The experimental design is simple. Yeast are grown for a day, the culture is
allowed to settle, and only the fraction that reaches the bottom first is carried forward.
That one procedure, repeated every day across fifteen populations in three metabolic
environments, is the only selection we impose. Everything the MuLTEE has produced, from
millimetre-scale bodies visible to the naked eye to a duplicated genome, has evolved under
it.
Populations
15
Treatments
3
Cycle
24 h
Archived every
25 d
Running since
2018
Fifteen initially isogenic populations, five in each metabolic
treatment. Transfers continue daily.
The organism
What a snowflake yeast is
Baker's yeast, Saccharomyces cerevisiae, reproduces by budding. A daughter cell
swells off the side of a mother, and when mitosis is finished an enzyme dissolves the
septum at the bud neck so the two cells come apart. Snowflake yeast are what happens when
that last step fails. Mother and daughter stay joined, the daughter buds in turn, and the
cluster grows outward from one founding cell into a branched, roughly radial body.
Because the cluster is built by cells staying attached after division rather than by
free-living cells finding each other, every cell in it carries the same genome. This is
clonal multicellularity, and it matters: with no genetic differences inside a cluster,
there is nothing for a cheater cell to gain, and cell fitness and group fitness point the
same way from the start.
The genetic basis is a single transcription factor. ACE2 regulates the genes that
degrade the bud neck septum, so a loss-of-function mutation leaves daughters attached and
produces the whole snowflake package at once. When we sequenced ten independently evolved
snowflake lineages, five carried non-synonymous ACE2 mutations, and in every case
both copies had been made identical by gene conversion. The mutation is repeatable. We do
not wait for it: all fifteen MuLTEE populations were founded from an
ace2Δ strain in the Y55 background, so the experiment starts at the
multicellular state rather than having to evolve into it.
Plate 01
Subject
Snowflake yeast cluster
Timepoint
Day 400
Stain
Cell wall / nuclei
Frame
60× magnification
One cluster, one genome. The cyan outlines are cell walls and the
orange points are nuclei, one per cell. Branch points are where a daughter budded and never
let go. The original frame carries a burned-in 20 µm scale bar along the bottom
edge, cropped here.
The daily cycle
Growth, settling, transfer
One round of the MuLTEE takes a single day, and none of it is automated. The
populations grow, they sediment under gravity, and the fraction that settles fastest is
transferred by pipette into fresh medium. The regime is therefore selection for rapid
growth followed by selection for larger group size, and the fact that these two act in
sequence within each day is central to the design.
01
Grow, for about twenty-four hours in liquid medium
The small volume carried over from yesterday expands back to a full culture. This
phase is not neutral. It rewards whatever divides fastest, and in the obligately
aerobic populations, where the measurement has been made, that is the smaller
clusters.
02
Settle, under gravity
The culture is left standing and biomass falls out of suspension. Larger and denser
clusters sink faster, so within a fixed settling window they arrive at the bottom of
the tube while smaller ones are still in the liquid above.
03
Transfer the fastest-settling fraction into fresh medium
Only the bottom fraction is moved forward. Everything still in suspension is
discarded. That transfer is the selection event, and tomorrow it happens again. Over
600 rounds this took mean cluster radius from 16 µm to 434 µm in
the anaerobic populations, roughly a 20,000-fold increase in volume, and took the
number of cells in a group from about 100 to about 450,000.
Interactive · Schematic, not to scale
The transfer engine
Forty clusters in a tube. Press play and they
fall, larger ones faster. Everything that reaches the transfer line goes into fresh medium
and regrows to a full culture; everything still in suspension is thrown away. The chart
beside the vessel is the size distribution of the whole culture, and the starting
distribution stays on it as a dashed outline. About fifteen transfers is enough to see the
whole distribution move right. The run stops after twenty so the drawing stays in
scale.
This section carries an interactive diagram of the selection step,
which needs JavaScript. The mechanism it draws is the one described above: clusters of
different sizes fall through a tube, larger ones sink faster and reach the bottom first,
only the bottom fraction is pipetted into fresh medium, and because size is inherited the
size distribution of the population shifts toward larger clusters transfer by
transfer.
Transfer 1 of 20.
Your system is set to reduce motion, so nothing
here animates. The frame shows the settled state of the transfer, with the transfer line
drawn and the retained clusters marked. Use Step one transfer to advance
the cycle one selection event at a time, and Reset to start the run
again.
What is schematic. Settling speed here rises with
cluster size by one simple monotonic rule, plus a small random difference between clusters
that stands in for everything else. Real snowflake clusters are branched and porous, so
they do not follow the simple relation that holds for a solid sphere: fluid moves through
and around a branched body, and evolved changes in shape and packing alter how fast it
sinks independently of how big it is. Sizes appear only as multiples of the starting mean,
never in micrometres.
The selection step is simplified too. In the widget the settling
window ends when the pipette volume is full, so about twelve clusters out of forty carry
forward every transfer. The real regime uses a fixed settling window, which is not quite
the same thing.
For scale, two numbers that are measured: mean cluster radius in the
anaerobic populations went from 16 µm
to 434 µm over
600 transfers, roughly a 20,000-fold
increase in volume (Bozdag et al. 2023, Nature
617:747–754). Those two endpoints come from the published record. Nothing
inside the frame above does.
Both phases are selective and they act in opposite directions. Growth favours small,
fast-dividing clusters, and settling favours large ones. A trait spreads only if it pays
for itself across the full 24-hour cycle, which is why the same selection regime produced
macroscopic bodies in one treatment and, in another, a stable split into small growth
specialists and large survival specialists that have coexisted for thousands of
generations. When we compete evolved and ancestral isolates under this regime and score
the two phases separately, snowflake yeast make most of their fitness gains during
settling, which means they are increasing fitness at the level of the group rather than
the cell.
Full culture parameters
Detailed culture parameters (vessel type, growth temperature, shaking speed, settling
window length, and transferred volume) will be posted here as part of the open protocol
documentation.
Why settling selection
How and why we select for size
Larger clusters sink faster through liquid. After 24 hours of growth, we let the
culture settle and transfer only the bottom fraction into fresh medium. Everything still
in suspension is discarded. Larger, denser clusters reach the bottom first, so the
regime is directional selection for increased group size.
We chose settling selection because it is experimentally tractable. Its strength can be
tuned simply by shortening the settling window, it is far more reproducible than
predation, and there is no predator to co-evolve or go extinct halfway through a
decade. Sedimentation rate is also an ecological trait in its own right for aquatic
organisms, not purely a laboratory convenience.
Many of the interesting phenotypes that evolved were not things we directly selected
for. Cells elongated, lowering how densely a cluster packs and delaying fracture.
Branches entangled. Division timing synchronised. All of those arrived because they
made clusters bigger or harder to break.
Plate 02
Subject
Single cluster, 3-D reconstruction
Colour
Depth coded, near to far
Frame
Scale bar burned in, lower centre
Plate 03
Subject
Many clusters, 3-D reconstruction
Colour
Depth coded, near to far
Reads as
Population, not individual
Selection acts on a whole distribution of cluster sizes, like the
one in Plate 03, and retains its upper tail every day. Cluster size is reported as a
biomass-weighted radius for this reason: a population containing many small propagules and a
few large adults would otherwise be summarised by its propagules. Plate 02 carries a
burned-in 20 µm scale bar.
Three metabolic environments
Three metabolic treatments
A big body has an interior, and the interior has to be fed. For an organism that
respires, that means oxygen has to diffuse in, and diffusion sets a hard ceiling on how
large a solid mass of metabolising cells can get. The three treatments exist to test
whether that ceiling is what limits multicellular size over the long run, by giving five
replicate populations each a different relationship to oxygen and then applying the
identical settling regime to all fifteen.
These treatments were established in 2018, and the first 145 transfers were published
in 2021 as a test of how oxygen affects size evolution. The result was counterintuitive:
intermediate oxygen suppressed size evolution, while both anaerobic conditions and high
oxygen permitted it. Fifteen of those populations have been transferred continuously ever
since, and they are the fifteen described here.
PA1–5 Anaerobic
Cannot respire, so pays no oxygen cost for size
Metabolism
Obligate fermentation
Mitochondria
Petite; respiration lost
Oxygen
Cannot be used at all
Petite mutants carry defective mitochondria and cannot use oxygen at all. Removing
respiration also removes the diffusion limit on size, so if oxygen delivery is what
constrains size, these lines were the ones expected to escape it. All five became
macroscopic.
PM1–5 Mixotrophic
Can ferment or respire
Metabolism
Fermentation and respiration
Mitochondria
Intact
Carbon
Glucose-based medium
Glucose with access to oxygen, which is how ordinary yeast grows. These populations
stayed microscopic, but they are not a passive control. All five underwent the same
whole-genome duplication as the anaerobic lines, and all five have since evolved hollow
toroidal morphologies whose central opening drives rapid flow with no cilia or flagella
(unpublished).
PO1–5 Obligately aerobic
Must respire, so oxygen is a contested resource
Metabolism
Obligate respiration
Mitochondria
Intact and required
Carbon
Glycerol-based medium
Glycerol cannot be fermented, so growth is impossible without dissolved oxygen. That
turns oxygen into something clusters compete over, and three of the five populations
split into coexisting small and large lineages held together by negative
frequency-dependent selection.
Population-by-population detail is on the
Populations page.
The frozen fossil record
Cryopreservation and the revivable past
Every 25 days, roughly 125 generations, we cryopreserve all fifteen populations at
−80 °C. Frozen samples neither die nor continue to evolve, so when they are
thawed they resume growing in the state the population was in on the day it was frozen.
This frozen fossil record now holds more than 3,000 samples.
A revivable past is what separates a long-term evolution experiment from an experiment
that simply runs for a long time. In most studies the past exists only as data. Here it
exists as a living organism that can be thawed, placed in a tube alongside its own
descendant, and competed directly. That is how we measure fitness: an evolved isolate is
competed against the ancestor it came from, under the same growth-and-settling cycle the
population experienced.
The archive also lets us test the order in which things happened. If a trait is present at
day 600, we can thaw day 400 and check whether it was already segregating there, rather
than inferring it. Reversibility can be tested the same way, by relaxing selection on a
revived evolved strain and seeing whether the trait is lost. That is how we showed that
tetraploidy in these populations is maintained by ongoing selection rather than by genome
stability.
Finally, the archive is built for instruments that do not exist yet. Some of the most
important results from Lenski's E. coli LTEE came from genome sequencing and
genetic reconstruction, neither of which was available at scale when that experiment began
in 1988. Freezing samples is inexpensive. Regenerating a decade of evolution is not
possible at all.
Working with the archive
Requests for any population at any timepoint go through the
Data & Strains page. We share strains with no authorship
requirement, provided the request does not overlap with ongoing work in our lab or that
of a collaborator. Nothing would make us happier than to foster a community of
researchers working on the MuLTEE.
What gets measured
Types of data we collect
The daily transfer is the experiment itself. Everything else is measurement, carried out
on isolates revived from the freezer. Because the archive is available, the same assay can
be applied to day 0, day 400 and day 1,000 in the same week, by the same person, on the
same instrument.
Morphology
Biomass-weighted cluster radius across evolutionary time, cellular aspect ratio, and
bud scar geometry, which records where every past division happened.
Fitness
Evolved isolates competed head to head against the revived ancestor under the
standard growth-plus-settling regime, with the growth phase and the settling phase
scored separately when the question calls for it.
Mechanics
Compression assays for the stress-strain response and for multicellular toughness. As
a material, these clusters went from around 100-fold weaker than gelatin to the
strength and toughness of wood.
Genomes
Whole-genome sequencing of evolved isolates, plus synthetic reconstruction: candidate
mutations are engineered back into a clean background to test whether they actually
cause the phenotype they correlate with.
Imaging
Serial block-face scanning electron microscopy to reconstruct the inside of a
cluster, time-lapse microscopy to follow division timing cell by cell, and
super-resolution panoramic integration to measure subcellular shape across whole
populations rather than a handful of picked cells.
