The fifteen evolving lines
We founded fifteen initially isogenic populations of ace2Δ snowflake yeast in 2018, all derived from the Saccharomyces cerevisiae Y55 background. Five replicates went into each of three metabolic environments. The daily selection is identical across all fifteen: about 24 hours of growth, then a settling step in which only the fastest-settling fraction is carried forward. Only the metabolism differs.
PA, anaerobic. PM, mixotrophic. PO, obligately aerobic. Five replicates per treatment. Click any treatment to view archived strains.
Treatment one · five replicates
Plate 01
The PA lines are petite mutants: mitochondrially dysfunctional, unable to grow on glycerol, and confirmed non-respiring by direct oxygen measurement with optodes. Because they never use oxygen, oxygen cannot become scarce inside a large cluster, so the diffusion cost that normally penalises large size is absent. This treatment was designed to test whether size evolution proceeds once that cost is removed. It did, in all five replicates, producing organisms roughly 20,000-fold larger by volume.
Biomass-weighted cluster radius increased about 27-fold in linear dimension, an estimated volume increase of roughly 2 × 104. Groups went from around 100 cells to around 450,000. All five anaerobic populations became macroscopic, visible without a microscope, and the clonal life cycle was retained throughout.
The ancestor fractures at about 240 Pa with toughness as low as 8.9 J m−3. Evolved macroscopic clusters exceed 0.6 MJ m−3. Individual-cell stiffness did not measurably change, so the gain is architectural rather than material. The published abstract reports about 104-fold.
Super-resolution panoramic integration imaging across the archived timepoints puts cellular aspect ratio at about 1.30 in the ancestor and 3.28 by 1,000 transfers, while cross-sectional module size grew from 30.7 to 87.1 µm. The two do not track each other perfectly: there are deviations from simple aspect-ratio scaling at t400 and t1000.
Segmented three-dimensional electron-microscopy reconstructions of evolved branches show branches held by two or more neighbours in configurations that rigid-body translation and rotation cannot undo without deformation or bond breakage. Preliminary and not yet published: clusters cut with a razor re-entangle and self-heal within about two hours, which non-entangled ancestral clusters do not do.
Day et al. 2024 · Physical Review X 14:011008. Self-healing: unpublished data from our lab.
All five anaerobic populations doubled their genome within the first 50 days, fixed tetraploidy by day 100, and held it for the next 950 days despite genomic instability. The anaerobic tetraploids then accumulated extensive aneuploidy that tracked the evolution of macroscopic size and helped sustain it. Preliminary and not yet published: total genome size after 1,000 days is about 2.2-fold larger than the ancestor's.
Tong et al. 2025 · Nature 639:691–699. Genome-size figure: unpublished data from our lab.
Reduced HSC82 expression destabilises the Hsp90 client Cdc28 at the protein level without changing CDC28 transcript, which delays mitosis and prolongs polarized growth, and that is what makes cells longer. Restoring either gene shortened cells, shrank clusters and lowered multicellular fitness. The focal line was PA5 at 600 transfers.
The petite ancestor divides asynchronously because daughter cells take about 25% longer to complete their first division than their later ones. The anaerobic populations had lost that delay by day 200 and were still dividing synchronously at day 1,000. Synchrony is favoured during growth and during settling, so one cell-level timing change is visible to selection at both levels.
Evolved PA5 isolates from days 200, 400, 800 and 1,000 keep growing exponentially in static liquid at millimetre scale, and only sub-exponentially on agar. Above a threshold cluster size their own metabolism generates buoyancy-driven flows that enter from the sides and exit from the top, at speeds comparable to those extant organisms produce with cilia. The flows stop in dead or glucose-starved clusters and reverse when gravity is reversed. Grant-stated and still unpublished: by 1,000 transfers this permits exponential growth up to about 1 cm across.
Narayanasamy et al. 2025 · Science Advances 11:eadr6399. The 1 cm figure: unpublished data from our lab.
Across the five replicate anaerobic populations the mutational targets were enriched in the same functional categories, including parallel changes in GIN4 and PHO81.
Reductive evolution of the mitochondrial genome in the anaerobic lines is a stated research priority, not a published finding. No paper in the record reports the extent of mitochondrial genome loss in PA1–5.
Treatment two · five replicates
Our imaging archive holds no micrograph that can be attributed to a mixotrophic line. Rather than illustrate this treatment with an image from another treatment, we leave the frame empty and note the gap here.
This is the regime our 2021 oxygen experiment identified as the one that selects against large size: oxygen is available, so cells use it, and using it makes the interior of a large cluster a worse place to be. These lines are not a passive control. They took the same whole-genome duplication as the anaerobic lines on the same schedule, and they have since evolved a morphology no other treatment produced.
Diploid ancestors became tetraploid by day 50, tetraploidy fixed by day 100, and it persisted for the next 950 days. Whatever drives the genome duplication, it is not specific to the lineages that went on to become macroscopic.
Where the anaerobic tetraploids accumulated extensive aneuploidy alongside increasing size, the mixotrophic tetraploids did not, and they stayed microscopic. The contrast between the two is the cleanest evidence in the experiment that the aneuploidy is tied to the size transition rather than to tetraploidy itself.
The mixotrophic line that remained microscopic did not show the reduced HSC82 expression that all five macroscopic anaerobic lineages converged on, which is the expected result if reduced Hsp90 is specifically associated with the evolution of large size.
The central opening of the torus drives rapid flow, measured at around 100 µm/s, with neither cilia nor flagella anywhere in the organism. A microscopic treatment arrived at a transport solution by a different route than the macroscopic one.
Unpublished data from our lab.
Treatment three · five replicates
Plate 02
Glycerol cannot be fermented. Growth on it requires functional mitochondria and dissolved oxygen, which makes oxygen a resource the population competes over during the 24-hour growth phase rather than a background condition. That is the ingredient the other two treatments lack, and it is what turned group size into an ecological axis: small clusters win the competition for oxygen, large clusters win the settling step, and neither can exclude the other. Group size alone, with no differentiated tissues, was enough to partition a niche.
From a single monomorphic ancestor, over 715 daily transfers. The genomic data imply the two forms diverged early and then persisted together for close to the full span, which the paper reports as roughly 4,300 generations.
Under the standard growth-plus-settling regime, mixtures started across a wide range of initial frequencies return to the same equilibrium. That is negative frequency dependence, not a sweep that has not finished. The focal Small and Large genotypes were isolated from population PO-4 after 715 daily transfers.
This identifies dissolved oxygen as the axis of the trade-off. Competition for oxygen is what keeps the small growth specialist in the population despite a daily selection step that rewards nothing but size.
Small clusters are the better oxygen competitors during the growth phase. Large clusters gain the survival advantage during settling selection. Group size alone, with no differentiated tissues of any kind, was enough to partition the niche.
Grown in monoculture without the opposite-sized competitor, both lineages evolve back toward intermediate size, and small-lineage monocultures re-evolve the two forms quickly. This is character displacement operating in both directions.
The aerobic ancestor divides synchronously, unlike the petite anaerobic ancestor, whose daughter cells take about 25% longer to complete their first division. The two treatments therefore started from developmentally different founders, which matters for any comparison of their branching topologies.
How to read a strain code
We use the following strain codes across our papers and method notes. Usage has not always been consistent between papers, so the conventions we will use going forward are set out here.
The imaging archive
Line and timepoint below are read from the original file names in our imaging archive. Where a file name carries no treatment or timepoint token, the label says so rather than guessing. Several plates carry a burned-in magnification label and scale bar; those are the microscope's own annotation, left in the frame.
Plate 03
Plate 04
Plate 05
Plate 06
Plate 07
Plate 08
Plate 09
Plate 10
All micrographs were made in the Ratcliff Lab. No plate on this site is attributed to a mixotrophic line, because the archive holds none. Please contact us before reuse.
Open
A few details are not yet settled in the published record.
Questions about any of these? Write to us, or see Data & Strains for the archive itself.