Nuclear microenvironments modulate transcription from low-affinity enhancers

1Genomic regions near coding genes, called enhancers, direct specific patterns of gene expression. Enhancers contain short DNA sequences that bind sequence-specific activating and repressive transcription factor proteins, and the integration of these positive and negative signals directs gene expression. Protein-DNA binding is often an ephemeral event; studies in mammalian cells demonstrate that transcription factors disassociate within seconds of binding to DNA. Furthermore, recent studies in animals ranging from fruit flies to mammals have revealed that low-affinity DNA-binding sites are critical to confer specificity between related transcription factors having binding sites with similar DNA sequences. Increasing the affinity of binding sites to more stably recruit transcription factors activates promiscuous gene expression, which leads to developmental defects. It is unclear how brief protein-DNA contacts can mediate efficient transcription from enhancers containing low-affinity binding sites.

2One possible mechanism that could mitigate their kinetic inefficiency is to increase the local concentrations of transcription factors. At the scale of a single enhancer over a few hundred base pairs long, multiple low-affinity binding sites for the same transcription factor in close proximity could increase the frequency of binding events by trapping the protein. Furthermore, interactions between transcription factors and cofactors with multiple binding sites within an enhancer could generate ‘microenvironments’ of high factor concentrations.

3We have explored this problem using the shavenbaby (svb) locus, which contains multiple enhancers that drive specific patterns of svb gene expression in developing Drosophila embryos. Each of three characterized svb enhancers contains clusters of low-affinity binding sites for the Hox gene Ultrabithorax (Ubx). These enhancers also require a Ubx cofactor Homothorax (Hth) to function. We have exploited robust transgenic tools in Drosophila, new fluorescent dyes, and new approaches to prepare embryos for microscopy to systematically perturb these svb enhancers and directly image the results at a sub-nuclear level. We find that microenvironments of high Ubx and Hth concentrations mediate transcription from low-affinity enhancers.

4We first examined whether nuclei in Drosophila melanogaster embryos possess Ubx microenvironments by performing immunofluorescence (IF) staining in fixed embryos and high-resolution confocal imaging using Airyscan (Carl Zeiss Microscopy, Jena, Germany). We found that Ubx protein was not distributed uniformly, but rather exhibited regions of high and low Ubx intensities (Figure 1A,B). To observe Ubx distribution at higher resolution, we expanded the size of the embryos by approximately four-fold in each dimension (Figure 1C). Nuclei of expanded embryos revealed distinct regions of high Ubx intensity separated by regions of low Ubx intensity. We observed, on average, 185 ± 25 (n = 12, three embryos) clusters per nucleus that were stronger than one-quarter of the maximum Ubx intensity within that nucleus (Figure 1D,E, and Figure 1—figure supplement 1).

5One explanation for the observed distribution of Ubx is that transcription factors localize generally to accessible regions of the nucleus that have high levels of transcriptional activity. This mechanism, if shared by transcription factors in general, should yield Ubx distributions that mostly overlap with that of other transcription factors. Engrailed (En), a transcription factor unrelated to Ubx, displayed non-uniform sub-nuclear concentrations, but its distribution only partially overlapped with that of Ubx (Figure 1—figure supplement 2A–C, white regions indicate overlap). We similarly observed only partial overlap between Ubx and Even-skipped (Eve) (Figure 1—figure supplement 2D–F). Abdominal-A (AdbA), a paralog of Ubx that is expressed mainly in separate cells from Ubx and that has similar DNA-binding specificity as Ubx, was excluded from Ubx regions in the few nuclei where both were expressed (Figure 1—figure supplement 2G–I). These results indicate that the distributions of these transcription factors do not result from a shared mechanism that limits the distribution of all transcription factors to the same sub-nuclear regions.

6We also examined whether Ubx simply occupies regions containing actively transcribed DNA. Both active RNA Polymerase II (Pol II, Ser5 phosphorylated CTD) and the methylated histone H3K4me3, which marks actively transcribed DNA, only partially overlapped with Ubx (Figure 1—figure supplement 3A–F). In contrast, the histone mark H3K27me3, which marks regions of repressed chromatin, displayed almost no overlap with the distribution of Ubx (Figure 1—figure supplement 3G–I). Thus, Ubx is not merely restricted to regions inside the nucleus that are available to transcription factors or to regions of high transcriptional activity.

7To understand if the heterogeneous distribution of Ubx is dynamic or stable over the timescale of seconds to minutes, as well as to rule out the possibility that our observations of Ubx microenvironments are an artifact of the fixation protocol, we examined the spatiotemporal dynamics of single Ubx molecules in live Drosophila embryos. Single-molecule imaging has been mostly performed in cell lines previously because live-imaging studies of transcription factor dynamics in embryos requires overcoming several new challenges, including imaging at lower signal-to-noise ratios, compensating for rapid morphological changes during embryonic development, and determining how to deliver fluorescent dyes. We overcame these challenges by generating a HaloTag-Ubx transgene that allowed precise control of fusion protein levels (Figure 1—figure supplement 4A) and coupling HaloTag-Ubx in vivo to new, strongly fluorescent dyes. The transgene we built can be expressed either from a heat-shock promoter (hsp70) or from a 20x UAS promoter by crossing with a GAL4 driver line.

8Over-expression of the HaloTag-Ubx transgene by incubating the embryos at 30°C transformed anterior segments to the fate of more posterior segments, indicated by the presence of additional trichomes. This result indicates that the HaloTag-Ubx protein retains the expected Ubx behavior (Figure 1—figure supplement 4D and E). We then expressed HaloTag-Ubx from the 20x UAS promoter with the nos::GAL4 (nanos promoter driving GAL4) driver line, which drives HaloTag-Ubx expression in all cells at early developmental stages. We injected the HaloTag ligand of Janelia Fluor 635 (JF635) into these live embryos. JF635 is minimally fluorescent in solution but its fluorescence increases by over 100-fold when bound to a HaloTag protein, allowing the detection of labeled Ubx molecules against a background of dim freely diffusing dyes. The fluorescence intensity of labeled Ubx scaled with distance from the site of dye injection (Figure 1—figure supplement 4B and C), consistent with dye diffusion from the site of injection. To measure the time-averaged density of HaloTag-Ubx in specific locations of nuclei in live embryos in early stage 5, we calculated the summed intensity over 100 s (1000 frames at 100 ms per frame). We observed regions of Ubx signal (3-10x background) similar to the high-intensity clusters observed in fixed embryos (Figure 1F). We examined the dynamics of HaloTag-Ubx in nuclei by plotting fluorescence intensity over time (Figure 1G and H and Figure 1—figure supplement 5). We found that fluorescence signals over time changed in discrete up or down steps, indicating that individual HaloTag-Ubx molecules bind to specific nuclear domains with residence times on the order of a second before dissociation. Most unbound Ubx molecules move too quickly to be captured with the 100 ms exposure time; they move in and out of a diffraction-limited region in significantly less than 100 ms on average. These timescales are consistent with transcription factor-DNA binding dynamics measured in live-cell imaging experiments using mammalian cell lines. These repeated binding events produced the high intensities observed in the time-averaged projections and indicate that Ubx concentrates and remains within specific nuclear regions.

9Observation of embryos at late stage 6 showed that total HaloTag-Ubx concentration continued to increase as the embryo ages (Figure 1—figure supplement 6A and B). Embryos at late stage 6 had nuclei containing high background concentrations of Ubx that masked single-molecule events, as well as displaying larger sites (>4 × 4 pixels) that constantly remained bright, possibly indicating the presence of multiple molecules or protein aggregation (Figure 1—figure supplement 6C and D). In contrast, the embryos observed during stage 5 did not contain areas that remained constantly bright, suggesting that we observed single molecule dynamics in stage 5 embryos.

10To determine whether regions of high Ubx concentration depended on DNA binding, we performed the same experiments with a version of the HaloTag-Ubx transgene where Arg3 and Asn51 of the homeodomain were mutated to Ala (R3A and N51A), abrogating DNA binding. Both the wild-type and DNA-binding deficient Ubx were expressed and imported into the nucleus (Figure 1—figure supplement 7A,B,D,E,G, and H), suggesting that the protein is stable. In contrast, an unstable HaloTag-NLS construct (NLS from H2B) serving as a negative control, neither increased JF635 fluorescence post injection nor became enriched into the nucleus (Figure 1—figure supplement 7C,F, and I). The mutant HaloTag-Ubx (R3A N51A) did not display spatial heterogeneity and exhibited only extremely brief fluctuations in intensity inconsistent with transcription-factor DNA-binding events (Figure 1I and J). These results suggest that binding of Ubx to DNA is required to generate restricted nuclear distributions of Ubx.

11The heterogeneous distributions of Ubx we observed are consistent with the hypothesis of nuclear ‘microenvironments’, whereby high local concentrations of transcription factors may drive transcription. Therefore, we examined whether these regions of high Ubx concentration co-localized with sites of active transcription. The svb gene is directly regulated by Ubx protein through binding of Ubx to low-affinity sites in multiple svb enhancers. We marked sites of active svb transcription by fluorescence in situ hybridization (FISH) and compared the localization of actively transcribed svb loci to Ubx protein concentration (Figure 2A and B). We observed high local Ubx concentrations surrounding active svb transcription sites (Figure 2C–F). To quantify Ubx distributions around these sites, we calculated the radially averaged Ubx intensity as a function of distance r from the point of maximum FISH intensity for each svb transcription site (Figure 2G–I). Ubx intensity was normalized to one at r = 0 (maximum FISH intensity) and averaged across all sites measured. To adjust for background fluorescence, we located the minimum intensity in the averaged Ubx distribution (r = 2–4 μm) and subtracted that value from the distribution. The first micrometer of the radially averaged 3D distribution is shown, with the shaded area representing the variance (Figure 2J). Within the first micrometer, svb transcription sites showed a relative enrichment of Ubx. Because these sites are on average within 200 nm of a local intensity maximum, Ubx intensity decreased monotonically away from the transcription sites, leading to a relatively constant variance after 200 nm. The normalized Ubx intensity after background subtraction at the site of svb transcription was 0.60 ± 0.17 (n = 59, four embryos, uncertainty is the variance of the background) and decreased approximately 250 nm away from the site. Thus, active svb transcription sites colocalized with areas of high Ubx concentration spanning approximately a few hundred nanometers.

12If Ubx protein co-localizes with actively transcribed svb loci because Ubx drives svb expression, then we would expect that transcription at a locus not regulated by Ubx should not co-localize with high Ubx concentrations. Indeed, we observed that active transcription sites driven by a synthetic enhancer containing binding sites for a TALEA transcription factor did not show Ubx enrichment on average despite wide fluctuations in Ubx levels, with a relative enrichment of Ubx at TALEA-driven enhancers of 0.02 ± 0.63 (Figure 3A–C, n = 29, three embryos). As these transcription sites are not close to maxima of Ubx intensity, the variance in these distributions incresed with distance from the site of transcription.

13In numerous nuclei actively transcribing svb on the X chromosome, we observed what appeared to be two transcription sites within 200 nm of each other (Figure 2—figure supplement 1A and B). This indicates that the svb locus on homologous X chromosomes often co-localizes to the same Ubx microenvironment. There are several possible mechanisms that could explain this observation. We consider two broad classes of mechanism. First, a unique chromosomal signature specific to the region containing the svb locus could facilitate localization of homologous alleles to the same transcriptional microenvironments. Second, microenvironments contain distinct combinations of transcription factors and enhancers localize to the relevant microenvironments to enable transcription. To distinguish between these alternative hypotheses, we examined the spatial distribution of the native svb locus, located on the X chromosome, and a single svb enhancer driving lacZ expression which we placed on chromosome 3.

14Double-FISH experiments revealed that the native svb locus and the ectopic svb enhancer co-localized often in nuclei in which both were transcribed (Figure 2—figure supplement 1C and D). In contrast, the transcription sites of forkhead (fkh, also on chromosome 3) did not colocalize with the svb locus (Figure 2—figure supplement 1E). The average distance between pairs of related transcription sites (svb-svb, svb-7H, and svb-E3N) within single nuclei is approximately 250 nm, near the resolution limit of AiryScan images (Figure 2—figure supplement 1F). On the other hand, fkh and svb transcription sites are on average 1 µm apart. These results indicate that related enhancers co-localize in transcriptional microenvironments independently of their chromosomal location. This suggests that transcription factor microenvironments are highly differentiated and that related enhancers often exploit the same transcriptional microenvironments.

15The experiments described so far showed that the actively transcribed native svb locus co-localizes with local concentration maxima of Ubx in the nucleus. We wondered whether the position of actively transcribed enhancers within Ubx microenvironments depended on Ubx binding site affinity. To address this question, we examined transcription driven by the individual svb enhancers DG3, E3N, and 7 hr, each of which contains a cluster of low-affinity Ubx-binding sites and can independently drive transcription of a reporter gene when moved from their native location. Transcription sites driven by these relocated enhancers also colocalized with regions of high Ubx concentration (Figure 3D). The relative Ubx enrichment for each of the three enhancers was 0.56 ± 0.16 for DG3 (n = 61, three embryos), 0.51 ± 0.19 for E3N (n = 142, 11 embryos), and 0.68 ± 0.10 for 7 hr (n = 38, three embryos) (Figure 3E–H,M,N). These results indicate that low-affinity enhancers actively transcribed far from the native svb locus also co-localize with microenvironments of high Ubx concentrations.

16Increasing the binding affinity of a site should increase its sensitivity to Ubx and allow transcriptional activation at lower Ubx concentrations. We found previously that replacing a single low-affinity Ubx site with one of a higher affinity led to higher levels of expression and sometimes drove promiscuous transcription, suggesting that more stable Ubx-DNA interactions allowed higher transcriptional activation. Consistent with these previous results, we observed that increasing the affinity of a single low-affinity binding site in the E3N enhancer decreased Ubx enrichment near transcription sites to 0.44 ± 0.27 (Figure 3I and J, E3N High Affinity, n = 36, three embryos).

17In contrast, we reported previously that deletion of low-affinity binding sites reduced transcription. Removing some Ubx-binding sites should lower the effective affinity of the enhancer, and we hypothesized that this might result in transcription only when genes are localized to areas of higher Ubx concentrations. Consistent with this model, when we deleted two low-affinity sites in E3N, active transcription was observed in regions of increased Ubx enrichment (0.65 ± 0.18, Figure 3K and L, E3N Mut23, n = 62, five embryos). Deletion of two low-affinity Ubx sites from the 7 hr enhancer did not alter Ubx enrichment around transcription sites (0.63 ± 0.37, Figure 3O and P, 7H Mut23 n = 81, six embryos). But, deletion of three Ubx-binding sites in the 7H enhancer increased relative Ubx enrichment, consistent with the pattern we observed for the E3N enhancer (0.91 ± 0.27, Figure 3Q and R, 7H Mut123, n = 52, eight embryos).

18Across all manipulations, we observed an inverse correlation between binding site affinity and the distribution of Ubx intensities at transcription sites (Figure 3—figure supplement 1). Thus, the number of Ubx-binding sites and their affinities determine the response of svb enhancers to local Ubx concentration. Lower affinity enhancers require higher Ubx concentrations to drive transcription. Conversely, higher affinity enhancers can drive transcription at lower local Ubx concentrations.

19Taken together, these data suggest that enhancers may be dynamically sampling local nuclear environments. A lower fraction of nuclei showing transcription from enhancers with binding site deletions (Figure 3K,O,Q) may occur because there are fewer areas of the nucleus in which peak Ubx levels are sufficient for weakened svb elements.

20Co-factors can stabilize low-affinity binding interactions through cooperative and scaffolding interactions with transcription factors. A co-factor-dependent enhancer would require sufficient concentrations of both the factor and the co-factor to drive transcription. The homeodomain proteins Extradenticle (Exd)/Pbx and Homothorax (Hth)/MEIS interact with Ubx during DNA binding, and Ubx and Hth regulate a partially overlapping set of genes. In vitro, Ubx requires Hth/Exd to bind to the low-affinity sites in 7H and E3N. In vivo, Hth deficiency led to the loss of expression for both 7H and E3N (Figure 4A–D). Consistent with this requirement for both Ubx and Hth, Hth was co-enriched with Ubx around active transcription sites driven by 7H or E3N (Figure 4E–T). The relative enrichment for Ubx and Hth, respectively, was 0.58 ± 0.14 and 0.41 ± 0.16 for 7H (n = 51, seven embryos) and 0.66 ± 0.13 and 0.39 ± 0.24 for E3N (n = 74, five embryos). These results suggest that transcription from co-factor-dependent enhancers requires microenvironments that contain high concentrations of both transcription factors and their co-factors. This observation provides further support for the model that transcription factor microenvironments are present as multiple highly differentiated transcription domains containing unique combinations of transcription factors.

21Biological systems often generate locally high concentrations of interacting molecules to increase the efficiency of biochemical reactions. This appears to be true also for transcription from low-affinity enhancers. Microenvironments of high local concentrations of transcription factors and their co-factors may circumvent the instability of low-affinity interactions by promoting more frequent DNA binding and cooperative interactions when enhancers are located within these domains (Figure 4U and V). These microenvironments may be relatively stable domains generated by rapid dynamics of individual molecules. For example, we observed interactions between transcription factors and DNA on the timescale of seconds, with transcription factors continuously arriving to and departing from specific loci. From the perspective of gene expression, transcription likely occurs intermittently, switching on and off as the gene locus samples different nuclear regions. These rapid dynamics ensure that, once the gene locus moves outside of a microenvironment, or the conditions to form microenvironments are no longer satisfied, then the transcription factors needed to sustain expression quickly depart from low-affinity binding sites. In contrast, the fact that svb enhancers placed on the third chromosome often co-localized with the native svb locus on the X chromosome suggests that unique microenvironments may have relatively long half-lives. One challenge for the future is to determine how rapid dynamics of individual molecules generates apparently stable sub-nuclear domains.

22Many mechanisms might work in concert to create these observed microenvironments. First, clustered binding sites for the same transcription factor could lengthen the dwell time of proteins near enhancers and increase effective local protein concentrations. Second, cooperative and scaffolding interactions between transcription factors and co-factors, each of which may bind independently to enhancers, can stabilize transcription factors at low-affinity sites. Finally, clustering of enhancers could trap transcription factors over longer length scales, perhaps generating the ~200 nm microenvironments that we observed. This last model is supported by recent findings that multiple promoters can share the same enhancer in a common local environment.

23Transcription factor microenvironments may be a general feature of eukaryotic transcription, as supported by studies showing mouse and human cells exhibiting RNA polymerase II crowding, transcription factors using local clustering to efficiently find their binding sites, and chromatin packaging in Drosophila cells generating distinct chromatin environments at the kilobase-to-megabase scale. Collectively, these findings are consistent with a phase-separated model of transcriptional regulation whereby distinct microenvironments contain different combinations of proteins inside the nucleus. These localized regions impose a spatial constraint on the expression of genes, allowing transcriptional activation from enhancers only when they are physically in regions with the correct combinations of transcription factors and co-factors. Multiple enhancers acting as DNA scaffolds for protein binding could provide the anchoring interactions that form transcriptional microenvironments. These microenvironments would, in turn, provide a mechanism to allow both efficient and specific transcription from low-affinity enhancers.

24D. melanogaster strains were maintained under standard laboratory conditions. All enhancer constructs were cloned into the placZattB expression construct with a hsp70 promoter. Transgenic fly lines were made by Rainbow Transgenic Flies Inc. E3 and 7H were integrated at the attP2 landing site. DG3 was integrated at ZH-86Fb.

25Flies were reared at 25°C and embryos were fixed and stained according to standard protocols. Primary antibodies were detected using secondary antibodies labeled with Alexa Fluor dyes (1:500, Invitrogen). In situ hybridizations were performed using DIG or biotin-labeled, antisense RNA-probes against a reporter construct RNA (lacZ) or the first intron of svb or fkh. DIG-labeled RNA products were detected with a DIG antibody: Invitrogen, 9H27L19 (1:200 dilution) and biotin-labeled RNA products are detected using a biotin antibody: Pierce, PA1-26792 (1:200).

26The following primary antibodies for proteins were used at the indicated concentrations:

27Ubx: Developmental Studies Hybridoma Bank, FP3.38-C (1:20)

28Hth: Santa Cruz Biotechnology (dN-19), sc-26186 (1:50)

29Eve: Developmental Studies Hybridoma Bank, 2B8-C (1:20)

30AbdA: Santa Cruz Biotechnology (dN-17), sc-27063 (1:50)

31En: Santa Cruz Biotechnology (d-300), sc-28640 (1:50)

32RNA PolII RPB1 (Ser5 phosphorylated): BioLegend, (920304), (1:200)

33Histone H3K27me3: Active Motif, 39157 (1:200)

34Histone H3K4me3: Cell-signaling technology C42D8 (1:200)

35LacZ: Promega anti-ß-Gal antibody (1:1000)

36Fixed Drosophila embryos mounted in ProLong Gold mounting media (Molecular Probes, Eugene, OR) were imaged on a Zeiss LSM 880 confocal microscope with Airyscan (Carl Zeiss Microscopy, Jena, Germany) using 3D Airyscan in SR mode to obtain images with 1.7-fold higher resolution compared to diffraction-limited confocal imaging (method supplements: imaging setup for Airyscan). Images presented in the figures were processed with ImageJ.

37To expand embryos, after fixation and staining, embryos were embedded into poly-acrylate gels and expended according to a previously published protocol (method supplements: handling expansion gels).

38Expanded gels containing embryos were imaged in 6-well glass bottom plates (Cellvis, Mountain View, CA) using a Zeiss LSM 800 confocal microscope (Carl Zeiss Microscopy, Jena, Germany) using standard settings (method supplements: imaging setup for expanded embryos).

39Transgenic fly lines containing HaloTag-Ubx under the control of both a hsp70 and a 20x UAS promoter was made by Rainbow Transgenic Flies Inc. The lines were made homozygous for the transgene.

40Embryos resulting from crossing the homozygous line with the HaloTag-Ubx transgene with a nos::GAL4 driver line were injected following previously established protocols with the HaloTag ligand of JF635. Briefly, embryos were collected for 30 min at 25°C and placed in oxygen permeable Halocarbon 27 oil. The stock dye solution of 1 mM JF635 with a HaloTag ligand in DMSO was diluted 1:100 into fly injection buffer and injected into the posterior end of the embryos. The embryos were then aged to stage 5 or late stage 6 and imaged in oxygen permeable Halocarbon 27 oil.

41Injected embryos were imaged on a customized inverted Nikon Ti-Eclipse (Nikon Instruments, Tokyo, Japan) with the appropriate settings (method supplements: imaging setup for live embryos).

42Embryos from the homozygous HaloTag-Ubx transgene line were exposed to 30°C to induce the heat shock promoter and cuticle preps were prepared following previously established protocols.

43To obtain the distributions of Ubx and Hth around a transcription site, the processed Airyscan stacks obtained from the Zeiss LSM 880 confocal microscope were analyzed in Fiji using native functions and the 3D ImageJ Suite plugin. Radially averaged distributions for individual transcription sites were computed using the 3D ImageJ Suite Plugin. Distributions for all sites were averaged and background offset in Matlab (MathWorks, Natick, MA) using a custom script (method supplements: settings for extracting radially averaged distributions).

44All Airyscan images were acquired using a Zeiss Plan-Apochromat 63x/1.4 Oil DIC M27 objective due to its well-characterized point spread function. First an embryo at the appropriate developmental stage (stage 15 for most embryos) and proper orientation was located. The band of mRNA expression in high Ubx regions of the first abdominal (A1) segment was then found. Within that band, areas containing transcription sites in nuclei of high Ubx expression were imaged. Images with both Ubx and Hth were acquired in the same manner by locating the proper area using the mRNA and Ubx. When Ubx was imaged together with RNA polymerase II, a histone marker, or other transcription factors, Ubx expression levels were used to locate the region of interest.

45The optimal setting suggested by Zeiss for the number of pixels in the x-y direction (40 nm per pixel) and displacement in the z-stack (190 nm) were used for all Airyscan images. The images from different fluorophores were acquired sequentially with the appropriate laser lines (405 nm, 488 nm, 561 nm, or 633 nm) and spectral filters. The laser power and gain were adjusted to maximize the signal to noise ratio within the dynamic range of the Airyscan detector. The acquired stacks were processed with Zen 2.3 SP1 (Carl Zeiss Microscopy GmbH, Jena, Germany) in 3D mode to obtain super-resolved images.

46To allow easier handling of expanded gels, the gels containing embryos were cast into eight-well silicone isolators without adhesives (eight round chambers with a diameter of 9 mm and a thickness of 0.5 mm, Grace Bio-Labs (Bend, OR)) and allowed to polymerize. The gels were transferred into a six-well glass-bottom cell culture plate (Cellvis, Mountain View, CA) and expanded using ultrapure water containing 500 nM DAPI. Before imaging, the water was removed and the gel encased in 3% low melting temperature agarose (NuSieve GTG Agarose, Lonza Group Ltd, Basel, Switzerland), taking care not to allow the agarose to flow under the gel and float the gel away from the cover glass bottom. Water was then added back into the wells to prevent drying.

47A long working-distance water immersion objective, the Zeiss LCI Plan-Neofluar 25x/0.8 Imm Korr DIC M27, was selected for index-matching with the gel and its ability to image up to 400 µm above the surface of the coverslip. Stage 15 embryos in the correct orientation were located using the DAPI and Ubx staining. Regions of low to high Ubx expression were imaged sequentially using the appropriate laser lines (405 nm, 488 nm, or 561 nm) with the proper spectral filters. The laser power settings and the gain were selected to maximize signal to noise within the dynamic range of the detector. The full field of view of the microscope was imaged with 2048 × 2048 pixels and with a z-step of 1 μm. The final images presented were processed in ImageJ.

48All videos were collected under a Nikon CFI Plan Apo NCG 100X Oil NA 1.41 objective with an Andor iXon 897 EMCCD camera (Andor Technology Ltd., Belfast, UK). Embryos at stage 5 and late stage 6 in the correct orientation were found and imaged. We selected an area in the middle of the embryo with enough dye-labeled Ubx molecules to observe single molecules and we avoided regions close to the injection site to avoid oversaturating the camera (compare with Figure S5A-D where there are too many labeled Ubx). The samples were illuminated with a 633 nm laser to image the JF635 tagged Halo-Ubx molecules with laser power and camera gain set to maximize signal from individual Ubx molecules without oversaturating the EMCCD detector. The 512 × 512 pixel videos were acquired at an exposure time of 100 ms per frame for up to 200 s. Images were processed using ImageJ to generate the time-averaged images and the intensity-over-time traces presented in the figures.

49To extract radially averaged protein distributions, we used Fiji to identify transcription sites inside nuclei by thresholding at a level that is roughly 50-fold above the background intensity. The center of a transcription site was defined as the pixel of maximum intensity in 3D in the mRNA channel inside a nucleus with high levels of Ubx expression. The radially averaged distribution out to a radius of 4 μm from transcription site for the transcription factor in 3D was computed using the 3D ImageJ Suite. The suite generates the distribution by computing the average intensity on the surface of a sphere with a radius r from the center in three dimensions for all the values of r ranging from zero to a desired outer limit (4 μm in this case).

50The individual distributions from each transcription site were normalized to have the intensity at the center (r = 0) equal to 1. The distributions were averaged and background offset in Matlab. To adjust for background Ubx intensity outside of the nucleus, the entire averaged distribution was offset by a constant value to bring the minimum intensity present in the distribution to zero to generate the distribution plots. The shaded area around the line represents the variance. The first μm of the distributions, where contributions from outside of the nucleus were minimal, are shown in the figures. The relative enrichment of Ubx or Hth for each enhancer variant is the intensity at r = 0 in the distribution and the cited uncertainty is the variance at the location of zero Ubx or Hth intensity (the site of minimum intensity before offsetting, between 2 and 4 μm from the transcription site).

51The initial dataset for 7H enhancers contained only a part of the deletion series. A subsequent dataset contained all the 7H deletion mutants. The 7H mutants present in both sets were compared and the distributions of Ubx intensity between the sets were found to differ by a multiplicative factor. When such factor was computed for each overlapping 7H mutant present in both datasets, the results were similar, indicating that there was a systematic shift in background noise. This could have resulted from differences in embryo handling during fixation, antibody staining, and other steps in sample preparation. Other characteristics such as the functional form of the distributions between the two sets and the trends between 7H mutants within each set remained unchanged after correcting for the difference in intensity. The wild-type 7H data from the first set with a correction factor and the rest of the deletion series uncorrected from the second set were used to minimize the normalization employed.