Can dense optical genome mapping assemble chromosome-scale yeast genomes?

Extending a bacterial assembler to eukaryotic genomes

The paper asks whether dense optical genome mapping (OGM) can be assembled into chromosome-scale maps of eukaryotic genomes, using two yeasts as test cases. OGM images individual DNA molecules and converts each into a one-dimensional fluorescence intensity profile, called a barcode. The authors' DOGMA pipeline had previously assembled bacterial genomes from such barcodes; here they extend it to S. cerevisiae and S. pombe, which are compact but structurally rich.

The labeling chemistry matters for what the barcode contains. The study uses competitive binding of YOYO-1 and netropsin, a dense protocol that yields continuous intensity profiles reflecting local AT/GC content along each molecule. That contrasts with sparse enzymatic labeling, which marks only specific motifs and therefore carries less information in poorly labeled or structurally complex regions. The trade-off is that dense profiles also encode genome-scale compositional structure, which can make unrelated regions look similar.

Near-complete coverage before any assembly

Before assembling, the authors aligned 1102 S. pombe and 551 S. cerevisiae barcodes to theoretical barcodes derived from the reference genomes, using a full-length Pearson-correlation alignment. Reference-based coverage exceeded 98% for both genomes, with mean coverage depths of 29.5x for S. pombe and 16.4x for S. cerevisiae, calculated only for molecules that passed quality control and aligned successfully.

That coverage estimate is not the same as assembly success, and the paper treats it as a proxy. Two artifacts inflated or depleted local coverage. Molecules approaching or exceeding full chromosome length could not be placed by full-length alignment, producing apparent gaps at their true origin, particularly on the smaller S. cerevisiae chromosomes. Separately, regions with similar intensity profiles attracted barcodes from other loci, creating local enrichment. As a specificity check, 76.4% of barcodes aligned preferentially to their own species, while cross-species matches clustered at specific loci enriched for long-range intensity trends or extended repeats.

Three structural features that break naive assembly

The authors identify three properties of eukaryotic barcodes that complicate correlation-based assembly. First, chromosome-spanning molecules: individual barcodes up to 800 kbp fully spanned ten of the sixteen S. cerevisiae chromosomes. For small chromosomes this can simplify assembly, but it breaks full-length alignment strategies that require the query to be shorter than the reference.

Second, repetitive barcode components. Because DOGMA groups molecules by local barcode-to-barcode alignment, shared repeats can dominate the matching and pull molecules into non-canonical groups independent of their unique content. Third, long-range intensity trends: broad low-frequency variations extending over hundreds of kilobases. In S. pombe these appear as GC gradients declining over roughly 300 kbp toward chromosome ends and low-GC motifs spanning about 200 kbp at centromeres. Local barcode structure is defined by 5–20 kbp variations, so these trends add a shared background that raises correlation scores between otherwise unrelated regions.

The paper's intragenomic self-similarity analysis supports this mechanism: regions of at least 150 kbp with a correlation score of at least 0.75 formed discrete hotspots shared between S. pombe chromosomes, many coinciding with or near regions showing pronounced long-range trends. The authors conclude that repeat composition and long-range organization, not genome size or insufficient coverage, are the primary obstacles.

Masking repeats lets unique sequence drive grouping

The updated pipeline identifies repetitive elements and masks them before assembly rather than deleting them, so the non-repetitive parts of barcodes drive grouping while repeat content is retained for later analysis. The paper reports that assembly without masking produced bargroups mixing misaligned repetitive and non-repetitive segments, and that masking corrected groupings that would otherwise be wrong or absent.

A second filter addresses a subtler failure. To prevent grouping driven by smooth long-range trends rather than local structure, the authors score each bargroup by signal-to-noise ratio, the standard deviation of the consensus profile divided by the mean per-barcode variability. Bargroups with a ratio below 1.0 were excluded; the paper illustrates a failing example at 0.9 and a passing one at 1.4. Only bargroups meeting the mean correlation, standard-deviation, and signal-to-noise criteria were kept.

These filters are selective. Roughly 20% of input barcodes for S. pombe and 30% for S. cerevisiae contributed to bargroup formation, while the rest remained ungrouped despite passing quality control. Those ungrouped molecules were not discarded: they were kept for coverage analysis and gap resolution.

Reference placement closes joins the assembly cannot

After de novo bargrouping, the authors aligned bargroups to the reference using local rather than full-length similarity, because assembled contigs can extend past reference chromosome edges when they contain insertions or unresolved repeats. Consistent placements were used to identify unresolved joins between bargroups, and adjacent bargroups with matching placements were merged by local bargroup-to-bargroup similarity. The paper calls this step reference-assisted refinement.

This is a hybrid strategy, and its dependence on a reference is a real boundary: the final chromosome-scale maps are not purely de novo. The authors also used high-confidence ungrouped barcodes, defined as those with strong reference alignment and low alignment ambiguity (a difference above 0.03 between the best and second-best correlation scores), to restore local contiguity where assembled contigs were absent. That threshold was applied only to reference-based analysis, not to de novo bargroup formation, and the paper notes the trade-off: raising it improves specificity while reducing the fraction of retained barcodes.

Repeat arrays longer than the reference

The final assemblies spanned both yeast genomes and showed local differences from the corresponding references. In S. cerevisiae, a large repetitive block interrupted the reference assembly of chromosome XII near 479 kbp; the authors observed this region extending at least 550 kbp, built from roughly 9 kbp repeat units corresponding to at least 60 tandem copies. In S. pombe, extended repeat arrays appeared at both ends of chromosome III beyond the reference length, spanning at least 1 Mb in total, at least 500 kbp from each telomeric end, with individual units of about 10.5 kbp consistent with rDNA repeats and at least 50 tandem copies per end.

The authors connect these observations to prior pulsed-field gel electrophoresis studies reporting substantial rDNA cluster size variability in both organisms, and interpret the regions as highly expanded rDNA arrays that current references represent incompletely. That interpretation is the authors', not a direct measurement of copy number by an independent method.

Two further discrepancies appeared on S. cerevisiae chromosomes VIII and III, each involving roughly 30 kbp of additional contiguous sequence. Because structural variant detection is not integrated into the DOGMA workflow, the paper did not determine the structure or breakpoints of these regions directly. Comparison with publicly available long-read assemblies showed substantially improved agreement with the DOGMA assemblies in both cases, which the authors read as evidence that the current reference may be incompletely resolved or collapsed there rather than that the strains carry novel variants.

A structural probe as much as an assembler

The paper's central claim is that repeat content is not a fundamental limit of dense optical assembly; what matters is detecting, classifying, and appropriately weighting repetitive and long-range signal features. The same long-range trends that bias correlation scores also capture chromosome-scale compositional organization that sparse motif labeling does not sample, which is why the authors describe dense OGM as both an assembly tool and a probe of large-scale genome structure.

The practical case for the method rests on two properties the paper documents. The intensity signal comes from YOYO-1 labeling, a fluorophore already widely used for DNA visualization, so competitive binding staining requires no additional excitation sources, emission filters, or major instrument changes. And because the signal is continuous over hundreds of kilobases, it preserves long-range context that could be combined with other labeling strategies along the same molecules.

The scope is yeast. The authors state that evaluation in human genomes lies beyond this study and frame extension to mammalian genomes, where repeats constitute over half of the human genome, as anticipated rather than demonstrated. The most defensible reading is that dense OGM with repeat-aware masking and explicit handling of long-range trends is a validated route to chromosome-scale assembly in these two yeasts, and a candidate complement to sequencing where copy number and long-range repeat organization are the questions that matter.

Sources used in this answer