Establishment and diversification of the coleoid genome architecture at coding and non-coding levels

Coleoid chromosomal complements are distinct from other molluscs30, yet an in-depth integrated assessment of both chromosomal and sub-chromosomal synteny has been lacking. For this, we conducted synteny analysis on orthologous coding genes to identify homologous chromosomes across several coleoid genomes. This analysis confirms general conservation of chromosomal synteny for the three focal species, E. scolopes, S. officinalis, and O. bimaculoides (Fig. 2)30. While S. officinalis and E. scolopes chromosomes showed almost one-to-one correspondence, their homology to O. bimaculoides was more complex, indicative of the proposed secondary modifications (fusions and translocations) to the octopod karyotypes33 (Fig. 2A). To further quantify the degree of fusions or translocations, we calculated how many 1 Mb, 10 Mb, and 50 Mb windows have single or multiple homologies between the species. For 10 Mb windows, we find 351 instances of pairwise (one-to-one) E. scolopes to O. bimaculoides homologous regions. Additional 80 O. bimaculoides regions had a mapping to this shared set, i.e., resulting in two or more correspondences to a single E. scolopes region (Methods). Conversely, there were 686 additional E. scolopes regions that corresponded to single regions in O. bimaculoides. This suggests that, on average, almost every O. bimaculoides 10 Mb homologous region corresponds to two distinct E. scolopes regions. This pattern, where single O. bimaculoides regions often correspond to two or more E. scolopes regions, is consistent across window sizes, though the number of detected events scales with the size of the window. This genome-wide observation can be explained either by duplication in the squid lineage or fusion-with-mixing in the octopus lineage. As we do not find any duplicated genes on the squid segments that correspond to a single octopus region, these results point to a high degree of within-chromosomal fusion-with-mixing in the Octopodiformes33.

Fig. 2: Genome topology and chromosome evolution across coleoids.Fig. 2: Genome topology and chromosome evolution across coleoids.

A Dot plots of orthologous gene content comparing E. scolopes chromosomes to S. officinalis (top) and O. bimaculoides (bottom). Each dot represents an orthologous gene. Orange dots indicate genes located on chromosome pairs inferred to be orthologous based on significant enrichment of shared orthologous genes (two-sided Fisher’s exact test, P ≤ 1 × 10⁻⁵), whereas grey dots indicate genes located on other chromosome combinations. Black squares highlight the chromosomes shown in panel (B). O. bimaculoides and E. scolopesin particular show extensive chromosomal rearrangements, including large-scale inversions (visible as reverse-diagonal synteny blocks), and chromosomal fusions and fissions, as shown by several one-to-many and many-to-one relationships across species. B Micro-C contact maps for orthologous chromosomes across the three coleoid species; chromosome 11 in E. scolopes, chromosome 6 (OZ199146.1) in S. officinalis, and chromosome 19 in O. bimaculoides, showing 3D genome organisation patterns at 150 kb resolution in the decapodiform species and 100 kb resolution in O. bimaculoides. The regions highlighted by grey boxes correspond to the zoomed-in regions shown in panel (C). Chromatin interaction domains and loops are visible across all species. C Cross-species comparisons of Micro-C contact maps and synteny relationships for zoomed-in regions of the orthologous chromosomes shown in panel (B), highlighting one-to-one orthologous genes (orange rectangles) between E. scolopes, S. officinalis, and O. bimaculoides. Black lines indicate orthology between genes across species. Genes are only included in the plot if they are one-to-one orthologs in all species and located within the regions pictured. Insulation scores at a 350 kb window are represented by orange tracks. Note that the crossed orthology lines in the top panel reflect opposite assembly orientations between the two species rather than a true inversion; gene order within this region is collinear.

To quantify the effect of fusion-with-mixing on local sequence conservation, we analysed the lengths and genomic dispersion of sequence alignments between species. Dispersion is defined as the proportion of translocated regions from homologous to non-homologous chromosomes and correlates with evolutionary distance35. Between E. scolopes and S. officinalis, we found 150,610 and 110,440 total alignments (including both coding and non-coding sequences of 100 bp or longer, 78 Mb total alignment length) on homologous and non-homologous chromosomes, respectively (42.3% dispersion). Between E. scolopes and O. bimaculoides, 35,273 and 29,948 alignments (19 Mb total alignment length) were on homologous and non-homologous chromosomes, respectively (45.9% dispersion), showing a similar dispersion rate with the absolute numbers scaling with the genetic distance. Distinguishing between alignments that overlapped coding regions and those that did not (CNEs), we found that the dispersion rate for CNEs was higher, with only 3185 alignments on homologous chromosomes, yet 22,462 alignments on non-homologous chromosomes for E. scolopes and S. officinalis (87.6% dispersion). Similarly, CNEs comprised only 528 alignments on homologous chromosomes compared to 6182 alignments on non-homologous chromosomes for E. scolopes and O. bimaculoides (92.1% dispersion). This suggests that across coleoids, there are more relaxed constraints on the translocation of putative regulatory non-coding sequences compared with coding regions, likely due to their smaller size, modular function, and reduced sensitivity to disruption compared with protein-coding loci.

While these results highlight the different evolutionary modalities in Octopodiformes compared to Decapodiformes, they cannot provide for a quantitative assessment of how different levels of genome organisation evolved in these clades. For this purpose, we applied a recently established evolutionary genome topology approach (specifically, multi-locus topology34) to 14 cephalopod species. Using pairwise distances and manifold projection, this approach allows us to study multi-scale changes in both coding and non-coding sequences across the coleoids. Coding gene pairwise distances showed several disconnected clusters for both Octopodiformes and Decapodiformes (Fig. S1A). Interestingly, the number of clusters on these topological maps were similar between the two clades and in the range of the decapodiform chromosome number, supporting its more ancestral state33. However, octopodiform genomes showed more connectivity between the orthologous genes, indicative of the proposed additional chromosomal fusion-with-mixing within the octopodiform lineage, leading to a higher degree of sub-chromosomal mixing (putative regulatory entanglement) and a more derived state. This is further supported by the observed higher entropic mixing of these fused chromosomes (Supplementary Note 1). We hypothesise that this more entangled genome architecture in Octopodiformes imposes topological and regulatory constraints that limit the capacity for further genome expansion in this lineage, in contrast to the more extensive genome expansion observed in Decapodiformes. Interestingly, these analyses also show that unlike coding genes that are scrambled across chromosomes and sub-chromosomal topological areas, CNEs are less mixed within the Octopodiformes compared to the coding genes, as evidenced by a line-pattern (as opposed to ‘cloud’, Fig. S1B, C). In Decapodiformes, the CNEs are, on the other hand, more scrambled than the coding genes. While this pattern may partly reflect the sparser octopodiform sampling, with fewer octopodiform species than decapodiform species, and therefore fewer CNEs spanning deeper evolutionary nodes, it remains consistent with the more regulatory-entangled state of the Octopodiformes after enhanced chromosomal mixing of more ancient genomic regions such as the conserved orthologous genes (Fig. 1A).

Together, these data quantify gene and CNE movements across coleoid chromosomes, revealing a pattern that is highly dependent on both the phylogenetic distance and the evolutionary history of chromosomal fusions and translocations. Our results indicate a more derived and mixed (putatively entangled) nature of octopodiform genomes, which is further assessed below.

Topological properties of coleoid genomes

To account for inconsistencies in TAD calling, we evaluated 3D genome organisation directly from the normalised Micro-C interaction signal. This approach captures both multi-scale and nested chromatin domain architecture. Throughout this manuscript, we use the term chromatin domain to describe genomic regions bordered by high insulation, and reserve the term TAD only for regions identified using TAD-calling algorithms36. Overlaying Micro-C contact maps onto genomic coordinates revealed conservation of sub-chromosomal regions across species (Fig. 2B, C). To quantify this conservation, we calculated pairwise interaction intensities (derived from the Micro-C matrices; see Methods) between intrachromosomal orthologous gene pairs in the coleoids. As expected, when calculating all possible gene pair combinations within chromosomes, the majority of gene pairs (53,538) were not in any conserved coleoid interactions. We identified 1592 gene pairs that interacted (KR normalised interaction strength ≥10, Fig. S2) in all coleoid species, 952 gene pairs interacting exclusively in Decapodiformes, and 3634 gene pairs interacting only in O. bimaculoides.

To further characterise pairwise chromatin interaction types, we assessed their co-expression (Fig. 3A), absolute expression (Fig. S3A–C, Supplementary Note 2, and Supplementary Data 3, 4), and functional enrichment properties (Fig. S4 and Supplementary Note 3). In E. scolopes, gene pairs interacting across all three coleoid species exhibited the highest correlation in gene expression across E. scolopes tissues, followed by those interacting in Decapodiformes only. Gene pairs with orthologs interacting exclusively in O. bimaculoides (i.e., not interacting in E. scolopes) showed the second weakest expression correlation across E. scolopes tissues, while those not in any conserved coleoid interaction displayed the weakest correlation (Fig. 3A). All comparisons were statistically significantly different (two-sided Wilcoxon rank-sum test with Benjamini–Hochberg [BH] correction, P < 0.05), with the strongest statistical support observed between the most evolutionarily distinct interaction categories. Gene pairs interacting across the coleoids were also found to be associated with significantly higher (Fig. S3B, C, Supplementary Note 2, and Supplementary Data 3) and broader (Supplementary Note 2 and Supplementary Data 4) expression patterns across tissues, as well as core housekeeping functions (Supplementary Note 3 and Fig. S4). These results suggest that gene pairs maintained in conserved spatial proximity across species are more likely to be functionally coordinated and under strong evolutionary constraints, possibly reflecting shared regulatory mechanisms and core biological roles.

Fig. 3: Co-expression, insulation and intergenic distances for gene pairs across different interaction categories, split by P. maximus chromosomal origin.Fig. 3: Co-expression, insulation and intergenic distances for gene pairs across different interaction categories, split by P. maximus chromosomal origin.

A Pearson’s correlation of gene expression across 19 tissues in E. scolopes for gene pairs grouped by interaction status. Gene pairs interacting across coleoids show the highest co-expression, and all pairwise comparisons were statistically significant (BH-corrected, two-sided Wilcoxon rank-sum test, P < 0.05). Adjusted P values were: no conserved interactions vs. O. bimaculoides only = 2.19 × 10⁻⁷; no conserved interactions vs. Decapodiformes only = 1.82 × 10⁻⁷; O. bimaculoides only vs. Decapodiformes only = 0.0201; interacting across coleoids vs. Decapodiformes only = 0.0415; no conserved interactions vs. interacting across coleoids = 3.72 × 10⁻²³; and O. bimaculoides only vs. interacting across coleoids = 4.87 × 10⁻⁸. B Density plot of average insulation score in E. scolopes for each interaction category. Insulation scores were computed from E. scolopes Micro-C data at 100 kb resolution using a 350 kb window. All pairwise comparisons were statistically significant (BH-corrected two-sided Welch’s ttest, P < 0.05). Adjusted P values were: no conserved interactions vs. interacting across coleoids = 9.92 × 10⁻⁸; no conserved interactions vs. O. bimaculoides only = 1.77 × 10⁻⁶; no conserved interactions vs. Decapodiformes only = 1.95 × 10⁻¹⁶; interacting across coleoids vs. O. bimaculoides only = 8.21 × 10⁻⁴; interacting across coleoids vs. Decapodiformes only = 0.0242; and O. bimaculoides only vs. Decapodiformes only = 8.84 × 10⁻¹⁰. C Proportion of gene pairs within chromatin domains (average insulation score between gene pairs ≥ 0.2) and located across boundaries (average insulation score between gene pairs ≤ −0.2) for each interaction group. D Proportion of gene pairs within chromatin domains (average insulation score between gene pairs ≥ 0.2) and located across boundaries (average insulation score between gene pairs ≤ −0.2) separated by their chromosomal status in P. maximus. E Genomic distances in E. scolopes between gene pairs grouped by P. maximus chromosome status and interaction status. F Proportion of gene pairs within chromatin domains (average insulation score between gene pairs ≥ 0.2) and located across boundaries (average insulation score between gene pairs ≤ −0.2) across P. maximus genomic distance categories. G Genomic distances in E. scolopes between gene pairs grouped by interaction status. H Schematics illustrating typical distances and chromatin domain configurations for gene pairs interacting across coleoids versus those not in conserved coleoid interactions, given their chromosomal origin in P. maximus, as suggested by the patterns observed in (E) (top) and (G) (bottom). Boxplots were generated from orthologous gene pairs, and sample sizes for each group are provided in Fig. S9. Boxplot significance values are based on BH-corrected two-sided Wilcoxon rank-sum tests: NS = non-significant, * = P < 0.05, ** = P < 0.005, *** = P < 0.0005. Exact P values are provided in Supplementary Data 7Aii, Bii. Centre line, median; box limits, upper and lower quartiles; whiskers, 1.5 × interquartile range; points represent outlying gene pairs. Source data are provided as a Source Data file.

When examining the average insulation score between gene pairs (Fig. 3B for E. scolopes; for S. officinalis and O. bimaculoides, see Fig. S5), we found that the vast majority of pairwise comparisons between interaction categories were significantly different (BH-corrected Welch’s t-test, P < 0.05). Gene pairs with no conserved interaction in the coleoids frequently had an insulation score of zero or just below, with little variation around these values. This suggests that these regions lack a distinct topological structure. In contrast, interacting gene pairs exhibited positively skewed insulation scores, indicating that they are often located within chromatin domains. However, the broader range of scores, including negative values, suggests that some interacting gene pairs overlap with chromatin domain boundaries. Consistent with this, when we examined the proportion of gene pairs with high or low insulation scores across species and interaction categories, we found that the vast majority of gene pairs with no conserved interaction in the coleoids were associated with insulation scores ≤ -0.2, whereas interacting gene pairs were much more likely to be found in regions with higher insulation ( ≥ 0.2; Fig. 3C). Notably, mixing entropy was higher for regions containing conserved interacting gene pairs (Supplementary Note 1), indicating greater mixing within chromatin domains and fewer cross-boundary translocations, suggesting that insulation has a direct evolutionary impact.

In the two coleoid species with comprehensive gene annotations, E. scolopes and O. bimaculoides, interacting gene pairs consistently showed higher gene density between them (defined here as the proportion of the intervening genomic region occupied by annotated genes) than pairs not in conserved coleoid interactions. For example, the median gene density within gene pairs not in conserved coleoid interactions was 33.0% in E. scolopes and 33.1% in O. bimaculoides, significantly lower than in the interacting gene pair categories, ranging from 37.7% to 46.2% (Supplementary Data 5). This enrichment suggests that interacting genes are more likely to reside in gene-dense regions, which may facilitate coordinated regulation, and further implies that these interactions are under stronger evolutionary constraint, likely reflecting their functional importance. Additionally, genomic regions between interacting pairs showed depletion of long interspersed repeats in all species and had a higher proportion of simple repeats, suggesting constraint against transposon-driven genomic expansion (Fig. S6).

Chromosomal origin and genome rearrangement shape regulatory interactions and gene pair proximity

To further investigate the relationship between chromatin interactions and ancestral coleoid genome reorganisation, we examined the chromosomal locations of orthologous gene pairs in Pecten maximus, a molluscan outgroup with a chromosome-scale genome (n = 19). P. maximus retains extensive ancestral molluscan linkage structure and therefore provides a useful baseline for evaluating genome reorganisation in the coleoids35. Patterns of chromosomal origin were consistent between P. maximus and Nautilus pompilius (Fig. S7), supporting its use as an appropriate reference for capturing genomic changes associated with coleoid genome evolution.

We found that chromosomal origin influences absolute expression patterns (Fig. S3D–F, Supplementary Note 2, and Supplementary Data 6), co-expression (Supplementary Note 2), and functional enrichment (Fig. S8 and Supplementary Note 3) of interacting gene pairs, but not tissue specificity (Supplementary Note 2). Furthermore, we found that gene pairs not in conserved coleoid interactions (Fig. S9A), as well as those that had lower insulation scores between them (Figs. 3D and S10A) were more frequently located on different P. maximus chromosomes than interacting gene pairs and those with higher insulation scores. This is consistent with the notion that chromatin domain boundaries frequently coincide with synteny breakpoints, regions of the chromosome that are more likely to break and reorganise14,15,16,17.

Across interaction categories, when examining genomic distance between gene pairs, we found that gene pairs interacting across the coleoids, despite large-scale genomic changes, tend to remain in close spatial proximity (Figs. 3E, G, S11, Supplementary Data 7Ai, Bi). In contrast, lineage-specific or species-specific interactions often form longer-distance interactions, potentially resulting from fusions of chromatin domains during genome rearrangement or from regions more prone to genome expansion due to somewhat relaxed regulatory constraints. As expected based on their lack of interaction, gene pairs not in conserved coleoid interactions are consistently separated by very large genomic distances.

Within each interaction category (interacting across coleoids, in Decapodiformes only, in O. bimaculoides only), the genomic distance between gene pairs varied significantly, depending on whether the orthologous genes were located on the same or different P. maximus chromosomes (Figs. 3E, S11A, and Supplementary Data 7Aii). In all species, interacting gene pairs that were located on the same P. maximus chromosome were significantly closer together than those located on different chromosomes. This suggests that interactions between genes that have remained syntenic since the ancestral molluscan genome are under stronger topological constraints, maintaining their proximity despite large-scale genome rearrangements. Alternatively, or in addition, interacting gene pairs may become incorporated into larger regulatory domains when brought onto the same chromosome through chromosomal fusions, which, after mixing, could result in entangled regulatory states, thereby favouring their maintenance over deep evolutionary timescales. Interestingly, this pattern was also significant for gene pairs not in conserved coleoid interactions, though the effect was less pronounced. This may reflect the fact that gene pairs not in conserved coleoid interactions that originate from different P. maximus chromosomes have had less time for intrachromosomal mixing following genome rearrangement, leading to a residual signal of their ancestral chromosomal organisation. An exception to this pattern was observed for decapodiform gene pair distances in decapodiform-specific interactions, which showed similar genomic distances regardless of their chromosomal origin. As these interactions are likely lineage-specific, this pattern may reflect the larger interaction distances in the more expanded genomes of this clade, or, if they are ancestral coleoid interactions that have been lost in O. bimaculoides, moderately relaxed constraints on the regulatory linkages of these gene pairs. Together, our results suggest that the majority of regulatory interactions and chromatin domains at sub-chromosomal levels were maintained following genome rearrangements in the coleoids, whereas many gene pairs without conserved coleoid interactions remained unlinked in their regulation once they were transferred to the same chromosome.

Generally, longer distances between intrachromosomal gene pair orthologs in P. maximus serve as a proxy for a higher chance that coleoid gene pairs were unlinked in their regulation prior to the large-scale chromosomal rearrangements in the coleoid ancestor (Fig. 1C). Topological organisation of coleoid gene pairs across different P. maximus distance categories were mostly evenly distributed (Figs. 3F, S9B, S10B, Supplementary Data 7Bi). However, gene pairs interacting across the coleoids were significantly closer together across all profiled coleoid species if their orthologs in P. maximus were ≤5 Mb apart (Figs. 3G, S11B, Supplementary Data 7Bii). Conversely, beyond this threshold, there were no significant differences in genomic distance between gene pairs interacting across the coleoids. This suggests that while proximity within 5 Mb may increase the likelihood of maintaining a regulatory interaction, interactions occurring at greater distances in the outgroup species do not appear to impact the conservation of spatial proximity. Consequently, coleoid-interacting gene pairs located more than 5 Mb apart in P. maximus are likely to represent post-coleoid-rearrangement linkages rather than ancestral regulatory interactions. For other interaction categories, there were no significant differences in genomic distance between gene pairs in any coleoid species across P. maximus distance categories, with the exception of E. scolopes gene pairs interacting only in O. bimaculoides. Furthermore, there were no significant differences in the genomic distances between gene pairs not in conserved coleoid interactions across P. maximus distance groups. This suggests that, regardless of how closely genes were linked in P. maximus, gene pairs without conserved coleoid interactions are equally affected by genome rearrangement and expansion, likely because their linkages are not functionally important for gene regulation or constrained by regulatory entanglement. Indeed, median genomic distances for these gene pairs were remarkably similar across P. maximus distance groups (Supplementary Data 7Bi), implying that they may evolve under consistently relaxed regulatory constraints. Consequently, the subtle but significant differences observed across categories of interacting gene pairs and chromosomal origin likely reflect meaningful, divergent evolutionary processes rather than random variation.

Together, our results highlight different evolutionary scenarios of chromatin domain evolution (as measured by interacting gene pair distributions), through genome rearrangement and expansion (summarised in Fig. 3H). These models provide a conceptual framework for understanding how chromosomal origin, genomic distances, and evolutionary changes in genome architecture influence 3D genome organisation, and how 3D genome organisation in turn shapes these features.

Chromatin loops represent evolutionarily dynamic structures associated with regulatory element enrichment

To investigate the role of genome topology in forming specific regulatory interactions, we assessed chromatin loop repertoires across coleoid species. In whole-embryo samples we identified 329, 912, and 663 chromatin loops in E. scolopes, S. officinalis, and O. bimaculoides, respectively, based on combined 50 kb and 100 kb resolution loop calls (Methods). After quantifying loops with genes present at both anchors and removing duplicates containing identical gene sets, 156 gene-associated loops in E. scolopes, 402 in S. officinalis, and 388 in O. bimaculoides remained. This indicates that a large proportion of loops lacked genes at one or both ends, potentially reflecting the presence of cis-regulatory elements such as enhancers or silencers, which may mediate both activation and repression of gene expression through chromatin looping37,38,39,40,41. Genes at loop anchors were found to be enriched for dynamic regulatory functions such as responses to stimuli and hormone signalling (Supplementary Note 4 and Fig. S12), as well as cadherin genes (Supplementary Note 4 and Fig. S13).

Across species, only a minority of gene-associated loops overlapped with gene pairs from the four interaction categories described above. Approximately 5% of loops contained gene pairs in any of the three types of conserved interactions (i.e., those interacting across the coleoids, or restricted to interacting across the Decapodiformes or in O. bimaculoides) while ~ 1% overlapped with gene pairs not assigned to any interaction category. Because the four interaction categories were restricted to intrachromosomal orthologs shared across all three coleoid species and the outgroup P. maximus, they represent an evolutionarily conserved interaction potential. This suggests that most chromatin loops are not formed around deeply conserved orthologous multi-gene loci, but may instead reflect novel regulatory configurations involving genes that are scattered across different chromosomes in other species or recently evolved genes.

Loops containing genes were considered to be conserved across species if they included at least one shared orthologous gene overlapping each loop anchor, irrespective of gene orientation. Across E. scolopes, S. officinalis and O. bimaculoides, we found just three loops fitting these criteria (Fig. S14). We identified 27 conserved loops shared exclusively between E. scolopes and S. officinalis (Fig. S15), 6 between E. scolopes and O. bimaculoides (Fig. S16), and 8 between S. officinalis and O. bimaculoides. This small proportion of conserved loops suggests a high degree of loop turnover. The higher number of conserved loops shared between E. scolopes and S. officinalis is consistent with their more recent common ancestry compared to the deeper divergence between decapodiform species and O. bimaculoides (Fig. 1A). Interestingly, it is not only the presence of conserved loops that is maintained across species, but also their internal structural features. For example, we observe conservation of complex loop architectures such as ‘double’ loops (e.g. Figs. S14C, S16F), ‘volcano’-shaped interaction patterns42 (e.g., Figs. S15A, B, S16A, B), and ‘double domain’-like structures within loops (e.g. Fig. S14B). This suggests that even small, subtle differences in local chromatin topology can be functionally relevant and under regulatory constraint.

Genes located in conserved chromatin loops exhibited distinct expression patterns across E. scolopes tissues, with clustering revealing both broadly and tissue-specifically expressed genes (Fig. S17). Genes in loops conserved across all three species were consistently highly expressed in neural tissues and in the white body, a cephalopod-specific hematopoietic and immune organ43. This suggests that these deeply conserved loops may regulate genes involved in cephalopod-specific innovations. Genes in loops conserved between E. scolopes and either O. bimaculoides or S. officinalis showed more variable patterns, potentially reflecting lineage-specific roles.

Motif enrichment analysis of loop anchor regions in whole-embryo samples across species (considering all loops rather than only conserved ones) revealed both shared and species-specific patterns (Supplementary Data 8–10). All three species showed strong enrichment for low-complexity and repeat-associated motifs, including poly-A/T tracts and GA/CA-rich sequences, likely reflecting the high abundance of LINEs in S. officinalis and E. scolopes, and SINEs in O. bimaculoides. Across species, motifs associated with key regulatory families such as Homeobox (Tgif1/2, Nkx3.1), Zinc finger (Maz, ZNF148), bHLH (SCL, Tcf12), Forkhead (Foxa2), and MYB were significantly enriched (hypergeometric test, BH-adjusted P < 0.05), suggesting a role for transcriptional control at loop anchors. CTCF-like motifs were also detected in all three species, supporting the idea that loop extrusion may be a conserved feature of cephalopod 3D genome organisation (as proposed in ref. 8).

We also tested whether CNEs were enriched within loop anchor regions across whole-embryo samples of the three coleoid species (Supplementary Note 5). Across all species, we observed consistent enrichment of CNEs in loop anchors compared to background regions. These findings indicate that loop anchors, where regulatory contacts are often established, are highly enriched for evolutionarily conserved putative regulatory elements. This suggests that although chromatin loops are evolutionarily dynamic and frequently rewired across species, they tend to recruit conserved regulatory elements, repurposing ancestral genomic features to drive lineage-specific patterns of gene regulation. In all three E. scolopes developmental stages with available ATAC-seq data, loop anchor regions showed significantly greater chromatin accessibility than background regions (Supplementary Note 5). These findings are consistent with previous studies and indicate that loop anchors represent focal points of elevated chromatin accessibility, supporting the hypothesis that loops function as organisational hubs for transcriptional control19,44. Together with our GO, CNE, and motif enrichment results, this supports a model in which loop anchors are functionally enriched loci that contribute to dynamic gene regulation patterns.

Loops also showed significantly lower gene density between them than conserved interacting gene pairs (Supplementary Note 5). This may reflect their roles as dynamic regulatory elements subject to fewer constraints, allowing them to emerge in gene-sparse regions that are more permissive to structural change. Gene sparsity may also facilitate the accumulation of repetitive elements between anchors, contributing to the low conservation of loops across species.

Chromatin loops show distinct scaling properties depending on their evolutionary history

To test whether conserved chromatin loop size scales with genome size across species, we calculated the linear regression slope between genome size and loop size for each loop conserved in at least two species. A two-sided Wilcoxon signed-rank test on the per-loop slopes showed a statistically significant positive median slope (P = 0.014), indicating that conserved loop size increases with genome size across species (Fig. S18). This is consistent with these loops being ancestral coleoid structures affected by species-specific genome expansions.

Next, we compared all chromatin loop sizes across whole-embryo samples of E. scolopes, S. officinalis, and O. bimaculoides, expanding the analysis beyond only conserved loops to include all loops with at least one gene at each of their anchor points. Unlike the patterns we observed for the conserved loops, despite having the smallest genome, O. bimaculoides exhibited the largest and most variable loop sizes, while E. scolopes and S. officinalis showed consistently smaller loops with narrower distributions (Fig. 4A). Median loop sizes were smallest in S. officinalis, followed by E. scolopes, with O. bimaculoides having the largest median (medians = 1.3 Mb, 1.6 Mb, 2.9 Mb, respectively). All pairwise comparisons were statistically significant. This pattern suggests that overall, loop size is negatively correlated with genome size, and may reflect lineage-specific differences in genome rearrangement rather than genome expansion. In particular, the large loops in O. bimaculoides may result from its extensive history of chromosomal fusions, representing entangled regulatory states, potentially requiring longer-range contacts between regions that were previously on separate chromosomes.

Fig. 4: Chromatin loop configuration is shaped by lineage-specific chromosomal rearrangements and an example of a dynamic loop across development.Fig. 4: Chromatin loop configuration is shaped by lineage-specific chromosomal rearrangements and an example of a dynamic loop across development.

A Boxplots of loop sizes across species for loops containing genes at loop anchors detected in whole-embryo samples (E. scolopes, n = 152 loops; S. officinalis, n = 402 loops; O. bimaculoides, n = 386 loops). Loops whose size could not be reliably estimated owing to likely genome misassemblies were excluded (Methods), so sample sizes differ slightly from the total number of gene-associated loops. Boxplot significance values are based on BH-corrected two-sided Wilcoxon rank-sum tests: * = P < 0.05, ** = P < 0.005, *** = P < 0.0005. Exact BH-adjusted P values were 0.020 (S. officinalis vs. E. scolopes), 0.0083 (O. bimaculoides vs. E. scolopes), and 1.22 × 10⁻⁸ (O. bimaculoides vs. S. officinalis). Centre line, median; box limits, upper and lower quartiles; whiskers, 1.5 × interquartile range; points represent outlying loops. B Percentage of loop anchor genes in E. scolopes, S. officinalis and O. bimaculoides with orthologous genes located on multiple different chromosomes in the other coleoid species’ genome. C Loop sizes in each species, grouped by whether genes at loop anchors have orthologs on the same or on different chromosomes in another coleoid species (E. scolopes and S. officinalis relative to O. bimaculoides; O. bimaculoides relative to E. scolopes and to S. officinalis). Sample sizes (n; same-/different-chromosome groups) were 83/69 loops for E. scolopes, 265/137 for S. officinalis, 156/230 for O. bimaculoides relative to E. scolopes, and 151/235 for O. bimaculoides relative to S. officinalis. Boxplot significance values are based on BH-corrected two-sided Wilcoxon rank-sum tests: NS = non-significant, * = P < 0.05, ** = P < 0.005, *** = P < 0.0005. Exact BH-adjusted P values were 0.11 (E. scolopes), 0.091 (S. officinalis), 0.019 (O. bimaculoides relative to E. scolopes), and 2.1 × 10⁻⁵ (O. bimaculoides relative to S. officinalis). Centre line, median; box limits, upper and lower quartiles; whiskers, 1.5 × interquartile range; points represent outlying loops. D Micro-C contact maps for a 5 Mb region of chromosome 5 for E. scolopes across three developmental stages: (Di) stage 20 (early organogenesis), (Dii) stage 25 (late organogenesis), and (Diii) stage 29 (pre-hatching). A significant reduction in loop strength was observed between stage 20 and 25 (FDR = 0.0032) and between stage 20 and 29 (FDR = 0.0094), as identified by Mustache’s differential loop calling approach. The loop present at stage 20 is indicated, and is missing at the later two developmental stages. Normalised ATAC-seq (from8) signal tracks show chromatin accessibility at each stage. E Insulation scores and the insulation score difference around conserved loop anchors across developmental stages in E. scolopes. Line plots show average insulation scores at a 750 kb window across stage 20, 25, and 29. Highlighted regions correspond to genes at loop anchors. ‘2x Chromatin assembly factor 1’ indicates 2 genes with the same name in that location. F Heatmap showing gene expression patterns of genes located at conserved loop anchors across 8 developmental stages in E. scolopes. Columns represent the average log-transformed TPM values across the three biological replicates for each developmental stage. Loop anchor genes include Prefoldin, Sirtuin, Sas10/Utp3/C1D, snRNA-activating complex (SNAPc), and two subunits of Chromatin Assembly Factor 1 (CAF-1). Source data are provided as a Source Data file.

To explore whether this increase in loop size is indeed linked to changes in genomic organisation, we examined the chromosomal origin of genes located at loop anchors. Specifically, we assessed whether loop-associated genes were located on the same or multiple different chromosomes in the other coleoid species and in P. maximus. A substantial proportion of loops in all species contained genes whose orthologs were located on different chromosomes in the other species, but this was particularly pronounced in O. bimaculoides, where over 50% of loops contained genes from different decapodiform chromosomes (Fig. 4B), and 70% from different P. maximus chromosomes (Fig. S19). Notably, this contrasts with our earlier finding that only ~ 40% of conserved interacting orthologous gene pairs were located across different P. maximus chromosomes (Fig. S9), suggesting that chromosomal fusions more strongly shape loop formation than conserved gene pair interactions.

Additionally, we compared gene-associated loop sizes in each species based on whether the genes at loop anchors were located on the same or multiple different chromosomes in the other coleoid species. In all three species, loops connecting genes from different chromosomes in the other coleoid species tended to be larger than those connecting genes from the same chromosomes in the other coleoid species (Fig. 4C). This trend was especially marked in O. bimaculoides, where the difference was statistically significant, consistent with the idea that extensive chromosomal fusions in the octopodiform genomes have created longer-range loop contacts. In E. scolopes and S. officinalis, the difference in loop size between same- and different-chromosome gene pairs was less pronounced and not statistically significant, suggesting that larger loops are most strongly associated with the extensively fused O. bimaculoides genome. Notably, loop size did not correlate with the intergenic distance between orthologous gene pairs on the same chromosome in the other species (R² <0.01 across all comparisons), indicating that the increased size of intrachromosomal loops is not simply due to ancestral gene spacing but is likely a direct consequence of chromosomal rearrangement events.

These results align with our previously outlined evolutionary and mixing analyses and corroborate findings on widespread chromosomal fusions in Octopodiformes33. Interestingly, this pattern also parallels the fusion-with-mixing scenario observed for gene pairs from different P. maximus chromosomes, where longer intergenic distances reflect residual signals of their original chromosomal configuration prior to rearrangement, but here occurring at a shorter evolutionary timescale. Together, these findings further support the dynamic nature and more recent evolutionary origin of chromatin loops in the coleoids.

Chromatin loop formation is highly context-specific

Since chromatin loop conservation across species is minimal, we next examined both the full set of detected loops (Fig. S20) and those specifically associated with genes (Fig. S21) across developmental stages and tissues. Chromatin loop abundance varied substantially in both contexts. A representative example of a significantly differential loop and its associated expression patterns across development is shown in Fig. 4D–F. This loop is present at stage 20 but significantly downregulated at stages 25 and 29 (FDR < 0.01 for both comparisons; Fig. 4D). At stage 20, the loop coincides with elevated chromatin accessibility (ATAC-seq peaks; Fig. 4D), local insulation at loop borders (Fig. 4E), and peak gene expression (Fig. 4F), suggesting it coordinates the activation of this regulatory module during the onset of tissue differentiation. The loop brings together six genes; Prefoldin, two subunits of Chromatin Assembly Factor 1 (CAF-1), the snRNA-activating complex (SNAPc), Sas10/Utp3/C1D, and Sirtuin, all of which are functionally linked to chromatin assembly, cytoskeletal organisation, and RNA processing, processes essential for early development. Each of these genes has well-established developmental roles: Prefoldin facilitates the folding of cytoskeletal proteins required for morphogenesis and cellular proliferation45; CAF-1 subunits organise chromatin and deposit histones to preserve epigenetic states during cell proliferation46; SNAPc regulates snRNA transcription necessary for pre-mRNA splicing during zygotic genome activation47. Sas10/Utp3/C1D proteins are involved in small subunit rRNA processing and ribosome biogenesis48,49; and Sirtuin modulates chromatin accessibility and epigenetic silencing during developmental patterning and cell fate specification50,51. Together, their activity likely orchestrates key transitions during early organogenesis. By stage 25, the disappearance of this loop is accompanied by a reduction in both insulation and expression, consistent with a shift from specification to more stable lineage commitment. Notably, while SNAPc and one CAF-1 subunit (cluster_18817) are syntenic in the outgroup P. maximus genome, the remaining genes are dispersed across the same or different chromosomes in P. maximus. This suggests that the looped configuration is a synapomorphic feature, likely facilitated by genome rearrangements in the coleoid ancestor, and may have provided a regulatory context that enabled the emergence of cephalopod novelties such as their unique developmental programme.

In E. scolopes, stage 25 embryos exhibited substantially more loops than either stage 20 or stage 29 (Figs. S20 Ai, S21Ai). As stage 25 marks the final stages of organ differentiation and functional maturation during late organogenesis52, this elevated loop abundance may reflect the widespread changes in gene expression occurring at this stage, as well as the regulatory complexity needed to establish stable, mature expression programmes. In O. bimaculoides, neural tissues, i.e. the arm crown, eyes and optic lobes, and the central brain, showed similar numbers of loops, while internal organs had about one third as many (Figs. S20Bi, S21Bi), consistent with the heightened regulatory demands of neural tissues31. In contrast, S. officinalis showed a distinct pattern: the eyes had by far the highest number of loops, followed by the optic lobes, with the arm crown, central brain, and internal organs exhibiting substantially fewer (Figs. S20Ci, S21Ci). Notably, in S. officinalis, retinal determination genes are expressed specifically in the eye and optic lobe during embryogenesis, linked to the formation of local regulatory architectures driving neural complexity53. Furthermore, the cuttlefish optic lobe continues to mature into adulthood54, paralleling ongoing increases in regulatory and circuit complexity. The contrasting tissue-specific patterns between O. bimaculoides and S. officinalis suggest that the regulatory role of chromatin loops is distributed differently across neural and peripheral tissues in these species, likely reflecting lineage-specific adaptations in gene regulation and differences in tissue complexity and function.

Pairwise comparisons across all developmental stages (Figs. S20Aii, S21Aii) and tissue combinations (Figs. S20Bii, Cii, S21Bii, Cii) revealed no statistically significant enrichment of shared loops, regardless of whether loops contained genes (SuperExactTest, BH-adjusted P > 0.05). This underscores the dynamic and context-dependent nature of chromatin loop architecture, further supported by the distinct GO and motif enrichment patterns observed across tissue-specific samples (Figs. S22, S23, S24 and Supplementary Note 4).

To gain deeper insight into how context-dependent loop formation relates to transcription, we examined expression patterns of genes within developmental stage-specific loops in E. scolopes. Genes in stage-specific loops tend to be more highly expressed at their corresponding developmental stage and exhibit lower expression at other stages (Fig. S26A). However, these overall trends are not always statistically significant, suggesting more complex regulatory dynamics. To investigate this further, we visualised gene expression within individual loops using heatmaps (selected example in Fig. 4F, heatmaps across all developmental stage-specific loops in Fig. S26B). These heatmaps reveal diverse expression patterns across developmental stages. In some cases, loops are associated with coordinated upregulation of genes at a specific stage, while in others, loops correspond to downregulation. Most frequently, we observe mixed patterns, with some genes upregulated and others downregulated within the same loop. Consistent with this, genes within stage-specific loops exhibited low co-expression across developmental RNA-seq data (median pairwise Pearson’s correlation coefficients across eight developmental stages = 0.044, 0.203, and 0.200 for E. scolopes stages 20, 25, and 29, respectively). Therefore, overall, context-specific loop-associated genes exhibit complex expression dynamics, suggesting an interplay between diverse regulatory elements, such as enhancers and repressors, operating within the same 3D chromatin context.

Together, these findings suggest that chromatin looping is not only poorly conserved across evolutionary time but is also highly modular within species, with distinct regulatory interactions likely tailored to the specific transcriptional needs of developmental stages or individual tissues. This context-specific organisation may enable precise temporal and spatial control of gene expression during complex processes such as organogenesis and tissue-specific functions, and may be facilitated by relaxed evolutionary constraints that allow greater plasticity in loop formation and usage. This is in contrast to the more stable chromatin domains as measured by the interacting gene pair approach or TAD predictors (Supplementary Note 5 and Fig. S25).

Knockout of putative regulatory sequence reveals long-range meta-loop-like interactions

Our genome-wide analyses indicated that although most chromatin loops are evolutionarily dynamic and highly context-specific, a small subset is deeply conserved across coleoid whole-embryo samples. Furthermore, loops are enriched for TF binding motifs, CNEs, and accessible chromatin, and are associated with GO terms and expression patterns associated with neural gene regulation, suggesting potential roles in orchestrating transcriptional programmes essential for nervous system development. To functionally test whether conserved loops harbour regulatory elements critical for gene expression, we used CRISPR-Cas9 to disrupt a candidate regulatory region within one such conserved loop in Euprymna berryi, a close relative of E. scolopes with established genetic tools55.

To this end, we targeted a ~ 1 kb region within an intron of the gene Quaking B, located in a conserved coleoid chromatin loop (Fig. 5A), which contains genes upregulated across neural tissues and developmental stages (Fig. S27A, B) but is scrambled within and across chromosomes in outgroup molluscs, suggesting a derived regulatory architecture in the coleoids. Quaking B was chosen as the host gene to identify a putative intronic regulatory sequence due to the availability of sufficiently long, non-repetitive non-coding regions within it and evidence for its role in neural development56,57,58. Furthermore, the targeted intronic region is conserved among Decapodiformes and also has a conserved region shared with O. bimaculoides located ~ 150 bp upstream (Fig. 5B). The repetitive nature of coleoid genomes makes it difficult to design CRISPR targets in non-coding regions, and the three guides selected (Fig. 5B) represented the only possible knockout scenario without predicted off-target effects. The deletion was confirmed by genotyping, using PCR amplification of the target region followed by Sanger sequencing (Fig. S28A, B).

Fig. 5: CRISPR knockout of a putative regulatory region reveals long-range meta-loop-like interactions and transcriptional effects in coleoid embryos.Fig. 5: CRISPR knockout of a putative regulatory region reveals long-range meta-loop-like interactions and transcriptional effects in coleoid embryos.

A Micro-C contact matrices showing the conserved chromatin loop between E. scolopes (top) and O. bimaculoides (bottom). The gene (Quaking B) containing the target intron is highlighted in yellow and indicated with a grey dotted arrow, with other genes within the loop shown in orange. Black lines indicate orthology between loop genes across species. Contact maps are shown at 50 kb resolution for E. scolopes and 25 kb for O. bimaculoides. To improve clarity, background genes were removed from the gene annotations overlaid on the contact maps. Gene colouring, orthology, and resolution are consistent with (E) and (F), where applicable. Note that the crossed orthology lines here and in (E) reflect opposite assembly orientations between the two species rather than a true inversion; gene order within this region is colinear. B Schematic representing the CRISPR targeting of the intron region in Quaking B. The top panel depicts the gene structure of Quaking B (EB05924.1) in yellow and the intron region targeted by CRISPR highlighted in red in E. berryi. The bottom panel shows a detailed view of the targeted E. berryi intron region, showing conserved BLAST hits from E. scolopes, O. bimaculoides, and D. pealeii aligned to the E. berryi genome. The red bar indicates the E. berryi target intron region, blue bars show orthologous BLAST hits in other coleoid species, and green boxes mark the CRISPR guide sites. The knockout region spans from guide 1 to guide 3. C Representative images of control (left) and knockout (right) E. berryi embryos at stage 28, showing morphological defects, including smaller optic lobes (arrow) and smaller or missing arms (star), following deletion of the target regulatory region. D Heatmap of expression levels (VST-normalised) for all transcripts of genes located within the conserved loop across control and knockout E. berryi embryos at stages 14 and 28. Columns represent the average VST-normalised expression across biological replicates for each condition and developmental stage (n = 4 for stage 14 knockout embryos; n = 3 for all other groups). Quaking B, the gene with the target intron, is indicated by a yellow star. Genes from top to bottom of the heatmap correspond to those from downstream to upstream of the loop in the E. berryi, but right to left along the loop in the E. scolopes contact map shown in (A). Transcripts with zero variance in expression were removed from the heatmap. E Contact matrices and topology of distal interactions (purple arrows) conserved across E. scolopes and O. bimaculoides that were significantly downregulated upon knockout in E. berryi. Significantly downregulated genes are depicted in purple. F Contact map of a region E. scolopes showing topological positioning of downstream downregulated genes in E. berryi knockout, most of which lack strong BLAST hits and are unannotated in O. bimaculoides, suggesting they are clade-specific novel genes. Downregulated genes are represented in purple and their topological linkage to the target intron is indicated by purple arrows. Three genes are depicted, though it appears as only two because the two on the right overlap visually. G Heatmap of (VST normalised) expression of significantly downregulated genes downstream of the target intron in knockout embryos, showing consistent downregulation at stage 28. Columns represent the average VST-normalised expression across biological replicates for each condition and developmental stage (n = 4 for stage 14 knockout embryos; n = 3 for all other groups). Sulfotransferase 1C4 is the downregulated gene depicted in purple in (E) and the three unknown genes are those shown in purple in (F). H Schematic summary of knockout results and conservation status across coleoids. This model illustrates how conserved regulatory introns can regulate distal genes through long-range chromatin looping, and how these loops, despite being conserved, may undergo topological mixing of their contacts over evolutionary time, resulting in the rewiring of regulatory networks. Grey arrows represent variable changes in gene expression following knockout of the regulatory element, while blue arrows indicate significant downregulation. Green ticks denote conserved interactions in both gene order and contact size; the single green arrow shows partial conservation (i.e., conserved interaction but with differences in region size or gene order); the red cross marks an interaction not conserved in other coleoid lineages. Source data are provided as a Source Data file.

To investigate the potential regulatory role of the ~ 1 kb intronic sequence deleted by CRISPR–Cas9 (Fig. 5B), we performed motif enrichment analysis using HOMER (Supplementary Data 11). After correction, a Hox-like motif (Hoxb4) was significantly enriched (BH-corrected P < 0.05), consistent with input from homeodomain transcription factors involved in body patterning and neural development59. Additional metazoan motifs, including Lhx6 (LIM-homeobox), bHLH factors, and IRF-associated motifs, were detected at nominal levels, suggesting diverse regulatory inputs also associated with neural and developmental processes60,61,62,63. Notably, MYNN (Myoneurin) motifs64 were present within the deleted sequence and in an adjacent intronic region conserved in O. bimaculoides (Fig. 5B), which lies within a repetitive interval not targetable by guide RNAs. This suggests that the targeted region forms part of a broader regulatory module, with the deleted sequence representing one functional component. Together, these findings support the interpretation that the targeted intronic region represents a putative regulatory element, likely contributing to neural and developmental gene regulation.

Among the 25 CRISPR-injected stage 28 embryos, all showed signs of developmental delay. Five embryos exhibited a clear reduction in brain size, particularly in the optic lobe. An additional five were severely malformed, with multiple disrupted body structures, including neural tissues, while six appeared smaller overall with notably shorter arms. Nine did not show an overt phenotype. Thus, depending on classification criteria, approximately one third to two thirds of embryos displayed brain-specific morphological defects, consistent with developmental impairment of neural structures (Figs. 5C and S28C, D).

To assess transcriptional consequences, we conducted whole-body RNA-seq at stages 14 and 28. Stage 14 was selected because Quaking B, the target gene, exhibits peak expression at this developmental stage (Fig. S27B). PCA plots (Fig. S29) showed separation between control and knockout samples based on gene expression at both stage 14 and stage 28. At stage 28, samples with a recorded phenotype cluster separately from those without, suggesting a stronger transcriptional response associated with the observed phenotype. Although genes within the conserved loop exhibited expression variability (Fig. 5D), no statistically significant changes were detected. The most pronounced downregulation within the loop was observed in the host gene Quaking B itself at stage 14 and stage 28, although this was not significant after BH correction.

Strikingly, significant transcriptional effects were observed at long range. Upstream of the target site, a single significantly downregulated gene conserved with O. bimaculoides was identified, located 14.5 Mb away in E. berryi and 30.5 Mb in E. scolopes. Although these genes are more distant in E. scolopes, they are positioned much closer to the core conserved loop in O. bimaculoides (1.6 Mb) and they appear to be topologically linked in all three species (Fig. 5E and Fig. S30). Downstream of the E. berryi target, the nearest significantly downregulated genes (Fig. 5F, G) were located 2.9–4.1 Mb away in E. berryi and 2.9–4.2 Mb in E. scolopes. In O. bimaculoides, these genes are unannotated or on different chromosomes to the target loop and most lacked strong homology to other metazoans, indicating they are likely coleoid-specific. The significantly downregulated genes showed high expression in E. scolopes adult neural tissues, as well as the mantle and gills (Fig. S27C) and at late stages during E. scolopes embryonic development (Fig. S27D). The conserved loop in E. berryi, along with the regions containing genes downregulated following knockout of the target intron, are shown in Fig. S30 for both E. berryi and S. officinalis.

Given the phenotypic and transcriptional consequences of the knockout, we infer that the deleted region integrates regulatory inputs critical for neural gene expression and development. These results suggest that the intronic sequence acts as a regulatory hub, contributing to the expression of genes within its chromatin interaction network. Long-range effects of the knock-out are consistent with a meta-loop architecture, in which regulatory elements exert influence across chromatin domain boundaries and over large genomic distances65. The downregulated genes span both conserved and lineage-specific targets, suggesting evolutionary flexibility in regulatory wiring despite structural conservation of the core loop. The more limited expression changes observed within the conserved loop, compared to the stronger transcriptional effects on genes beyond the loop, suggest that genes embedded within highly conserved and likely regulatorily entangled loop topologies are more buffered against perturbation of single CNEs, whereas genes involved in more distal and less conserved regulatory interactions show stronger transcriptional responses when regulatory regions are disrupted. Supporting this, the corresponding loop in O. bimaculoides shows tighter co-localisation between enhancer and promoter, suggesting additional regulatory entanglement following chromosomal fusion and mixing in the Octopodiformes. Together, these results support a model in which conserved loop topologies retain core regulatory functions while allowing flexibility in their gene targets over evolutionary time. Figure 5H illustrates this scenario, showing how chromatin loops can mediate long-range regulatory interactions that are constrained and evolvable at the same time.

Concluding remarks

How chromatin structures and regulatory interactions change throughout evolution has only been studied in a few model systems65,66. Recent work highlights diverse patterns of chromatin architecture across animal lineages9,11,19. Coleoid cephalopods offer a powerful model for exploring how macroevolutionary processes shape and maintain regulatory landscapes, owing to their ancient and extensively reorganised and expanded genomes and striking phenotypic novelties.

By analysing 3D genome evolution across three coleoid species in the context of their unique evolutionary history, our study provides insight into the origin, maintenance, and plasticity of chromatin structures and their associated regulatory patterns. We show that in the expanded genomes of Decapodiformes, the regulatory interactions have more ancestral signatures. In the less expanded O. bimaculoides genome, our data on syntenic comparisons, chromatin domain evolution, and loop dynamics support a model of lineage-specific chromosomal fusion-with-mixing, resulting in potentially more entangled regulatory configurations22 and a more derived topological state for Octopodiformes. Furthermore, we show that while higher-order chromatin domain structures are generally conserved, chromatin loops are flexible and enriched in dynamic regulatory features. Knockout of a putative regulatory site confirmed the functional relevance of long-range loop-mediated interactions with both conserved and novel genes.

Crucially, our results reveal a hierarchy of regulatory constraints across the 3D genome (summarised in Fig. 6), in which different classes of topological interactions, defined by their ancestral chromosomal state, clade specificity, and 3D structure, show distinct levels of evolvability. This variation in constraint influences how genome rearrangement and expansion affect different parts of the genome. Rather than uniformly reshaping regulatory architecture, these processes selectively rewire, retain, or eliminate interactions based on their underlying level of constraint. This allows some structures to maintain phenotypic stability, while others remain more flexible and enable regulatory innovation. Furthermore, mixing of multiple regulatory interactions can lead to entangled regulatory configurations, such as the particularly pronounced pattern observed in O. bimaculoides, which, depending on their complexity, may become increasingly resistant to reorganisation22,34. This dynamic interplay between topological flexibility and constraint gives rise to a distinct regulatory landscape in the coleoids, shaped not only by the scale of genomic change, but also by the functional importance, regulatory context, and structural organisation of the affected regions.

Fig. 6: Hierarchy of regulatory constraint across chromatin interactions in coleoid cephalopods.Fig. 6: Hierarchy of regulatory constraint across chromatin interactions in coleoid cephalopods.

Schematic representation of regulatory constraint levels across different classes of topological interactions in the coleoids. This continuum reflects a hierarchy of stability shaped by the interplay between ancestral chromosomal state, clade specificity, and type of 3D structure. Differential constraint modulates the extent to which large-scale genomic processes, such as rearrangements and expansions, reshape regulatory architecture, ultimately contributing to the emergence of the distinct gene expression patterns and phenotypic innovations in the coleoids. In the first column, rectangles represent genes, triangles represent chromatin domains, and dots within triangles mark chromatin loops; the schematic below depicts the corresponding three-dimensional chromatin loop structure. Interaction classes are colour-coded as in previous figures: pink indicates interactions shared across all coleoids; green and purple indicate Decapodiformes- and O. bimaculoides-specific interactions, respectively; red denotes gene pairs located on the same P. maximus chromosome, and blue indicates those on different P. maximus chromosomes. Orange indicates chromatin loops. Median distances between interacting gene pairs and gene expression statistics (fourth column) are reported for E. scolopes unless otherwise stated.

Taken together, our findings establish coleoid cephalopods as a model system for understanding how 3D genome architecture, differential evolutionary constraints, and large-scale chromosomal changes interact to drive regulatory innovation, illustrating how spatial genome organisation both shapes and is shaped by evolution over deep timescales.