Introduction

Chromosomes in multicellular animals are organized into a series of topologically independent looped domains, called TADs (topologically associated domains: (Cavalheiro et al., 2021; Chetverina et al., 2017; Jerković et al., 2020; Matthews and White, 2019; Rowley and Corces, 2018). The arrangement of TADs in a given chromosomal DNA segment is generally (though not precisely) similar in different tissues and developmental stages, and this is a reflection of the mechanism underlying TAD formation—the endpoints of TADs are determined by a special class of elements called chromatin boundaries or insulators. While boundary-like elements have been identified in a wide range of animals and plants, the properties of this class of DNA elements have been most fully characterized in Drosophila (Cavalheiro et al., 2021; Chetverina et al., 2017). Fly boundaries have one or more large (100-400 bp) nucleosome-free nuclease hypersensitive sequences that are targets for multiple DNA binding chromosomal architectural proteins. While only a single chromosomal architectural protein, CTCF, has been characterized in mammals, there are almost two dozen such proteins in flies and the list is still growing (Heger et al., 2013; Heger and Wiehe, 2014; Schoborg and Labrador, 2010). In addition to subdividing the chromosome into a series of loops, fly boundary elements have insulating activity. When placed between enhancers or silencers and their target promoters, boundaries block regulatory interactions (Bell et al., 2001; Chetverina et al., 2014; Chetverina et al., 2017). This activity provides a mechanism for delimiting units of independent gene activity: genes located between a pair of compatible boundaries are subject to regulatory interactions with enhancers/silencers present in the same chromosomal interval, while they are insulated from the effects of enhancers/silencers located beyond either boundary in adjacent regulatory neighborhoods. It is currently thought that their ability to organize the chromosome into topologically independent loops is important for their insulating activity (Cai and Shen, 2001; Gohl et al., 2011; Muravyova et al., 2001).

Studies dating back to the ‘90’s have suggested that fly boundaries subdivide the chromosome into loops by physically pairing with each other (Chetverina et al., 2014; Chetverina et al., 2017). In these first experiments, regulatory interactions were observed for transgenes inserted at distant sites in chromosome that were carrying either the gypsy transposon boundary su(Hw) or the bithorax complex (BX-C) boundary Mcp (Muller et al., 1999; Sigrist and Pirrotta, 1997; Vazquez et al., 1993). Further support for the idea that boundaries function by pairing has come from chromatin immunoprecipitation, chromosome conformation capture (CCC), MicroC, and direct imaging experiments (Chen et al., 2018; Li et al., 2011; Vazquez et al., 2006). More recent studies have revealed physical interactions in the CNS, such as those found for su(Hw) and Mcp, which “reach over” multiple intervening TADs, consistent with them playing an important role in cell type-specific gene regulation (Mohana et al., 2023) by bringing distant enhancers and promoters together.

The parameters governing pairing interactions have been defined using insulator bypass, transvection and boundary competition assays. These studies have shown that fly boundaries are able to pair not only with heterologous boundaries but also with themselves. Moreover, the pairing interactions typically exhibit a number of characteristic features: promiscuity coupled with clear partner preferences, and orientation dependence.

Partner preferences depend upon the chromosomal architectural proteins that interact with each boundary. For example, in the boundary bypass assay, a set of enhancers are placed upstream of two reporters (Cai and Shen, 2001; Kyrchanova et al., 2008a; Muravyova et al., 2001). When multimerized CTCF sites are placed between the enhancers and the closest reporters, both reporters are insulated from the enhancers. When a second set of multimerized CTCF sites are place downstream of the closest reporter, bypass is observed. In this case the closest reporter, which is bracketed by the multimerized CTCF sites, is still insulated from the enhancers; however, the downstream reporter is activated (Muravyova et al., 2001).

Heterologous combinations give a different result: when multimerized dCTCF sites are placed upstream of the closest reporter and multimerized Zw5 sites are placed downstream, no bypass is observed. Endogenous fly boundaries also show partner preferences in bypass assays and in boundary competition experiments (Gohl et al., 2011; Kyrchanova et al., 2011; Kyrchanova et al., 2008b). On the other hand, while boundaries have partner preferences, they are also promiscuous in their ability to establish functional interactions with other boundaries. For example, the Fab-8 insulator can pair with scs’ from the Drosophila heat shock locus (Gohl et al., 2011).

In addition, to partner preferences, pairing interactions between endogenous fly boundaries are, with a few exceptions, orientation dependent. Self-pairing interactions are head-to-head. This seems to be a common feature of fly boundaries and has been observed for scs, scs’, iA2, wari, Mcp, Fab-8, AB-I, homie and nhomie (Fujioka et al., 2016; Kyrchanova et al., 2008a). In contrast, pairing interactions between heterologous boundaries can be head-to-head or head to tail. The two boundaries bracketing the even-skipped (eve) locus, homie and nhomie, pair with each other head-to-tail, while boundaries in the Abdominal-B (Abd-B) region of BX-C pair with their neighbors head-to-head. The topology of the loops (TADs) generated by head-to-tail and head-to-head pairing between neighboring boundaries in cis are different. As illustrated in Fig. 1, head-to-tail pairing generates stem-loops, while head-to-head pairing generates circle-loops. The loops could be connected to each other by unanchored loops (Fig. 1A and C), or they could be linked directly to each other if boundaries can pair simultaneously with both neighbors (Fig 1B and D). An alternating pattern of TADs connected by DNA segments that crosslink to each other with reduced frequency (c.f., 11 DNA below) is not observed in MicroC experiments. Instead, most TADs appeared to be directly connected to both of their neighbors without an intervening unanchored loop (Batut et al., 2022; Bing et al., 2023; Levo et al., 2022): see also below). This would suggest that TAD boundaries are typically linked to both neighbors, either simultaneously or as alternating pair-wise interactions.

Figure 1.

Key to understanding the 3D organization of chromosomes in multicellular eukaryotes will be the identification of TADs that are stem-loops and TADs that are circle-loops. In the studies reported here, we have used MicroC to analyze the contact maps generated by stem-loops and circle-loops. Stem-loops and circle-loops TADs are expected to interact differently with their neighbors and this should be reflected in the patterns of crosslinking events between neighboring TADs. As illustrated for linked stem-loops in Fig. 1B, TAD2, is isolated from its next-door neighbors, TAD1 and TAD3. In this configuration, crosslinking events between sequences in TAD2 and sequences in TAD1 and TAD3 will be suppressed. On the other hand, TAD1 and TAD3 are in comparatively close proximity, and crosslinking between sequences in these two TADs will be enhanced. A different pattern of neighborly interactions is expected for circle-loop TADs. In this case, the TAD in the middle, TAD2, is expected to interact with both of its neighbors (Fig. 1D). To test these predictions, we have first compared the MicroC contact profiles for stem-loop and circle-loop TADs. For stem-loops we selected the eve TAD, while for circle-loops we chose the four TADs that comprise the Abd-B parasegment-specific regulatory domains. We show that these stem-loop and circle-loop TADs have distinctive crosslinking signatures. To confirm these MicroC signatures we converted the topology of the eve TAD from a stem-loop to a circle-loop. In addition to changing the MicroC signature of the eve TAD, the change in topology is accompanied by changes in the regulatory interactions between eve and its neighbors.

Results

Stem-loops versus circle-loops

The distinctive loop topologies of stem-loops and circle-loops are expected to be reflected in the contact maps that are generated in MicroC experiments. To determine if this is the case, we compared the MicroC contact maps for the eve TAD and the TADs that correspond to the four Abd-B parasegment specific regulatory domains, iab-5, iab-6, iab-7 and iab-8. The eve TAD is generated by pairing interactions between the nhomie boundary upstream of the eve transcription unit and the homie boundary downstream. Since nhomie and homie pair with each other head-to-tail, the eve TAD has a stem-loop topology. Unlike the eve boundaries, the boundaries that delimit the Abd-B regulatory domains are thought to pair with their neighbors head-to-head (Chetverina et al., 2017; Kyrchanova et al., 2008a; Kyrchanova et al., 2011; Kyrchanova et al., 2008b)). This means that the parasegment-specific regulatory domain TADs, iab-5, iab-6, iab-7 and iab-8, are expected to have a circle-loop topology.

As shown in Fig. 2, the eve TAD and the four Abd-B TADs have distinctive MicroC contact patterns. The eve TAD is a volcano triangle with a plume. The endpoints of the volcano triangle are delimited by nhomie on the left and homie on the right, and within the eve locus (the volcano), there are additional enhanced interactions. While the volcano triangle is generated by contacts between sequences within the eve stem-loop, contacts between sequences in eve and in the neighboring TAD on the left, TL (which contains multiple sub-TADs and six genes: CG15863, CG1418, Pal1, CG12133, eIF3j and CG12134), are much reduced (L-ev in Fig 2A). There is a similar suppression of contacts between sequences in the eve TAD and sequences in the large neighboring TAD on the right, TM (which contains TER94 and Pka-R2) (ev-M in Fig 2A). On the other hand, as expected from the regulatory interactions observed for stem-loops in boundary bypass experiments (Kyrchanova et al., 2008a), physical contacts between sequences in TL and TM are enhanced compared to those between eve and TL (L-ev) or TM (ev-M). Because of the preferential interactions between TADs to either side of the eve stem-loop, the eve volcano triangle is topped by a plume (L-M in Fig. 2A). TM also interacts with the two of TADs farther to the left of eve, TK (K-M in Fig. 2A) and TJ (J-M in Fig. 2A) (see also Fig. Supplemental 1).

Figure 2.

Like the eve boundaries, the TADs in the Abd-B region of BX-C region are connected to their neighbors by the boundaries at their base. As predicted from genetic studies on BX-C boundaries, each TAD corresponds to one of the four parasegment-specific regulatory domains. While the MicroC contact maps for the four Abd-B TADs resemble the contact patterns in the eve TAD, these Abd-B TADs differ from eve in that there are no plumes at the top of their volcano triangle (Fig. 2B). Instead, the Abd-B TADs are overlaid by a series of rectangular interlocking low-density contact (LDC) domains, or clouds. As illustrated in Fig 2B, the iab-6 regulatory domain is surrounded by clouds generated by crosslinking with its next door neighbors iab-5 (5-6) and iab-7(6-7), followed by crosslinking with neighbors that are a TAD away from iab-6, iab-4 (4-6) and iab-8 (6-8). The other regulatory domains also form a unique set of interlocking LDCs/clouds with their immediate neighbors, their next-next door neighbors, and their next-next-next door neighbors.

TAD formation in a nhomie deletion

To further investigate the pairing properties and functioning of the eve boundaries, we used CRISPR-Cas9 to add two attP sites flanking the nhomie region, replacing the region with a mini-white gene. Using mini-white as an exchange maker, <λC31 recombinase-mediated cassette exchange (RMCE) was used to restore the sequence of the region, with nhomie modifications.

As a control for possible effects of the sequences introduced in generating the modification, we reinserted a 597 bp nhomie DNA fragment in the same orientation as the endogenous nhomie boundary (nhomie forward). To maintain roughly the same distance between eve and the neighboring TAD in the nhomie deletion, we introduced a 606 bp DNA fragment from phage 11 (λ DNA). Fig. 3 (and Fig. Supplemental 1) shows the MicroC contact profiles for the nhomie forward and 11 DNA replacements. Except that the sequencing depth of the nhomie forward replacement is not as great as the WT shown for the eve locus in Fig. 2, the profile is quite similar. Like WT, there are sub-TADs within the eve TAD. One of these appears to link homie and the neighboring PRE to the eve promoter-proximal PRE (Fujioka et al., 2008), and is marked by an interaction dot (asterisk in Fig. 3A). Another links nhomie to the eve promoter region (blue arrow in Fig. 3A). The eve TAD is topped by a plume which is generated by interactions between sequences in neighboring TADs TL with TM (L-M). On the other hand, interactions between eve and its neighbors are suppressed. Like eve, there are sub-TADs in the neighboring TAD, TL. The TL sub-TAD closest to eve (TL4) corresponds to the CG12134 transcription unit (green arrowhead marks the boundary: Fig. 3), while the neighboring sub-TAD (TL3) encompasses the eIF3j transcription unit (blue arrowhead: Fig. 3).

Figure 3.

The MicroC profile of the 11 DNA replacement (Fig. 3D) is quite different from that of either nhomie forward or WT, which are similar (Figs. 2A and 3A). While homie still defines the distal (relative to the centromere) end of the eve locus, the 11 DNA replacement does not function as a TAD boundary, and the leftward endpoint of the eve TAD is no longer well-defined. One new “endpoint” for the eve locus maps to sequences between CG12134 and eIF3j (green arrowhead), which in wild type correspond to the left boundary of the TL sub-TAD TL4. The other endpoint maps to sequences between eIF3j and CG12133 (blue arrowhead), which in wild type define the left boundary of the TL sub-TAD TL3. These interactions are not as stable as those between nhomie and homie, as the density of interaction dots is lower. Furthermore, they appear to flip back and forth between alternative endpoints (as indicated by the green and blue double arrows in Fig. 3F) based on the MicroC contact profile, which is consistent with a mixture of (at least) two conformations. The eve TAD also interacts with sequences in the two other TL sub-TADs, TL1 and TL2. In addition, the eve promoter appears to interact with sequences located upstream of CG12134 (purple arrowhead in Fig. 3D and double arrow in Fig. 3F), while this interaction is not observed in the nhomie forward replacement. While the TL TAD (from the TK:TL boundary to nhomie) is also disrupted by the nhomie deletion (it has a much less distinct “volcano apex”, and its right-most sub-TAD TL4 is now fused with the eve TAD), the left-most TL sub-TADs (TL1, TL2, and TL3) are still present, indicating that their formation does not depend on nhomie. As shown in the virtual 4C at homie viewpoint in Fig. 3B and Fig. 3E, the homie boundary interacts with the nhomie forward replacement, but does not contact the 11 DNA replacement.

eve enhancers activate eIF3j expression in the nhomie deletion

In transgene assays, boundary elements block regulatory interactions when interposed been enhancers (or silencers) and reporter genes (Chetverina et al., 2014; Chetverina et al., 2017; Kellum and Schedl, 1992). To determine whether this is also true in their endogenous context, we compared the expression in syncytial blastoderm embryos of the two genes that flank nhomie at the eve locus, eIF3j and CG12134. As the nhomie deletion eliminates homie’s pairing partner and disrupts the eve TAD, we also examined the expression of eve and of the gene just beyond the homie boundary, TER94, which has strong maternal expression through stage 11 (Fig. Supplemental 3B, WT). Consistent with the seemingly normal MicroC profile on the homie side of the eve TAD, we did not detect evidence of eve-like TER94 expression (Figs. 4 and 5). Thus, the formation and functioning of the TM TAD does not appear to be impacted by the loss of nhomie and the fact that left end of the eve TAD is no longer properly anchored.

Figure 4.

Figure 5.

In the case of the gene closest to the nhomie deletion, CG12134, we were unable to consistently detect transcription driven by the eve enhancers in either nhomie forward or 11 DNA embryos. In some 11 DNA embryos, there were hints of stripes at the blastoderm stage (see Fig. Supplemental 2); however, these “stripes” were not observed in most embryos. Since CG12134 (which forms the TL4 sub-TAD in wild type) is closest to the eve enhancers and interacts most strongly, it is possible that the promoter is not compatible with the eve enhancers. A different result was obtained for eIF3j in 11 DNA embryos. As shown in the HCR-FISH experiment in Fig. 4 and quantitated in Fig. 5, we observed a series of eIF3j stripes over a dark background in pre-cellular blastoderm embryos. As this background hybridization is evident in earlier stages, much of it is likely to be of maternal origin. In contrast to the 11 DNA replacement, these stripes are not visible in the nhomie forward (control) replacement (Figs 4 and 5). While it is possible to detect all seven stripes in 11 DNA blastoderm stage embryos, their levels of expression are not equal.

The highest levels correspond to eve stripes 1, 2, 3 and 7, while eve stripes 4, 5 and 6 are expressed at much lower levels. Since the stripe enhancers for 1, 2, 3 and 7 are located between the eve promoter and nhomie, they are closer to the eIF3j promoter than the enhancers for stripes 4, 5 and 6, which are located downstream of the eve transcription unit. In addition to possible effects of distance, the subdomain linking the eve PRE and homie to the promoter is still observed in the 11 DNA replacement, and this could partially sequester the enhancers located downstream of the eve promoter.

We also used digoxigenin in situ hybridization to analyze eIF3j expression. With this procedure we were able to detect a very low level of eve-activated eIF3j stripe expression in WT and nhomie forward embryos at stage 5 and stages 7-8 (Fig. Supplemental 3). In stage 5 embryos when eve expression is driven by specific stripe enhancers, eIF3j expression appears to be similar in all seven stripes (Fig. Supplemental 3). In contrast, as was observed in the HCR-FISH experiments (Fig. 4), there is a clear bias for enhancers located upstream of eve in the 11 DNA replacement at this point in development. In stage 7-8 embryos, the seven-stripe enhancer drives eve expression. It is located close to nhomie and, not surprisingly, high levels of eIF3j expression are observed in all seven stripes in the 11 DNA replacement (Fig. Supplemental 3). At later embryonic stages, eve expression is driven by tissue-specific neurogenic, mesodermal, and anal plate enhancers. However, eIF3j is expressed at high levels in a complex pattern in older embryos, and we were unable to unambiguously detect expression driven by the eve enhancers over this background mRNA. This is also not surprising, given that all of the enhancers for these aspects of eve expression are located downstream of the eve promoter, like the enhancers driving stage 5 stripes 4, 5, and 6.

While the nhomie deletion did not have any obvious impact on the level or pattern of eve expression in blastoderm stage embryos (see Fig. 4), it seemed possible that eve activity was not entirely normal. To test this possibility, we mated males homozygous for either 11 DNA or nhomie forward to females heterozygous for a chromosomal deficiency that includes the eve gene. We then quantitated the number of missing denticle bands in embryonic cuticle preps. As shown in Fig. 5, the frequency of larvae with “severe” defects (2 or more missing ventral denticle bands) in nhomie forward is similar to that in a WT yw control. In contrast, in the 11 DNA replacement, the frequency of larval cuticles with 2 or more missing denticle bands is increased nearly two-fold. Taken together, the increase in severity of the cuticle defects is significant at the P > 0.01 level (one-tailed t-test). The A6 denticle belt is missing most frequently followed by A2, A4, and then A8. These even-numbered denticle bands are those that are lost in eve deficiency mutants (from which the name even skipped comes), suggesting that eve stripe expression at blastoderm stages is compromised in the embryos that produce these defective cuticles.

The eve TAD is converted from a stem-loop to a circle-loop by inverting nhomie

The orientation of boundary:boundary pairing interactions determines the topology of each chromatin loop (Bing et al., 2023; Fujioka et al., 2016). Since nhomie and homie pair with each other head-to-tail, the endogenous eve TAD is a stem-loop. This orientation dependence means that one can convert the eve TAD from a stem-loop to a circle-loop by inverting the nhomie boundary. If our expectations are correct, the MicroC contact pattern will also be transformed from a volcano triangle with a plume to one in which sequences in the eve TAD is surrounded by a cloud of crosslinked sequences from both neighboring TADs (TL and TM), like that observed in the Abd-B region of BX-C.

We tested this prediction by inserting the nhomie boundary in the reverse orientation (nhomie reverse). Fig. 6A shows that the eve TAD is reconstituted by nhomie reverse (compare with Fig. 3). The sub-TAD evident in the nhomie forward replacement linking nhomie to the eve promoter is also re-established (blue arrow). In addition, consistent with our expectation, the plume topping the eve TAD is gone and is replaced by a much more sparsely populated LDC domain (purple double arrow and above). The more prominent LDC TAD-TAD interactions (the clouds) are between sequences in the eve TAD and the neighboring TADs. On the right, eve forms an LDC interaction domain with TM (ev-TM). On the left, eve interacts strongly with sequences in TL4, and progressively less strongly with sequences in the sub-TADs TL3, TL2, and then TL1. In addition to restoring the eve TAD, the TL TAD is also re-established, indicating that the nhomie boundary is important in defining both endpoints of the TL TAD. On the other hand. with the exception of the CG12133 sub-TAD, nhomie does not play a role in generating the three other sub-TADs in the TL TAD. The other interesting feature is a 45° band of interaction (to the left and below the purple double-arrow) between sequences in TL4 and sequences in eve that appear to be located near the left edge of homie, and likely correspond to the distal eve PRE (Fujioka et al., 2008). A similar though weaker 45° band is also seen in nhomie forward (Fig. 3).

Figure 6.

Insulation is reduced in nhomie reverse

Consistent with the models for stem-loops and circle-loops in Fig. 1, the neighborly interactions evident in the MicroC contact patterns for nhomie forward and nhomie reverse are quite distinct. The nhomie forward TAD is isolated from its neighbors, and crosslinking between eve and the neighboring TADs is suppressed (Fig. 1B and 2A). This is not true for the circle-loop TAD generated by nhomie reverse: in this configuration, the eve TAD is not sequestered from neighboring TADs, but instead interacts much more frequently with sequences in next-door TADs than in the stem-loop configuration. Since the eve TAD is no longer isolated from its neighbors, this could increase the frequency of “productive” interactions between eve regulatory elements and genes in nearby TADs, and vice versa.

To test whether the circle-loop topology has an impact on the regulatory landscape, we hybridized nhomie reverse embryos with HCR-FISH probes for eIF3j, TER94 and eve. In early blastoderm stage nhomie forward embryos, there is little evidence of eIF3j or TER94 expression driven by eve stripe enhancers, and the HCR-FISH hybridization pattern is uniform (see Fig. 4). In contrast, it is possible to discern individual stripes of both eiF3j and TER94 mRNA over the background signal in a subset of nhomie reverse blastoderm stage embryos in HCR-FISH (Fig. 7). Since these genes are assembled into their own topologically independent looped domains rather than being in the same domain as eve, the level of eiF3j and also TER94 stripe expression is lower than that observed in the nhomie deletion (11 DNA). The nhomie reverse circle-loop also differs from the nhomie deletion (11 DNA) in that there is not such an obvious preference for which eve enhancers activate expression. In addition, eve-dependent eiF3J and TER94 stripes are detected in only about half of the blastoderm stage embryos (Fig. 5A). It is possible that the frequency of productive inter-TAD contacts differs from one embryo to the next; however, a more likely reason is that the high background of eIF3j and TER94 transcripts obscures the low level of eve enhancer-driven expression at this stage. Once the seven-stripe enhancer is activated, eve-dependent TER94 expression in nhomie reverse is elevated, and all seven stripes are observed (Fig. Supplemental 3B). This fits with the MicroC contact profile. As shown in Fig. 6, the TER94 gene is preferentially crosslinked to eve sequences located between nhomie reverse and the eve promoter as compared to sequences spanning the eve gene and the downstream enhancers. By contrast, there is much less seven-stripe enhancer-driven expression of eIF3j (Fig. Supplemental 3A) than TER94. Consistent with this observation, crosslinking between eIF3j and sequences in eve close to nhomie reverse and the seven-stripe enhancer appear to occur less frequently than crosslinking to sequences on the other side of the eve TAD (though this crosslinking bias is not as pronounced as that observed between TER94 and the eve sequences near the seven-stripe enhancer relative to downstream eve sequences.)

Figure 7.

While eve stripe expression is not discernably different from wild type, the circle-loop topology still impacts eve function. As shown in Fig. 5B, the fraction of nhomie reverse embryos with 2 or more missing denticle bands is nearly twice that in yw or the nhomie forward replacement. Taken together, the increase in severity of the cuticle defects is significant at the P > 0.01 level (one-tailed t-test). As was observed for the 11 DNA replacement, the A6 denticle band is missing most frequently, followed by A2, A4 and then A8.

homie forward converts the eve stem-loop to a circle-loop

While the findings in the previous section show that loop topology impacts how sequences in TADs interact with each other and with their neighbors, one might argue that the effects we observed are a reflection of some novel properties of the nhomie boundary when it is inverted. To test this possibility, we took advantage of the fact that in addition to pairing with nhomie, the homie boundary pairs with itself (Fujioka et al., 2016). However, unlike nhomie:homie pairing which is head-to-tail, homie:homie pairing is head-to-head. This means that it is possible to convert the eve TAD into a circle-loop by inserting homie into the nhomie deletion in the forward orientation.

As shown in Fig. 6C, the MicroC contact profile of homie forward is similar to that observed for nhomie reverse. The plume toping the eve TAD in wild type (Fig. 2) or in nhomie forward (Fig. 4) is absent. Likewise, instead of being isolated from its neighbors, the eve TAD contacts TL and TM. Also like nhomie reverse, homie forward forms a subdomain within the eve TAD linking it to sequences in the eve promoter. There is also a 45° band of crosslinking between sequences in TL4 and sequences on the distal side of eve that appears to correspond to the eve 3’ PRE.

The MicroC pattern is not the only similarity between homie forward and nhomie reverse. The functional properties of the homie forward eve TAD are also similar. Fig. 7 shows that the eve enhancers weakly activate both eIF3j and TER94. As was the case for nhomie reverse, expression levels at the blastoderm stage are low and are observed in only about half of the embryos (Fig. 5A). After the blastoderm stage, when the seven-stripe enhancer drives eve expression, an even higher level of TER94 expression is observed (Fig. Supplemental 3B). In addition, the functioning of the eve gene when it is in the circle-loop configuration is not as efficient, and the frequency of homie forward embryos with 2 or more missing denticle bands is twice that of nhomie forward (Fig. 5B). Taken together, the increase in severity of the cuticle defects is significant at the P > 0.01 level (one-tailed t-test).

Discussion

Two different though overlapping classes of chromosomal architectural elements have been identified in flies. One class is the PREs found in many developmental loci. PREs were first discovered because they induce pairing sensitive silencing of reporter genes (Americo et al., 2002; Kassis et al., 1991). More recent studies have shown that the ability of these elements to physically pair with each other may be their most important function (Batut et al., 2022; Levo et al., 2022). The other class of architectural elements are chromatin boundaries (also called insulators). PRE pairing in cis typically takes place within the context of a larger chromosomal domain, or TAD. In contrast, boundary elements are responsible for defining the endpoints of these looped domains (Arzate-Mejía et al., 2020; Batut et al., 2022; Bing et al., 2023; Chetverina et al., 2017; Ibragimov et al., 2023; Stadler et al., 2017). While not much is known about the parameters governing PRE pairing, the pairing interactions of fly boundaries have been studied in some detail. The key features include an ability to engage in promiscuous pairing interactions, distinct partner preferences, and orientation dependence. Of the endogenous (non-gypsy) boundaries whose functional properties have been studied in detail, only one, Fab-7, appears to be able to pair in both orientations. However, Fab-7 may be unusual in that its boundary activity depends upon factors that have been implicated in the functioning of PREs (Kyrchanova et al., 2018). For all of the other boundaries studied so far, pairing interactions are orientation dependent. When fly boundaries pair with themselves, the interactions are head-to-head (Kyrchanova et al., 2008a). This make sense, as the available evidence suggests that self-pairing interactions in trans may be largely responsible for the pairing of homologs in precise register (Erokhin et al., 2021; Fujioka et al., 2016). In this case, head-to-tail self-pairing would uncoupled the loops on the two homologs.

Unlike self-pairing in trans, pairing interactions between heterologous boundaries in cis can be head-to-head or head-to-tail. The topological consequences are quite distinct. The former generates a circle-loop, while the later forms a stem-loop (Chetverina et al., 2017). In the studies reported here, we have investigated how these two different topologies impact the local chromatin organization. We have also determined whether circle-loops and stem-loops alter the ability of boundary elements to define units of independent gene activity and insulate against regulatory interactions between neighboring TADs.

nhomie deletion disrupts the eve TAD

As would be predicted from many different studies (c.f., (Cavalheiro et al., 2021; Chetverina et al., 2017), deletion of the nhomie boundary and replacement with a control 11 DNA disrupts the eve TAD and alters the regulatory landscape. The MicroC profile shows that disruption of the eve TAD and the neighboring TADs is one sided. Within the eve TAD, the subdomain linking homie and the nearby PRE to the eve promoter is unaffected. Likewise the large TAD, TM, which encompassess both TER94 and pka-R2, and is defined at one end by homie and at the other by an uncharacterized boundary element upstream of the pka-R2 promoter, is intact (Fig. Supplemental 1B). In contrast, on the nhomie side of the eve TAD, the sub-TAD linking nhomie to the eve promoter is absent, and is replaced by a less well-defined sub-TAD linking the eve promoter to an element near the CG12134 promoter. However, the end-point of the eve TAD is no longer distinct, and eve sequences are now crosslinked to the eIF3j sub-TAD and to sequences in the more distant TL sub-TADs, TL3, TL2 and TL1.

Consistent with the alterations in the physical organization of the eve and neighboring TADs, the TER94 gene is still insulated from the eve enhancers. While sequences in the eve TAD physically interact with the gene closest to nhomie, CG12134, only the next gene over, eIF3j, is clearly activated by the eve enhancers. Since crosslinking between sequences in the CG12134 and the eve TAD is more frequent than crosslinking between the eIF3j sub-TAD TL3 and the eve TAD, it seems likely that CG12134 is refractory to activation by the eve enhancers. This could be due to an incompatiblity between the CG12134 promoter and the eve enhancers.

Alternatively, the promoter may not be active at this stage. While eIF3j is activated by the eve enhancers in stage 5 embryos in the nhomie deletion, the stripe enhancers located upstream of the eve gene drive a higher level of expression than those located downstream. Two factors in addition to the effects of distance could potentially account for this finding. Since the eve promoter is located between the downstream enhancers and the eIF3j gene, activation of eIF3j by the downstream enhancers could be suppressed by promoter competition. Alternatively, or in addition, the sub-TAD formed between the 3’ PRE/homie and the eve promoter/proximal PRE (Fujioka et al., 2008) could tend to sequester the downstream eve stripe enhancers away from interactions with eIF3j.

Topology impacts local 3D genome organization and the potential for regulatory interactions

In boundary bypass experiments using endogenous fly boundaries, the ability of the upstream enhancers to activate the downstream reporter depended on the topology of the loop generated by the paired boundaries (Kyrchanova et al., 2008a). Activation is observed for stem-loops, as this configuration brings the upstream enhancers into close proximity with the downstream reporter. In contrast, the enhancers and downstream reporter are not brought into contact when the topology is a circle-loop. As would be predicted from these bypass experiments, the stem-loop formed by the head-to-tail pairing of nhomie and homie physically isolates the eve TAD from its neighbors, and this is reflected in the low density of contacts between sequences in eve and the neighboring TADs (Fig. 2A). Conversely, the TADs that flank eve are brought together and contacts between them generate the plume that is observed above the eve volcano triangle.

The physical isolation afforded by the stem-loop topology is lost when the eve TAD is converted to a circle-loop, either by inverting the nhomie boundary or by replacing nhomie with the homie boundary inserted in the forward direction (Fig. 6). In the former case, head-to-tail nhomie:homie pairing generates a circle-loop. In the later case, head-to-head pairing of homie (inserted in the forward orientation) with endogenous homie generates a circle-loop. The alteration in the local 3D organization induced by the conversion of eve to a circle-loop is evident from the changes in the MicroC contact pattern. Instead of being isolated from neighboring TADs, the eve TAD interacts not only with its immediate neighbors, but also with more distant TADs. As a result, the plume of enhanced contacts linking TM to TL, TK, and TJ is absent and is replaced by contacts between these TADs and the eve TAD.

As might be expected from the MicroC contact patterns, the conversion to a circle-loop topology is accompanyied by alterations in regulatory interactions between eve and the genes in the neighboring TADs (Fig. 7 and Fig. Supplemental 3). Unlike the nhomie forward replacement, the eve stripe enhancers in both of the circle-loop replacements are able to weakly activate expression of two neighboring genes, eIF3j and TER94. This pattern of activation mirrors the enhancement in contacts between the eve TAD and the neighboring TL and TM TADs evident in MicroC experiments. Thus, though eIF3j and TER94 are clearly shielded from the eve enhancers when the eve TAD is a circle-loop, a greater degree of isolation from the action of the eve enhancers is afforded when the eve TAD is a stem-loop.

In addition to reducing insulation from regulatory interactions with genes in neighboring TADs, the circle-loop topology impacts the functioning of the eve gene (Fig. 5B). For both nhomie reverse and homie forward, the frequency of multiple denticle band defects compared to the nhomie forward control is enhanced in a sensitized genetic background. While we did not detect any obvious reductions in the eve stripes in blastoderm stage embryos, the circle-loop provides less insulation than the stem-loop, and it is possible that the neighboring promoters suppress eve expression by competing for the eve enhancers. Another (non-mutually exclusive) possibility comes from the studies of Yokoshi et al. (Yokoshi et al., 2020), who used live imaging to examine the effects of flanking a reporter with the nhomie and homie boundaries. In their experiments, reporter expression was enhanced over two-fold when the reporter was flanked by nhomie and homie; however, the enhancement was greater when the paired boundaries formed a stem-loop than when they formed a circle-loop.

Stem-loops versus circle-loops

The results we have reported here raise several questions. One of these is, what is the relative frequency of stem-loops versus circle-loops in the fly genome? The MicroC contact patterns for the stem-loop and circle-loop versions of the eve TAD are quite distinct. The former is a volcano triangle with a plume while the latter is a volcano triangle flanked by clouds. A survey of the MicroC contact patterns elsewhere in the non-repetitive regions of the fly genome indicates that volcanoes with plumes are rare (∼30). For example, there are two volcano triangles with plumes in the Antennapedia complex and they encompass the deformed and fushi-tarazu genes (Levo et al., 2022). However, since most of the “euchromatic” regions of the fly genome are assembled into TADs whose MicroC profiles resemble that observed for eve circle-loops and the Abd-B region of BX-C, it possible that much of the fly genome is assembled into circle-loops, not stem-loops.

While this suggestion is consistent with the available data, it is based on contact patterns between neighboring TADs, and important caveats remain. For one, the contact patterns between neighboring TADs can deviate in one way or another from that seen in the Abd-B region. For example, there are TADs in which interactions with one set of neighbors appears to be suppressed as expected for stem-loops, but the classic plume is absent, as interactions are not suppressed with the other neighbors (c.f., Fig. Supplemental 4A). In other cases, there are a series of complicated TAD-TAD interactions topped by a rectangular plume (Fig. Supplemental 4B: purple arrow). For this reason, it will not be possible to draw firm conclusions about the frequency of stem-loops versus circle-loops genome-wide until the relative orientation of the paired boundaries themselves can be determined directly. On the other hand, it is clear from our studies that both classes of TADs must exist in flies. If, as seems likely, a significant fraction of the TADs genome-wide are circle-loops, this would effectively exclude cohesin-based loop extrusion as a general mechanism for TAD formation in flies. In addition, though stem-loops could be generated by a cohesin-dependent mechanism, it is unlikely that this mechanism is operational in flies, as we have shown here and in Bing et al. (Bing et al., 2023) that stem-loops in flies are formed by orientation-dependent boundary:boundary pairing.

Another important question is whether our findings have any relevance to the formation and topology of TADs in mammals. In the most common version of the loop extrusion model, the mammalian genome is assembled into an alternating pattern of stem-loops and unanchored loops (c.f., (Davidson and Peters, 2021; Higashi and Uhlmann, 2022; Perea-Resa et al., 2021). In this case, one might expect to observe volcano triangles topped by plumes alternating with DNA segments that have a considerablly lower densisty of internal contacts. However, this crosslinking pattern is not observed in published MicroC data sets (Hsieh et al., 2020; Krietenstein et al., 2020). Instead of an alternating pattern of high-density TAD triangles separated by regions of low density contacts, the TAD triangles are generally linked to both neighbors, just as in flies. Moreover, also like in Drosophila, there are very few stem-loop volcano TADs topped by plumes. Instead, the crosslinking pattern between neighboring TADs appears similar to that observed for circle-loops in flies. Of course, one problem with these MicroC studies is that the resolution may not be sufficient to detect volcanos with plumes or the other features predicted by the loop extrusion model. However, there are no obvious volcanoes with plumes in the much higher resolution RCMC studies of Goel et al. (Goel et al., 2023).

Instead the MicroC profiles most closely resemble those seen in the Abd-B region of BX-C (c.f. the Ppm1g locus in Fig. 4 of Goel et al., (Goel et al., 2023)). Moreover, compromising cohesin activity has minimal impact on the TADs in this region of the mouse genome, as evidenced from the MicroC pattern before and after knockdown. Based on these observations, one can reasonably question whether cohesin-mediated loop extrusion is deployed in mammals as the mechanism for not only generating TADs but also determining TAD boundaries. Clearly validation of the loop extrusion/CTCF road-block model as currently formulated will require a direct demonstration that mammalian TADs are exclusively either stem-loops or unanchored loops, and that the endpoints are always (or almost always) determined by CTCF roadblocks.

Acknowledgements

We would like to thank the Gordon Grey for running the fly food facility at Princeton, members of the Lewis Sigler Genomics Core facility for their invaluable assistance with DNA sequencing, and Qing Liu for excellent technical assistance. We would also like to thank members of MOL431 for creative input. Special thanks to Olga Kyrchanova, Daria Chetverina, Maksim Erokhin, Pavel Georigev, Tsutomu Aoki, Girish Deshpande, Airat Ibragimov, Sergey Ryabichko, Yuri Pritykin, Alex Ostrin, Xinyang Bing, Xiao Li and Mike Levine for stimulating discussions and sharing unpublished data.

Materials and methods

Creation of nhomie deletion flies

To modify nhomie at the eve locus, we used recombinase-mediated cassette exchange (Bateman et al., 2006). First, we inserted two nearby attP sites, using CRISPR. The donor plasmid for this was constructed as follows. First, a mini-white (mw) gene with Glass binding sites (Fujioka et al., 1999) was inserted into BlueScript. From the standard mw gene, the Wari insulator (Chetverina et al., 2008) was deleted. Then, two 102bp attP sequences (Venken et al., 2011) were inserted, one just 5’ of the Glass binding sites and the other at the 3’ end of the modified mw, creating the plasmid P-attPx2-mw. 5’ and 3’ homologous arms were added to both ends. Two gRNA sequences were cloned into plasmid pCFD4 ((Port et al., 2014), Addgene).

The donor and gRNA plasmids were injected into a Cas9 line (y[1] M{vas-Cas9.S}ZH-2A w[1118]), Bloomington Drosophila Stock Center). This chromosomal modification resulted in one attP site being inserted in the intron of CG12134, and the other being inserted between the eve 7-stripe enhancer and the 3+7 stripe enhancer regions. This also deleted 2.2kb of endogenous sequence, including nhomie and the eve 7-stripe enhancer.

After identifying a successful insertion (NattPmw), mw was replaced by each of the following using RMCE: 1) the previously deleted 2.2kb, restoring nhomie and the eve 7-stripe enhancer, to create “wild-type nhomie” (nhomie forward), 2) the same 2.2kb sequence, but with 600bp of phage 11 DNA in place of 600bp nhomie (11 DNA), 3) the same 2.2kb sequence, but with 600bp nhomie inverted (nhomie reverse), and 4) the same 2.2kb sequence, but with 600bp nhomie replaced by a copy of ∼600bp homie in its native orientation in the chromosome (homie forward). Each of these changes was confirmed by sequencing of genomic DNA from the transgenic fly lines.

Analysis of embryonic cuticle patterns and in situ hybridization

To identify defects in developing embryos, embryos were collected for 2.5 hr., and allowed to develop for an additional 20-21 hr. at 25°C. Embryos were dechorionated and mounted in a 1:1 mixture of Hoyer’s medium and lactic acid. Mounted embryos were left at room temperature until they cleared (12-14 days), and the patterns of ventral abdominal denticles were examined and tallied as follows. Loss of at least one-fifth of a denticle band (in A1-A8) was counted as “missing”. Fused denticle bands, which rarely occurred, was also counted as a “missing” band. Minor defects such as in individual denticle rows were not counted.

Digoxigenin (DIG) in situ hybridization was performed using DIG-labeled anti-sense RNA against CG12134, eIFj3, and TER94. RNA expression was visualized using alkaline phosphatase-conjugated anti-DIG antibody (Roche), using CBIP and NBT as substrates (Roche). Each set of experiments was carried out in parallel to minimize experimental variation. Representative expression patterns are shown in each figure.

Hcr fish

The sequences of target genes were obtained from Flybase (flybase.org)(Gramates et al., 2022). To design probes, the target gene sequences were submitted to the Molecular Instruments probe design platform (www.molecularinstruments.com/hcr-rnafish) (Choi et al., 2016), with parameters set to a 35 probe set size for Drosophila melanogaster. A similar method was designed based on published smFISH methods (Little and Gregor, 2018; Trcek et al., 2017). 100- 200 flies were placed in a cage with an apple juice plate at the bottom of the cage. For early stages, the embryos were collected for 7 hours, while for later-stage embryos, collection were overnight. Embryos from each plate were washed into collection mesh and dechorionated in bleach for 2min, then fixed in 5mL of 4% paraformaldehyde in 1X PBS and 5mL of heptane for 15min with horizontal shaking. The paraformaldehyde was then removed and replaced with 5mL methanol. The embryos were then devitellinized by vortexing for 30s, and washed in 1mL of methanol twice. Methanol was then removed and replaced by PTw (1X PBS with 0.1% Tween-20) through serial dilution as 7:3, 1:1, and 3:7 methanol:PTw. The embryos were washed twice in 1mL of PTw and pre-hybridize in 200μL of probe hybridization buffer for 30 min at 37°C. 0.4pmol of each probe sets were added to the embryos in the probe hybridization buffer and the embryos were incubated at 37°C for 12-14h. The embryos were then washed 3X with probe wash buffer at 37°C for 30min and 2X with 5X SSCT(5X SSC+0.1% tween) at room temperature for 5min. Then the embryos were pre-amplify with 300μL amplification buffer for 10 min at 25°C. Meanwhile, 6 pmol of hairpin h1 and h2 were snap cooled separately (95°C for 1:30, cool to RT with 0.1°C drop per second), and then mixed in 100μL of amplification buffer at room temperature. After that, the pre-amplification solution was removed from the embryos, and 100μL of hairpin h1/h2 mix were added to the embryos. Next, the embryos were incubated for 12-14h at room temperature in dark. To remove the excess hairpins, the embryos were then washed in SSCT as follows: 2X for 5min, 2X for 30min and 5X for 5min. Then, the embryos were washed with 1mL PTw for 2min and stained with DAPI/Hoechst at 1μg/mL for 15min at room temperature in the dark. The embryos were then washed with PTw 3X for 5min. Finally, the embryos were mounted on microscope slides with Vectashield and a #1.5 coverslip for imaging.

Imaging, image analysis and statistics

Embryos from hcrFISH were imaged by using a Nikon A1 confocal microscope system, Plan Apo 20X/0.75 DIC objective. Z-stack images were taken at interval of 2μm, 4X average, 1024×1024 resolution, and the appropriate laser power and gain were set for 405, 488, 561, and 640 channels to avoid overexposure. Images were processed by ImageJ and the maximum projection was applied to each of the stack images. To determine the presence of stripes of early embryos, multi-channel images were first split into single channels and the stripe signal was highlighted and detected by the MaxEntropy thresholding method. GraphPad Prism was used for data visualization and statistical analysis. Two-way ANOVA with Tukey’s multiple comparisons test for each pair of groups was used to determine the statistical significance for the percentage of embryos carrying stripes in eIF3j and TER94 channels in each group.

MicroC library construction for the nhomie replacements

Embryos were collected on yeasted apple juice plates in population cages for 4 hours, incubated for 12 hours at 25°C, then subjected to fixation as follows. Embryos were dechorionated for 2min in 3% sodium hypochlorite, rinsed with deionized water, and transferred to glass vials containing 5mL PBST (0.1% Triton-X100 in PBS), 7.5mL n-heptane, and 1.5mL fresh 16% formaldehyde. Crosslinking was carried out at room temperature for exactly 15min on an orbital shaker at 250rpm, followed by addition of 3.7 mL 2M Tris-HCl pH7.5 and shaking for 5min to quench the reaction. Embryos were washed twice with 15mL PBST and subjected to secondary crosslinking. Secondary crosslinking was done in 10mL of freshly prepared 3mM final DSG and ESG in PBST for 45min at room temperature with passive mixing. The reaction was quenched by addition of 3.7mL of 2M Tris-HCl pH7.5 for 5min, washed twice with PBST, snap-frozen, and stored at ™80°C until library construction.

Micro-C libraries were prepared as previously described (Batut et al., 2022) with the following modifications: 50µL of 12-16h embryos were used for each biological replicate. 60U of MNase was used for each reaction to digest chromatin to a mononucleosome:dinucleosome ratio of 4. Libraries were barcoded, pooled and subjected to paired-end sequencing on an Illumina Novaseq S1 100nt Flowcell (read length 50 bases per mate, 6-base index read).

Micro-C data processing

MicroC data for D. melanogaster were aligned to custom genomes edited from the Berkeley Drosophila Genome Project (BDGP) Release 6 reference assembly (dos Santos et al., 2015) with BWA-MEM (Li and Durbin, 2009) using parameters −S −P ™5 −M. The resultant BAM files were parsed, sorted, de-duplicated, filtered, and split with Pairtools (https://github.com/mirnylab/pairtools). We removed pairs where only half of the pair could be mapped, or where the MAPQ score was less than three. The resultant files were indexed with Pairix (https://github.com/4dn-dcic/pairix). The files from replicates were merged with Pairtools before generating 100bp contact matrices using Cooler (Abdennur and Mirny, 2020). Finally, balancing and Mcool file generation was performed with Cooler’s Zoomify tool.

Virtual 4C profiles were extracted from individual replicates using FAN-C (Kruse et al., 2020) at 400bp resolution. The values were summed across replicates and smoothed across three bins (1.2kb). The homie Viewpoint was set to the 549nt homie sequence that was defined in previous studies.

Figures Supplemental

Fig. Supplemental 1. MicroC contact profiles for nhomie forward, λDNA, nhomie reverse, and homie forward in larger scale. N(replicates)=2. Resolution=200. A) MicroC contact maps for the nhomie forward replacement. B) MicroC contact maps for the λDNA replacement. C) MicroC profile of the nhomie reverse replacement. D) MicroC profile of the homie forward replacement.

Fig. Supplemental 2. Expression of CG12134 in WT (yw) and the four nhomie replacements.

Digoxigenin in situ hybridization was used to detect expression of CG12134 during development in the indicated genetic backgrounds. Approximate developmental stages of the embryos in each genetic background are shown on the right. As controls, embryos of similar stages were hybridized with an eve probe (top to bottom: stages 5, 7, 10, 11, and 13).

Fig. Supplemental 3. Expression of eIF3j (Adam) and TER94 in WT (yw) and the four nhomie replacements. Digoxigenin in situ hybridization was used to detect expression of A) eIF3j and B) TER94 during development in the indicated genetic backgrounds. Approximate developmental stages of the embryos in each genetic background are shown on right. As controls embryos of a similar stages were hybridized with an eve probe. Note that unlike the HCR-FISH results shown in Fig. 4, we can detect a low level of eIF4j stripe expression in WT (yw) and nohomie forward St 5 embryos. The difference is likely due to the fact that the signal amplification in the digoxigenin in situ hybridization procedure is not linear.

Fig. Supplemental 4. MicroC patterns of DNA segments on left and right arm of chromosome 2. A) MicroC contact profiles on left arm of chromosome 2 around 3,000,000 bp. Black arrow indicates that interactions between neighboring TADs are suppressed as might be expected for a stem-loop configuration. The red arrow in immediately above the black arrow points to contacts between next-next-(next) door neighbors that are enhanced. However, the plume that is formed is one sided as indicated by the green arrow. B) MicroC contact profiles for TADs on the right arm of chromosome 2. The TADs indicated by the double headed blue arrow at the bottom have a complex pattern of neighborly interactions. The asterisk and purple arrow indicate a potential volcano plume; however, unlike the eve volcano triangle and plume, this plume appears to be generated by crosslinking between sequences to either side of the collection of TADs indicated by the doubleheader blue arrow. The TAD-to-TAD interaction patterns are further complicated by a band of enhanced crosslinking (red arrow).