A key step in the evolution of complex multicellularity is the emergence of regulated life cycles that coordinate growth and reproduction. One potential route toward regulation involves co-opting intrinsic information: cues generated by routine cellular activities such as aging or mechanical stress from growth. Here, we model the simplest form of multicellular organization, linear filaments, to investigate whether intrinsic information can be harnessed to produce regular multicellular life cycles. Based on our analyses, we find that these information sources face an inherent trade-off between flexibility and regularity. Some sources, such as mechanical stress, precisely regulate when reproduction occurs but generate only a single reproductive mode. Others, such as cell age, can in principle produce diverse life cycles but fail to generate any of them reliably. Combining information sources through simple genetic circuits reduces variance in some cases, but the range of achievable life cycles remains constrained. Together these results suggest that while intrinsic information may facilitate early multicellular evolution, there are significant limitations on the degree to which it can be harnessed to evolve tightly-regulated, flexible life cycles. Our work highlights the constraints faced by nascent multicellular organisms and the evolutionary innovations likely required for coordinated multicellular development.

The evolution of reproductive specialization, in which somatic cells forfeit reproduction, represents a fundamental innovation in complex multicellular life. This specialization imposes a fitness cost: because somatic cells do not produce offspring, organisms that invest in soma have reduced fecundity. The magnitude of this cost might be expected to depend simply on the proportion of cells allocated to soma. Here, we show that these costs also decrease with the logarithm of organism size, because larger organisms require proportionally more cell divisions for development, diluting the rate at which reproductive costs compound across multicellular generations. We derive this result analytically and validate it with data from the volvocine green algae. When somatic cells provide a compensating survival benefit, a positive feedback emerges: larger organisms can afford greater somatic investment, which in turn favors further size increases. This size-scaling relationship helps explain the broad association between large organism size and multicellular complexity.

Abstract ‘Complex’ multicellularity has evolved in just five lineages (animals, plants, brown algae, red algae, and fungi) and in each case, these organisms develop clonally and are obligately multicellular. While prior work has shown that clonal development plays a critical role in the evolution of complex multicellularity, none has disentangled this from the impact of obligate vs facultative multicellular life cycles. Here we use experimental evolution with engineered ‘snowflake yeast’ ( Saccharomyces cerevisiae ) to directly test how life cycle structure affects multicellular adaptation. We created isogenic strains capable of switching between unicellular and clonal multicellular phases, then evolved populations for 192 days under obligately multicellular, facultatively multicellular, and obligately unicellular regimes. Obligately multicellular populations rapidly evolved larger size, primarily driven by a whole genome duplication, in all five replicates. Facultative populations showed dramatically constrained evolution, with tetraploidy evolving in only 2/10 facultative populations despite experiments demonstrating that it is strongly beneficial across the full life cycle. Mathematical modeling reveals the mechanistic basis for this constraint: facultative life cycles create establishment barriers through two population genetic effects. Group formation dramatically reduces the number of units of selection, making beneficial multicellular mutations vulnerable to drift. This asymmetry in population size between life cycle phases also allows cell-level selection to overpower group-level selection, eliminating mutations that provide group-level benefits but carry cell-level costs. These findings demonstrate that obligate multicellularity circumvents fundamental population genetic barriers to collective-level adaptation, helping explain why complex multicellularity has evolved exclusively in obligately multicellular lineages, and suggesting similar constraints may operate in other evolutionary transitions in individuality.

Many bacteria rely on efflux pumps to survive antibiotic stress and exposure to antibiotics often leads to mutations in pump genes or their regulators that increase pump expression. Predicting the spectrum of these mutations is important for designing effective antibiotic treatments, but the underlying regulatory networks are large and complex, making them difficult to map experimentally. To address this challenge, we developed a mathematical framework that integrates dynamical equations for efflux pump regulation with a genetic algorithm for parameter estimation and evolutionary simulations. Using this framework, we simulated in silico evolution of Pseudomonas aeruginosa under exposure to the antibiotics meropenem, tobramycin, and ciprofloxacin. The simulations revealed mutational spectra affecting the expression of four Resistance-Nodulation-Division efflux pumps and their shared regulatory network. The most frequently mutated genes were single-target regulators that matched well with previous observations in clinical and in vitro studies. The model also showed that the shared use of the OprM protein by two pumps is a key factor shaping their distinct mutational patterns. Mutations often produced multitrait phenotypes, manifesting as collateral sensitivity or cross-resistance to antibiotics not used for selection. While cross-resistance evolved readily, its extent depended on initial pump expression levels and thus may vary between strains. Finally, simulations of changing environments showed that efflux pump genes tend to be lost in the absence of antibiotics, suggesting a potential strategy to steer bacterial evolution toward reduced capacity to re-evolve resistance.

Bacteria live in diverse communities, forming complex networks of interacting species. A central question in bacterial ecology is whether species engage in cooperative or competitive interactions. But this question often neglects the role of the environment. Here, we use genome-scale metabolic networks from two different open-access collections (AGORA and CarveMe) to assess pairwise interactions of different microbes in varying environmental conditions (provision of different environmental compounds). By computationally simulating thousands of environments for 10,000 pairs of bacteria from each collection, we found that most pairs were able to both compete and cooperate depending on the availability of environmental resources. This modeling approach allowed us to determine commonalities between environments that could facilitate the potential for cooperation or competition between a pair of species. Namely, cooperative interactions, especially obligate, were most common in less diverse environments. Further, as compounds were removed from the environment, we found interactions tended to degrade towards obligacy. However, we also found that on average at least one compound could be removed from an environment to switch the interaction from competition to facultative cooperation or vice versa. Together our approach indicates a high degree of plasticity in microbial interactions in response to the availability of environmental resources.

A fundamental challenge in microbial ecology is determining whether bacteria compete or cooperate in different environmental conditions. With recent advances in genome-scale metabolic models, we are now capable of simulating interactions between thousands of pairs of bacteria in thousands of different environmental settings at a scale infeasible experimentally. These approaches can generate tremendous amounts of data that can be exploited by state-of-the-art machine learning algorithms to uncover the mechanisms driving interactions. Here, we present Friend or Foe, a compendium of 64 tabular environmental datasets, consisting of more than 26M shared environments for more than 10K pairs of bacteria sampled from two of the largest collections of metabolic models. The Friend or Foe datasets are curated for a wide range of machine learning tasks – supervised, unsupervised, and generative – to address specific questions underlying bacterial interactions. We benchmarked a selection of the most recent models for each of these tasks and our results indicate that machine learning can be successful in this application to microbial ecology. Going beyond, analyses of the Friend or Foe compendium can shed light on the predictability of bacterial interactions and highlight novel research directions into how bacteria infer and navigate their relationships.

Endosymbiotic relationships have fueled the evolution of complex life. The persistence of these relationships relies on coordinating the reproduction of hosts and their guests. Yet, such coordination may come at a cost, as one or both partners may need to slow their growth to synchronize. Here, we examine the fitness consequences of synchronization by using a combination of mathematical and bacterial metabolic models. Analyzing millions of putative host–guest pairings, we find that synchronization is typically costly for hosts but beneficial for guests. We show that synchronization can occur when hosts relinquish metabolic resources to guests. Given the costs to hosts, we investigate whether hosts can benefit from growing faster, at the risk of losing their endosymbionts. Our mathematical model suggests this outcome is theoretically possible, but metabolic modeling consistently indicates that synchronization increases overall fitness— provided the formation of new endosymbioses is rare. When new relationships can readily form, the optimal strategy shifts: hosts maximize fitness by growing as fast as possible, leading to regular cycles of loss and reacquisition of their guests. Overall, our findings indicate that while synchronization is costly for hosts, it offers long-term fitness benefits in contexts where endosymbionts are not easily replaced. ### Competing Interest Statement The authors have declared no competing interest. Kempestiftelserna

The evolution of reproductive specialization represents a fundamental innovation in multicellular life, yet the conditions favoring its evolution remain poorly understood. Here, we develop a population genetic framework that examines the fitness cost of reproductive specialization as a function of organism size. We show analytically that the costs of specialization decrease dramatically with organism size. For example, while a 4-cell organism with 50% somatic cells experiences a 50% reduction in population-level exponential growth rate (the same as the two-fold cost of sex), a hundred-cell organism faces only a 15% reduction, a thousand-cell organism 10%, a million-cell organism 5%, and a billion-cell organism merely 3.3%. This scaling relationship arises from the fact that proportionally more cellular growth in larger organisms is required for development, reducing the rate at which the fitness costs of specialization are compounded over multicellular generations. We contextualize our mathematical model with data from the volvocine green algae, showing that simple theoretical predictions closely match empirical measurements. While cellular differentiation demands that somatic advantages compensate for lost reproductive potential, we demonstrate that these compensatory requirements diminish with the logarithm of organism size, fundamentally altering the cost-benefit landscape for large organisms and potentially driving the evolution of a size-differentiation ratchet. This size-scaling relationship helps explain the broad association between large organismal size and multicellular complexity. ### Competing Interest Statement The authors have declared no competing interest. John Templeton Foundation, https://ror.org/035tnyy05, 63580 United States Department of Education, https://ror.org/021adze67, P200A210046

Multicellularity spans a wide gamut in terms of complexity, from simple clonal clusters of cells to large-scale organisms composed of differentiated cells and tissues. While recent experiments have demonstrated that simple forms of multicellularity can readily evolve in response to different selective pressures, it is unknown if continued exposure to those same selective pressures will result in the evolution of increased multicellular complexity. We use mathematical models to consider the adaptive trajectories of unicellular organisms exposed to periodic bouts of abiotic stress, such as drought or antibiotics. Populations can improve survival in response to the stress by evolving multicellularity or cell differentiation-or both; however, these responses have associated costs when the stress is absent. We define a parameter space of fitness-relevant traits and identify where multicellularity, differentiation, or their combination is fittest. We then study the effects of adaptation by allowing populations to fix mutations that improve their fitness. We find that while the same mutation can be beneficial to populations of different complexity, e.g., strict unicellularity or life cycles with stages of differentiated multicellularity, the magnitudes of their effects can differ and alter which is fittest. As a result, we observe adaptive trajectories that gain and lose complexity. We also show that the order of mutations, historical contingency, can cause some transitions to be permanent in the absence of neutral evolution. Ultimately, we find that continued exposure to a selective driver for multicellularity can either lead to increasing complexity or a return to unicellularity.

It has been argued that the historical nature of evolution makes it a highly path-dependent process. Under this view, the outcome of evolutionary dynamics could have resulted in organisms with different forms and functions. At the same time, there is ample evidence that convergence and constraints strongly limit the domain of the potential design principles that evolution can achieve. Are these limitations relevant in shaping the fabric of the possible? Here, we argue that fundamental constraints are associated with the logic of living matter. We illustrate this idea by considering the thermodynamic properties of living systems, the linear nature of molecular information, the cellular nature of the building blocks of life, multicellularity and development, the threshold nature of computations in cognitive systems, and the discrete nature of the architecture of ecosystems. In all these examples, we present available evidence and suggest potential avenues towards a well-defined theoretical formulation.

With climate change, droughts are expected to be more frequent and severe, severely impacting plant biomass and quality. Here, we show that overexpressing the Arabidopsis gene AtFtsHi3 (FtsHi3OE) enhances drought-tolerant phenotypes without compromising plant growth. AtFtsHi3 encodes a chloroplast envelope pseudo-protease; knock-down mutants (ftshi3-1) are found to be drought tolerant but exhibit stunted growth. Altered AtFtsHi3 expression therefore leads to drought tolerance, while only diminished expression of this gene leads to growth retardation. To understand the underlying mechanisms of the enhanced drought tolerance, we compared the proteomes of ftshi3-1 and pFtsHi3-FtsHi3OE (pFtsHi3-OE) to wild-type plants under well-watered and drought conditions. Drought-related processes like osmotic stress, water transport, and abscisic acid response were enriched in pFtsHi3-OE and ftshi3-1 mutants following their enhanced drought response compared to wild-type. The knock-down mutant ftshi3-1 showed an increased abundance of HSP90, HSP93, and TIC110 proteins, hinting at a potential downstream role of AtFtsHi3 in chloroplast pre-protein import. Mathematical modeling was performed to understand how variation in the transcript abundance of AtFtsHi3 can, on the one hand, lead to drought tolerance in both overexpression and knock-down lines, yet, on the other hand, affect plant growth so differently. The results led us to hypothesize that AtFtsHi3 may form complexes with at least two other protease subunits, either as homo- or heteromeric structures. Enriched amounts of AtFtsH7/9, AtFtsH11, AtFtsH12, and AtFtsHi4 in ftshi3-1 suggest a possible compensation mechanism for these proteases in the hexamer.

The evolution of multicellular life spurred evolutionary radiations, fundamentally changing many of Earth’s ecosystems. Yet little is known about how early steps in the evolution of multicellularity affect eco-evolutionary dynamics. Through long-term experimental evolution, we observed niche partitioning and the adaptive divergence of two specialized lineages from a single multicellular ancestor. Over 715 daily transfers, snowflake yeast were subjected to selection for rapid growth, followed by selection favouring larger group size. Small and large cluster-forming lineages evolved from a monomorphic ancestor, coexisting for over ~4,300 generations, specializing on divergent aspects of a trade-off between growth rate and survival. Through modelling and experimentation, we demonstrate that coexistence is maintained by a trade-off between organismal size and competitiveness for dissolved oxygen. Taken together, this work shows how the evolution of a new level of biological individuality can rapidly drive adaptive diversification and the expansion of a nascent multicellular niche, one of the most historically impactful emergent properties of this evolutionary transition.

Endosymbiotic relationships are pervasive across diverse taxa of life, offering key avenues for eco-evolutionary dynamics. Although a variety of experimental and empirical frameworks have shed light on critical aspects of endosymbiosis, theoretical frameworks (mathematical models) are especially well-suited for certain tasks. Mathematical models can integrate multiple factors to determine the net outcome of endosymbiotic relationships, identify broad patterns that connect endosymbioses with other systems, simplify biological complexity, generate hypotheses for underlying mechanisms, evaluate different hypotheses, identify constraints that limit certain biological interactions, and open new lines of inquiry. This Essay highlights the utility of mathematical models in endosymbiosis research, particularly in generating relevant hypotheses. Despite their limitations, mathematical models can be used to address known unknowns and discover unknown unknowns.

Undoubtedly, Earth s first redox revolution, which culminated 2.4 billion years ago in the Great Oxidation Event (GOE), fundamentally altered the resources available to microbial communities, leading to novel ecological competitions and evolutionary innovations. These eco-evolutionary dynamics are largely unexplored, particularly at the molecular level. Here, we hypothesize that such dynamics in the wake of the GOE explain the otherwise paradoxical evolutionary history of metal use in nitrogen fixation by nitrogenase. This ancient metalloenzyme exists in three isozymes, with distinct metal cofactors. Recent research demonstrates that the most ancient isozyme, emerging a billion years or more before the GOE, required a molybdenum (Mo) based cofactor. Alternative nitrogenases using iron (Fe) or vanadium (V) cofactors evolved after the GOE. This history is puzzling because Mo availability in the environment increased after the GOE, while Fe availability decreased, due to the contrasting environmental redox behaviors of these elements. Why, then, did the alternatives emerge only after the GOE? Using a simple model constrained by known microbial Mo quotas, we demonstrate that a strong selection pressure for use of metals in nitrogenase other than Mo is a likely consequence of competition between nitrogen fixing prokaryotes and nitrate reducing microbes, which require Mo for nitrate reduction and assimilation. This competition would have intensified after the GOE due to increasing environmental availability of nitrate, explaining the evolutionary timing of the nitrogenase isozymes. Ecological resource competition therefore emerges as a third driver of metallome evolution in deep-time, alongside the relative environmental availabilities and adaptive advantages of particular metals. ### Competing Interest Statement The authors have declared no competing interest.

The evolution of the mitochondria was a significant event that gave rise to the eukaryotic lineage and most large complex life. Central to the origins of the mitochondria was an endosymbiosis between prokaryotes. Yet, despite the potential benefits that can stem from a prokaryotic endosymbiosis, their modern occurrence is exceptionally rare. While many factors may contribute to their rarity, we lack methods for estimating the extent to which they constrain the appearance of a prokaryotic endosymbiosis. Here, we address this knowledge gap by examining the role of metabolic compatibility between a prokaryotic host and endosymbiont. We use genome-scale metabolic flux models from three different collections (AGORA, KBase, and CarveMe) to assess the viability, fitness, and evolvability of potential prokaryotic endosymbioses. We find that while more than half of host-endosymbiont pairings are metabolically viable, the resulting endosymbioses have reduced growth rates compared to their ancestral metabolisms and are unlikely to gain mutations to overcome these fitness differences. In spite of these challenges, we do find that they may be more robust in the face of environmental perturbations at least in comparison with the ancestral host metabolism lineages. Our results provide a critical set of null models and expectations for understanding the forces that shape the structure of prokaryotic life.

Simple multicellularity evolves readily in diverse unicellular species, but nascent multicellular groups are prone to reversion to unicellularity. Successful transitions to multicellularity therefore require subsequent mutations that promote the entrenchment of the higher-level unit, stabilizing it through time. Here we explore the causes of entrenchment using digital evolution. When faced with a trade-off between cellular metabolic productivity and information fidelity, digital “multicells” often evolve reproductive division of labor. Because digital “unicells” cannot circumvent this trade-off, unicellular revertants tend to exhibit low fitness relative to their differentiated multicellular ancestors. Thus, division of labor can drive the entrenchment of multicellularity. More generally, division of labor may play a crucial role in major transitions, enriching the complexity and functionality of higher-level units while enhancing their evolutionary stability.

Multicellularity has evolved several independent times over the past hundreds of millions of years and given rise to a wide diversity of complex life. Recent studies have found that large differences in the fundamental structure of early multicellular life cycles can affect fitness and influence multicellular adaptation. Yet, there is an underlying assumption that at some scale or categorization multicellular life cycles are similar in terms of their adaptive potential. Here, we consider this possibility by exploring adaptation in a class of simple multicellular life cycles of filamentous organisms that only differ in one respect, how many daughter filaments are produced. We use mathematical models and evolutionary simulations to show that despite the similarities, qualitatively different mutations fix. In particular, we find that mutations with a tradeoff between cell growth and group survival, i.e. “selfish” or “altruistic” traits, spread differently. Specifically, altruistic mutations more readily spread in life cycles that produce few daughters while in life cycles producing many daughters either type of mutation can spread depending on the environment. Our results show that subtle changes in multicellular life cycles can fundamentally alter adaptation.

Full text Figures and data Side by side Abstract Editor's evaluation Introduction Results Discussion Methods Data availability References Decision letter Author response Article and author information Metrics Abstract A key step in the evolutionary transition to multicellularity is the origin of multicellular groups as biological individuals capable of adaptation. Comparative work, supported by theory, suggests clonal development should facilitate this transition, although this hypothesis has never been tested in a single model system. We evolved 20 replicate populations of otherwise isogenic clonally reproducing ‘snowflake’ yeast (Δace2/∆ace2) and aggregative ‘floc’ yeast (GAL1p::FLO1 /GAL1p::FLO1) with daily selection for rapid growth in liquid media, which favors faster cell division, followed by selection for rapid sedimentation, which favors larger multicellular groups. While both genotypes adapted to this regime, growing faster and having higher survival during the group-selection phase, there was a stark difference in evolutionary dynamics. Aggregative floc yeast obtained nearly all their increased fitness from faster growth, not improved group survival; indicating that selection acted primarily at the level of cells. In contrast, clonal snowflake yeast mainly benefited from higher group-dependent fitness, indicating a shift in the level of Darwinian individuality from cells to groups. Through genome sequencing and mathematical modeling, we show that the genetic bottlenecks in a clonal life cycle also drive much higher rates of genetic drift—a result with complex implications for this evolutionary transition. Our results highlight the central role that early multicellular life cycles play in the process of multicellular adaptation. Editor's evaluation This study provides fundamental insight into the evolution of multicellularity by experimentally demonstrating that yeast strains that form clonal groups evolve stronger group traits than ones that aggregate into non-clonal groups. Through modeling and analysis of the mutations accumulated during the experiment, this paper provides compelling evidence in support of clonal development favoring selection at the group rather than the cellular level. https://doi.org/10.7554/eLife.84336.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Multicellularity has evolved more than 50 times across the tree of life (Umen and Herron, 2021; Grosberg and Strathmann, 2007). Multicellular organisms vary widely in their life cycles, ecological niches, and traits, and there are few features shared by all members of this diverse group (O’Malley, 2022). However, it is generally accepted that nascent multicellular lineages start out relatively simple (e.g. as undifferentiated groups of cells), gradually evolving increased organismal integration and functionality (Grosberg and Strathmann, 2007; Brunet and King, 2017; Pfeiffer and Bonhoeffer, 2003; Knoll, 2011; Herron et al., 2022). In some cases, multicellular groups that gain the capacity for group reproduction undergo a shift in evolutionary mode: the origin of multicellular Darwinian individuality (Michod, 2007; Godfrey-Smith, 2009; Buss, 2014; West et al., 2015; Maynard Smith and Szathmary, 1997). This shift, from cell-level traits being the focal point of adaptation in the unicellular ancestor, to group-level traits becoming the focal point of adaptation in a nascent multicellular lineage, represents a crucial tipping point in evolutionary dynamics. In this paper, we examine the impact of early multicellular life cycles on the capacity for groups to serve as evolutionary units capable of adaptation. There are two basic routes to forming a multicellular body. Nascent multicellular organisms employing either of these routes to group formation possess distinct multicellular life cycles. Individual cells can ‘stay together’ by forming permanent cell-cell bonds, forming clonal clusters that exhibit little within-group genetic variation. Alternatively, free-living single cells can ‘come together’, or aggregate, and often evolve to do so in response to some stimuli, such as starvation (Bonner, 1998; Bonner, 2001; Crespi, 2001; Tarnita et al., 2013; Velicer and Vos, 2009). Multicellularity has evolved multiple times via both staying together and coming together (Grosberg and Strathmann, 2007; Bonner, 1998; Du et al., 2015), but ‘complex multicellularity’ (Knoll, 2011; e.g. plants, animals, fungi, red algae and brown algae) has only evolved in lineages that develop clonally (Brunet and King, 2017; Knoll, 2011; Fisher et al., 2020). Evolutionary theory explains this observation as a consequence of social evolution: clonal life cycles are a simple and powerful mechanism that ‘de-Darwinizes’ cells (Michod, 2007; Godfrey-Smith, 2009; Godfrey-Smith et al., 2013) while Darwinizing multicellular groups. By limiting within-group genetic diversity, clonal development prevents intra-organism genetic conflict, as there is little standing within-group genetic variation for selection to act on Buss, 2014; Maynard Smith and Szathmary, 1997; Clarke, 2014. Any genetic variation that arises due to mutation gets partitioned among multicellular offspring, allowing selection to act on the group-level effects of de novo mutations (Dahaj et al., 2021). Clonal groups align the fitness interests of cells and groups, allowing cells to evolve altruistic social traits necessary for cellular differentiation (Michod, 2007; Michod, 2006). Finally, organisms that aggregate from a free-living state may experience contrasting selection on the fitness of free-living single cells and the fitness of these cells in a multicellular group (Márquez-Zacarías et al., 2021). While the numerous origins of multicellularity provide a uniquely rich set of natural experiments to examine correlations between developmental mode and organismal complexity (Umen and Herron, 2021; Brunet and King, 2017; Bonner, 2004; Fisher et al., 2013), there is no direct evidence that clonal development has played a causal role in the transition to multicellular individuality, and the subsequent evolution of organismal complexity. Indeed, a number of alternative explanations exist. For example, clonal development has evolved more frequently in aquatic environments, while aggregation has evolved more often in terrestrial environments (Bonner, 1998; Fisher et al., 2020). The lower complexity of aggregative organisms may be due simply to different life history selection in these fundamentally different environments, rather than evolutionary constraints due to within-organism genetic diversity. Alternatively, the potential for cellular differentiation appears to be highly contingent on the cell biology and behavioral repertoire of the unicellular ancestor (Brunet and King, 2017; Hanschen et al., 2016; King, 2004; King et al., 2008; Suga et al., 2013). The fact that complex multicellularity evolved in just five lineages may reflect historical contingency in developmental plasticity in these specific lineages, rather than evolutionary consequences of early developmental mode. Progress has been limited by the fact that all known transitions to multicellularity occurred in the deep past (>200 MYA Grosberg and Strathmann, 2007; Herron et al., 2009), obscuring the evolutionary dynamics of early multicellular evolution in extant lineages. Experiments directly comparing the evolutionary consequences of developmental mode have not yet been conducted, in part due to a lack of experimentally tractable model systems that can be induced to undergo either clonal or aggregative development. Here, we circumvent these constraints by synthetically generating an isogenic yeast model system capable of either clonal or aggregative development. We created clonal ‘snowflake yeast’ by knocking out ACE2 in a unicellular ancestor (Ratcliff et al., 2017; Pentz et al., 2020), and aggregative floc yeast by placing the dominant FLO1 gene under transcriptional control of the GAL1 promoter (Pentz et al., 2020; Smukalla et al., 2008). While these strains differ only in these two genes, these differences produce two fundamentally different life cycles: obligately multicellular snowflake yeast undergo a unicellular genetic bottleneck during ontogeny, making them clonal (Ratcliff et al., 2017; Ratcliff et al., 2012; Ratcliff et al., 2015), while floc yeast can form genetically diverse aggregates (Smukalla et al., 2008). We evolved 20 populations of each genotype for 24 weeks with galactose as the main carbon source, ensuring robust flocculation, selecting daily for both faster growth and increased multicellular size by selecting for rapid sedimentation in liquid media (Ratcliff et al., 2012). We chose this selective regime because it is a simple and powerful way to examine the evolutionary consequences of selection acting simultaneously on both cell-level traits (such as growth rate) and group-level traits (such as settling rate). While buoyancy regulation may be an important driver of multicellularity in some lineages (Dudin et al., 2021), we chose this selective regime because it is an efficient way to select on multicellular size, a fundamentally important multicellular trait (Tong et al., 2022). In our experiment, both floc and snowflake yeast adapted to this fluctuating environment, settling faster and growing faster. However, competition experiments with their ancestors reveal fundamentally different modes of adaptation. In clonal snowflake yeast, groups of cells served as the primary evolutionary units, gaining nearly all their increased fitness during the settling selection phase of the experiment. In contrast, aggregative floc gained most of their fitness as the cell-level, with evolved floc growing faster but possessing no measurable advantage during settling selection when competing against their ancestor. Snowflake yeast, but not floc, are evolving as primarily multicellular Darwinian individuals. Clonal development, however, resulted in far more genetic drift than aggregation. Mathematical modeling suggests that this is the result of a key difference in their life cycles. In floc, a rare beneficial mutant can disperse into many groups, while in clonal snowflake yeast it is constrained to a few, exposing them to greater sampling error during group selection. This is a general property of clonal multicellular life cycles, and this disparity scales with the number of cells within the organism prior to reproduction. Together, our results show how a simple difference in the mechanism of group formation, whether cells adhere with reformable bonds or adhere permanently, can fundamentally change their subsequent evolutionary dynamics and impact the level of Darwinian individuality (Rose and Hammerschmidt, 2021). Results Experimental evolution Our selection regime involved 24 hr of batch culture followed by daily selection for rapid sedimentation in liquid media (Ratcliff et al., 2012). This selective scheme has previously been shown to promote multicellular adaption in snowflake yeast (Ratcliff et al., 2012; Ratcliff et al., 2015; Ratcliff et al., 2013; Bozdag et al., 2023) and has led to increases in cluster size of up to 20,000-fold over 600 consecutive rounds of selection (Bozdag et al., 2023). We have previously quantified the effect of settling selection on snowflake yeast by using a variety of tools (i.e. microscopy and flow cytometry Ratcliff et al., 2012; Ratcliff et al., 2015; Ratcliff et al., 2013; Bozdag et al., 2023) that cannot be used for floc, because floc aggregates form dynamically as the clusters are settling. Thus, we developed a method to measure the settling rate of both floc and snowflake yeast populations during sedimentation (Pentz et al., 2020), calculating the displacement of biomass as they settle via high-resolution video (see Methods). We measured the settling speed of each of the 40 populations weekly over the course of the 24week experiment (Figure 1A and B). Both snowflake yeast and floc evolved to settle significantly faster (one-way ANOVA; floc: F23,456=16.57, p<0.0001; snowflake: F23,456=13.65, p<0.0001, pairwise differences assessed with Tukey’s post-hoc HSD with α=0.05). Floc, however, exhibited a much larger increase in settling rate than snowflake yeast (12-fold increase vs 1.3-fold increase after 24 weeks, respectively, Figure 1C). Figure 1 Download asset Open asset Dynamics of settling rate evolution in snowflake and floc populations. The settling rate (as determined by time-lapse imaging, see Methods) of 40 independently evolving populations of snowflake yeast (A) and floc yeast (B) over 24 weeks of evolution. Settling rate was measured for each population weekly. Each data point shows the mean of 4 biological replicates for each population. Snowflake yeast evolved to settle 30% faster over 24 weeks of evolution, while floc evolved to settle an average of 12-fold faster (C). We next examined the traits underlying increased settling speed. Using flow cytometry, we measured the size of 24-week evolved snowflake yeast. Biomass weighted mean forward scatter (a proxy for group size) increased by an average of 33% (Figure 2A and B; one-sided t-test, t=7.48, n=24, p<0.0001; non-overlayed histograms for individual snowflake isolates shown in Figure 2—figure supplement 1). Evolved snowflake isolates retain the same basic growth form (Figure 2D and E). Examining one lineage through time, we found that size appeared to plateau after 8–10 weeks of settling selection (Figure 2C), which is consistent with previous work in this model system where aerobic metabolism, and the corresponding reliance on diffused oxygen for growth, strongly inhibits the evolution of increased size (Bozdag et al., 2023; Bozdag et al., 2021). By 24 weeks, floc yeast appeared to be aggregating far more efficiently than their ancestors. To quantify this, we measured the coefficient of variation in pixel opacity in a well-mixed population just prior to settling selection (Figure 2F). This standardized variance measurement is as a proxy for flocculation efficiency, as when more cells are in flocs, the biomass will be more heterogeneously distributed within the cuvette. Thirteen populations showed significantly increased flocculation efficiency relative to the ancestor (one-way ANOVA, F(20,46)=45.53, p<0.0001, pairwise differences assessed with Tukey’s honestly significant difference [HSD] with ɑ=0.05). This resulted in a noticeable reduction in the density of free un-flocculated unicells, see examples of the ancestor (G) and a representative 24-week isolate (H). We confirmed that increased flocculation was not a product of cell density in floc cultures (Figure 2—figure supplement 2) Thus, floc yeast evolved to settle more rapidly by forming larger aggregative groups and reducing the proportion of non-aggregated cells in the population (Figure 2G and H). Figure 2 with 4 supplements see all Download asset Open asset Both floc and snowflake yeast evolved to form larger groups. (A, B) Snowflake yeast increased their settling rate by evolving larger cluster size. Shown in (C) is the size distribution of a representative snowflake population (S8) every three weeks for the duration of the experiment. Relative to their ancestor (D), evolved snowflake yeast (E) are larger but retain a similar pattern of cellular attachment (insets). (F) We estimated the flocculation efficiency of ancestral and evolved floc by measuring the coefficient of variation in pixel intensity within cuvettes of a well-mixed population taken from shaking incubation (data on top, images from each population on the bottom). Flocculation efficiency was significantly higher in 13/20 evolved populations (significance at the overall α=0.05 level denoted by asterisks). Shown are the mean and standard deviation of four biological replicates normalized to the mean of the ancestor. Representative images of ancestral (G) and evolved (H) floc yeast (genotype F4), showing fewer planktonic cells in the evolved isolate with higher flocculation efficiency. Partitioning fitness between growth and multicellular-dependent survival A common way to analyze ETIs is to use the Price equation to partition fitness arising from selection acting at distinct levels, i.e., cells and groups (Rose and Hammerschmidt, 2021; Shelton and Michod, 2020). Our system is not amenable to this kind of decomposition given the dynamic nature of flocs: groups rapidly form and fuse during settling selection, changing in size and composition until they either succeed at joining the pellet at the bottom of the tube, where they rapidly adhere to these cells, or fail to do so and are discarded. The fluid nature of flocs, and corresponding difficulty of measuring their traits without changing the traits we seek to measure, prevents us from quantifying the genetic composition and fitness of flocs during settling selection - data that is necessary for a Price equation fitness decomposition. Fortunately, we can still gain insight into how selection is acting on cell and group-level traits by leveraging the biphasic nature of our experiment. Fitness in our experiment depends on growth rate during the 24 hr of batch culture when the population is competing for resources, and survival during the subsequent settling selection step (Pentz et al., 2020; Ratcliff et al., 2012; Conlin and Ratcliff, 2016). Growth rate during batch culture is a classic trait under strong selection during experimental evolution (Lenski et al., 1991). The main way that a lineage can increase growth rates is by increasing the rate at which cells divide. In contrast, the main way in which a lineage can increase survival during settling selection is through changes in the phenotype of multicellular groups (i.e. increasing their size, packing density, or hydrodynamic profile Ratcliff et al., 2013). Thus, if populations evolve higher fitness during the growth phase, this can be taken as evidence that selection is acting effectively at the cell-level. Correspondingly, fitness improvement during the settling phase, where strong between-group selection determines survival, can be taken as evidence that selection is acting at the group-level. We measured the fitness of evolved isolates (one isolate per 24-week evolved population, or 40 isolates in total) in competition against their unevolved ancestor one 24 hr period of growth, and one round of settling selection, allowing us to partition fitness increases amongst the two factors that affect fitness in our system: growth and settling. Both floc and snowflake yeast showed overall fitness increases in our experimental regime. However, floc yeast showed a significantly higher increase in fitness than snowflake yeast, increasing by an average of 15% as opposed to 7%, respectively (Figure 3A first panel: one-sample t-test, n=20, p<0.0001). Floc yeast showed a significant fitness advantage during growth (Figure 3A; mean increase = 24%, one-sample t-test, n=20 t=10.55, p<0.0001), but despite settling an average of 12-fold faster than their ancestor when grown in monoculture, were only slightly significantly more fit than their ancestor during settling selection (Figure 3A; mean increase = 3%, one-sample t-test, n=20, t=1.633, p=0.12). Snowflake yeast displayed the opposite behavior, possessing a 5.5% fitness benefit during growth (Figure 3A; one-sample t-test, n=20, t=8.374, p<0.0001) and a 28% fitness advantage over their ancestor during settling selection (Figure 3A; one-sample t-test, n=20, t=10.29, p<0.0001). Despite floc as an overall treatment group not having higher fitness during settling selection, isolates from 2/20 replicate populations did have detectably higher fitness than their ancestor when competing during settling selection (F20,63=4.528, p<0.0001, multiple comparisons controlled by a Bonferroni correction with overall α=0.05, the same two populations were identified using a Dunnett’s test against an ancestor:ancestor control). In contrast, 12/20 snowflake populations significantly increased fitness during settling selection (F20,63=4.89, p<0.0001, multiple comparisons controlled using a Dunnett test against ancestor:ancestor control). To determine why floc obtained such marginal fitness benefit during settling selection despite evolving such a large increase in their rate of flocculation, we measured the genetic structure of flocs formed by five 24-week isolates from across the range of settling speeds by calculating their assortment (Yanni et al., 2019), a scaled statistic of genotypic enrichment relative to what would be expected by chance, which ranges from –1 to 1 (representative images of flocs, compressed to a single-cell thickness, shown in Figure 3C and D). Multicellular adaptation requires a positive correlation between group phenotype and underlying cell-level genotype; without this, selection acting on groups cannot drive changes in allele frequency (Clarke, 2014; Fletcher and Doebeli, 2009; Pepper and Smuts, 2002). Genotypes that achieve high assortment should thus have greater potential for multicellular adaptation. Figure 3 Download asset Open asset Examining fitness during growth and group-dependent competition. (A) When competing against their ancestors, both snowflake and floc yeast increased fitness in our experimental regime (first panel). Floc yeast had a 15% average fitness increase over three days with daily selection for settling, while snowflake yeast had a 7% increase. However, snowflake and floc yeast evolved distinct differences in their fitness during the growth and settling selection phases of our experiment. Floc yeast obtained a 30% fitness advantage during growth, but showed no detectable increase in fitness during settling selection. Alternatively, snowflake yeast had a 28% fitness advantage during settling but only a 6% advantage during growth. Data points show the mean of four biological replicates for each evolved isolate. (B) When competing against their ancestor, representative floc isolates formed groups with low genetic assortment. This may explain the decoupling between dramatically increased group sedimentation rate (Figure 1B), and the negligible increase in fitness during settling selection. Shown are the mean and standard deviation of three and six biological replicates for the evolved isolates and ancestor, respectively. Representative flocs are shown in (C&D), with the ancestor competing against itself, or a RFP-labeled ancestor competing against an GFP-labeled 24-week evolved isolate (genotype F10). Overall assortment in floc was relatively low (mean of five randomly-selected strains from Figure 3B was 0.06), though there was significant among-strain variation (one-way ANOVA, F4,10=4.159, p=0.008). Low assortment impedes the potential for selection acting on groups to drive changes in allele frequencies, explaining why evolved floc yeast obtained little fitness benefit from their remarkably improved sedimentation rates. In contrast, snowflake yeast canonically have an assortment of 1 (when competing two strains, every group is entirely clonal Figure 2—figure supplement 3; Pentz et al., 2020), allowing for selection acting on emergent group-level traits (i.e. settling speed) to act on underlying genetic mechanisms. Genomic analysis provides insight into evolutionary dynamics We sequenced the genome of one isolate per 24-week population (40 isolates in total), representing about 700 generations of evolution (Figure 4—figure supplement 1). Floc yeast accumulated more mutations than snowflake yeast, an average of 5 vs 3 mutations per genotype, respectively (Figure 4A; two-tailed t-test, t=2.49, n=40, p=0.017). All mutations are listed in Supplementary file 1 (Snowflake) and Supplementary file 2 (Floc). To develop insight into how each life cycle affected the potential for selection to act on mutations, we compared the frequency of different types of point mutations (which constituted the large majority of total mutations; Figure 4B) to those predicted by a null model of evolution lacking selection. Specifically, we generated a distribution of the expected frequency of different types of mutation (i.e. missense, nonsense, synonymous, or intergenic) in yeast by simulating 100,000 mutations using the Python package Mutation-Simulator (Kühl et al., 2021), then ran 1000 bootstrap simulations in which we sampled the actual number of mutations identified in both floc (104 mutations total) and snowflake (69 mutations total) populations. Then we compared the observed frequency of these four types of mutation in both floc and snowflake populations to this null distribution, which is the distribution of mutations expected without selection (i.e. under genetic drift). Figure 4 with 2 supplements see all Download asset Open asset Examining mutations for signatures of selection. After 24 weeks of evolution, floc yeast accumulated more mutations on average than snowflake yeast (5 vs 3, respectively, A). We categorized these mutations into four broad classes (B), then compared the number of each type of mutation to expectations based on a simulation of neutral evolution (C). We cannot distinguish the overall pattern of mutations in snowflake yeast from neutral expectations. In contrast, floc yeast showed strong evidence of selection, with more missense mutations and fewer synonymous and intergenic mutations than expected. The number in the upper left-hand corner of each subplot reports the shaded area of each distribution, which is the proportion of simulated runs in that observed a result at least as extreme as the observed number of mutations of that type we identified, pooling across all 20 floc and snowflake genotypes. The overall pattern of mutations observed in snowflake yeast were not distinguishable from drift, with the number of mutations in each of the four categories being near the mean of the expectation under selective neutrality (Figure 4C). In contrast, the mutations in floc yeast show a strong signature of selection. We observed more missense mutations than expected based on drift (there was a 3% chance of observing at least this many missense mutations in our null model). Similarly, we observed fewer synonymous mutations and fewer mutations in intergenic regions than expected based on our null model (in both cases, there was a 3% chance of observing this many mutations or fewer in our null model). Nonsense mutations occur infrequently, and do not provide a sufficiently large sample size for analysis. Because we did not have a prior expectation about the deviation from the null, each of the above tests should be treated as a two-sided test, meaning there is a 6% chance of observing a deviation in either direction at least as large as the one observed for the three categories of mutation described above, even when the null hypothesis of no selection is true. Taken together, our results show that selection was acting strongly on the mutations found in floc, but not snowflake yeast. The analysis above only examined the overall statistical properties of mutations, and should not be interpreted to mean that snowflake yeast did not undergo adaptive evolution. Indeed, we see that all evolved isolates taken from all 20 populations have evolved to form larger groups (Figure 2A–E, Figure 2—figure supplement 1) that settle faster (Figure 1C) and are much more fit (Figure 3A). Some of the mutations identified within snowflake yeast are putatively adaptive, such as missense mutations in the cell cycle (i.e. ULP1, SLK19; Supplementary file 1) or mutations affecting cellular morphology (i.e. AYR1, ACM1, GSC2, CHC1, ARP7, HKR1), which prior work in snowflake yeast has shown are a common mode of evolving larger group size (Dahaj et al., 2021; Bozdag et al., 2023; Bozdag et al., 2021; Jacobeen et al., 2018). We saw only a single GAL2 mutant (in contrast to seven arising independently in floc, Supplementary file 2), which is consistent with less growth rate adaptation in snowflake yeast. Indeed, it is plausible that selection acting on multicellular traits also contributed to genetic drift through the hitchhiking of otherwise non-adaptive alleles (an outcome that has been seen previously in yeast selection experiments Lang et al., 2013). Our results (Figure 3) suggest that faster settling in floc, in contrast to snowflake yeast, is only marginally adaptive. Yet if that is the case, why would floc convergently evolve to settle an average of 12 times faster than their ancestor (Figure 1)? We see two potential explanations, which are not mutually exclusive. The first is that low genetic assortment during flocculation reduces, but does not eliminate, the benefit of faster settling. Indeed, we found that 24 week evolved floc were an average of 10% more fit than their ancestor during settling selection, although this difference was not statistically significant for the treatment as a whole. Consistent with flocculation itself being under selection, we saw five parallel missense mutations in FLO1, and four in MSN1, a transcriptional activator of another major flocculin gene, FLO11 (Fichtner et al., 2007; van Dyk et al., 2005; Bayly et al., 2005; Supplementary file 2). Greater statistical power during competitions may be required to detect the benefit of increased flocculation if it exists, however. Alternatively, it may be that faster settling has evolved as a pleiotropic consequence of increased growth rate. In our experiments, mutations that increase growth rate by altering carbon metabolism may also increase flocculation efficiency as a side-effect, because FLO1 expression is under the control of a GAL1 promoter. We saw seven parallel missense mutations in GAL2, a transmembrane gala