Multiple allelic configurations govern long-range Shh enhancer-promoter communication in the embryonic forebrain

SUMMARY

Developmental gene regulation requires input from enhancers spread over large genomic distances. Our understanding of long-range enhancer-promoter (E-P) communication, characterized as loops, remains incomplete without addressing the role of intervening chromatin. Here, we examine the topology of the entire Sonic hedgehog (Shh) regulatory domain in individual alleles from the mouse embryonic forebrain. Through sequential Oligopaint labeling and super-resolution microscopy, we find that the Shh locus maintains a compact structure that adopts several diverse configurations independent of Shh expression. One configuration contained most of all distal E-P contacts and came at the expense of those more proximal to Shh, consistent with an interconnected loop. Genetic perturbations demonstrate that this long-range E-P communication operates by Shh expressionin-dependent and dependent mechanisms, involving CTCF binding sites and active enhancers, respectively. We propose a model whereby gene regulatory elements secure long-range E-P interactions amid an inherent architectural framework to coordinate spatiotemporal patterns of gene expression.

eTOC blurb

Harke et al. use sequential DNA-FISH to achieve single-allele resolution of the Sonic hedgehog (Shh) regulatory domain in the mouse forebrain. They show that multiple allelic configurations occur independently of Shh expression, while CTCF binding sites are critical for enhancer-promoter interactions and Shh expression within a compact chromatin structure.

Graphical Abstract:

INTRODUCTION

Mammalian gene expression requires precise communication between enhancers and their target promoters that frequently span large genomic distances.^1–6 Despite 100s to 1000s of kilobases of intervening DNA, experiments measuring long-range enhancer-promoter (E-P) interactions by DNA fluorescent in situ hybridization (DNA-FISH) often find close physical proximity between these regulatory elements.^7,8 Similarly, sequencing-based chromosome conformation capture methods reveal high-frequency E-P contacts relative to other pairs of neighboring DNA.^9–11 These data support a prevailing model in which the physical looping of distal DNA elements facilitates E-P communication.^12–14 However, this model may overlook the organization of chromatin between the E-P pair and what role it may play in mediating gene expression. Moreover, pairwise data cannot distinguish among individual looping events in a population of alleles or multiway interactions among single alleles. Recent data emerging from multiplexed and live imaging experiments in single alleles paint a more variable and dynamic landscape of chromatin organization, warranting reevaluation of the strict looping model.^15–26 Notably, developmental genes typically respond to multiple enhancers to elicit spatiotemporal expression patterns^2,27–33 and do not always abide by a simple looping paradigm.^17,34 These findings are largely substantiated upon ensemble data of entire cell populations, either from cell culture models or dissociated tissues, removed from their developmental context. Hence, embryonic gene expression may hinge upon a more complex orchestration of E-P communication events within developing tissues.

An alternative model proposes that multiway 3-dimensional (3-D) interactions poise loci for gene expression and restrict E-P communication within insulated domains. Supporting this model, architectural proteins like CTCF bind near enhancers, promoters, and regions flanking mapped gene regulatory units. The propensity of CTCF to cluster enables long-range interactions across short length-scales, and renders the gene regulatory unit robust to perturbations of select CTCF binding sites.^35–40 Formation of the regulatory unit precedes gene expression and is largely maintained across different cell and tissue types suggesting that at least some components provide uniform structural stability.^8,37,41,42 These observations have led to a hypothesis that CTCF provides the framework underlying E-P communication. While global depletion of CTCF in cultured cells dramatically reduces chromatin interactions, the effect on gene expression is less pronounced indicating that E-P communication may be retained through other mechanisms.^16,43–47 In particular, determining the role of enhancers in locus topology remains challenging due to the paucity of loci with fully mapped and functionally validated enhancer landscapes.

Herein, we use the well-characterized Shh locus to interrogate the topologies of single alleles in the mouse embryonic forebrain. Employing three super-resolution microscopy modalities, we measured precise nanoscale metrics of chromatin and found that the Shh locus is compact across forebrain regions. Sequential DNA-FISH experiments reveal that multiple Shh locus configurations underlie the ensemble depiction of heatmaps aggregated from thousands of alleles. We found that equal proportions of configurations exist in Shh-expressing and non-expressing forebrain regions, and that the dominant structure exhibited multiway interactions throughout the locus with subtle shifts in a distal E-P contact. We propose that developmental genes, like Shh, may prefer this condensed arrangement since they rely on a multitude of E-P interactions to direct their spatiotemporal patterns of expression. Further, our genetic perturbation experiments demonstrate that long-range gene regulation depends on a combination of CTCF binding sites and active enhancer sequences. Our results implicate a model whereby E-P communication is securely executed by the collective architectural framework of chromatin interactions and enhancer activities that further stabilize relevant connections for gene expression.

RESULTS

The Shh regulatory domain confers a compact structure independent of Shh expression

The Shh locus embodies model features for dissecting long-range E-P communication. Each of the mapped Shh regulatory elements lie within the bounds of a ~1-Mb topologically associating domain (TAD), flanked by convergent CTCF binding sites (Figure 1A–C).^37,42 Enriched distal interactions in the form of a corner dot appear across multiple Hi-C datasets of different cell and tissue types, suggesting the presence of an invariant loop structure (Figures 1A and S1A).^48–50 Coincident with this distal interaction are Shh Brain Enhancers 1 and 5 (SBE1/5), which co-regulate Shh expression in the zona limitans intrathalamica (zli), a signaling center that partitions the thalamic and prethalamic regions in the mouse embryonic forebrain (Figure 1B–E).^2,51,52 Shh expression first appears in the zli around embryonic day 10 (E10.0), reaches maximal expression at E12.5, and is no longer detected by E14.5.^51–54

Figure 1. Compaction of the Shh locus occurs independent of Shh expression.

(A) Contact frequency matrix of the Shh locus as depicted by Hi-C data from mouse embryonic stem cells (mESCs; Dixon et al., 2012⁴²). Dashed box indicates enriched distal interactions with the bin containing Shh. (below) Oligopaint probe design for STORM imaging of the Shh domain (magenta) and its neighboring 5’ domain (cyan). CTCF ChIP-seq from E10.5 midbrain (Paliou et al., 2019³⁷). (B) Schematic of Shh expression in the mouse embryonic forebrain at E12.5 (gray) and sites of SBE1 and SBE5 activity in the ventral midbrain, caudal diencephalon and zli (orange). Dashed line indicates the coronal plane of section through the forebrain used in this study. (C) Distribution of Shh enhancers within the Shh domain including the positions of SBE1 and SBE5 (magenta) relative to the Shh promoter. (D) Coronal section through a wildtype mouse embryonic forebrain (E12.5) co-labeled for Shh-GFP in the zli (green) and Olig2 in the surrounding thalamic and prethalamic regions (red). White box indicates the region of interest for this study. (E) Single-molecule RNA-FISH quantification of Shh mRNA in the zli (orange) and prethalamic region (violet) of wildtype embryos (E12.5). Two independent biological replicates are plotted. (F) STORM images of the 5’ domain (cyan) and Shh domain (magenta) showing single-molecule (above) and corresponding convex hull (below) representations. Red, green and blue arrows indicate x-, y- and z-planes, respectively. Scale bar = 500nm. (G) Quantification of domain volumes (μm³) normalized to genomic size (bp) across forebrain regions. zli: median 5’ domain=1.64×10⁻⁷, median Shh domain=1.19×10⁻⁷; prethalamus: median 5’ domain=1.82×10⁻⁷, median Shh domain=1.09×10⁻⁷ (****P<0.0001, two-tailed Mann-Whitney U-test, n = number of alleles). Data are from two biological replicates combined, with independent replicates showing the same effect (see Figure S1).

To determine if the 3-D morphology of the Shh TAD differed with Shh expression, we generated Oligopaint probes that span the entire Shh domain and another that labels the neighboring 5’ domain.^55–57 We then performed DNA-FISH on coronal sections through the mouse forebrain at E12.5, coinciding with peak Shh expression in the zli. Olig2 immunostaining was used to distinguish the Shh-expressing zli (Olig2⁻) from surrounding forebrain regions (Olig2⁺) that do not express Shh (Figure 1D–E). We used Stochastic Optical Reconstructive Microscopy (STORM)⁵⁸ to generate single-molecule super-resolution images and measured the precise volumes of each domain, normalizing for their corresponding genomic content (Figures 1F–G and S1B–C). The Shh domain was significantly more compact and exhibited less variance than the neighboring 5’ domain in both the Shh-expressing zli and non-expressing prethalamus, 4.25-fold and 3.4-fold, respectively (****P<0.0001, Figures 1G and S1C, Table S6B). These observations suggest that the Shh locus maintains a relatively compact and restricted topology independent of Shh expression.

Sequential DNA-FISH uncovers multiple allelic configurations for the Shh domain

To resolve the sub-structure of the Shh domain, we performed sequential DNA-FISH^59–61 and mapped the 3-D configuration of the Shh locus in the mouse embryonic forebrain (Figure 2A). We divided the Shh TAD into 32 30-kb segments (hereafter referred to as “bins”; Figure 2B). Sequential labeling and imaging allowed for super-resolution localization of each 30-kb bin within the gross morphology of the Shh domain. Two independent biological replicates showed a high correlation between the DNA probe hybridization efficiencies (r=0.63) and their median distance matrices (Figure S2A–D). Over 80% of the median distance measurements exhibited less than 50 nm difference between replicates; therefore, the data were pooled into one larger dataset for subsequent analyses (Figure S2D–F).

Figure 2. Sequential DNA-FISH analysis of the Shh locus in the mouse embryonic forebrain.

(A) Schematic of sequential DNA-FISH probe design and protocol. The primary probes also contained the fiducial signal. Rainbow pseudocoloring of 30-kb bins illustrates the order of bridge-secondary oligo hybridization (hyb) and imaging. The “hyb 1” signal is subsequently removed by strand-displacement and the next segment (bin) is illuminated. The process is repeated iteratively for n desired hybs until the entire locus has been imaged. All images were collectively aligned to a fiducial signal prior to measuring the nanometric distances for each individual allele. The distances were then tabulated into contact frequencies at a set distance threshold of 200 nm. (B) Ensemble contact frequency heatmap of sequential DNA-FISH alleles (112 ≤ n ≤ 2,351; n = number of allelic distances). Red shading is separated by the data quartiles. Dashed bracket indicates enriched distal interactions with the bin containing Shh. (C) Virtual 4C-like plot of sequential DNA-FISH data with the bin containing Shh as the anchor. Dashed line indicates the ensemble Shh domain minimum (i.e. lowest detected) contact frequency. The bins containing the SBE/ZRS enhancers (squares) and CTCF sites (circles) are indicated below. SBE1 and SBE5 are highlighted in magenta (bins 1 and 27, respectively). Bins with sites deleted in subsequent experiments are underlined (see Figure 5).

On average, we collected 1,109 distance measurements for each pairwise combination of loci. Median distances ranged from 201 – 317 nm, consistent with our initial finding that the locus largely maintains a compact topology (Figure S2F). To identify enriched interactions within the compacted locus, we calculated contact frequencies based on a distance cutoff of 200 nm. The ensemble contact frequency heatmap created from our sequential DNA-FISH data resembled key features found in Hi-C datasets from other cell and tissue types, including the enrichment of distal interactions with the Shh gene region (Figures 2B–C, 1A, S1A and S3A).^48–50 These distal contacts corresponded to bins containing long-range enhancers, including SBE5 (bin 27) and an SBE5-proximal cluster of CTCF sites (bins 28–30; Figure 2C). In particular, the contact frequency between the bins containing Shh and SBE5 was 24% across all alleles, compared to 16% for a region 450-kb closer to the Shh gene (bins 27 and 11, respectively; Figure 2C). Taken together, the ensemble sequential DNA-FISH heatmaps unveiled a compact and interconnected locus with patterned interactions occurring throughout.

To delineate the multiway interactions among individual alleles, we created virtual 4C-like plots, by anchoring all distances to the bin containing Shh. We performed a principal component analysis (PCA) and k-means clustering to define five groups of allelic configurations (Figure 3A–B). As expected, the individual Shh alleles were relatively compact yet displayed differential enrichments of contacts within each group (Figures 3C–E and S3B–D). The most abundant configuration, denoted Group 1, represented 40.4% of alleles and exhibited several multiway contacts across the locus, consistent with a highly compacted structure. Group 2 (27% of alleles) had enriched contacts near the bin containing Shh, in contrast to Group 3 (18.2% of alleles), which harbored more distal contacts. Finally, Group 4 (12.2% of alleles) and Group 5 (2.2% of alleles) showed few scattered or no contacts, respectively, and represented the minority of alleles and an open configuration. Hereafter, we focused on Groups 1–4 that contained a minimum of 30 alleles each.

Figure 3. The Shh locus displays multiple configurations independent of Shh expression.

(A) Elbow plot to determine the number of defined clusters. (B) PCA plot colored by 5 defined clusters (groups). (C) (left) Proportion of individual alleles within a given cluster: Group 1 (n=109), Group 2 (n=73), Group 3 (n=49), Group 4 (n=33), Group 5 (n=6); n = number of alleles. (right) Contacts with the bin containing Shh for individual alleles in Groups 1–4 (distances < 200 nm, red). Note the contact frequency for Group 5 is zero across the locus. (D) Contact frequencies (distances < 200nm) quantified as a percentage of the grouped alleles that exhibit contact with the bin containing Shh. (E) Ball-and- stick representations of allelic configurations for each Group. (F) (left) Schematic depicting the region of interest (ROI, yellow box) on a coronal section (x) through the embryonic forebrain at the level of the zli (black). (middle) Representative field of view for the adjacent tissue section (x+1) showing the sequential DNA-FISH fiducial signal. A combination of gross ventricular morphology and intronic Shh RNA-FISH signal on the immediately adjacent tissue section (x) were used to delineate the zli on the tissue section used for sequential DNA-FISH (x+1). (right) Group configurations mapped back to the zli and non-zli regions of the embryonic forebrain. (G) Proportions of configurations in the zli and non-zli were not significantly different (two-sided Fisher’s Exact test, see Table S2). (H) Distances between the bins containing Shh and SBE5 (magenta) are indicated from the allelic configurations shown in (E). (I) Fraction of alleles across the groups where the bins containing Shh and SBE5 are in contact. Group 1 (n=52), Group 2 (n=3), Group 3 (n=12), Group 4 (n=1), where n = number of alleles. (J) Differences in contact frequency between Group 1 subgroups: those with SBE5-contact (*) versus those without SBE5-contact (n = 52 and 57 alleles, respectively). The location of bins containing SBE/ZRS enhancers (gray squares) and CTCF binding sites (red circles) are indicated below. Bins containing SBE1 and SBE5 are highlighted in magenta. (K) Models for how locus compaction may facilitate distal interactions. Linear: distal interactions depend upon intermediate contacts along the length of the locus. Displaced: distal interactions occur at the expense of more proximal contacts.

To determine if the configurations differed with Shh expression, we mapped Groups 1–4 back onto the imaged tissue and quantified the abundance of each configuration (Figure 3F). We found all configurations across each forebrain region with no significant differences in their proportions (Figure 3G and Table S2). These findings indicate that the various Shh locus configurations are not dependent on Shh expression.

Distal enhancer-promoter interactions are associated with a shift in configurations within a compact Shh locus

We next focused on the specific contact between bins containing Shh and the distal zli enhancer, SBE5. Representative images of each allelic configuration expose the multiway compaction of Group 1, while Group 3 appears to have a more selective interaction around these bins of interest (Figure 3H and S3D). Groups 1 and 3 exhibited the greatest contact frequencies among the bins containing Shh and SBE5 (47.7% and 24.5%, respectively; Figure S3E and Table S2). Moreover, Group 1 accounts for the overwhelming majority of all the alleles with contact between the bins containing Shh and SBE5 (76.5%, Figure 3I). Of the Group 1 alleles, approximately half exhibit contact between the bins containing Shh and SBE5 (48%, Figure S3F). We therefore divided the Group 1 alleles into SBE5-contacting and SBE5 not-contacting subgroups and then compared the subgroups against each other by subtracting their Shh-anchored contact frequency profiles. We found that SBE5-contact comes with an enrichment of local contacts immediately surrounding the enhancer; whereas in the absence of SBE5-contact, there is a subtle enrichment of contacts spread along regions more proximal to the bin containing Shh (Figure 3J). This configuration more closely resembles that of Group 3 and indicates a relative tradeoff: contact with distal SBE5 emerges at the expense of Shh-proximal interactions. Indeed, we performed hierarchical clustering of each allele based on their distances to the Shh bin and found that Group 3 represents a transition state between Groups 1 and 2 (Figure S4). Hence, rather than linear compaction of the locus, our single allele configurations favor a model of shifting interactions within a confined space, where a strong distal interaction displaces the intervening chromatin (Figure 3K).

The zli displays enhanced spatial proximity between the distal enhancer and Shh promoter

To interrogate the distal interaction at the Shh locus with greater precision, we designed higher-resolution Oligopaints that labeled 10 kb surrounding each of the partially redundant zli enhancers (SBE1 and SBE5) and the Shh promoter. We then measured the distances between select probe pairs using super-resolution structured illumination microscopy (SIM, Figure 4A).^62,63 The distance between the Shh promoter and SBE1, located in the second intron of the Shh gene, was the closest of all E-P interactions examined (median _SBE1 = 188nm; inter-probe distance = 7 kb; Figures 4B and S5A–C, and Table S3). Consistent with our sequential DNA-FISH data, the Shh promoter and SBE5 also exhibited short distances despite their greater genomic separation (median _SBE5 = 237nm; inter-probe distance = 769 kb). For comparison, an inactive enhancer (SBE2) showed relatively large spatial distances to the Shh promoter despite being nearly half the genomic distance as SBE5 (median _SBE2 = 416nm; inter-probe distance = 394 kb). SBE5 was closer to the Shh promoter and exhibited greater contact frequency in the zli, where Shh is expressed, compared to the non-expressing prethalamus, suggesting that SBE1/5 activity is associated with a small but significant increase in distal E-P interactions (zli = 37%, prethal. = 30%, **P=0.0012; Figures 4C–D and S5C, and Table S3). Only SBE5 exhibited zli-specific enhanced spatial proximity with the Shh promoter, compared to other interactions tested, including an equidistant control (“−800 kb”) probe and between the two Shh TAD boundaries (Figures S5A–C and S6, and Table S3).

Figure 4. Proximity between the Shh promoter and SBE5 is greater in the zona limitans intrathalamica.

(A) Schematic depiction of the relationship between the sequential DNA-FISH (30-kb) probe set, precise genetic elements, and high-resolution Oligopaint probes (10-kb) used to measure pairwise distances between the Shh promoter and SBE1, 2, and 5. (B) Cumulative distributions of pairwise distance measurements in the zli. Distances between the Shh promoter and SBE1 (teal; median=188nm, n=393), SBE2 (gray; median=416nm, n=817), and SBE5 (magenta; median=237nm, n=1091) ****P<0.0001; two-tailed Mann-Whitney U-test, n = number of alleles. Data are from two biological replicates combined, with independent replicates showing the same effect (see Figure S5A–C). (C) Comparison of contact frequencies between the Shh promoter and SBE1 (teal; zli: 56%, non-zli: 54%), SBE2 (gray; zli: 7.2%, prethalamus: 6.6%), and Shh promoter and SBE5 (magenta; zli: 37%, prethalamus: 30%) **P=0.0012; two-sided Fisher’s Exact test, ns = not significant. Data are combined from two biological replicates, with independent replicates showing the same effect (see Figure S5A–C and Table S3). (D) Representative images of two DNA-FISH probes with distances in contact (< 200 nm) and not in contact (> 200 nm). Scale bar = 500 nm.

We next evaluated how these E-P interactions associate with Shh activation for individual alleles within the zli by combining 3-color DNA-FISH (SBE1, SBE5 and Shh promoter) with Shh nascent RNA-FISH (Figure S5D–J). SBE1 was significantly displaced from the Shh promoter at transcribing alleles compared to non-transcribing alleles (Figure S5F, H–I). In contrast, SBE5 remained in close proximity with the Shh promoter regardless of whether Shh was being actively transcribed (Figure S5G–H, J). Taken together, these data demonstrate an association between E-P interactions and Shh expression in the zli, while underscoring the existence of a Shh expression-independent locus structure.

Shh transcription mediates distal enhancer-promoter proximity in the zli

We next sought to identify the sequence features governing the enriched distal E-P interaction between the Shh promoter and SBE5. Two genomic elements stand out as putative regulators of this interaction: the active enhancers themselves and a cluster of CTCF binding sites adjacent to the distal enhancer, SBE5 (Figure S7A). We first asked whether the active zli enhancers, SBE1 and SBE5, collectively contribute to the distal E-P interaction. Deletion of both elements (SBE1/5^−/−, Figure 5A) increased the distance between the 10 kb probes overlapping the Shh promoter and the SBE5 deleted region (Δ2kb) by 40nm (medians: zli ^WT = 212nm, zli ^{SBE1/5 −/−} = 252nm, ****P<0.0001) and decreased the contact frequency by an average of 9.8% specifically within the zli (zli ^WT = 45.1%, zli ^{SBE1/5 −/−} = 35.3%; Figures 5B–C and S7B, and Table S5). Of note, the increased distances in the SBE1/5^−/− mutant zli resulted in a distribution indistinguishable from that observed in the non-Shh expressing wildtype prethalamus (medians: zli ^{SBE1/5 −/−} = 252nm, prethal ^WT = 252nm, ns; Figure 5B). Surprisingly, the individual deletion of either SBE1 or SBE5 did not affect this distal E-P interaction (Figure S8). These findings suggest that the increased contact between SBE5 and the Shh promoter in the zli, compared to the prethalamus, is not dependent on the individual enhancers per se. Instead, they imply that Shh transcription further promotes long-range E-P interactions, above the default contact probability.

Figure 5. Contributions of active enhancers and CTCF sites to distal E-P interactions and Shh expression.

(A) Schematic representation of SBE1 and SBE5 deletion mutants (SBE1/5^−/−). (B) Cumulative distributions of pairwise distance measurements between high-resolution DNA-FISH probes (10-kb) overlapping the Shh promoter and SBE5 in wildtype (WT) and SBE1/5^−/− forebrain tissues. WT zli (solid orange; median=212nm, n=496); SBE1/5^−/− zli (dotted orange; median=252nm, n=382); WT prethalamus (solid violet; median=252nm, n=456); SBE1/5^−/− prethalamus (dotted violet; median=252nm, n=449) ****P<0.0001; two-tailed Mann-Whitney U-test, ns = not significant, n = number of alleles). Data are combined from two biological replicates, with independent replicates showing the same effect (see Figure S7B). (C) Contact frequencies for the data in (B). Each dot represents a biological replicate; independent and combined replicates show the same trends (see Table S5). Average contact frequencies: zli +/+ (45.1%); zli −/− (34.3%); PTh +/+ (33.4%); PTh −/− (36.8%). Contact (distances < 200nm); PTh = prethalamus. (D) Schematic representation of CTCF binding site deletions (i4:i5:zrs:i9) adjacent to SBE5. (E) Cumulative distributions of pairwise distance measurements between high-resolution DNA-FISH probes (10-kb) overlapping the Shh promoter and SBE5 in WT and CTCF^i4i5zrsi9 forebrain tissues. WT zli (solid orange; median=237nm, n=1091); CTCF^i4i5zrsi9 zli (dotted orange; median=286nm, n=640); WT prethalamus (solid violet; median=264nm, n=898); CTCF^i4i5zrsi9 prethalamus (dotted violet; median=327nm, n=438) ****P<0.0001; two-tailed Mann-Whitney U-test, n = number of alleles). Data are combined from two biological replicates, with independent replicates showing the same effect (see Figure S7C). (F) Contact frequencies for the data in (E). Each dot represents a biological replicate; independent and combined replicates show the same trends (see Table S5). Average contact frequencies: zli +/+ (37.7%); zli −/− (25.7%); PTh +/+ (30.0%); PTh −/− (19.5%). Contact (distances < 200nm); PTh = prethalamus. (G) Single molecule RNA-FISH quantification of Shh mRNA using an exonic probe (Shh_ex) in the zli of WT and SBE1/5^−/− mutants. Data are normalized to the WT zli expression. Each dot represents a biological replicate. Average Shh mRNA expression in SBE1/5^−/− mutants is 3.3% of WT. (right) Representative images showing Shh_ex in the zli (dashed white line). Scale bar = 10 μm. (H) Intronic RNA-FISH quantification of bursting Shh alleles using an intronic probe (Shh_in) in the zli of CTCF^i4i5zrsi9 and SBE5^−/− single enhancer deletion mutants. Bursting fractions were normalized to their respective WT zli for comparison across genotypes (see Figure S7F for raw values). Each dot represents a biological replicate; independent and combined replicates show the same trends (see Table S5). Average mutant expression relative to WT: CTCF^i4i5zrsi9 (30.6%) and SBE5^−/− (38.6%). (right) Representative images showing Shh_in in the zli (dashed white line). Scale bar = 10 μm.

CTCF sites adjacent to SBE5 license long-range enhancer-promoter proximity independent of Shh expression

Next, to evaluate whether the CTCF binding sites neighboring SBE5 independently contribute to the distal E-P contact, we deleted four sets of CTCF binding sites (“i4:i5:zrs:i9”, underlined in Figure 2C) and measured the distal E-P distances (Figure 5D). Loss of the CTCF sites significantly increased the Shh promoter to SBE5 distances within both the zli and the prethalamus by 49nm and 63nm, respectively (medians: zli ^WT = 237nm, zli ^i4:i5:zrs:i9 = 286nm, ****P<0.0001; prethal. ^WT = 264nm, prethal. ^i4:i5:zrs:i9 = 327nm, ****P<0.0001; Figures 5E–F and S7C, and Table S5). Likewise, we observed reduced contact frequency in the mutant zli and the mutant prethalamus by 12% and 10.5% on average, respectively (zli ^WT = 37.7%, zli ^i4:i5:zrs:i9 = 25.7%, prethal. ^WT = 30%, prethal. ^i4:i5:zrs:i9 = 19.5%; Figure 5F and Table S5) indicating that these CTCF sites play an integral role in licensing the distal E-P interaction across forebrain regions independent of Shh expression. The distance between the Shh promoter and SBE2 remained unaffected in the CTCF mutant, indicating that the CTCF sites specifically uncouple the long-range E-P interaction without affecting the general locus architecture (Figure S7D and Table S5). We corroborated this finding among individual alleles with all three sequence elements co-labeled (Shh promoter – SBE2 – SBE5), demonstrating that the loss of contact with the Shh promoter occurs specifically with SBE5, in both the zli and prethalamus of CTCF mutant embryos (Figure S7E, compare yellow to green groups).

To determine the effects on Shh expression from altered E-P contact frequencies, we measured both the fraction of bursting alleles in the zli and the mRNA product by single-molecule RNA-FISH, using intronic and exonic probes, respectively. We consistently observed that ~20% of wildtype alleles were bursting within the zli (CTCF ^WT = 22.3%, SBE5^+/+ = 21%, SBE1^+/+ = 19.8%; Figure S7F). Deletion of both SBE1 and SBE5 resulted in near complete loss of Shh expression in the zli (96.7% reduction of wildtype level; Figure 5G and Table S5). We then examined if loss of the SBE5-adjacent CTCF binding sites affected Shh expression in the zli, using a mutant null for the distal enhancer alone (SBE5^−/−) for comparison. Deletion of CTCF^i4:i5:zrs:i9 reduced Shh bursting by 69.4% on average, similar to the 61.4% reduction we observed in SBE5^−/− single enhancer mutants (Figures 5H and S7F, and Table S5). Total Shh mRNAs were also reduced by ~50% or more in both mutants (reduced expression by: CTCF^i4:i5:zrs:i9 = 78%, SBE5^−/− = 49.7%; Figure S7G). These results support a model in which CTCF sites license long-range E-P communication by facilitating distal E-P interactions.

DISCUSSION

Mammalian development requires the complex orchestration of E-P communication events. For developmental loci with elaborate regulatory inputs, such as Shh, multiple enhancers converge on the promoter to drive spatiotemporal gene expression patterns throughout the embryo. By labeling the entire Shh regulatory domain (i.e. TAD) across forebrain regions, we found the locus to be compact with low variability in size independent of Shh expression status (Figure 1). Using the STORM measurements, we calculated the median diameter across the Shh domain to be 315nm, indicating that the enhancers spanning ~1Mb are all within relatively close physical proximity to the promoter (Table S6A). This may help explain the ‘regulatory potential’ of TADs as observed by sensor assays, where independent reporter insertions at multiple locations throughout the Shh TAD responded equally to enhancer activities and mimicked the endogenous gene expression pattern.^11,64,65 Our data support the idea that enhancers’ regulatory reach is not limited by genomic distance and may be facilitated by their compact physical structure.

The looping model of E-P communication is predicated on an abundance of data identifying high-frequency contacts or physical proximity between pairs of loci and oversimplifies the configuration of the intervening chromatin. Many techniques bolstering the looping model fail to capture simultaneous, multiway interactions of individual alleles in expressing tissues. Using sequential DNA-FISH, we discovered diverse configurations that underlie a compact Shh locus (Figures 2–3). We found that only 18% of the allelic configurations resembled a classic loop structure, and of those, only 10% had a singular, exclusive contact with the bin containing Shh. Instead, the vast majority (77%) of alleles exhibited diverse multiway contacts along the entire length of the locus. Even within our most compact allelic configuration, we observed a toggling of distal E-P interactions at the expense of more proximal chromatin, an observation that may reconcile some conflicting models relating de-compaction to E-P activity at the Shh locus.^8,34 Our allelic configurations challenge the simple looping model by revealing a compact topology with multiple connections.

We found that the spectrum of allelic configurations spanned both Shh-expressing and non-expressing forebrain regions (Figure 3), suggesting that loci may sample different configurations awaiting further instruction from tissue-specific transcription factors. A recent study at the human SOX9 locus likewise identified variable configurations in a cell culture model of differentiation.²⁵ As fixed samples, our results are consistent with two possibilities: alleles may conform to one configuration over their lifetime, or alleles may cycle through configurations. Instead of a set of discrete configurations, we identified variability within groups of configurations as well as shared features between them as seen by hierarchical clustering of alleles. This observation is more consistent with a fluid population of alleles, a notion greatly supported by recent live imaging data showing that distal DNA elements actively move through a range of possible trajectories^26,66,67 (Figure 6, left). Whether the surveying of configurations is dependent on cohesin-mediated loop extrusion or a property of steady-state chromatin mobility, remains an exciting opportunity to explore.

Figure 6. Model depicting multiple Shh locus topologies with the sequence features governing 3-D architecture and gene expression.

(Left)The Shh locus samples four interconnected groups of allelic configurations in the mouse forebrain. Linear representation of the Shh locus with the small arrow representing the Shh transcription start site. SBE1/5 and CTCF sites are highlighted below in magenta and red, respectively. Arcing lines represent connections between the Shh promoter and other (distal) chromatin regions. Large arrows between configurations indicate the dynamic sampling of allelic configurations with the size of the arrow proportional to the frequency of configurations observed in the mouse embryonic forebrain. (Right) Model of distal E-P communication: CTCF licenses the locus for distal interactions while transcription factors (TFs), deposited at active enhancers, bolster E-P proximity and commission gene expression.

Underlying the assortment of configurations was an enriched distal E-P contact (Figures 3–4). We, and others, have found that E-P contact is greatest where the enhancer activity and gene expression align.^8,68 However, contact alone is insufficient to drive gene expression, as evidenced by the high contact observed between SBE5 and the Shh promoter in forebrain regions adjacent to the zli where Shh expression is silent. Rather, we propose that E-P communication requires a combination of Shh expression-independent and dependent mechanisms (Figures 4–5). In this context, CTCF would occupy a set of distal binding sites to bring SBE5 and the Shh promoter into close proximity in a manner independent of Shh-expression. In a transcriptionally permissive environment, such as the zli, Shh promoter engagement with SBE5 depends on distal CTCF occupancy and the transcription factor collective² bound at active enhancers to yield the Shh expression-dependent component of the proposed E-P communication mechanism. Our data add to a growing body of literature parsing out the intertwined contributions that support 3-D genome architecture and gene expression.^69–72 We suggest that the Shh locus employs a combination of structural and transcriptional features to ensure proper spatial regulation of gene expression (Figure 6, right). As loci sample different configurations, multiple inputs would safeguard against aberrant gene regulation. We envision that this model will apply to other developmental loci that depend on complex E-P interactions to coordinate spatiotemporal patterns of gene expression.

Limitations of the study

While our study provides significant insights into the range of allelic configurations and enhancer-promoter (E-P) communication within the Shh regulatory domain, several limitations should be acknowledged. First, our use of sequential DNA-FISH and super-resolution microscopy, while powerful, is limited by the resolution and efficiency of probe hybridization, which may result in an underestimation of certain chromatin interactions. Additionally, our study is based on fixed tissue samples, which precludes real-time observation of dynamic chromatin movements and the temporal progression of enhancer-promoter interactions during development. Finally, our findings are centered on the mouse embryonic forebrain, and it remains to be seen how these mechanisms generalize to other tissues or species. Future studies utilizing live-cell imaging and cross-species comparisons will be necessary to address these limitations and further elucidate the complexities of long-range gene regulation.

RESOURCE AVAILABILITY

Lead Contact

Requests for further information and resources should be directed to the Lead Contact, Eric F. Joyce (erjoyce@upenn.edu).

Materials Availability

All unique and stable reagents generated in this study are available from the Lead Contact without restrictions.

Data and Code Availability

• All processed data presented in this study are publicly available as of the date of publication with accession numbers and DOIs listed in the Key Resources Table.

• All original code has been deposited on GitHub (https://github.com/melikelakadamyali/SequentialDNA-FISH) and is publicly available (https://doi.org/10.5281/zenodo.13958549).

• Any additional information required to reanalyze the data reported in this paper is available from the Lead Contact upon request.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal anti-Olig-2	Millipore	Cat#AB9610
Rabbit polyclonal anti-Olig2	Novus	Cat#NP1-28667
Goat anti-Rabbit AlexaFluor-488	Invitrogen	Cat# A11008
Chemicals, peptides, and recombinant proteins
Antibody Diluent Reagent Solution	LifeTechnologies	Cat#003118
Formamide	EMD, Millipore Sigma, VWR, Fisher	Cat#3442061L, NC9569627, 97062010, AM9342
Dextran sulfate	Sigma	Cat#D8906-100G
RNase A	Fisher	Cat#FEREN0531
SlowFade Gold	Fisher	Cat#536936
Polyvinylsulfonic acid (PVSA)	Sigma	Cat#278424-250ML
Deposited data
Hi-C from mESCs	Dixon et al.42	GSE35156
CTCF ChIP-seq from E10.5 midbrain	Paliou et al.37	GSE123388
Hi-C from whole mouse brain	Deng and Ma et al.50	GSE59779, GSE68992
Hi-C from mESCs	Bonev et al.48	GSE96107
Hi-C from activated B-cells	Kieffer-Kwon, Nimura, Rao, Xu et al.49	GSE82144
Experimental models: Cell lines
Mouse ESCs	Paliou et al.37	N/A
Mouse ESCs with CTCFi4:i5:zrs mutation	Paliou et al.37	N/A
Experimental models: Organisms/strains
Mouse: SBE1+/−; SBE5+/−	Yao et al.2	N/A
Mouse: SBE1+/−	Jeong et al.51	N/A
Mouse: SBE5+/−	Yao et al.2	N/A
Oligonucleotides
sgRNA: CTCF_Lmbr1_i9_cenF: AGGGCGTCAGGAAATTCCAC	This study	N/A
sgRNA: CTCF_Lmbr1_i9_telF: TGAACTGCCAATCACCTGGG	This study	N/A
Oligopaint probes	This study; Custom Array, Twist Bioscience	Table S1; sequences available upon request
Secondaries (fluorescent oligos)	IDT	Sequences available upon request
Bridge oligos	IDT	Sequences available upon request
Strand-displacement oligos	IDT	Sequences available upon request
RNA-FISH probes	Stellaris	Sequences available upon request
Software and algorithms
CRISPR Design Tools	Zhang lab	http://www.genome-engineering.org/crispr/
CRISPR Guide RNA Design Tool	Benchling	https://benchling.com/
OligoMiner	Beliveau et al.57	N/A
Bruker SRX	Bruker	N/A
DBScan	Ester et al.76	N/A
Prism (9.2.0)	Graph Pad	N/A
Custom-built MATLAB (R2022b) analysis pipeline	https://github.com/melikelakadamyali/SequentialDNA-FISH	https://doi.org/10.5281/zenodo.13958549
Adobe Suite	Adobe	N/A
Python (2.7.15): Average_Clustering.py, Imaris_min- dist.py	This study	Scripts available upon request
MetaMorph	Molecular Devices	N/A
Imaris (9.3.1)	Oxford Instruments	N/A

STAR METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice

SBE1^+/−; SBE5^+/− double-heterozygous mice, SBE1^+/− and SBE5^+/− mice were housed at the University of Pennsylvania in accordance with the ethical guidelines set forth by the National Institutes of Health and approved by the Institutional Animal Care and Use Committee of the University of Pennsylvania (Protocol #: 803565). Heterozygous mice were intercrossed in to generate homozygous mutant embryos. Littermate WT embryos (SBE1^+/+; SBE5^+/+, SBE1^+/+ and SBE5^+/+) were used as the respective controls. Timed evening breeding and subsequent detection of a mucous plug, assumed conception at midnight (E0). Embryos were harvested at E12.5. Embryos were not sexed due to the early developmental timepoint.

CTCF ^i4:i5:zrs:i9 mice were generated at the University of Geneva in accordance with institutional, state, and government regulations (GE/196/19). Briefly, mouse embryonic stem cells (mESCs) were CRISPR-engineered to generate the desired ‘i9’ mutation on top of the existing CTCF^i4:i5:zrs mutant allele³⁷ (see below). Embryos were generated from mESCs using tetraploid complementation. Blastomeres of harvested two-cell stage embryos were fused by an electric pulse to create tetraploid embryos. Concurrently, mESCs were cultured, trypsinized into small clumps, and then aggregated with tetraploid embryos. The WT littermate controls were generated from WT mESCs aggregated in parallel. Upon successful aggregation, blastulas were implanted into pseudo-pregnant adult female surrogates. Embryos were harvested after 12.5 days gestation, fixed, and shipped overnight on ice. Embryos were not sexed due to the early developmental timepoint. Upon receipt, embryos were further processed, as described herein, in preparation for tissue sectioning.

CRISPR-Cas9 Generated Allele

The CTCF ‘i9’ motifs targeted for deletion were as follows:

CTCF motifs deleted at ‘i9’	(mm10):
ACAGTGACATCAAGGGACTA	chr5: 29,271,026 – 29,271,045
AGTGCCACCTGTTGGCTGAA	chr5: 29,274,938 – 29,274,957
GGCCAGAAGAGGGTGTCAG	chr5: 29,275,547 – 29,275,565

The Zhang lab’s CRISPR Design Tool (http://www.genome-engineering.org/crispr/) and Benchling website (https://benchling.com/) were used to design single-guide RNAs (sgRNAs) with a quality score greater than 80% and fewer than two exonic off-target regions.

saRNAs:
CTCF_Lmbr1_i9_cenF:	AGGGCGTCAGGAAATTCCAC
CTCF_Lmbr1_i9_telF:	TGAACTGCCAATCACCTGGG

The sgRNAs were cloned as described in Paliou et al. (2019)³⁷ and transfected into mESCs accordingly. The following mutations were confirmed by genotyping and Sanger sequencing:

CTCF4:i5:zrs:i9 deletion alleles (mm10):	size (kb)
chr5: 29,270,430 – 29,276,257	5.8
chr5: 29,270,818 – 29,276,296	5.4

METHOD DETAILS

Oligopaint DNA Probe Design and Synthesis

We present an overview of all the super-resolution designs and methods utilized in Figure S9. OligoMiner⁵⁷ was used to extract oligo sequences of desired length and characteristics for the genomic regions of interest listed in Table S1. Oligopaints were designed at the maximal density allowed based on the input parameter restraints and yielded probes with an average of 5.8 probes per kb. Oligos were purchased from Custom Array and Twist Bioscience and amplified for STORM and (sequential) DNA-FISH experiments as described in Luppino et al. (2020)¹⁶ and Park et al. (2023)⁷³, respectively.

Tissue Preparation

Yolk sac tissue was harvested at the time of dissection for genotyping. Embryonic heads (E12.5) were fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) at 4°C for two hours immediately upon dissection. Heads were thoroughly washed in cold PBS three times for 5-mins each, then placed in 30% D-sucrose in PBS for cryoprotection at 4°C overnight, or until the tissues sank. The heads were measured along their length, from the forehead to the back of the head, for size-matching controls and mutants. Heads were embedded in Tissue Tek O.C.T. Compound on dry ice and stored at −80°C until sectioning.

Coronal tissue sections were adhered to poly-L-lysine coated glass slides or coverslips at 10–16μm. Each slide hosted wildtype/control and mutant tissue sections which were anatomically matched according to morphological landmarks during sectioning. After sectioning, slides were briefly dried a room temp for 60-mins and then stored at −20°C until use in downstream assays.

Immunofluorescence

Samples were removed from −20°C and allowed to warm to room temp for one hour, then fixed in 4% PFA for 5-mins, then washed in PBS twice for 5-mins each and once in PBS with 0.1% Triton (PBT) for 5-mins. Tissue was blocked in 10% normal sheep serum (NSS) in PBT for one hour at room temperature in a humidified chamber. The samples were then incubated with primary antibody in 1% NSS in PBT in a humidified chamber overnight at 4°C (1:400 for Millipore Anti-Olig-2: Cat. AB9610 and 1:200 for Novus* Anti-Olig2: Cat. NP1–28667). Tissue was rinsed in PBS twice and PBT twice, for 5-mins each, followed by incubation with the secondary antibody for one hour at room temperature in a humidified chamber and shielded from light (1:200 for goat anti-rabbit AlexaFluor-488: Cat. A11008). Samples were washed in PBS three times for 5-mins each, post-fixed in 4% PFA for 10-mins and washed one final time in PBS for 5-mins prior to mounting/imaging or moving forward with DNA-FISH.

*The Novus Anti-Olig2 antibody was diluted in LifeTechnologies Antibody Diluent Reagent Solution (Ref #003118) and its use required an additional antigen retrieval step at 80°C for 30-mins in 10mM sodium citrate buffer (pH6.0), prior to blocking. Samples were cooled for 10-mins at room temp then transferred to water using the same (warm) Coplin jar for an additional 10-mins, before proceeding with the two PBS washes and one PBT wash.

DNA-FISH

Samples were removed from −20°C and allowed to warm to room temp and immediately fixed in 4% PFA for 5-mins, washed in PBS for 5-mins, and hydrated in 10mM sodium citrate buffer. Antigen retrieval was performed in 80°C sodium citrate buffer for 25-mins then cooled to room-temp for 30-mins. Samples were transferred to 2x SSCT (2x saline-sodium citrate buffer with 0.1% Tween-20) for 5-mins then placed in 50% FMM in 2x SSCT for 1-hr. Samples were dehydrated in an ethanol row of 70%, 95%, and 100% for 3-mins each and air-dried for >90-mins, or overnight. Samples were pre-treated with 0.1M triethanolamine (TEA) for 10-mins, then 0.25% acetic anhydride in TEA for 5-mins, followed by a wash in 2x SSCT for 10-mins.

Primary Oligopaint probes were combined with hybridization mix (50% formamide and 10% dextran sulfate in 2x SSCT), 10μg RNase A and 5.6mM dNTPs, in water. STORM and conventional DNA-FISH experiments used 50pmol of each probe per 25μL of hybridization mix; sequential DNA-FISH used 300pmol of the amplified probe set pool per 40μL of hybridization mix. Upon application to the sample, the probe solution was sealed with glass and rubber cement. Samples remained in a humidified chamber at 42°C for 3-hrs to allow probe infiltration. Sample DNA was denatured at 80°C for 5-mins and then allowed to hybridize with the probes overnight (>16-hrs) at 42°C. Samples were washed first in 60°C then room-temp 2x SSCT, for 15- and 10-mins. respectively, followed by a wash in 0.2x SSC for 10-mins.

For sequential DNA-FISH experiments, samples were placed in PBS then fixed once more in 4% PFA for 10-mins. Samples were washed in PBS then 2x SSC for 5-mins each, mounted with SlowFade Gold, and sealed with clear nail polish. Samples were kept shielded from light at 4°C until imaging. Bridge-secondary labeling occurred iteratively at room temperature on the Vutara VXL (see Sequential DNA-FISH Imaging).

STORM samples were directly labeled with 10pmol of secondaries (purchased from IDT) per 25μL of modified hybridization mix (10% formamide and 10% dextran sulfate in 2x SSCT). Conventional DNA-FISH samples utilized bridge-secondary labeling with 2pmol of each bridge (purchased from IDT) and 10pmol of secondaries per 25μL of modified hybridization mix. After application to the sample, the mixture was sealed with glass and allowed to hybridize for 2-hrs in a covered, humidified chamber at room-temp. Washing in 60°C then room-temp 2x SSCT, for 10- and 5-mins, respectively, followed by a wash in 0.2x SSC for 5-mins. The sample mounted with SlowFade Gold and sealed with clear nail polish. Samples were kept shielded from light at 4°C until imaging.

STORM Imaging and Analysis

Images were acquired on a Bruker Vutara 352 super-resolution microscope with an Olympus 60x/1.2 NA water objective and Hamamatsu ORCA Flash 4.0 v3 sCMOS camera. Imaging buffer was freshly prepared at the time of use and contained 10% glucose, 2x SSC, 0.05M Tris, 2% glucose oxidase solution, and 1% 2-mercaptoethanol. The glucose oxidase solution consisted of 8440 AU glucose oxidase and 70200 AU catalase from bovine liver dissolved in 50mM Tris pH 8.0 and 10mM NaCl buffer.

Fields of view were selected in widefield mode to identify zli and non-zli (i.e. prethalamic) regions based on an antibody stain in 488nm. The 561nm and 640nm lasers were used to acquire samples labeled with CF568 and AlexaFluor-647, respectively. Imaging of samples occurred in a sequential manner, with AlexaFluor-647 imaged first followed by CF568, capturing 10,000 frames for each channel at 15% laser power. A 405nm activation laser was used in conjunction with CF568 at 0.0015% for all 10,000 frames to optimize photoswitching. Bruker’s proprietary biplane imaging technology captures localizations for 1μm of image depth in z.

Calibration, image acquisition, and analysis were done with the Bruker SRX software. Image processing parameters were selected in accordance with the user manual (https://guide.vutara.bruker.com/login) and with input from Bruker representatives. Single-molecule localization precision was determined by fitting the data to a B-spline PSF interpolation since this method is more robust to background commonly observed in tissue samples.^74,75 Localizations were assigned to clusters by running the density-based scan (DBScan) algorithm with a maximum particle distance of 0.100μm, minimum particle count of 40, and 0.100μm hull alpha shape radius (Figure S10).⁷⁶ Measurements were exported for plotting and statistical analysis in GraphPad Prism 9.2.0. Domain volumes with particle counts less than 1000 were excluded from the analysis due to incomplete labeling or imaging. The latter can result from the random nature of slicing through nuclei inherent to cryosectioning tissue. Normalized volumes were calculated as the volume (μm³) divided by the genomic content (bp). Representative images of domain clusters were visualized and exported from the Bruker SRX software as both ‘point cloud’ and ‘wireframe’ options.

Sequential DNA-FISH Imaging

Images were acquired on a Bruker Vutara VXL super-resolution microscope with an Olympus 60x/1.3 NA silicon oil immersion objective and Hamamatsu ORCA Flash 4.0 v3 sCMOS camera. The sample was fit into a Bioptechs fluidics flow chamber and placed onto the microscope. Fields of view were selected based on Shh_in RNA-FISH signal on the immediately adjacent tissue section and local morphological landmarks to distinguish the zli and surrounding non-zli forebrain regions. The 488nm and 638nm lasers were used to acquire samples labeled with AlexaFluor-488 and AlexaFluor-647 (fiducial and segment signals, respectively). Channels were acquired interleaved at 0.2μm z-step size.

Iterative cycles of imaging and washes were carried out by the PlexFlo system under the operation of Bruker SRX software. Each bridge-secondary hybridization reaction contained 0.1μM bridges, 0.2μM secondaries (conjugated with AlexaFluor-647), and 1uM strand-displacement probes (to remove the bridges of the previous segment) in buffer (2x SSC, 30% formamide, and 4% PVSA). The sequence of events was as follows: (1) hybridize the bridge-secondary mix for 3.5-mis flowing with a 30-mins pause, (2) wash three times with 30% formamide in 2x SSC, for 10-mins flowing and pausing for 5-mins each, (3) apply imaging buffer* for 5-mins flowing pausing for 2-mins, (4) acquire images, (5) wash with 50% formamide in 2x SSC for 5-mins. Events 1–5 were repeated n times until all segments of the locus were imaged (i.e. n = 32).

*Imaging buffer contained 0.4% glucose, 2x SSC and 10mM Tris-HCl, filter-sterilized and temporarily stored at 4°C until use. Glucose oxidase solution was added fresh to the imaging buffer immediately before use; 0.37 mg glucose oxidase and 64166.4 AU catalase from bovine liver (dissolved in 50mM Tris pH8.0 and 10mM NaCl buffer), was added per 10mL of imaging buffer.

Calibration and image acquisition were done with the Bruker SRX software; images were exported for analysis using a custom-built pipeline in MathWorks MATLAB (see Sequential DNA-FISH Analysis).

Sequential DNA-FISH Analysis

The HP Z440 with 4 cores and 64 GB of RAM was used to process the images. Images were analyzed using a custom-built toolbox in MathWorks (https://github.com/melikelakadamyali/SequentialDNA-FISH). In brief, the intensity of each image was min-max normalized to correct for differences across images. Foci for both the fiducial and bridge-secondary signals were then identified in the images in x/y using the Laplacian of Gaussian method, and further refined using Gaussian signal fitting to exclude non-Gaussian foci (e.g., background signal). Gaussian size (σ) and signal threshold (α) parameters were optimized and set across images: σ = 2 (fiducial and bridge-secondary); α = 0.00001 (fiducial) and 0.001 (bridge-secondary). Once all foci were called in each z-slice image, they were combined in the axial direction with a search radius in x/y of 5-pixels (490 nm) and in the axial direction of 5 z-slices (1000 nm), with no gaps in slices allowed. The z-position was set to the slice that had the maximum intensity for that combined spot. An additional filtering was performed to exclude signal intensities below thresholds of 0.3 and 0.1 for the fiducial and bridge-secondary images, respectively, to increase robustness of the signal investigated.

Next, the bridge-secondary spots were linked together in sequential order by tracking the fiducial signals, as this signal remains constant across all sequential bridge-secondary hybridizations (total = 37). First, the bridge-secondary spots were linked to the fiducial spots by using a search radius in x/y of 5-pixels (490 nm) and in the axial direction of 5 z-slices (1000 nm). Next, the fiducial spots were linked together sequentially by using a search radius in x/y of 5-pixels (490 nm) and in the axial direction of 7 z-slices (1400 nm), with 2 sequential gaps permitted to compensate for axial drift in the tissue. These linked fiducial ‘tracks’ were then filtered for those with 30 or more sequential links to ensure that the data investigated were relatively complete. Finally, the coordinates of the linked fiducials were drift-corrected by mapping them onto the first-detected fiducial coordinate. Since each bridge-secondary spot is linked to a fiducial, the corresponding fiducial drift correction applies to the coordinates of the sequential bridge-secondary spots, as well. Efficiencies for each bridge-secondary hybridization were calculated as the fraction of fiducial spots with a linked bridge-secondary spot. Together, the sequentially linked bridge-secondary spots formed a ‘trace’ for each individual allele. The allelic distances between all coordinates of the trace were then calculated as the Euclidean distance between subsequent hybridizations, which created a 37×37 table (matrix) for each allele (missing distances were represented as NaNs). Distance tables were exported for downstream analyses and display in GraphPad Prism 9.2.0. Contact frequency was calculated as the proportion of distances less than 200nm. Of the 37 total sequential hybridizations, only the 32 representing the Shh domain (4–35) are presented in this work and are renumbered (1–32) for clarity. Representative alleles were exported from the custom-built MathWorks toolbox using the ‘plot traces’ function and color-coded in Adobe Illustrator.

Clustering Analysis

From the individual allele distance matrices, we created virtual 4C-like datasets by anchoring the distances to the bin containing Shh. These data were smoothed by taking a rolling average distance, where each 30-kb bin represents the average of itself and its neighboring bins. The datasets with distance information across all bins were further analyzed with the scikit-learn package (version 0.22), using 6-component PCA analysis to explain > 80% of variance and performing k-means with 5 clusters based on plotting the sum of square errors. All steps were performed using the Average_Clustering.py script.

Structured Illumination Microscopy Imaging

Conventional DNA-FISH and RNA-FISH samples were imaged using a VisiTech instant structured illumination microscope (iSIM) with a 100x/1.49 NA oil immersion objective and Hamamatsu ORCA Flash 4.0 sCMOS monochrome and ORCA-Quest qCMOS cameras. The 405nm, 488nm, 561nm, and 640nm lasers were used to acquire samples labeled with Hoechst, AlexaFluor-488, CF568/Quasar570 and AlexaFluor-647/Quasar670, respectively.

For conventional DNA-FISH, fields of view were selected to identify zli and non-zli regions (including prethalamic tissue) based on either staining with an antibody in 488nm or Shh RNA-FISH signal on the same or immediately adjacent tissue section, with guidance from local morphological landmarks. For RNA-FISH, fields of view directly captured the Shh signal in the zli plus the surrounding non-zli tissue. For SBE1/5^−/− RNA-FISH, fields of view were selected to identify zli with antibody staining in 488nm on the same or immediately adjacent tissue section, with guidance from local morphological landmarks. For RNA-/DNA-FISH, the RNA-FISH signal was imaged first then the samples removed from the microscope, DNA-FISH performed, and the samples re-imaged in the same positions using morphological landmarks. Images were acquired as interleaved at a z-step size of 0.15μm for a total depth of ~6–9μm. Image acquisition was performed with MetaMorph software and exported for analysis using Imaris (see Conventional DNA-FISH Analysis and RNA-FISH Analysis).

Conventional DNA-FISH Analysis

DNA-FISH images were analyzed using Imaris software. Individual foci labeling 10-kb segments (i.e. Shh promoter, SBE5, etc.) were called using the ‘Spots’ function based on control tissue thresholds for ‘Size’ and ‘Quality’ parameters (Figure S10). ‘Size’ was determined based on measuring the diameter of foci which was typically fell between 0.2–0.4μm; ‘Quality’ was thresholded to each dataset. Together, the ‘Spot’ calls were applied uniformly across each dataset and visually confirmed to fit the raw data. The zli was delineated based on either an antibody stain or Shh RNA-FISH signal on the same or immediately adjacent tissue section, with guidance from local morphological landmarks. Using the ‘Surface’ function, contour lines were manually drawn on the images to partition the zli volume; the non-zli volume is everything outside of the zli volume. Prethalamic images were partitioned as the central block of cells of approximately equal size to the corresponding zli. ‘Spots’ within the ‘Surfaces’ were identified by using the ‘Find Spots Close to Surface’ function with the distance threshold set to zero. The x/y/z coordinates for all ‘Spots’ were then exported. Distances were calculated between pairs of ‘Spots’ coordinates using a custom script in Python that searches for the nearest neighbor without replacement, Imaris_min-dist.py. For 3-color FISH experiments, pairs of distances were then linked to individual alleles based on a common (linking) ‘Spot’ (i.e. the Shh promoter) coordinate. All distances were filtered for those less than 1μm for all conditions. Measurements were plotted and statistically analyzed in GraphPad Prism 9.2.0. Contact frequency was calculated as the proportion of distances less than 200nm.

RNA-FISH Probe Design

RNA-FISH probes were custom-designed and ordered from Stellaris. The Shh intronic probes (Shh_in) were designed to a 4.7-kb of intron 1 (chr5:28,461,600–28,466,383, mm10) and labeled with Quasar670, while the Shh exonic probes (Shh_ex) were designed to exons 2, 3, and part of the 3’ UTR (chr5:28,457,158–28,461,583, mm10) and labeled with Quasar570.

RNA-FISH

Samples were removed from −20°C and allowed to warm to room temp and immediately fixed in 4% ice-cold PFA for 15-mins and washed twice in PBS for 5-mins each. Samples were then dipped in water, then TEA, followed by 10-mins in 0.126% acetic anhydride in TEA. Next, samples spent 3-mins each in 2x SSC, an ethanol row at 70%, 95%, and 100%, 5-mins in chloroform, then 3-mins each in 100% and 95% ethanol, and dried overnight at room-temp.

RNA probes were combined with hybridization mix (10% formamide and 10% dextran sulfate in 2x SSCT), 10% sodium dodecyl sulfate and 5.6mM dNTPs, in water. Fresh 1:4 RNA probe dilutions in water were made from the 12.5 uM stocks in TE buffer (10mM Tris-HCl, 1mM EDTA, pH8.0); 1.25uL of the dilution was used per 125uL hybridization mix. Upon application to the sample, the probe solution was sealed with glass and rubber cement. Samples remained in a humidified chamber at 37°C overnight (>16-hrs). Samples were washed in 37°C Buffer A (10% formamide in 2x SSC) for 30-mins, then again in 37°C Buffer A with Hoechst stain (1:10,000) for 30-mins. Finally, samplers were washed in 2x SSC for 3-mins, PBS for 5-mins, fixed in 4% PFA for 10-mins, briefly re-stained with 1:10,000 Hoechst in PBS for 5-mins, then mounted with SlowFade Gold and sealed with clear nail polish. Samples were kept shielded from light at 4°C until imaging.

RNA-/DNA-FISH was performed consecutively: first RNA-FISH and imaging, then DNA-FISH and imaging.

RNA-FISH Analysis

RNA-FISH images were analyzed using Imaris software. Median filters and 3-D deconvolution were applied to optimize images for analysis. Individual foci (i.e. ‘Spots’) were called for Shh_in and Shh_ex based on control tissue thresholds for ‘Size’ and ‘Quality’ parameters, as described above. Transcriptional bursts were classified as colocalized (within 0.4μm) Shh_in with Shh_ex signal. The zli was delineated based on the scope of the total Shh_ex signal using the ‘Surfaces’ function, as described above (Figure S10). The total number of cells within the zli was manually quantified by counting the Hoechst-stained nuclei in the middle of the z-stack. Measurements were exported for plotting and statistical analysis in GraphPad Prism 9.2.0.

RNA-/DNA-FISH Analysis

Images were processed as described above for independent RNA- and DNA-FISH experiments in Imaris, resulting in x/y/z positional data for RNA-FISH ‘Spots’ and 3-D distances between the DNA-FISH for the Shh promoter-SBE1 and the Shh promoter-SBE5. The two DNA-FISH distances were linked together based on the (shared) Shh promoter. The rest of the analyses were done in 2-D (x/y) by negating the z-dimension of the RNA-FISH data and recalculating 2-D distances for the DNA-FISH. Note: DNA-FISH ‘Spots’ were first linked together in 3-D and then the 2-D distances recalculated afterwards. This avoids incorrectly linking foci from different alleles due to the stacking of nuclei in tissue; this does not hold true if the max projection is done first in the analysis pipeline.

The RNA- and DNA-FISH images were manually aligned in X/Y in Adobe Photoshop using tissue morphology. Each pixel is 0.046 × 0.046μm. The pixel shift was applied to the RNA-FISH x/y positions. The distance between the RNA and Shh promoter ‘Spots’ x/y positions were calculated to link bursting alleles to the DNA-FISH data using Imaris_min-dist.py. An RNA burst was linked to the DNA-FISH data if the distance between the RNA ‘Spot’ and the Shh promoter was < 1μm; all other alleles were presumed non-bursting. All plots and statistical analyses were generated from GraphPad Prism 9.2.0.

Publicly Available Datasets

Gene Omnibus Datasets (GEO) accession numbers used in this study include: Figure 1: Dixon et al. (2012)⁴² GEO: GSE35156 and Paliou et al. (2019)³⁷ GEO: GSE123388; Figure S1: Deng and Ma et al. (2015)⁵⁰ GEO: GSE59779 and GSE68992, Bonev et al. (2017)⁴⁸ GEO: GSE96107, and Kieffer-Kwon, Nimura, Rao, Xu et al. (2017)⁴⁹ GEO: GSE82144. Juicebox v1.11.08 was used to pull the Shh TAD Hi-C data presented in Figures 1, S1, and S3. To compare published Hi-C datasets to the sequential DN-FISH data generated in this study, we plotted the observed contacts and calculated contact frequencies by the distance from the Shh-containing bin.

QUANTIFICATION AND STATISTICAL ANALYSIS

The statistics software used is listed in the appropriate methods subsections and all statistical tests are indicated in the figure legends. The main figure data represents the sum of measurements (n) from at least two independent biological replicates, with each replicate showing the same effect in the respective Supplementary Figure. Biological replicates are defined as separate embryos, generated from unique parents. Where mutant mice are assayed, the controls are littermates (i.e. control and mutant embryos are born of the same mother) and heads are size-matched to avoid subtle differences in development within litters.

Highlights.

1. Multiple configurations underlie Shh locus topology in the embryonic forebrain.

2. Long-range interactions depend on transcription dependent- and independent inputs.

3. CTCF binding sites facilitate distal enhancer-promoter communication at Shh.