We spend most of the paper looking at how the level of selection shifts from the uni- to the multicellular level. I’ll briefly summarize our arguments. Once clusters evolve, we see a shift to among cluster selection: they either get to the bottom of the tube and survive as a whole, or they fail to do so and die as a whole. If clusters vary in traits that affect their fitness, and if this variation is heritable, then cluster level traits can evolve in response to between-cluster selection. We have two different experimental results that confirm this:

1) Divergent selection experiment. We see that in response to an increased strength of selection for rapid settling, snowflake yeast evolve faster settling by growing to a larger size. This is accomplished by delaying the onset of ‘reproductive maturity’, the minimum size of propagule production. This is a change in cluster-level traits that only makes sense in the context of a cluster of cells (single cells from larger, faster settling clusters probably don’t settle any faster than single cells from smaller, slower settling clusters).

2) The evolution of increased apoptosis. Apoptotic cells die, reducing fitness at the single-cell level, but increase the fitness of the cluster. Here the two levels of selection are directly opposing each other, and the cluster-level benefit beats out the single-cell level cost. There is no better test over which level selection is acting.

Of course, you can model this all thorough Hamiltonian kin selection (single cells that commit suicide help their clonemates in the same cluster increasing their inclusive fitness). This is not a problem for major transitions, though, as you can also do the same thing for frogs or redwood trees.

Thanks for your follow-up, I appreciate the response and positive tone. A few brief responses to your additional questions:

1) We already know from classical yeast genetics that there are multiple ways to get a cluster. Since this work was carried out in different strains than the one we used in our experiment, I’d be confident in guessing that snowflake yeast could be evolved in any Saccharomyces cerevisiae strain. Simply getting clusters through post-division adhesion strikes me as quite easy and general (and has been seen in other microbes), which is why Carl quoted me as saying that it’s not “some freaky yeast thing”.

2) The evolution of clusters and subsequent cluster-level adaptation occurs very quickly, which is why we call it rapid. Even if you think this is affected by ancient multicellular ancestors (which I do not), it is still rapid. The link you provide is looking at convergent evolution among different animals, but that is not the definition of convergent evolution, it is a special case. The definition of convergent evolution (according to Doug Futuyma) is “Evolution of similar features independently in different evolutionary lineages, usually from different antecedent features or by different developmental pathways”.

3) All we’re saying by “division of labor” is that dead cells play a beneficial, functional role that is different from live cells (like the outer layer of skin, say). We see that the rate of apoptosis increases through evolutionary time, showing that selection acts on this benefit.

4) Yes, cluster size is limited by branches breaking off. You can see this in some of the videos (like S2).

Finally, I’d like to note that language is an imprecise form of communication. You may not like the term ‘apoptosis’, but that’s what yeast people call the process by which cells produce tons of ROS then quickly die. If they called it ROS-Mediated Catastrophic Yeast Implosion or something like that I’d have used that term.

Cheers,

Will

The other point is that this is not about yeast: it’s about the evolution of multicellularity. The fact (or non-fact) of ancestral yeast multicellularity is irrelevant. This is an illustration of how multicellularity can evolve, and the conditions under which it evolves (I mean the entire body of research, not just this particular experiment). Every “major transition” in evolution began within a single species.

1-If the multicellular ancestor of yeast became unicellular through a gain of function mutation, then loss of this function could easily result in multicellularity again. The problem is that your paper states: “The potential for the evolution of multicellularity may be less constrained than is frequently postulated.” No matter how you put it, starting with an organism that has multicellular ancestors is different from evolving multicellularity entirely de novo.

2- See above why rapid is ill fitting. See this link (http://www.pbs.org/wgbh/evolution/library/01/4/l_014_01.html) regarding convergent evolution. (“This is a dramatic example of convergent evolution, when organisms that aren’t closely related evolve similar traits as they both adapt to similar environments.”) You started with the same strain of one species. Did you test different strains? Is Y55 potentially predisposed to this phenotype?

3- I did not intend to define multicellularity, but rather what one might understand as “division of labor”. ROS maybe a step in apoptosis, although apoptosis in yeast certainly isn’t a clear cut case, but regardless showing the presence of ROS is not equivalent to demonstrating apoptosis.
Thanks for sharing the comparison to the similarly cultured single cell yeast. I know you are limited in what you can include in a paper, but showing this is actually a heritable trait seems very important.
To me a semi-penetrant lethal mutation that stochastically leads to cell death is the simplest explanation. Is that division of labor? Maybe, and with subsequent modifications it certainly could become it.

4- Above you mentioned that “Even cells that are just one division old (about 1.5 hours) can begin the process of apoptosis, so it’s not restricted to ‘old aged’ cells.” and parent size does vary a lot. More importantly, you are selecting for something that settles in the culture. Your scheme likely selected against clusters smaller than a minimal size as they would have been to soluble to be captured by your transfer.
Along those lines: Is the size of clusters limited by branches breaking off? I.e. does the size of a propagule plateau as the rate of cell death and branch dispersal reaches equilibrium with cluster growth. Also, does the transfer of the settled cells by pipette limit their size? Presumably clusters could get too large to be transferred in this manner.

5- I included this as Carl stated above “”Flocculating” refers to the clumps that unrelated yeast cells form when they starve.”

Will, I greatly appreciate your explanations and am looking forward to Michael’s and your future publications. As mentioned above, it would be very interesting to see if other S. cerevisiae isolates and close relatives of this species can make the transition as easily.
Best of luck with your experiments.

@ST. Saccharomyces cerevisiae lost multicellularity hundreds of millions of years ago. I explained in the prior post why we are confident that the traits that we see evolving are not a reversion to ancestral multicellularity.

I’ll address your points above sequentially. My responses in bold.

1- ST- In his response above, Will Ratcliff points out that many loss of function mutations can give rise to this phenotype. Exactly. But unfortunately the paper suggests that a novel trait was evolved and not that multiple loss of function mutations were isolated.

Only the first step, the evolution of clusters through post-division adhesion appears to be the result of a loss of function (the loss of the ability to separate off daughter cells after mitosis). After this, we see several adaptations that improve the fitness of clusters. We see an increase in cluster size, due to an increase in both the number of cells in the cluster as well as the size of those cells. We see the evolution of higher rates of apoptosis (I’ll explain below why this is not synonymous with ‘aged’ cells). These traits are novel, and are interesting to us because they improve the fitness of whole clusters, not their component cells. This means that adaptation is occurring in cluster-level traits, and that the yeast are evolving in a multicellular manner.

2-ST-“Rapid and convergent evolution…”. The rapid is already debunked and convergent would require starting with different organisms.

The rapid part is debunked? What? We started with a single genotype that did not form clusters. Mutations that result in clustering occur, and then increase in frequency due to selection. This is textbook evolution, and it occurs rapidly. As for convergence, 10 independent populations that started with the same ancestor evolved the snowflake phenotype, not, say huge cell size or flocculation. This is textbook convergent evolution.

3-ST-Secondly, the argument that these clusters are indeed a multicellular organisms depends on the authors’ point that they see “division of labor”. What do they mean by division of labor? That cells differentiate, have different functions and in the end are dependent upon each other? No. Rather some cells die while others don’t. Cells dying isn’t much of a specialization, hence the authors emphasis on cells dying as part of *Programmed* cell death (apoptosis). *Programmed* cell death means that cells do differentiate and make a cell fate decision to die.

I disagree with your assessment about what defines multicellularity, but that’s a matter for another time. A key step in programmed cell death in yeast is the production of reactive oxygen species (Frank Madeo has many papers discussing the finer points of yeast apoptosis, we’re using his procedure to measure ROS). Here’s what we see- the unicellular ancestor and snowflake yeast have very low levels of ROS induction, and very low levels of cell death during the first 24 hours of growth. During our experiment, they evolve a nearly 10-fold increase in the number of cells that are producing high levels of ROS during the exponential phase of growth. Because it was measured at the same time in the culture cycle, we know it’s not due to, say, differences in the number of old cells in the cluster. Further, we showed that this trait segregates independently of cluster size during classical mating experiments, so it’s heritable (almost certainly genetic). Even cells that are just one division old (about 1.5 hours) can begin the process of apoptosis, so it’s not restricted to ‘old aged’ cells. We show that by dying, these cells serve a function that benefits the cluster, allowing a cluster to produce a greater number of small, fast growing propagules. This function is different from the other genetically-identical cells in the same cluster that reproduce: it is thus division of labor. From an evolutionary perspective, the fact that the cells that perform this function give up their lives is significant, demonstrating that selection is acting at the level of the whole cluster, not it’s component cells.

4-ST-“No propagules were produced by clusters less than a minimal size, demonstrating that the snowflake phenotype exhibits juvenile/adult life stage differentiation (Fig. 3B).” – Yeast cells die as they age. The snowflake cells grow outwards, meaning old cells will be towards the center and result in the release of “branches” when they do die. Saying this demonstrates a juvenile/adult life stage differentiation is a huge leap.

We’re not claiming that this trait evolved specifically; rather we think that the juvenile/adult distinction is an emergent property of the way that clusters form. The fact of the matter is that propagules don’t make their own propagules until they grow up to their parent’s size. This is a exerts a real, tangible constraint on their evolution. For example, we see adaptation occurring in this emergent multicellular life history. In our divergent selection experiment, we found that snowflake yeast adapt to stronger selection for settling speed (less time to get to the bottom) by delaying reproduction until the cluster is larger, which increases settling speed.

5- ST-Flocculation also occurs between genetically identical cells that express the same cell surface proteins.

It can, yes, but nonrelatives that are express the same surface proteins can also stick together. See the excellent paper by Scott Smukalla et al., “FLO1 Is a Variable Green Beard Gene that Drives Biofilm-like Cooperation in Budding Yeast.” With post-division adhesion, this possibility for within-group chimerism is removed.

If you are not satisfied with these explanations, you can simply wait for more data. We’re working out the molecular genetics of the transition described in the paper, including apoptosis (preliminary results suggest that it is indeed an active process, not simply the result of ‘old age’). We’re redoing our experiment with other microbes that never had a multicellular ancestor, such as Chlamydomonas. Finally, and perhaps most importantly, we’re continuing to evolve our yeast in a long-term experiment- my expectation is that this is just the beginning of their multicellular evolution.

Other than that the paper is very much oversold.
Given that S. cerevisiae arose from a multicellular ancestor, it appears the paper describes the reversion of a unicellular organism to multicellularity, rather than the evolution of multicellularity. The multicellular ancestor makes the last sentence of the paper (“The potential for the evolution of multicellularity may be less constrained than is frequently postulated.”) rather ludicrous. In his response above, Will Ratcliff points out that many loss of function mutations can give rise to this phenotype. Exactly. But unfortunately the paper suggests that a novel trait was evolved and not that multiple loss of function mutations were isolated.
In that regard the title of Figure 1 states “Rapid and convergent evolution…”. The rapid is already debunked and convergent would require starting with different organisms (http://www.pbs.org/wgbh/evolution/library/01/4/l_014_01.html). Mere reproducibility does not equal convergent evolution.

Secondly, the argument that these clusters are indeed a multicellular organisms depends on the authors’ point that they see “division of labor”. What do they mean by division of labor? That cells differentiate, have different functions and in the end are dependent upon each other? No. Rather some cells die while others don’t. Cells dying isn’t much of a specialization, hence the authors emphasis on cells dying as part of *Programmed* cell death (apoptosis). *Programmed* cell death means that cells do differentiate and make a cell fate decision to die. But unfortunately the authors provide no evidence that these dying cells undergo *Programmed* cell death (staining for reactive oxygen species, a marker of cell “sickness”, and cell permeability, a late marker of cell death, is no evidence of a cell fate decision). Within any population of yeast, a fraction of cells will undergo precisely this form of cell death and eventually every single cell will die this way.

Other points: “No propagules were produced by clusters less than a minimal size, demonstrating that the snowflake phenotype exhibits juvenile/adult life stage differentiation (Fig. 3B).” – Yeast cells die as they age. The snowflake cells grow outwards, meaning old cells will be towards the center and result in the release of “branches” when they do die. Saying this demonstrates a juvenile/adult life stage differentiation is a huge leap.
Lots of talk about “cluster-level traits”, but no discussion of how they may simply arise from variations in single cell fitness. All these “cluster-level traits” appear to boil down to the rate at which single cells die.
Flocculation also occurs between genetically identical cells that express the same cell surface proteins.
Through their use of anthropomorphisms (“division of labor”, “suicide”, “junvenile/adult stage”) the authors suggest that their yeast isolates evolved novel and highly complex traits. But at no point do they acknowledge that there are simpler explanations.

Each participating class would attempt (perhaps with baker’s, vintner’s , or laboratory yeast) to replicate your results. Those classes which successfully replicate an evolution to a strain exhibiting multicellular characteristics could report their results (with photos) on a website set up for the purpose.

Going further, dried samples of the resulting strains could be sent to some central point (such as your lab, or perhaps to a participating university) where characteristics might be examined of the snowflakes produced by each culture, and perhaps further experiments performed, including multicultural environments where the relative fitness of various snowflake cultures in different simulated environments could be examined.

Beyond engaging the students in the science, this might produce results with direct bearing on your current project.