Are dinoflagellate chromosomes really cholesteric liquid crystals?

What the cholesteric model predicted for Hi-C

The cholesteric liquid crystal (CLC) model, proposed more than fifty years ago, describes a dinoflagellate chromosome as a stack of flat disks. Inside each disk, chromatin fibers run parallel to a director, and that director rotates by a constant angle between adjacent disks, defining the cholesteric pitch. Disks follow one another in the same order as the primary DNA sequence, and extrachromosomal loops protrude outward, connecting either the same disk or adjacent disks. Because those loops were thought to be more accessible than the condensed core, they were also proposed to host the transcriptionally active genes.

The paper's first move is to turn this geometric description into a testable prediction. The authors generated a population of 10^6 simulated CLC structures, roughly matching the number of cells sequenced in a typical bulk Hi-C experiment, and added randomly placed extrachromosomal loops with exponentially distributed lengths. They then computed the Hi-C contact map this population would produce. The result is a grid-like pattern: stripes of high contact probability parallel to one another and perpendicular to the central diagonal, with the number and spacing of stripes set by the number and diameter of the disks. That grid survives changes in loop fraction, pitch, and disk tapering, so it is a robust signature rather than a tuning artifact.

The simulated grid does not match dinoflagellate Hi-C

Empirical Hi-C maps from Fugacium kawagutii and Symbiodinium microadriaticum look qualitatively different. Their dominant feature is topologically associating domains (TADs): blocks of primary sequence that contact each other far more often than they contact the rest of the chromosome. The CLC simulations produce no such blocks. Quantitative TAD metrics behave accordingly — the insulation score and directionality index oscillate smoothly with a constant period for CLCs, but vary irregularly and abruptly in the real chromosomes.

The mismatch is also quantitative. Contact maps from the CLC model predict experimental data only modestly, with Pearson correlations of r = 0.62 for F. kawagutii and r = 0.59 for S. microadriaticum. This is the paper's first line of evidence, and it is a comparison between a model-generated map and an experimentally measured one, not a direct observation of molecular structure.

Two reconstruction methods both drop the disks

To ask what conformation the real Hi-C data actually support, the authors used two modeling approaches rather than one, explicitly to reduce algorithm bias. Consensus modeling with CSynth treats Hi-C contacts as spring-like forces between monomers and outputs a single best-fitting conformation; it was applied to all F. kawagutii and S. microadriaticum chromosome scaffolds at 5 kb per monomer resolution. Population-based modeling with Integrative Genome Modeling (IGM) uses maximum likelihood estimation to produce a population of conformations statistically consistent with bulk Hi-C data; it generated 100 conformations for the largest chromosome scaffold of each species, also at 5 kb resolution.

Neither approach recovers cholesteric disks. CSynth conformations are compact and ellipsoidal, while IGM conformations can be elongated or spherical, with multiple compact domains and long extrachromosomal loops. The two methods also disagree with each other in instructive ways: CSynth's ellipsoids resemble shapes seen in transmission electron microscopy, whereas IGM sometimes produces dumbbell and irregular shapes that are less common in the microscopy literature. Yet IGM conformations reproduce the empirical Hi-C maps and contact probability curves far better, with correlations of r = 0.91 for F. kawagutii and r = 0.84 for S. microadriaticum, and their simulated maps show squares along the diagonal rather than stripes. Notably, CSynth can be made to reproduce CLC-like conformations while IGM cannot.

A third species supports the pattern with a caveat

Breviolum minutum provides a partial replication. Its available Hi-C data cover an estimated 39% of the genome, compared with 70% for F. kawagutii and 72% for S. microadriaticum, so its scaffolds may not correspond to full chromosomes; the authors therefore reconstructed the 10 largest Hi-C scaffolds rather than whole chromosomes. Consensus modeling gave spherical conformations with many extrachromosomal loops (5.3 per scaffold), unlike the ellipsoids with few loops seen in the other two species, and population-based modeling gave similar spheres with 3.7 loops per conformation. The authors attribute these differences to incomplete scaffolds rather than to genuine structural differences in B. minutum. Even so, the IGM population reproduced the observed Hi-C map with r = 0.93, and both showed TAD-like squares rather than CLC-like stripes.

TADs can be grafted onto a cholesteric stack

The paper does not claim the CLC geometry is incompatible with TADs in principle. When the authors grouped cholesteric disks according to observed dinoTAD sizes and introduced locus-specific boundaries between groups, permuting disk order within groups but not between them, the simulated contact map developed squares of high contact probability along the diagonal — the signature of TADs. The insulation score and directionality index then varied irregularly, as in the real data.

This is a compatibility result, not a rescue of the model. The authors' point is that locus-specific boundaries plus conformational heterogeneity, the ingredients of loop extrusion in mammalian cells, are what generate TAD-like patterns; the cholesteric stack alone does not. The experiment shows which added ingredient is necessary, and it also shows that the presence of TADs cannot by itself be used as evidence for cholesteric disks.

Order is moderate and local, not cholesteric

The authors quantified orientational order with an order parameter (OOP) computed from tangent vectors between monomers at varying primary-sequence separation, plus its Fourier transform. A simple helix and consecutive CLC disks give an oscillatory signature; equilibrium globules, which are confined random walks without orientational order, decay monotonically to zero with no peaks. Consensus conformations of F. kawagutii, S. microadriaticum, and B. minutum match neither extreme. They show a negative peak at roughly 10^5 bp, indicating partial anti-parallel orientation at that scale, which corresponds to the average length of gene arrays. No orientational order is detected at the scale of dinoTADs (~10^6 bp), likely because those domains are structurally heterogeneous.

A second measure, the local nematic order parameter, was computed within small and large voxels across 75 structures or conformations, capturing order that depends on 3D spatial proximity rather than sequence separation. Together these measures support the paper's summary description: dinoflagellate chromosomes carry moderate orientational and nematic order, more than expected by chance but less than CLCs. The authors also note that the OOP peaks at gene-array length suggest gene arrays may fold in an anti-parallel orientation, and that transcription of co-oriented genes in divergent arrays could produce such an arrangement — an interpretation offered as a possibility, not a demonstrated mechanism.

Transcription is not concentrated at the chromosome edge

The CLC model makes a functional prediction as well as a structural one: if extrachromosomal loops at the chromosome surface are the accessible sites, transcriptionally active genes should be enriched there. The authors tested this by mapping RNA-seq reads onto their simulated conformations. The largest 75 chromosome scaffolds per species were scaled to the same size with the long axis horizontal, and reads were counted in voxels of a cylindrical coordinate system, integrated azimuthally, and divided by the cylindrical Jacobian to control for the varying amount of DNA per voxel.

Transcription levels at the chromosome surface were comparable to those in the interior, not elevated. The authors read this as evidence that although dinoflagellate chromosomes are highly condensed, they remain accessible to RNA polymerase. This is a statement about the spatial distribution of transcription in the modeled conformations, and it directly contradicts the accessibility rationale that motivated the extrachromosomal-loop part of the CLC model.

What the reconstruction cannot settle

The conclusions rest on polymer conformations inferred from Hi-C data, and the authors are explicit about the limits of that inference. Different computational methods can generate different outputs from the same input, which is why two packages were used; their disagreement over chromosome shape is reported rather than hidden. The B. minutum analysis is qualified by incomplete genome coverage. And the broader structural debate has been shaped by two-dimensional transmission electron microscopy cross-sections, in which fiber orientation has never been imaged in 3D; interpreting banding patterns has relied on projecting 3D CLC models onto a 2D plane. The authors note that chromosomes producing such banding appear helical after whole-mounting, and that a recent study found fibers not parallel within transverse sections.

The paper's own priority for follow-up follows from this gap: 3D imaging and chromatin fiber tracing inside dinoflagellate chromosomes, together with 3D imaging of nascent RNA and transcription. It also points to integrating orthogonal data types such as FISH, DamID, and SPRITE into population-based modeling, which IGM can already do but which are not yet available for dinoflagellates. The authors further note that their conformations are inconsistent with the other proposed models too — interwoven double helices, twisted toroids, polytene bands, hierarchical helices, and helical cores with abundant loops — since the Hi-C maps lack the X-shape of circular chromosomes, show no variable contact probability along the diagonal, and lack significant trans-chromosomal contacts.

The judgment this paper supports

The most defensible reading is a split verdict rather than a flat rejection. The global claim — that dinoflagellate chromosomes are stacks of cholesteric disks with transcription confined to peripheral loops — fails on four independent comparisons: the absence of TADs in CLC-simulated maps, contact probability scaling inconsistent with CLCs, conformations less compact and less ordered than CLCs, and transcription not enriched at the surface. The weaker claim survives: these chromosomes are condensed polymers with local order at gene-array scale and no global cholesteric architecture, and that condensation does not by itself block transcription.

For readers weighing structural claims about dinoflagellate chromosomes, the practical consequence is that TAD presence is not diagnostic of cholesteric organization, since the authors could generate TAD-like maps by adding locus-specific boundaries to a cholesteric stack. The discriminating evidence is the grid-like stripe pattern and the contact probability scaling, both of which the real data fail. The authors restrict their conclusion to the three species examined, and the B. minutum scaffold incompleteness is the clearest reason to treat that species' shape results as provisional rather than as a species difference.

Sources used in this answer