentiation of select developmental lineages. Patterning 66 is initiated by four distinct maternal pathways that alone are sufficient to initiate zygotic regulatory 67 networks that specify all of the primary cell identities that arise along the major embryonic axes 68 [10][11][12][13][14][15][16][17]. At the outset of patterning, nuclei have what can be considered a 'ground state' of 69 chromatin structure that contains the initial set of accessible CRMs and promoters that will define 70 the first regulatory network interactions [7]. The ground state effectively provides a baseline for 71 determining the influence of epigenetic mechanisms of gene regulation on developmental 72 processes. The early Drosophila embryo therefore provides an ideal starting point to observe both 73 how regulatory networks are constrained by chromatin states, and how these states evolve as a 74 function of progression through the developmental program. 75 Before embryos can respond zygotically to maternal patterning cues, they must first 76 undergo a series of 13 rapid, synchronous mitotic divisions that serve to amplify the single nucleus 77 formed after fertilization into a set of ~6000 largely uncommitted, pluripotent cells [18,19]. These 78 mitotic divisions occur in a state of general transcriptional quiescence that effectively prevents 79 nuclei from responding prematurely to regulatory stimuli [20][21][22][23]. The shift from the initial 80 proliferative phase to later periods of differentiation comes at a major developmental milestone 81 chromatin state contains all the accessible cis-regulatory information required to complete this 127 well-characterized developmental patterning task. 128 To evaluate chromatin accessibility states between ZGA and gastrulation, we collected 129 single embryos aged either 12 or 72 minutes into NC14 and performed ATAC-seq. Mapped reads 130 were assigned to peaks, which were subsequently cross-referenced against the Redfly database of 131 previously characterized CRMs within the segmentation network [40]. These were then scored for 132 accessibility either shortly after ZGA (NC14+12') or one hour later at the onset of gastrulation 133 (NC14+72') ( Figure 1A; see Materials and Methods). 134 We find that CRMs within each tier of the segmentation network are not constitutively 135 accessible and have distinct temporal chromatin accessibility profiles that correlate with the 136 activity periods associated with these regulatory elements. All of the known early gap gene CRMs 137 are open at ZGA and typically lose accessibility by the onset of gastrulation ( Figure 1A, B). In 138 contrast, pair-rule CRMs separate into two distinct temporal classes of chromatin accessibility. All 139 of the early, stripe-specific CRMs within the pair-rule network are open at ZGA, whereas later, 140 seven-stripe (or 14-stripe) specific CRMs generally lack open chromatin at ZGA, and gain 141 accessibility by the onset of gastrulation ( Figure 1A, C). The majority of the known segment 142 polarity CRMs lack accessible chromatin at ZGA and undergo significant gains in accessibility by 143 gastrulation ( Figure 1A, D). Taken together, these results demonstrate that during the one-hour 144 period between ZGA and gastrulation patterns of chromatin accessibility within segmentation 145 network CRMs are dynamic, correlating with the early or late activity of gene expression patterns 146 within the network. We conclude from this that the ZGA chromatin state contains insufficient 147 accessible cis-regulatory information to sustain the function of the segmentation gene network. 148 Chromatin accessibility patterns continue to evolve over the one-hour period between ZGA and 149 gastrulation to support the later-acting components of the network, particularly the segment 150 polarity and late pair-rule systems. This raises the possibility that the hierarchical networks that 151 drive embryonic segmentation derive timing information from regulated chromatin accessibility. 152 Notably, binding of pioneers [28,41] implicated in the establishment of the initial ZGA chromatin 153 state is low or absent at sites that gain accessibility late (Figure 1     2B) that are all fated to become posterior endoderm. We predicted that, compared with wild-type 204 embryos, bcd osk cic tsl Tl RM9 (hereafter, "mutant") embryos would enrich for posterior 205 endodermal chromatin states at the expense of all others, allowing for the unambiguous 206 determination of sites that undergo patterning-dependent versus -independent changes in 207 accessibility. 208 We therefore performed ATAC-seq comparing single wild-type or mutant embryos 209 precisely staged at 12 and 72 minutes into NC14. As expected, comparison of differentially 210 accessible regions between stage and genotype identifies regulatory elements with spatially 211 restricted expression patterns. For example, at the wingless (wg) locus, two closely apposed 212 regions (wg -2.5 and -1) show differential behavior between genotypes. The wg -2.5 region 213 undergoes a modest increase in accessibility in wild-type embryos but not in mutants ( Figure 3A, 214 green shading). In contrast, the neighboring wg -1 region shows greatly increased accessibility in 215 mutant samples ( Figure 3A, cyan shading). These two regions both comprise a single previously 216 identified CRM (wg WLZ4L) controlling early wg expression in both a segment-polarity multi-217 stripe pattern as well as a single posterior endodermal stripe [46]. On the basis of our ATAC data, 218 we individually cloned the wg -2.5 and -1 regions into reporter constructs and compared expression 219 between wild-type and mutant embryos. wg -2.5 becomes active just prior to gastrulation in wild-220 type but not in equivalently staged mutant embryos, and exclusively drives expression of the 221 segment-polarity multi-stripe pattern ( Figure 3B and data not shown). In constrast, wg-1 is active 222 earlier in NC14 and exclusively drives expression of the endodermal stripe pattern. As expected, 223 whereas wg -1 expression is restricted to a single posterior endodermal domain in wild-type 224 embryos, all cells in a mutant embryo have strong wg -1 expression ( Figure 3B). We therefore 225 performed differential enrichment analysis using DESeq2 [47] to identify the complete set of 226 regions with patterning-dependent or -independent changes in chromatin accessibility.
We identify significant sources of both pattern-independent as well as patterning-228 dependent changes in chromatin accessibility over the period between ZGA and gastrulation. In  variance, Figure 3C). The second principal component separates samples according to genotype 236 and therefore resolves patterning-dependent variance (PC2: 21% variance, Figure 3C). There is 237 less of a patterning-dependent difference between NC14+12' samples compared with the +72' 238 timepoint, supporting the conclusion that cells initiate the zygotic phase of development with a 239 large degree of chromatin state homogeneity and that heterogeneity emerges over the period 240 leading up to gastrulation from both patterning-dependent and -independent sources. 241 To relate the observed changes to a single, discrete developmental process, we returned to 242 the set of known segmentation network CRMs and plotted scaled chromatin accessibility over time 243 between wild-type and mutant samples. Within all three tiers of the segmentation network, we find 244 evidence for extensive patterning-dependent chromatin accessibility, at both early and late 245 timepoints ( Figure 3D). As previously shown, the gap gene network receives extensive patterning-246 dependent chromatin accessibility cues from a pioneer activity of Bcd (7/18 CRMs, 38.9%) [32]. 247 Early pair-rule CRMs receive both patterning dependent and independent inputs, however the 248 majority of late pair-rule CRMs gain accessibility in a patterning-dependent manner (25/30 pair-249 rule CRMs with late accessibility, 83.3%). Segment polarity CRMs (e.g., wg -2.5 and wg -1, Figure  250 3A) likewise have extensive patterning-dependent accessibility states (16/33, 48.5%). Therefore, 251 these results indicate that although overall changes in accessibility tend to occur independently of 252 embryonic patterning, the networks dedicated specifically to embryonic patterning display a 253 disproportionate reliance on patterning systems for determination of their chromatin accessibility 254

states. 255
Next, we quantified the types of changes in chromatin accessibility that we observed in our 256 analysis. Similar to the PCA analysis, we find fewer patterning-dependent differences at ZGA than 257 at gastrulation (408 versus 1871, Figure 3E, left two panels). In contrast, a greater number of time-258 dependent differences are observed for both genotypes (5190 for wild-type, 8655 for mutant, 259 Figure 3E right two panels). We note that these numbers represent above-threshold statistical 260 significance for tests on only one of two critical parameters in this experiment, either time or 261

genotype. 262
To comprehensively classify the types of changes that any single region undergoes over 263 time and relative to patterning inputs, we took an clustering-based approach to identify groups of 264 similarly-behaved regions and then used the output from paired DESeq2 tests to assign regions to 265 each identified category (see Materials and Methods). By this approach, we identify a total of 2917 dynamic regions that classify into one of ten distinct dynamic categories with respect to time and 267 patterning-dependence ( Figure 4). Overall, roughly similar numbers of sites gain and lose 268 accessibility over time with 1351 of sites that are open early losing accessibility over time (46.3%, 269 "late repression" classes, Figure 4) and 1446 sites gaining accessibility by the onset of gastrulation 270 (49.6%, "late accessibility" classes, Figure 4). In total we identify 1635 (56.1%) strictly pattern-271 independent regions that either gain (n = 725 (24.9%)) or lose (n = 910 (31.2%)) accessibility     Figure 4). In addition to 298 these, we frequently observe across all categories enrichment for three maternally supplied factors 299 with expected repressive activity, Tramtrack (Ttk), Adult Enhancer Factor 1 (Aef1), and Combgap 300 (Cg). While Ttk has long been hypothesized to play a broad repressive role over the maternal-to-301 zygotic transition [35] as well as in regulation of embryonic patterning [51-54], much less is known 302 about potential early embryonic roles of Aef1 and Cg (see Discussion). We also note that although 303 motifs for Cg are not included in the available DNA binding motif databases used to compile these results, we include Cg here on the basis of previous identification of Cg binding to a (CA)n motif 305 [55], maternal expression of Cg, and frequent recovery of an orphan motif (CA)n in our analysis. 306 The recovery of Bcd and Fkh motifs in our dataset suggests that enrichment analysis could 307 identify potential patterning-dependent pioneer activities responsible for driving differential 308 accessibility states. We note here that recovery of motifs in our analysis is similar to those 309 recovered in another recent report [56], which measured changes in accessibility between early 310 and late NC14 samples but did not distinguish between patterning-dependent and independent 311 events. Bcd has been demonstrated to pioneer accessibility at a subset of its targets [32], and Bcd-312 motif enrichment in this analysis correlates with the set of previously identified bcd-dependent 313 regions (e.g., gt -10). Fkh is the Drosophila homolog of a well-characterized pioneer factor 314  patterning-dependent accessibility late in NC14, and because many pair-rule and segment polarity 350 CRMs also display late patterning-dependent accessibility, we tested the hypothesis that Opa 351 functions as a pioneer factor to confer changes in chromatin structure downstream of maternal 352 patterning cues. (data not shown). Llama-tagged Opa first becomes detectable above background by live imaging 362 at 35 minutes into NC14 reaching an apparent steady-state expression level at 60 minutes, shortly 363 before gastrulation ( Figure 5B). These measurements indicate that Opa expression is consistent 364 with an exclusively late NC14 role in regulating gene expression over nearly all cells of the 365 embryonic segmental primordium.
By measuring chromatin accessibility in single opa +/+ , opa -/+ , and opa -/embryos at NC14 367 + 72', we find that Opa is necessary to pioneer chromatin accessibility at a subset of its direct 368  We have demonstrated that following ZGA, both local and global changes in chromatin 428 accessibility patterns continue to take place. The identification of pioneers like Opa raises the 429 possibility that distinct zygotic factors function to establish accessibility states conditional on prior, maternal patterning information. We wished to investigate the hypothesis that the sequence of 431 chromatin accessibility changes are themselves critical for the proper execution of the 432 developmental patterning program. To address this, we therefore quantified opa-dependent 433 chromatin accessibility within the segmentation network and evaluated the consequences of 434 premature opa expression on patterning. We predicted that maternal mis-expression of opa would 435 effectively conflate a ZGA chromatin state with a gastrulation chromatin state. Opa is necessary 436 for full chromatin accessibility at a set of late pair-rule and segment polarity CRMs ( Figure 6A-437 C). Late-acting opa-dependent CRMs within the pair-rule network include the late eve seven-stripe 438 element (eve-late also referred to as eve-autoregulatory, [79][80][81], see also    Notably, although opa is sufficient to induce accessible chromatin at most of the late pair-458 rule opa-dependent targets (odd-late, eve-late, slp1 "5", slp1 i1523, prd 01, prd A8 repressor) as 459 well as three regions within the opa locus itself, segment polarity targets show a distinctly reduced 460 sensitivity to gain accessibility in response to premature opa expression, with only en H2 and an   Figure 6E). In addition, compared with wild-type, the regular spacing of 483 odd-and even-parasegmental stripes at positions 4-6 is disrupted with maternal opa expression. 484 Despite the fact that tub>opa is expressed uniformly across the entire embryo, premature odd-late 485 expression appears not uniformly, but within the spatial domains to which its later expression will 486 be restricted ( Figure 6D may play a more global role in ZGA timing by limiting pioneer factor activity at target sites until 549 they are cleared from the embryo. In this respect, it is interesting to note that the subset of opa-550 dependent targets that are insensitive to maternal opa expression demonstrate an enrichment for 551 Ttk and Cg motifs. One possible explanation for this apparent developmental competency to 552 respond to Opa pioneering activity is that binding of maternal repressors can antagonize pioneer 553 factor activity. Future work will include testing the role of these maternal factors in the context of 554 ZGA timing and regulation of coordinated, global chromatin remodeling events.
In contrast with these mechanisms for uniform regulation of accessibility, while there is 556 relatively little influence of maternal patterning systems directly on chromatin accessibility status, 557 we observe that certain zygotic targets of maternal pathways, such as Opa, can have a major impact 558 on chromatin accessibility states. Opa's primary role in the pair rule network is to facilitate the 559 transition, termed a 'frequency doubling', from early to late expression patterns. In the absence of 560 Opa, pair-rule loci (primarily odd, slp, run, and prd) fail to undergo the transition from early seven-561 stripe to late 14-stripe patterns. Additionally, late 7-stripe expression of eve is also strongly 562 accessibility of the CRMs that drive these late expression patterns. We predict that Opa pioneer 568 activity will therefore result in conditional cis-regulatory interactions of the remaining pair-rule 569 factors with late CRMs. This mechanism can help explain the previously observed 'conditional 570 regulation' between network components (e.g., Odd repression of prd to yield anterior and 571 posterior stripes) [71], which we propose is largely mediated through opa-dependent CRM 572 accessibility states. The set of opa-dependent CRMs within the pair-rule network that we identify 573 strongly support this conclusion. Incorporating such 'time-gated' pioneering events into a 574 regulatory network may therefore allow for a system to generate multiple patterning outputs from 575 a limited set of input transcription factors. Further investigation of the opa-dependence for 576 conditional cis-regulatory interactions amongst pair-rule factors, as well as identification of 577 additional zygotic pioneer factors will address these predictions. 578

A critical distinction that arises between transcription factors within a network, then, is 579
what effect they have on chromatin accessibility states. It is likely that not all transcription factors 580 have pioneer activity, or that the ability of a factor to pioneer is context specific. For instance, loss 581 of grainyhead (grh) has minimal effects on the pre-gastrula chromatin accessibility state, despite 582 the fact that grh has been demonstrated to function as a pioneer in other biological contexts [56]. were found to be enriched for previously identified bcd-dependent regions [8]. If dl were to pioneer 608 chromatin to a similar extent as bcd, we would therefore expect that single-cell ATAC-seq would 609 have also identified early "dorsal" and "ventral" clusters, but in this study, all DV-specific clusters 610 (e.g., mesoderm) were only found associated with cells presumed to be staged later based on 611 'pseudotime' analysis. So, although maternal systems may not drive differential accessibility along 612 the DV axis, it nevertheless likely that, similar to opa, specific zygotic targets within the DV In our study, we demonstrate distinct patterning-dependent gains in chromatin accessibility 642 over time, and a significant fraction of these we link to opa activity. However, within the 643 segmentation network specific CRMs, Opa is not the sole pair-rule factor predicted to bind to these 644 sites. At the odd-late enhancer, for instance, in addition to our demonstration of direct Opa binding, 645 Qiagen Midiprep grade DNA preparations. Following injection, surviving embryos were raised to 729 adulthood and single founders were crossed to five or six w; Dr e/TM6c, Sb flies. F1 progeny were 730 scored for the proportion of ebony and vials with >50% ebony progeny were scored as "jackpots" 731 in which we induced greater-than-monoallelic targeting of the ebony locus. Up to six individual 732 jackpot males from up to six jackpot founders were subsequently crossed to w; Dr e/TM6c, Sb 733 females and stocks were established from F2 progeny. We favored jackpots that produced 85-734 100% ebony progeny. We observed a 32% survival rate of injected embryos (effectively 64% and 110 Hz scan rate with bidirectional scanning. Series were collected at 30 seconds per xyz 761 stack. We required that a movie begin within NC12 and for data collection to continue until at least 762 NC14 + 72 minutes. This ensured not only that we quantified opa expression from the beginning 763 of NC14, but also controlled for variation in developmental rates by allowing us to measure the 764 duration of NC13. We required that NC13 would last no longer than 22 minutes and no shorter 765 than 17 minutes for imaging to proceed. Fluorescence intensity was quantified by segmenting 766 nuclei in the RFP channel and quantifying average, per nucleus intensity in the GFP channel.   Pseudoreplicates were generated by randomly splitting each replicate sample into two .bed files. 856 Overall pseudoreplicates were generated by pooling all replicates for either w or opa-3xmyc and 857 randomly splitting reads into two separate .bed files. Several peaks files were generated using 858 MACS2 in preparation for IDR analysis. First, peaks for each individual opa-3xmyc sample (n = 859 3) were called against the pooled w sample data as a control. Having previously determined the 860 MACS2 parameter -d, we bypassed model building (--nomodel) and manually specified the 861 expected fragment size (--extsize 149). Relaxed conditions were specified by designating the 862 option -p 1e-3 as recommended by IDR. Samples were scaled to the larger dataset (--scale-to 863 large). A second peaks list (n = 1) was generated for the pooled opa-3xmyc samples against the 864 pooled w samples using the same MACS2 options. A third set of peaks (n = 6) were called for 865 each pseudoreplicate for each biological replicate of opa-3xmyc against the pooled w samples, using the same MACS2 options. Finally, a fourth set of peaks (n = 2) were called for the two 867 pseudoreplicates of the pooled opa-3xmyc data against the pooled w data, using the same MACS2 868 parameters. We then performed IDR analysis on this set of peaks with threshold values of 0.02 for 869 individual replicates and 0.01 for the pooled data sets. IDR was performed for pairwise 870 comparisons between replicates (e.g., rep 1 vs rep 2, rep 2 vs rep 3, and rep 3 vs rep 1) using as a 871 reference the peak list from pooled opa-3xmyc samples. IDR options were --input-file-type 872 narrowPeak --rank p.value --soft-idr-threshold 0.02 and --use-best-multisummit-idr. IDR was then 873 subsequently performed on each pair of pseudoreplicates for each individual biological replicate 874 (e.g., rep1 pseudo 1 vs rep 1 pseudo 2…) using the same IDR options. Finally, IDR was performed 875 on the pooled sample pseudoreplicates using the same IDR options except --soft-idr-threshold was 876 0.01 instead of 0.02, per the recommendation of the IDR instructions. The largest number of 877 reproducible peaks between replicates was 881, and so we took the top 881 peak regions (ranked 878 by p-value) from the list of pooled peaks and filtered out two regions that mapped to non-canonical 879 chromosomes, leaving 879 high confidence peaks. Consequently, this is a conservative estimate 880 of true Opa binding sites but is determined by more rigorous criteria (reproducibility between 881 biological replicates) than either arbitrary p-value thresholding or simply taking the top N peaks. 882

ATAC-seq Analysis 884
Demultiplexed reads were trimmed of adapters using TrimGalore! and mapped to the dm6 885 assembly of the Drosophila genome using Bowtie2 with option -X 2000. To determine the different dynamic peak classes (Figure 4), we first undertook a clustering 906 based approach to explore how many different classes we could identify within the dataset. To do 907 this, we first averaged the number of reads per peak region for each sample and then scaled the 908 data for each peak region by dividing the mean reads for a peak region by the sum of all mean 909 reads for a peak region (i.e., so that the sum of the scaled reads for a peak region, sumregion(wt 12, 910 wt, 72, mut 12, mut 72), would equal 1). Next, we performed k-means clustering with a variable 911 number of cluster centers, and plotted the average per-cluster profile to visualize average behavior 912 of clustered regions. We found that 10 was the minimum number of cluster centers that would 913 capture all the unique patterns present in the data, that fewer clusters would combine similar but 914 qualitatively different classes, and more clusters would subdivide clusters into relatively similar 915 subsets. We note that over the two-year course of this study, attempts were made to replicate this 916 analysis on three different Apple (Apple Computer Inc., Cupertino, California) computers running 917 different base operating systems, versions of R, and versions of dependent libraries. For reasons 918 that are not clear, clustering based approaches were not strictly reproducible across differently 919 configured computers despite identical input datasets and identical scripting of the analysis code. 920 To be clear, similar cluster types, and minimum cluster number were called across different 921 systems, however the numeric order of the clusters, as well as the number of peaks assigned to 922 each cluster would vary between systems. Because of this, we could not rely on clustering alone 923 to reproducibly describe our results. Therefore, we took an alternative approach to categorizing 924 peak classes that depended on the statistical output of DESeq2, whose output was identical across 925 different computer systems. We paired statistical tests from DESeq2 to define classes. For 926 instance, to define regions that were uniformly open early and uniformly lost accessibility by 927 gastrulation, we required both wild-type and mutant samples to have a statistically significant difference across timepoints, with a log2 fold change of -1 or less. On the basis of these paired 929 DESeq2 criteria, we reproduced each of the 10 peak class types predicted by clustering. We note 930 that the final number of categorized peaks (2917) is substantially lower than the number of peaks 931 that score above the significance + log2 fold change thresholds of 0.01 + |1| (6775). This is due to 932 the fact that we require pairs of DESeq2 tests to score above significance thresholds for assignment 933 to a class. The remaining 3858 regions only score as significantly different in one of the four tests. 934 935

Generating reporter constructs based on ATAC-seq data 936
We used ATAC-seq coverage to delineate the sequence to test for potential enhancer 937 activities associated with wg-2.5, wg-1, odd-late, slp1 "5", 18w 1 and 18w 2. In general, coverage 938 at peak regions was plotted on the UCSC genome browser and views were zoomed out to identify 939 flanking regions of low accessibility. We hypothesized that functional genomic elements would 940 be defined by extended regions of accessible chromatin flanked by inactive, low-accessibility 941 regions. Primer pairs were designed to amplify peak regions plus a small amount of flanking 'low-942 accessibility' DNA and were cloned into a Gateway entry vector pENTR (Thermo Fisher 943 Scientific, Waltham Massachusetts). The exception to this overall strategy was in the case of the 944 wg-2.5 and -1 regions which are adjacent to one another ( Figure 3A). in which case we delineated 945 the two regions by the midpoint between the two major peaks of chromatin accessibility. Once we 946 had generated one pENTR-enhancer clone by TA-cloning of a PCR product, it was no longer 947 necessary, efficient, or desirable to perform TA cloning. Subsequent pENTR clones were made by 948 excising the original insert and replacing with new candidate enhancer fragments via Gibson 949 Assembly (NEB HiFi Assembly Kit, New England Biolabs). The candidate enhancer fragments 950 were shuttled to the transgenic vector pBPGUw [105] upstream of a minimal synthetic promoter 951 sequence driving Gal4 using standard Gateway cloning techniques (LR Clonase, Thermo Fisher 952 Scientific). Transgenic lines were established as described above by insertion into the attP40 953 landing site. 954 The genomic coordinates corresponding to regulatory elements tested in this study are:

In situ hybridizations 960
Probe sequences for in situ hybridizations were generated by PCR against cDNA clones for the 961 target transcript. Reverse primers included a leading T7 promoter sequence to facilitate probe 962 synthesis by in vitro transcription using an Ambion MegaScript T7 kit (Thermo Fisher Scientific) 963 supplemented with digoxigenin-11-UTP (Roche, Sigma-Aldrich) following standard procedures. 964 In situ hybridizations were performed according to standard procedures using a hybridization 965 temperature of 65°C, which was found to significantly reduce background compared with 56°C.