Disadvantages and benefits of evolved unicellularity versus multicellularity in budding yeast

Abstract Multicellular organisms appeared on Earth through several independent major evolutionary transitions. Are such transitions reversible? Addressing this fundamental question entails understanding the benefits and costs of multicellularity versus unicellularity. For example, some wild yeast strains form multicellular clumps, which might be beneficial in stressful conditions, but this has been untested. Here, we show that unicellular yeast evolve from clump‐forming ancestors by propagating samples from suspension after larger clumps have settled. Unicellular yeast strains differed from their clumping ancestors mainly by mutations in the AMN1 (Antagonist of Mitotic exit Network) gene. Ancestral yeast clumps were more resistant to freeze/thaw, hydrogen peroxide, and ethanol stressors than their unicellular counterparts, but they grew slower without stress. These findings suggest disadvantages and benefits to multicellularity and unicellularity that may have impacted the emergence of multicellular life forms.

Understanding the costs and benefits of social traits in yeast could elucidate general forces that maintain or convert unicellularity to multicellularity (Maynard Smith & Szathmáry, 1998) and back. The existence of unicellular and clumpy yeast in nature (Wloch-Salamon, Plech, & Majewska, 2013) suggests condition-dependent benefits and disadvantages, and bidirectional transitions between unicellularity and multicellularity. Could clumps provide protection from environmental stress as flocs do (Smukalla et al., 2008) while being disadvantageous in normal conditions? More broadly, could reverse transitions to unicellularity occur without cheaters, and what are the evolutionary forces that aid or prevent such reverse transitions?
To address these questions, here we compared how various environmental stressors affect the growth of genetically similar clumpforming and unicellular "EvoTop" yeast cells that we obtained by reversing the strategy of "snowflake" yeast evolution (Ratcliff et al., 2012). Sequencing and comparing the genomes of the clumping ancestor and single-celled "EvoTop" lines revealed unique missense and nonsense mutations in the AMN1 gene, which is associated with multicellularity (Li et al., 2013;Yvert et al., 2003). Clump-forming ancestral cell lines grew faster relative to untreated controls than EvoTop lines after exposure to rapid freeze/thaw, 1% ethanol, and 150 µM hydrogen peroxide stressors, indicating that clumping provides resistance to chemical and physical stresses. On the other hand, clumping hampered growth in the absence of stress, suggesting a trade-off between the benefits and downsides of multicellularity versus unicellularity. Overall, this work sheds light on the genetic bases, as well as on the disadvantages and benefits of unicellularity versus clumping multicellularity in yeast, with implications for bidirectional transitions between other unicellular and multicellular life forms.

| Yeast strains
We used three strains of the budding yeast S. cerevisiae in this study.

| Fluorescent labeling of TBR1
The integration of GFP and mCherry reporters into TBR1 chromosomes was described previously (Chen et al., 2014). Briefly, Escherichia coli strains with either the pDN-G1Gh (GFP) or the pDN-G1Ch (mCherry) plasmids (both of which harbor the ampicillin resistance selection marker) were incubated in LB media with ampicillin (1:1,000) at 37°C for 6-8 hr. The plasmids were then extracted by midi prep (QIAGEN), linearized, and purified. They were then transformed into the native GAL1 locus of the TBR1 strain, using the histidine auxotrophic marker. Transformation was performed using a modified lithium acetate procedure as described before (Chen et al., 2014). Synthetic drop-out (SD-his-tryp) plates were then used for selection (all reagents from Sigma, Inc.). Once established, TBR1-GFP and TBR1-mCherry strains were incubated in SD-his-tryp + 2% galactose at 30°C, shaking at 300 rpm (LabNet 311DS shaking incubator). The TBR1-GFP and TBR1-mCherry strains were imaged using a Nikon TE2000 fluorescence microscope. Cells were counted with a Becton-Dickinson FACScan flow cytometer during the experiment and manually from microscope images afterward.

| Selection for unicellular yeast
To initiate three replicate lines of the haploid yeast strain TBR1, we inoculated three tubes of 2 ml yeast peptone dextrose (YPD) medium (10 g yeast extract, 20 g bacto peptone, 2% glucose per L) with single TBR1 colonies. We allowed each culture to grow overnight, froze aliquots of these "ancestral" cultures (TBR1 A, B, and C), and then prepared two 100× dilutions of each culture in 2 ml YPD to obtain matched pairs of TBR1 A, B, and C for starting the EvoTop and EvoControl treatment lines. Each line reached stationary phase by growing for 24 hr in a shaking LabNet 311DS incubator at 300 rpm at 30°C. After removal from the shaking incubator, large clumps should settle faster than single cells and small clumps. For the EvoTop treatment lines, we vortexed each tube before allowing them to remain in a 30°C MyTemp Mini Digital static incubator for 45 min before taking a 20 µl sample from the top of the liquid culture to inoculate a new tube with 2 ml of YPD for growth overnight under the previously described conditions. We performed the selection procedure each day over 4 weeks, for a total of 28 rounds of selection and resuspension. Over the course of the selection experiment, we maintained parallel EvoControl lines by vortexing each strain before selecting 20 µl samples for each transfer into new tubes with 2 ml YPD. We froze ancestral samples at the beginning of the experiment and samples of each EvoTop and EvoControl line (700 μl cell solution with 300 μl 80% glycerol) in a −80°C freezer every 3-4 days, including the final cultures. We followed the same protocol with another haploid clump-forming strain, KV38 (Smukalla et al., 2008), and with the haploid unicellular laboratory strain BY4742, an S288c derivative (Brachmann et al., 1998), as a control.

| Estimating cell and clump sizes
To determine cell size and clump size of each EvoTop, EvoControl, and ancestral line, we used a Nexcelom Cellometer Auto M10 automated cell counter to analyze 10× dilutions in YPD of overnight cultures (grown in YPD solution in a 300 rpm shaking LabNet 311DS incubator at 30°C) started from frozen samples of each line, from the ancestral state through to the final frozen sample. Diameters from 10× dilutions of "live" samples were also measured a number of times toward the beginning and end of the selection experiment.
By default, the Nexcelom Cellometer software declusters clumps and measures and counts individual cells within them. Obviously, clumps are larger than cells, so the cell counting parameters must have smaller maximum diameters than the clump counting parameters. It is also very important to note that clumps were measured with the "Do not Decluster Clumps" parameter selected, which counts and measures each clump as a whole unit. Specifically, for cell size, we set the Cellometer Auto cell type parameters as follows: cell diameter minimum of 2.0 μm and maximum of 9.0 μm, roundness of 0.10, and contrast enhancement of 0.40, with a decluster edge factor of 0.5 and Th factor of 1.0. We measured clumps with the following parameters: cell diameter minimum of 2.0 μm and maximum of 40.0 μm, roundness of 0.10, and contrast enhancement of 0.40, with "Do not Decluster Clumps" selected.
We increased the maximum cell diameter of clumps to 100 μm for samples where the program indicated that some clumps were larger than 40 μm in diameter.
We combined clump and cell diameter "live" data gathered during the experiment with data from the lines taken out of the freezer and then averaged together all clump or cell data points by line (within strain and treatment) for days that had both, including the initial "ancestor" and final "EvoTop" and "EvoControl" days. We used the userfriendlyscience version 0.7.2 package in R version 3.6.0 (Peters, 2018;R Core Team, 2018) to perform one-way ANOVAs analyzing the effect of treatment on clump diameter or the effect of treatment on cell diameter for TBR1, BY4742, and KV38, followed by Games-Howell post hoc tests.

| Estimating cell and clump sizes of KV38
Flocculation is another form of multicellularity that could potentially confuse the results. It is well known that yeast flocs, but not clumps, can be separated by ethylenediaminetetraacetic acid (EDTA). When we started the selection experiment, the KV38 ancestor did not appear to flocculate, so EDTA was not added to the sample we obtained live ancestral clump and cell size data from. However, KV38's tendency to flocculate fluctuated throughout the selection experiment, so we added 5 mM EDTA to break up flocs before measuring clump and cell size. We also imaged the cultures started from frozen samples of each line in YPD with 5 mM EDTA.

| TBR1 sequencing analysis
The DNA from the ancestor and each of the three TBR1 EvoTop stress tested isolates was extracted from cultures started from individual colonies that were anticipated to be clonal. Clonal mutations are expected to be present at a minimum of 80%-90% in clonal population samples analyzed as if they were polymorphic (Saxer et al., 2014). Since the TBR1 A EvoTop isolate appeared to be founded by two clones, with the lower frequency clone present at about 18%, we reasoned that this clone could have other clonal mutations present at as low as 14% (=0.8 × 18%; Table 1).
We used high-throughput whole-genome sequencing to identify the mutations underlying the change from clumping to "unicellular" phenotypes in our TBR1 EvoTop lines and isolates. We followed QIAGEN's "Purification of total DNA from yeast using the DNeasy Blood & Tissue Kit" protocol to obtain high-quality genomic DNA from the TBR1 ancestor, EvoControl, and EvoTop lines and isolates.
Columbia Genome Center used these samples to run whole-genome sequencing on an Illumina HiSeq 2000 v3 instrument, and they aligned the obtained reads in fastq files to the Σ1278b reference genome (Dowell et al., 2010) using BWA-mem. Then, we added alignment qualities with lofreq_star-2.1.2 (Wilm et al., 2012). Because the EvoControl and EvoTop populations were heterogeneous, we looked for low frequency mutations in the genomic DNA of all 10 samples (TBR1 ancestor, TBR1 A, B, and C EvoControl, EvoTop, and EvoTop isolates) with lofreq_star-2.1.
In addition, we used the breseq-0.26.1 pipeline (Deatherage & Barrick, 2014) with bowtie2-2.2.6 and the -j2, -p, -c options to align fastq files to the Σ1278b reference genome (Dowell et al., 2010) in order to independently identify variants in each of the 10 lines. To identify mutations in the EvoTop isolates that were not called in the ancestor, we used the vcf-isec command of VCFtools  (Danecek et al., 2011) with option -c for LoFreq* files and gdtools SUBTRACT for breseq files. We used vcf-isec -f -n+2 to identify which of these mutations were called by both LoFreq* and breseq and then ran LoFreq*'s uniq command to identify which of these mutations were really unique from the ancestor. To help determine the potential impact of these mutations, we used gdtools ANNOTATE and the Sigma1278b_ACVY01000000.gff file deposited at yeastgenome.org by Dowell and colleagues (2010).
We analyzed all 10 lines as polymorphic populations so that we could directly compare the results. Table 1 (adapted from breseq's "Predicted mutations" tables) lists the mutations in the TBR1 EvoTop isolates that were unique from the ancestor and were called by both LoFreq* and breseq with a minimum frequency of 14% (see above).

| Assays of stress resistance
To compare the differences in stress response between the clump-forming TBR1 ancestor and its "single-celled" descendants, we isolated colonies from each ancestral and EvoTop TBR1 line (A, B, and C) that were good representatives of these phenotypes (see "TBR1 isolates" section below and Figure 5). We diluted exponentially growing cells started from individual colonies and placed 1.5 ml of each dilution into two separate 2 ml microcentrifuge tubes to start a control and experimental stress treatment as described below. We used R version 3.6.0 and R package emmeans version 1.3.4 to run three-way full-factorial ANOVAs and Tukey post hoc tests to analyze variant, line, and stress effects on the relative growth of TBR1 and BY4742 cells and contrast estimated marginal means.

| TBR1 isolates
For each TBR1 ancestor and EvoTop variant and line (A, B, and C), we took six colonies and used each to inoculate 1 ml SD-tryp + 2% glucose (6.7 g nitrogen base, 1.92 g drop-out supplements without tryp, 2% glucose per L) tubes for overnight growth in a shaking 30°C 300 rpm incubator. We froze isolates (700 μl SD with 300 μl 80% glycerol) and prepared dilutions in SD-tryp + 2% glucose for exponential growth overnight. The following day, we used the Nexcelom Cellometer Vision CBA "Clumpy Yeast" parameters to measure clump diameters, distributions, and concentrations of isolates. Isolates with extremely low concentrations (<1 × 10 6 clumps/ ml) were not considered for further use. We took average clump diameter and clump size distribution into account to determine which ancestor and EvoTop isolates to use in our stress tests. We chose EvoTop isolates with small clumps that lacked larger clumps in their distribution data. For ancestor isolates, we chose ones with large clumps and distributions that lacked very small clumps. Isolates shown in purple ( Figure 5) were used for stress tests. TBR1 C ancestor isolate 2 (green) was used in the first freeze/thaw experiment, but TBR1 C ancestor isolate 5 (purple) was used in all other stress test replicates.

| Testing for stress resistance: Freeze/thaw
For the freeze/thaw treatment, we rapidly froze cells in a dry ice and ethanol bath for 5 min and immediately thawed them in room temperature water. During this time, the remaining 1.5 ml cell samples stayed on the bench top as controls. We prepared 3× dilutions of each culture by adding 1 ml of cells to 2 ml SD-tryp + 2% glucose in 5 ml polystyrene round-bottom yeast culture tubes and transferred 1 ml of each sample into new yeast culture tubes. We placed the 2 ml samples back into a 300 rpm Labnet 311DS shaking incubator at 30°C, and prepared Nexcelom automated cell counter slides from each ancestral and EvoTop cell line from the 1 ml aliquots. In order to break up clumps and obtain accurate cell counts, we used a Qsonica Q55 sonicator to administer 20 pulses of 30% amplitude sonication to each of the 1 ml aliquots, which were placed on ice between each set of 10 pulses. We subsequently acquired clump diameter data before and after sonication, along with cell concentration data with a Nexcelom Cellometer Vision automated cell counter using the same parameters as described above for the Cellometer Auto (with the additional background adjustment parameter for cell measurements set at 1.0). After 1.5 hr of growth in the incubator, we prepared new 1 ml aliquots of each culture and followed the previously described protocol to obtain clump diameter and cell concentration data. We then calculated the proportion change in cell concentration for the EvoTop single cells and ancestral clumps under freeze/thaw and control conditions as follows: (final concentration − initial concentration)/initial concentration.
To obtain the relative growth of stressed cells compared to their controls, we divided the proportion change of each freeze/thaw sample by the proportion change in its control counterpart: (proportion change freeze/thaw)/(proportion change control).
We followed the same protocol to examine the effects of freeze/thaw stress on our BY4742 A, B, and C lines, but we determined that sonication was not needed to obtain countable cells.

| Testing for stress resistance: 150 µM hydrogen peroxide and 1% exogenous ethanol
In order to determine whether clumps of yeast are more resistant than single cells to the chemical stressors hydrogen peroxide (H 2 O 2 ) and ethanol, we followed a similar protocol to the one described for freeze/thaw. We diluted exponentially growing cells started from individual colonies of TBR1 A, B, and C EvoTop and ancestor isolates to 3.5 × 10 6 cells/ml in SD-tryp + 2% glucose. We made 2× dilutions of the cells for control and 150 µM H 2 O 2 or 1% ethanol treatments by adding 1 ml of cells to prepared 5 ml yeast culture tubes containing 1 ml SD-tryp + 2% glucose for controls and either 300 µM H 2 O 2 in 1 ml SD-tryp + 2% glucose or 2% ethanol in 1 ml SD-tryp + 2% glucose for stress treatments. To obtain aliquots for initial counts, we transferred 1 ml of each culture into new yeast culture tubes.
Then, we placed the remaining samples back into the 300 rpm shaking LabNet 311DS incubator at 30°C. We allowed the single-cell and clump lines to grow for 1.5 hr in a 300 rpm LabNet 311DS shaking incubator at 30°C. To calculate the proportion change in concentration of EvoTop single cells and ancestral clumps, we used sonicated cell counts obtained from a Nexcelom Cellometer Vision automated cell counter in the manner previously described for the freeze/thaw experiment. We also calculated and analyzed the relative growth of stressed cells compared to their unstressed controls for hydrogen peroxide or ethanol samples as previously described for the freeze/ thaw experiment. We followed the same protocols with our BY4742 A, B, and C EvoTop and ancestral lines, but we did not sonicate the cells for counting.

| TBR1 and BY4742 unstressed growth rates
We ran a three-way ANOVA (lm) in R version 3.6.0 to analyze the effects of variant, phenotype, and line, and variant by phenotype, and phenotype by line interactions on the growth (proportion change over an hour and a half) of unstressed TBR1 and BY4742 clumps and single cells. As we will discuss in the Results section, TBR1 ancestral clusters were significantly larger than TBR1 EvoTop clusters, while BY4742 ancestral multiplets were smaller than BY4742 EvoTop multiplets. Therefore, for this analysis, we considered TBR1 ancestors and BY4742 EvoTops to be "clump-forming", and we considered TBR1 EvoTops and BY4742 ancestors to be "single-celled".

| The TBR1 yeast strain shows the clumping phenotype
To explore the bidirectional transitions and fitness effects of unicellularity versus multicellularity in yeast, we focused on the haploid S. cerevisiae strain TBR1 (see the Materials and Methods section), a segregant obtained by multiple crosses of baking strains that carries thousands of polymorphisms relative to the standard laboratory strain S288c (Dowell et al., 2010) and develops wrinkly patterns on soft agar plates (Chen et al., 2014;Reynolds & Fink, 2001). We asked whether TBR1 cells would also be capable of clump formation by incomplete separation. Thus, we compared phenotypes related to clump formation in the TBR1 strain and the standard laboratory strain BY4742, a haploid-derivative of S288c.
Strains capable of clump formation tend to settle to the bottom of the culture tube over time. Therefore, we first visually tested the settling of these two yeast strains 45 min after removal from the shaking incubator. We noticed that the entire liquid culture medium was still uniformly turbid for the laboratory strain BY4742. In contrast, the idea, we used calcofluor white to stain bud necks within the multicellular structures. Indeed, we observed bud necks only inside multicellular structures, providing further support for clump formation by incomplete separation (Figure 1d).
Diploid multicellular "snowflake" yeast were found to divide as multicellular units, splitting into two smaller "snowflakes" (Ratcliff et al., 2012). To see whether this applied to TBR1 clumps, we tested how clumps multiply without shaking, to avoid the shedding of single cells due to physical shear. The TBR1 strain existed nearly exclusively in the form of clumps and grew by clump (rather than single-cell) division in these conditions (Figure 1e). This further supported the notion that the TBR1 haploid strain predominantly exists in the form of multicellular clumps that originate from incomplete daughter cell separation and divide as units.

| Multicellular to unicellular transition by laboratory evolution
Next, we asked if TBR1  To induce a reverse transition and obtain two yeast strains that differ in their clumping phenotype but are otherwise genetically similar, we derived a nonclumping, unicellular variant from the "ancestral" TBR1 strain by laboratory evolution. In 2012, Ratcliff and colleagues selected for multicellular diploid "snowflake" yeast by continuously propagating the yeast that settled most rapidly to the bottom of their cultures (Ratcliff et al., 2012). Here, we reversed that selection process by continuously selecting for single cells or smaller clumps from the tops of our cultures, which remained suspended after the larger clumps settled over 45 min (Figure 2a). We refer to these evolving cell lines as the "EvoTop" variant. We also propagated in parallel an "EvoControl" variant, with cell lines that had the same    Finally, as a control, we also conducted an identical selection experiment on the standard laboratory strain BY4742. As opposed to TBR1 and KV38, the ancestral cultures of this unicellular strain contained mainly single cells, intermixed with occasional multiplet structures of two or three cells (doublets or triplets). As expected, the mean "multiplet diameter" of the EvoTop lines did not decrease over time (Figure 3b). Surprisingly, however, the presence of small  vortexing (Chaudhari, Stenson, Overton, & Thomas, 2012).
In summary, by utilizing the settling rate differences among clumps of varying size within a population, we were able to select for unicellular yeast from clump-forming haploid ancestors. Surprisingly, even in the absence of settling-based selection, clump sizes decreased slightly. In addition, we observed a mild tendency toward multiplet formation for the initially unicellular BY4742 laboratory strain.

| Genetic bases of unicellularity
To determine the genetic changes underlying the multicellular-tounicellular transition, we performed whole-genome sequencing on the TBR1 ancestor and the three TBR1 EvoTop stress tested isolates.
We found that each TBR1 EvoTop isolate contained unique mutations in the "Antagonist of Mitotic exit Network" (AMN1) gene. The AMN1 gene (Li et al., 2013;Yvert et al., 2003) is part of the ACE2 regulon (Di Talia et al., 2009) that mediates the forward transition to multicellularity in yeast (Ratcliff, Fankhauser, Rogers, Greig, & Travisano, 2015). All isolates were monoclonal, except for the TBR1 A EvoTop isolate, for which sequencing indicated 2 clones. The FLO11 gene of TBR1 encodes a surface flocculin that is required for biofilm cell-surface adhesion and involved in cell-cell adhesion (Lo & Dranginis, 1998;Reynolds & Fink, 2001 However, the distribution of TBR1 EvoTop isolate clump diameters shifted away from the TBR1 ancestral distribution and closer to that of BY4742 (Figure 4a). The TBR1 B and C EvoTop isolates shared similar distributions of clumps to the TBR1 A EvoTop isolate ( Figure 4b). So, while the effects of the AMN1 mutations in the TBR1 B EvoTop isolate (Thr405Arg) and the TBR1 C EvoTop isolate (Lys496Asn) are harder to predict, they likely resulted in decreased or lost function, given the dramatic shift in clump-forming to "unicellular" phenotypes we observed (Figure 4).
Taken together, these findings suggest an essential role of the AMN1 gene in the transition to unicellularity, which is consistent with the known role of Amn1p in regulating a cytokinesis gene network (Fang et al., 2018;Wang, Shirogane, Liu, Harper, & Elledge, 2003).
Finally, to investigate the possible role of the AMN1 sequence in the unicellularity of laboratory strains, we also compared the sequence of the TBR1 ancestor to the published sequence of the standard laboratory strain BY4742. This revealed an Asp368Val polymorphism that changes an acidic residue to a hydrophobic one and likely impairs the functionality of the Amn1p protein (Yvert et al., 2003).

| Unicellularity weakens stress resistance but might aid growth in normal conditions
Next, we asked how unicellularity affects stress resistance compared to multicellularity. To compare the stress response of singlecelled and clump-forming strains, we isolated colonies from each TBR1 EvoTop and ancestral line that typified their respective unicellular and clump-forming phenotypes (see "TBR1 isolates" section and Figure 5). Then, we exposed each of these TBR1 isolates to three different stresses: rapid freeze/thaw, 150 μM hydrogen peroxide, and 1% exogenous ethanol, and measured their growth relative to unstressed cells over 1.5 hr. We also similarly compared the relative growth of BY4742 EvoTop and ancestral cells (Figure 6a). F I G U R E 6 Clumping protects yeast from stress but hinders growth in normal conditions. (a) Experimental procedure for testing the effects of three different forms of stress. Cells were diluted to equal concentrations, exposed to each stress, and then allowed to grow for 1.5 hr. The stress effect (relative growth) was the ratio of growth with stress to growth without stress. (b) Significant differences in the relative growth of stressed cells compared to unstressed controls indicated that single-celled TBR1 EvoTops were less stress resistant than the ancestral clump-forming TBR1 cells ( Figure 6c).
Overall, we concluded that clumping hampered growth, implying that clumping multicellularity could be costly in the absence of stress. Implicitly, then unicellularity would be beneficial without stress. However, these potential benefits were insufficient to significantly reduce clump size for the TBR1 EvoControl variant relative to the ancestral variant after 4 weeks without counter-gravitational selection (Figures 2 and 3b).
In summary, the multicellular clump-forming yeast phenotype offers the benefit of increased stress resistance over single cells.
Conversely, we found that clumping might be costly in normal conditions, arguing for some potential advantage of unicellularity over multicellularity in the absence of stress.

| Other mutations in EvoTop isolates do not appear to affect stress resistance
In addition to the AMN1 mutations we identified, TBR1 EvoTop isolate sequencing results revealed a few mutations that could potentially affect stress response (Table 1). The most notable of these is the Gln1442Lys missense mutation in the adenylate cyclase CYR1 gene of the TBR1 B EvoTop isolate. CYR1 is an essential gene that encodes adenylate cyclase, and null mutants are inviable (Giaever et al., 2002).
To help determine the impact of such mutations on stress resis-   (Figure 6b), and the overall relative stressed growth of the TBR1 EvoTop isolates was higher than that of the BY4742 ancestral and EvoTop variants ( Figure 6b). This suggests that stress tolerance of the TBR1 EvoTop isolates has not been compromised by random mutations.

| D ISCUSS I ON
While the transition from unicellularity to multicellularity has been studied extensively, the reverse transition from multicellularity to unicellularity (Hammerschmid et al., 2014) has received less attention, especially without cheaters in eukaryotes. We obtained a reverse transition by EvoTop selection as a new exception from Dollo's Law (Dollo, 1893). We found that clumping protects from stresses but is costly in normal conditions. This may imply a trade-off between growing fast and dying in stress (unicellular phenotypes) versus growing slower but resisting stress (multicellular phenotypes).
A similar trade-off was recently observed in the evolution of engineered unicellular yeast cells under stress (Gonzalez et al., 2015).
Such trade-offs emerge from the pressure to satisfy two conflicting tasks, resolved by Pareto optimality in biological evolution (Shoval et al., 2012). Overall, these findings suggest that environmental stress could play a major role in the maintenance, or possibly even the emergence of multicellularity in other species, such as social amoebae (Gregor, Fujimoto, Masaki, & Sawai, 2010). These findings corroborate the protective role that clumping or other forms of multicellularity provide against predation (Brunke et al., 2014;Pentz et al., 2015) and environmental stress (Smukalla et al., 2008).
Both physical shielding and physiological changes appear to play a role in the increased stress tolerance of flocs compared to single cells (Smukalla et al., 2008). Along with being blocked from external stressors by exterior floc cells, internal floc cells have limited access to nutrients and oxygen. This leads to decreased growth as indicated by the downregulation of mitotic genes. There is a corresponding upregulation of genes associated with stress response (Smukalla et al., 2008). Smukalla and colleagues (2008) were able to test the effects of different stressors on flocculating and nonflocculating cells by breaking up flocs with EDTA after exposure to stress and measuring colony forming units. Our clumps formed from incomplete daughter cell separation do not lend themselves well to the colony forming unit assay, which led us to examine the effects of stress by calculating the relative growth of stressed cells compared to unstressed controls. Clumps are not easily broken up, and the mechanical method of sonication we used could lyse and kill the cells. While these cells are still countable in our initial and final time points, they would not show up in a colony forming unit assay. Similarly, very small clumps with countable cells existed in the TBR1 samples after sonication, but these multiple cells would have only formed a single colony. We sonicated samples of the TBR1 cultures that we took to count, but the cells we measured for growth were not exposed to sonication until we were ready to obtain single-cell counts after they had undergone 1.5 hr of growth.
We found that haploid yeast clumps are more resistant than single cells against both physical (freeze/thaw) and chemical (ethanol and hydrogen peroxide) stressors. Interestingly, protection from freeze/thaw appears to be a benefit that the flocculating multicellular form does not possess (Smukalla et al., 2008). Strikingly, not only did the TBR1 ancestral cells tolerate freeze/thaw treatment, they grew ~110% better than their corresponding untreated controls. An explanation is that freeze/thaw cycles impose mechanical stress (Harju, Fedosyuk, & Peterson, 2004), which could cause clump splitting, improving access to nutrients. Interestingly, the unicellular BY4742 strain also started forming more multiplets during evolution, which may have been selected for because multiplets mitigate bud scars' vulnerability to mechanical stress from vortexing (Chaudhari et al., 2012). The higher stress tolerance of clumps may be due to either physical shielding of interior cells from external stressors, replicative aging, physiological changes, or a combination of such factors (Brachmann et al., 1998;Smukalla et al., 2008).
Our findings suggest that clumping may create phenotype switching and deterministic heterogeneity, which plays a role in the drug resistance of microbial pathogens (Aldridge et al., 2012).
Indeed, some S. cerevisiae strains emerging as opportunistic pathogens (Wei et al., 2007) contain the same AMN1 allele as the clumpforming TBR1 ancestor. Moreover, long-term evolution of Candida glabrata with macrophages causes the evolution of a filamentous multicellular form due to incomplete daughter cell separation (Brunke et al., 2014).

ACK N OWLED G M ENTS
We would like to thank the members of the Balázsi laboratory and Dr. Gerda Saxer, Dr. Chris Kuzdzal Fick, and Dr. Michael Lorenz for discussions. We also thank Dr. Kevin Verstrepen, Dr. Todd B.
Reynolds for strains, and Dr. Junchen Diao for time-lapse images.

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Whole-genome sequence data are available from NCBI assembled at BioProject PRJNA388338.