Single-cell phenotypic plasticity modulates social behavior in Dictyostelium discoideum

Summary In Dictyostelium chimeras, “cheaters” are strains that positively bias their contribution to the pool of spores, i.e., the reproductive cells resulting from development. On evolutionary time scales, the selective advantage; thus, gained by cheaters is predicted to undermine collective functions whenever social behaviors are genetically determined. Genotypes; however, are not the sole determinant of spore bias, but the relative role of genetic and plastic differences in evolutionary success is unclear. Here, we study chimeras composed of cells harvested in different phases of population growth. We show that such heterogeneity induces frequency-dependent, plastic variation in spore bias. In genetic chimeras, the magnitude of such variation is not negligible and can even reverse the classification of a strain’s social behavior. Our results suggest that differential cell mechanical properties can underpin, through biases emerging during aggregation, a “lottery” in strains’ reproductive success that may counter the evolution of cheating.


INTRODUCTION
A recognized function of multicellular organization is division of labor, emblematically represented by somatic cells, whose death contributes to the reproductive success of the germline. Such extreme differences in the fate of cells that belong to the same multicellular structure are also found in a number of unicellular organisms-both prokaryotes and eukaryotes-that have independently evolved the capacity of generating multicellular, differentiated structures by aggregation of formerly free-living cells. The most spectacular examples of such aggregative multicellular life cycles are provided by cellular slime molds, among which Dictyostelium discoideum has become a model organism for evolutionary biology. 1,2 When they run out of food, cells of D. discoideum converge to form multicellular aggregates, which subsequently differentiate in two main terminal fates. One type dies forming the stalk that lifts the other-the spores-above the ground and favors their dispersion. 3,4 In paradigmatic multicellular organisms, where the body derives from the clonal growth of the zygote, the coexistence of germ and somatic cells is facilitated by their genetic uniformity. In aggregative microbes, conversely, where barriers to co-aggregation of cells with different genetic background are weaker, multicellular structures often harbor different lineages. [5][6][7] Conflicts in reproductive investment among genetically diverse cells are expected to threaten collective functions, and to have been particularly acute at the transition to multicellular organization. 8 In D. discoideum, such conflicts are evidenced by comparing, for one of the co-aggregating types, the fraction of cells that are passed on to the following generation (the spores) to the fraction of cells that were present before aggregation (that sets the null expectation for spore pool composition, in the absence of differential reproductive success). ''Spore bias'' is; thus, used to identify strains that have qualitatively different social behavior 6,[9][10][11] : cheaters increase their representation in the following generation with respect to cooperators, who on the contrary reduce it.
Theory predicts that, all else being equal, a positive spore bias results in an increase in the frequency of cheaters across multiple social cycles of aggregation-development-dispersal. Hence, in the absence of mechanisms that produce positive assortment between cells with different social investment, cheaters should prevail on the evolutionary timescale, in what is known as the ''tragedy of the commons''. 12,13 This conclusion is based on two assumptions: first, that spore bias profiles (spore bias as a function of frequency of the focal strain) are genetically determined, so that they are maintained over the timescale (of These assumptions are violated in some cases at least. It is for instance well known that the probability of forming spores, hence spore bias, is affected by phenotypic variation also when cells that co-aggregate are isogenic. For instance, cells have been shown to form spores with different propensity, depending on nutritional history, 14 cell cycle phase, [15][16][17][18][19][20] and duration of starvation 9 among others (reviewed in Chattwood et al. 21 and Forget et al. 2 ). Therefore, spore bias is not exclusively a function of the genetic background of cells, but also of the environment-biotic and abiotic-and of cell physiology. Moreover, several studies of pairwise chimeras showed that the sign of the spore bias can change with the frequency of the focal type in D. discoideum 22 or in closely related species. 23 Models show that biases between co-aggregating populations can result from ''strategic'' decisions taken by cells within multicellular aggregates 22 or be the outcome of differential aggregation, for instance when cells have distinct motility. 24 Such a scenario, where the outcome of competition between two strains depends on their proportion in the chimera, leads to predict that, over multiple rounds of co-aggregation, an evolutionary equilibrium would be reached, where the two partners coexist. 25 These observations challenge the idea that genotypes that have been classified as ''cheaters'' in a given experimental context always pose an actual problem to the evolution of multicellular function. However, the extent to which phenotypic variation modifies genetically established spore bias profiles is unclear. Indeed, experiments have dealt separately with co-aggregation of populations whose phenotypic differences were of plastic or genetic origin. 1,2,26 Despite recent advances in understanding gene regulation in the course of development, 20,[27][28][29] the relation between gene expression and general organizing principles that have been proposed to link cell-level properties to spore bias in chimeras is still largely uncharted.
We have designed an assay to study within a common framework how plastic variation influences social behavior in binary mixtures of cells with the same or different genetic background. The phenotypic state of cells is (continuously, in principle) modulated by changing the growth phase of cultures when starvation-induced aggregation begins. On one hand, we can thus affect the nutritional state of cells-a factor that was suggested to be primordial in defining cell fate. 14,30,31 On the other hand, differences in aggregation timing seem to be more physiological than those imposed by well-distinct culture conditions, even though the amplitude of the difference is enhanced here for effects to be measurable despite unavoidable experimental variability. The ensuing ''chronochimeras'', obtained by co-aggregation of cells harvested at different growth phases, can be realized both when the two populations have the same or a different genotype, so that cell phenotypic variation is driven by both plastic and genetic differences.
First, we show that, for two populations with the same genotype mixed in various proportions, the time of harvesting affects quantitatively, and sometimes even qualitatively, the spore bias profile, to an extent comparable to that observed when mixing genetically different populations. We next address how physiological variation acquired in the course of growth, which defines the state of the populations at the beginning of aggregation, gives rise to spore bias. Differences in the proportion of non-aggregated cells suggest that early established biases can impact reproductive success-coherently with previous studies on ''loner'' cells 32-36independently of other effects that they may have during multicellular development. The observation that single-cell physical properties change during growth, moreover, supports the idea that cell self-organization in the very first phases of the multicellular cycle may impact evolutionarily relevant biases in more general circumstances, as indicated by numerical models. 24,37,38 We verify that, according to this hypothesis, spore bias modulation by a change in aggregation timing also occurs when mixing genetically distinct strains. Not only phenotypic effects combine with genetic differences in determining the social behavior of cells, but-by modifying the frequency-dependence of spore bias-they can change the qualitative nature of the ensuing evolutionary dynamics. Our results confirm that understanding how cells self-organize into aggregates can be as important as deciphering multicellular development for predicting the evolution of social strategies in facultatively multicellular microbes. Moreover, they suggest that simple and general phenotypic differences, such as in cell motility, could translate a multiplicity of molecular mechanisms in their evolutionary effects-something that may illuminate on the emergence of aggregative multicellularity from ancestral unicellular microbes.
is the axenic lab strain AX3. 39 Plastic variation in cell phenotype at the moment of aggregation was induced by changing the phase of vegetative growth where cultures were harvested. Populations progression in the growth cycle is indeed known to induce changes in single-cell properties mediated by the accumulation of secreted factors (see Gomer et al. 40 for a review). For instance, cells harvested at low density are round whereas cells harvested at higher density have multiple pseudopodia, hence possibly different mechanical interactions with the environment and other cells. 41 Parameters that correlate with spore bias, e.g., cell-cycle phase distribution and nutritional status, 9,14,19 are moreover expected to change as a growing population moves from exponential to stationary phase. 40,42 A same initial density of cells was grown in standard culture medium for different time intervals. In order to approach natural conditions, minimize the presence of polynucleated cells, 43 and avoid drastic changes in their interaction with surfaces, cells were grown in unshaken culture flasks for periods ranging from 24 to 92 h before harvesting (see STAR methods and growth curves in Figure S1). At the beginning of aggregation, cultures were thus in one of four growth phases: Early, Mid, Late Exponential (EE, ME, LE, respectively) and Early Stationary phase (ES). We expect that the phenotypic properties of cells change continuously from one phase to the other, so that the time of harvesting acts as a tunable control parameter.
As illustrated schematically in Figure 1, two populations harvested at different phases of growth were starved by replacing the growth medium with buffer, and then mixed in different proportions at a reference time t = 0 h. We call chrono-chimeras such binary mixes, whether they belong to the same or different Figure 1. Schematic representation of the experimental protocol to produce chrono-chimeras Cell populations carrying green or red fluorescent markers were harvested at a specific phase of their logistic growth. Cultures were started at different discrete times before the beginning of the experiment. When the culture was started 24 h before, cells were called Early Exponential (EE), and they were called Mid-Exponential (ME), Late Exponential (LE), or Early Stationary (ES) when they had been cultured for 46, 68, or 92 h, respectively. Figure S1 illustrates where in the culture's growth curve these time are located. As detailed in the STAR methods, different measures were realized in order to characterize the way biases got established in the course of the multicellular cycle, which was triggered at time t = 0 by cell starvation and plating on petri dishes covered with Phytagel. Before the start of aggregation, the fraction f of cells of one population was measured by flow cytometry. Similarly, the fraction f S of the same population in the spores was performed after completion of development, 24 h later. Time-lapse movies of the aggregation were recorded on an inverted microscope, allowing to count the proportion f L of each population within the fraction of cells that remained outside aggregates (the so-called ''loners''). Moreover, measures of single cell properties (discussed later in the text) were realized at t = 0 in order to connect initial phenotypic variability to realized biases. iScience Article strains. In order to count individual cells belonging to different populations, we transformed AX3 strains with plasmids that bear a fluorescent protein gene (GFP or RFP, see STAR methods). This fluorescent labeling is maintained in the course of the multicellular cycle, allowing us to quantify, by flow cytometry (see STAR methods), the proportion of cells of a focal type before aggregation (f), and among the spores (f S ).
The social cycle is started by plating a binary cell mix on Phytagel. The aggregation (about 8 h long) is followed by multicellular development, which results in the formation of mature fruiting bodies. Spores were collected by washing whole dishes after 24h. The spore bias of the focal strain (specified for every assay) could be thus quantified as the deviation f S À f of its proportion in the spores from that in the initial mix.

Variable frequency-dependent biases in isogenic chronochimeras
First, we quantified biases in spore production induced by growth phase differences in isogenic chronochimeras. Frequency-dependence in spore bias is the basis for inferring evolutionary trajectories, as it connects the composition of the population at the beginning of a social cycle-i.e., the onset of aggregationto its composition at the beginning of the following. Hence, we assessed several initial compositions of the binary mixes, so as to derive spore bias as a function of the fraction of cells of the focal type f (referred to, in the following, as spore bias profile). Our observations indicate that phenotypic differences induced by the phase of population growth at the beginning of aggregation bias spore production. The spore bias profiles are frequency-dependent and resemble in shape, and magnitude those observed in genetic chimeras. Even though we never observed spore bias profiles that changes sign at intermediate frequencies, 22,23 it is possible that this may manifest if we sampled more extensively the growth curve.
ME and LE populations are reproducibly associated with a positive spore bias when co-aggregating with EE populations (as was previously observed when mixing cultures that had been starved for different periods 9 ). However, such advantage appears to wane for older ES cultures, which can display a negative bias. Different isogenic populations would be therefore alternatively classified as cheaters or cooperators depending on their ecological history at the onset of aggregation. iScience Article Efficiency of aggregation links phenotypic variation to spore bias Growth phase at the onset of aggregation is thus able, together with other documented sources of plastic phenotypic variation, 14,15,19,44 to affect the probability that a cell will turn into a spore. How early phenotypic heterogeneity results in biases that manifest themselves many hours later is generally unknown. In general, genetically determined phenotypic differences persist on time scales much longer than a single social cycle. Plastic variation, on the other hand, is likely to be remodeled during development. Indeed, as revealed by recent transcriptomic studies on Dictyostelium, cells experience important changes in genes expression during the social cycle. 27,29 The variations in the multicellular environment that underpin such changes are likely to act similarly on cells of different populations, so that initial plastic differences can be quickly erased.
If growth-phase-induced spore biases were chiefly due to processes acting within multicellular aggregates, one should suppose that social interactions were primed by phenotypic differences present several hours before the aggregation is completed. Such a long-lasting phenotypic imprinting may then result in differential sorting within the slug, as commonly observed when mixing different genotypes. 16,[45][46][47] In our case, however, no noticeable segregation or differential sorting of the two populations was observed either in the mound stage or during slug migration ( Figure S3 A and B), suggesting that social, strategic interactions within the aggregates may not be the main factor determining spore bias differences in chronochimeras.
Alternatively (or additionally) biases may get established early enough, so that initial differences among coaggregating cells matter, even if these differences are inconsequential at later developmental stages. This hypothesis was proposed in models that stressed the evolutionary relevance of cells that remain outside aggregates, called ''loners''. [32][33][34][35][36] Different strains were shown to leave a different proportion of non-aggregated cells; 32,36 however, spore bias was not directly assessed in those experiments.
We therefore considered whether the spore biases observed in isogenic chrono-chimeras could reflect a disproportional representation of phenotypically different populations within aggregates; thus, also in the fraction of non-aggregated cells. We measured, at a time when mounds were completely formed (about 8 h into the social cycle, see Figure 1), the loner bias. This was quantified (see STAR methods) in isogenic chrono-chimeras by subtracting the proportion of the focal population in the pool of loners to its expected proportion, that is the initial fraction f $ 50%. We decided to focus on this relative measure because it is very complicated to count the absolute number of loners over a whole field of aggregation, and moreover the loner bias compares directly to the spore bias. Figure 3 shows that the loner bias is negatively correlated with spore bias when chrono-chimeras with different growth differences are taken into account (Pearson correlation coefficient = À 0:58, p = 0:01). It should be noted; however, that the loner bias of ES populations varied when the experiment was repeated, possibly reflecting the difficulty of precisely controlling the conditions that cells meet at the entry in stationary phase.
Cultures that leave more cells as loners are thus under-represented in the pool of spores, as would be expected in the absence of strong developmental biases. A substantial part of spore bias variation might then be attributed to cells being more or less efficient in aggregating, depending on their growth phase. In particular, it appears that aggregation is maximized at intermediate times during logistic growth, while cultures that are close to the stationary phase tends to leave more cells behind.
A go-or-grow single-cell mechanism may link growth phase to aggregation efficiency In order to understand how initial phenotypic differences lead to the observed biases in aggregation efficiency, we looked for relevant single-cell parameters that might operate early in the social cycle. Differences in sensitivity to an external, diffusing signal was proposed to underpin differential aggregation propensity, and a mathematical model confirmed that this mechanism can result in variable proportions of loner cells. 36 If it is biologically reasonable to assume that signaling gets affected by growth phase, and that differences in the perception of signals may last long enough to cause differential aggregation, the specific molecules involved in this process have not yet been identified.
Another possible-and by no means alternative--explanation of loner biases is that different aggregation propensity stems from changes, along the culture growth, of cell mechanical properties. Numerical models for cell self-organization into groups indeed show that differences in cell-to-cell adhesion or velocity can lead to loner biases and be potentially involved in the evolution of aggregative iScience Article multicellularity. 24,37,48,49 We thus characterized the variation during vegetative growth of three parameters that are involved in single-cell behavior and in short-range cell-cell interactions.
First, we considered surface and cell-cell adhesion (that can also encompass systems evolved for selfrecognition 50 ). The former was quantified (see STAR methods) by measuring the fraction of cells that adhere to a culture vial at low cell density, so that direct interactions among cells should be negligible and our measures pertain to traits of individual cells. Cell-cell adhesion was quantified by measuring the percentage of cells that formed multicellular clusters in shaken cultures (see STAR Methods). Cell adhesion was found to change gradually during demographic growth. Cell-substrate adhesion significantly increases in the course of vegetative growth (Figure 4 A). Cell-cell adhesion, on the contrary, significantly decreases as a population ages (Figure 4 B). Studies on the role of differential adhesion in the evolution of social behavior focused on cell-cell interactions, and predict that less adhesive cells are found more often among the loners, 48 which is not what we observe. Another expectation that is not met in chrono-chimeras (Figure S4 A) is that if cells were able to recognize the internal state, differential cell adhesion would, like for kin recognition, induce segregation of the two co-aggregating populations into aggregates that are mainly composed by one or the other type. We therefore presume that, beyond ''social'', cell-cell contacts, cellsubstratum adhesion plays a key role in establishing social behavior, as also supported by recent directed evolution experiments. 51 Then, we addressed single-cell motility. As amebae crawl by extending pseudopods, 52 variations in adhesion to the substratum may alter the probability of aggregation by altering the speed of displacement. This hypothesis is supported by numerical models for binary mixes of self-propelled particles, which showed that differential motility can result into assortment within the aggregated phase. 24,53 Moreover, cell motility has been recently invoked as the basis of differentiation biases observed in cells with different intracellular ATP concentration. 44,54 We measured individual cell motility in populations harvested in EE, ME, LE, and ES phase.
Individual cell motility can be studied by tracking single cells in diluted cultures, so that encounters are rare. We analyzed a large number of cell trajectories ($ 600 for each condition) in populations harvested at different growth phases and diluted before realizing time-lapse movies (see STAR methods). of Dictyostelium AX2-cells just before starvation 55 ). Cell speed within these two motility classes is not significantly different in different growth phases (total displacement of migrating cells was compared with a mixed effects model ANOVA, p value = 0.4691, Figure S6). However, growth phase alters population partitioning between slow and fast cells, the percentage of low-motility cells decreasing as cell culture ages ( Figure 4C). By comparing these results to variations in adhesion, we can speculate that cells belonging to the slower class also adhere less to the surface and are more likely to end up as loners. A progressive decrease in the fraction of non-migrating cells, which are more likely to remain outside aggregates, could thus explain why ME and LE cultures tend to have a positive bias relative to EE/EE chrono-chimeras. However, in order to explain the decrease of the bias observed in EE/ES chrono-chimeras, other yet uncharacterized mechanisms should be identified.
The observed variations in cell motility and surface adhesion are consistent, as we discuss later, with a change during population growth of cells distribution among different phases of the cell cycle. Older cultures are indeed enriched in cells that have stopped their progression through the cell cycle and are blocked in G2. 42 It is known that G2 cells have a higher chance of becoming spores. 15,16,20,56 Our results suggest that, by the relation of cell cycle phase and motility, aggregation efficiency mediates between the initial cell cycle phase differences and spore bias. Like the go-or-grow hypothesis, 57 cells that are dividing may make up the low motility subpopulation. Rearrangement of the cytoskeleton during mitosis indeed would cause cells to detach from the substratum 58,59 and therefore reduce their migration efficiency, underpinning the concomitant increase in surface adherence and motility during culture growth.
The go-or-grow mechanism is independent of possible later effects of cell cycle phase on differentiation within the multicellular aggregates. As suggested by theoretical models, 24,53 it may thus be relevant more broadly, whenever two cell types with different mechanical properties co-aggregate. In particular, we expect that growth-phase-induced biases manifest also when the populations that are mixed belong to distinct strains. They could; however, be negligible with respect to genetically induced biases, thus making the spore bias profile largely independent of the specific experimental settings, and thus predictive of long-term competition among strains. iScience Article Growth phase affects social behavior also in genetic chronochimeras We considered chrono-chimeras obtained by mixing strains whose synchronous co-aggregation resulted in differential spore production. We compared spore bias profiles in three conditions: when the two strains were both harvested in the ME phase (see Figure S1 for the growth curves of different strains), when the focal strain was in EE and the other in LE, and vice-versa. The first condition corresponds to standard settings, where the two populations are harvested at the same time, and is used to set the ''baseline'' spore bias profile. If growth phase had the same effect as for isogenic chronochimeras, the latter two conditions would induce changes of the profile in opposite directions. Quantification of these effects allows us to determine if and when such changes are comparable to those induced by distinct genetic backgrounds.
In line with the hypothesis that adhesion plays a key role in determining the efficiency of aggregation, we started examining chimeras composed of strains with highly divergent cell-substratum adhesion. The AX3-Bottom strain was evolved from the ancestral AX3 strain by imposing selection for increased adherence to a culture vial, and has a strong spore bias relative to the ancestor. 51 Differences between AX3 and AX3-Bottom are not known at the genotypic level, but these strains are expected to diverge chiefly for the phenotype that was under selection (single cell adhesion to the substratum). Compared to the ''baseline'' iScience Article profile, spore bias of the focal strain increased or decreased for all frequencies, depending on whether it is harvested later or earlier ( Figure S6A). Similar results are obtained when the more advanced population is harvested in ME phase rather than in LE phase ( Figure S6B). Spore bias is thus concomitantly affected by plastic variation and phenotypic differences resulting from selection on single-cell adhesion. Even though growth differences modify the bias; however, they are not sufficient to alter the qualitative classification of social behavior. The ancestor strain keeps being classified as a cheater, so that it is predicted to outcompete the evolved strain over multiple rounds of aggregation and dispersal.
A similar growth phase-dependent variation in the bias was obtained in chrono-chimeras where the reference strain AX3 was mixed with another strain derived from an AX3 ancestor, the well-known cheater strain chtA. 45 ChtA displays the most extreme and disruptive form of selfish behavior, obligate cheating: it is unable to form spores when developing clonally, but induces its ancestor to differentiate into stalk. Consistent with this classification, we found that the AX3 strain had a negative bias when harvested at the same time as chtA ( Figure 6A). Such bias tends to increase when AX3 is harvested in EE phase and chtA in LE phase, but is reduced to almost zero for the reversed growth phase relation (when AX3 is in high proportion in the chrono-chimera), confirming again that populations in different growth states can produce variable contributions to the spore pool.
AX3-Bottom and AX3-chtA were obtained by artificial selection (directional selection of single-cell adhesion and mutagenesis followed by screening for strong spore biases, respectively). Correspondingly, iScience Article they manifest extreme social behaviors. However, in natural conditions co-aggregating strains may not be as phenotypically divergent, having evolved under selective pressures that likely were both weaker and acting on many traits simultaneously. We examined chrono-chimeras for another pairwise combination of strains, where AX3 was mixed with the axenic strain AX2. Both these strains derive from the same natural isolate (NC-4), but have genome-wide differences: AX3 carries a large duplication, corresponding to 608 genes. 60 Despite such large genomic differences, AX2 is not known to display a marked social behavior with respect to AX3. Indeed, we found that, in the absence of growth phase heterogeneity, AX3 cells co-aggregating with AX2 cells are under-represented in the spores when in low proportion in the chimera, whereas they are associated with a positive spore bias when prevalent ( Figure 6B).Spore bias profiles that change sign with the frequency of one strain in the mix are commonly observed in natural isolates, 22,23 and are likely to be more representative of interactions in the wild than the previously considered chimeras. In the AX2/AX3 chrono-chimera the focal strain AX3 shifts from behaving like a cheater (when it is harvested in EE while AX2 is in LE) to behaving as a cooperator (in the opposite case).
Taken together, results on chrono-chimeras involving different strains indicate that the effects on social behavior of growth phase-induced phenotypic differences at the time of aggregation combine with those due to genetic diversity. In some cases, growth phase-induced phenotypic differences can reverse the classification of a strain from cheater to cooperator. The contribution of these two sources of variation to spore bias is not an additive. Instead, the direction of change in spore profile as a function of the growth phase of the two cultures depends on the chimeras genetic background, as summarized in Figure S7.

DISCUSSION
Division of labor within multicellular structures, whereby different cells take up different tasks, is essential for sustaining collective functions, but is often associated with differences in reproductive success among distinct cell types, e.g., between germ and somatic lineages. Such differences are particularly disruptive when cell heterogeneity is transmitted across generations of the collective association. In aggregative microbes like D. discoideum, where multicellular groups are formed by gathering previously isolated cells, cell types that are overrepresented in the spore pool have the potential to get, over successive aggregation cycles, progressively enriched in the population. Crucial for this to occur is; however, a heritable relation between cell genotype and its reproductive success. This is realized when the outcome of social interactions is by and large genetically determined, as assumed by the theory of sociobiology. 1 Several recent studies have started to question the relevance of this assumption for Dictyostelium and revealed complex relationships between properties of single cells and their reproductive success. Madgwick et al. 22 proposed that frequency-dependent spore bias profiles in pairwise chimeras of natural strains are explained by cells adjusting their probability of sporulating as a response to the multicellular context. In this perspective, cheating would result from a ''strategic'' choice of each cell, influenced for instance by the diffusion of morphogens during multicellular development. 61 Therefore, a same cellular genotype could give rise to multiple possible biases depending on how many and what kind of cells happen to surround a focal cell, giving rise to frequency-dependent spore bias profiles.
Our results show that spore bias profiles depend on the nature of the social partner also in isogenic chronochimeras, where recognition of genetic identity is not an issue. Not only the intensity of the bias depends on the composition of the mix (Figure 2), but a population harvested in early exponential phase can be associated with a positive or a negative spore bias, depending on the growth phase of the co-aggregating population. The correlation between spore and loner bias suggests that biases do not necessarily require ''negotiations'' involving a multitude of cells, and can get established ahead of multicellular development, as a result of cell self-organization during aggregation.
Similar mechanisms acting at early stages of the social cycle were suggested to underpin the role of loner cells in the evolution of cooperative behavior. 32,33,36 While early models proposed that the probability of aggregation (hence the ensuing spore bias) was determined by strain genotype, additional measures revealed that the situation is more complex, and context-dependence widespread. 32,36 In the absence of direct measures of spore bias, a mathematical model was used to show that differences in the sensitivity to an aggregation signal induce frequency-dependent loner bias profiles, which can be leveraged for main- iScience Article If biases are broadly set before multicellular development, then pre-existing differences in single-cell phenotypic properties can matter as long as they persist throughout the aggregation phase, whether their origin is genetic or plastic. Models for cell aggregation indeed point to the possible role of cell-cell adhesion and motility 24,34,35,37,48,62 -on top of the chemotactic response to diffusing signals 36 -in establishing loner biases. With this hypothesis in mind, we looked for single-cell mechanical properties that varied with the growth phase of the cell culture.
Cell-surface attachment, cell-cell adhesion and single-cell motility all show related changes during the initial phases of culture growth. We focused in particular on characterizing variation in the distribution of single-cell motility because, on one hand, motility appears to be regulated by ATP independently from cyclic adenosine monophosphate (cAMP) oscillations. 54 On the other hand, it can be connected more directly to the observed increase in proportion of aggregated cells as the growth phase advances. When a population ages, indeed, the fraction of actively dividing cells (in the M phase of the cell cycle) is known to decrease. 15,42 Dictyostelium cells entering cytokinesis tend to round up and to be less adhesive to the substratum. 58,59 As a consequence, they may contribute disproportionately to the pool of non-migrating cells that end up not joining any aggregate, with a mechanism analogous to the ''go-or-grow'' hypothesis proposed for cancer cells 57 and recently applied to Dictyostelium motility under hypoxia. 63 Such a mechanism, moreover, roots the previously reported negative correlation between the fraction of M/S cells and spore production in mechanical processes occurring at the onset of the multicellular cycle, when cell behavior is least influenced by social interactions. [15][16][17][18][19][20] Mathematical and numerical models showed that heterogeneity in single-cell motility can result in differential partaking of the multicellular organization, with consequences that extend to the evolutionary timescale. 24,36,49 It is therefore possible that some of the principles illustrated by our observations in controlled lab settings may apply more broadly.
Another cell cycle-based source of bias in aggregation efficiency may stem from differences in the timing of division during Dictyostelium social cycle. Every cell divides once, but, depending on the position within the cell cycle at starvation, some cells tend to divide more during aggregation while others divide mostly in the mound (see Maeda 64 for a review). When cells dividing mostly during aggregation are mixed with cells dividing mostly in the aggregates, the former population is expected to leave more loners than the latter, i.e., to be associated with a negative loner bias. Understanding the importance of differences in division timing as a source of aggregation bias will require an accurate description of populations' cell cycle distribution.
Growth phase differences at the moment of starvation are expected to occur in natural populations, where the history of cells during vegetative growth may vary greatly even within a single clone, due to different timing of spore germination, which sets the onset of demographic growth. As cAMP, the main signal driving aggregation of D. discoideum, diffuses very fast, it seems moreover likely that, in the soil, the aggregation domains of a few centimeters encompass micro-scale variation in biotic and abiotic factors.
Here, we have used growth phase as a control parameter to continuously tune the phenotype of cells in a population, and explored large time lags in order to quantify differences with more ease. Whether the hypothesis of synchronous or asynchronous aggregation is closer to natural aggregative cycles would require additional studies in the wild.
Temporal differences are just one possible non-genetic source of cell phenotype variation that affects representation in the pool of spores. Plastic heterogeneity could be caused by both environmental and physiological variation. 9,14,44,65 How it gets transmitted across the aggregation phase and through multicellular development is however, still unclear. Indeed, given the fast changes in gene expression during Dictyostelium's social cycle, 66,67 phenotypic variation forged during vegetative growth should in principle fade shortly after the beginning of the social cycle. On the contrary, non-aggregated cells are irreversibly excluded from multicellular development, so that differences in aggregation efficiency might explain biases in isogenic populations derived from different sources of plastic variation. The possibility that initial mechanical heterogeneity may compete in more general settings with genetically established social behavior will require further exploration.
Single-cell phenotypes can influence aggregation probability irrespective of their genetic or non-genetic origin. However, their underpinning and the way they get transmitted along the multicellular cycle are important to understand the evolution of social behaviors. In particular, the prediction that ''cheaters'' iScience Article have a long-term advantage could be upended if short-term measures of reproductive success do not carry over to successive generations, so that relevant variation is effectively neutral. 2,62,68 Our results point to a role of unpredictable variation that may be much larger than previously considered when associating a genotype to a social behavior. Single-cell properties at the moment of aggregation and the derived biases are not only shaped by the genetic identity of a strain, but also by factors that are not under direct genetic control. Such factors have more to do with the ecological history of individual cells, and its consequences on cell mechanics than with the genetically determined behavior in a multicellular, social context. We can thus speculate that spore bias variability may have similar underpinnings, whatever the origin of phenotypic variation. If such explanation holds true also in natural populations, it may contribute to understand how aggregative multicellular life cycles persist on evolutionary times despite unavoidable genetic conflicts.

Limitations of the study
A first limitation of our experimental approach comes from the high level of variability in spore bias measurements. Such variability seems to be irreducible even in controlled experimental conditions, and constrains the extent to which the effect of small phenotypic differences can be quantified. For this reason, we have pushed temporal differences in chronochimeras to extreme levels, which may not be attained in natural settings.
Second, our experimental setup does not allow counting the total number of loners produced at the end of aggregation, but only their composition. As a result, we could not test if the bias established during chronochimeras aggregation explains exactly the bias in the spores pool composition observed at the end of development (which would anyway require to assume a fixed proportioning of cell types in the multicellular body).
Finally, when exploring the mechanistic basis of loner bias, we did not explore single-cell heterogeneity in signaling and chemotaxis. Characterizing cAMP signaling dynamics of populations harvested in different growth phases would allow testing the hypothesis that variation in population signaling properties is a source of loner bias. 36 In particular, differences in signaling may explain the inconsistency between the loner bias observed in EE/ES chronochimeras and the motility properties of ES-cells.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:   32 Both plasmids carry a gene for antibiotic resistance (Gentamicin 418, Sigma-Aldrich: G418). Vegetative growth was started from frozen aliquots thawed every week to prevent the accumulation of undesired mutations due to prolonged culturing. Cells were grown in autoclaved HL5 medium (per L, 35.5 g HL5 from formedium, pH = 6.7) at 22 C with a concentration of 300 mg mL À1 Streptomycin. Additional 20 mg mL À1 G418 were supplemented when growing transformed strains. Pre-cultures were prepared by thawing frozen aliquots and growing in 25 cm 2 TC treated flasks (CytoOne CC7682-4825) with 10 mL culture medium for 30 h, which allows the population to restart and enter the exponential growth phase. Cells were grown in static cultures to limit the risk of impaired cytokinesis as observed in shaken suspension. SorC buffer was prepared with 0.0555 g CaCl 2 ; 0.55 g Na 2 HPO 4 7 H 2 O; 2 g KH 2 PO 4 per Liter.

Strains growth kinetics
Growth kinetics of AX3-GFP, AX3-RFP, AX3-Bottom, AX3-chtA and AX2 were characterized to estimate the timing of their growth cycle under the experimental conditions ( Figure S1).

Starvation protocol
In order to trigger Dictyostelium social cycle, cell populations were starved by washing out the nutrient medium via three successive centrifugations with buffer at 4 C (2000 rpm for 7 min). Cells were kept on ice between successive rounds of centrifugation. After the last centrifugation, the pellet was re-suspended in buffer and the cell density adjusted to 2:10 7 cells mL À1 .

Chrono-chimeras preparation
Chrono-chimeras are composed of a mix of starved cells from two populations harvested at different times during vegetative growth, i.e in different growth phases at the time t = 0 when the experiment was begun (Figure 1). Each population was started from a pre-culture diluted into fresh medium to a density of 10 5 cells mL À1 . In order for them to attain different growth phases (established based on the growth kinetics displayed in Figure S1) at t = 0, cultures were started a fixed number of hours before the beginning of the experiment: 24 h h for early exponential phase (EE); 46 h for mid-exponential phase (ME); 68 h for lateexponential phase (LE) and 92 h for early stationary phase (ES). Beforehand, we made sure that transformed cells used in isogenic chronochimeras had indistinguishable growth curves ( Figure S1) to confirm that populations harvested after the same growth duration were in the same growth phase.
At t = 0 h (Figure 1), the two cultures were starved as described in the previous section. Starved cells from the two populations were then mixed so as to attain a target proportion. Deviations from the target proportion sometimes occurred due to fluctuations in dilution. The actual mix composition f was thus quantified by measuring the proportion of labeled cells by flow cytometry (Cube8 cytometer, using Forward Scatter (FSC), Side Scatter (SSC), fluorescence channels: FL1 (GFP) and FL3 (RFP)). The accuracy of this measurement was first validated by comparison with manual countings with a hemocytometer. A volume of 40 mL of the mix (corresponding to 8 10 5 cells) was then plated on 6 cm Petri dishes filled with 2 mL of 2% Phytagel (Sigma-Aldrich), following Dubravcic et al. 32 Cells were then incubated for 24 h at 22 C. For each mix, three technical replicates were performed by plating three 40 mL droplets of cell suspension on 3 different Petri dishes.

QUANTIFICATION AND STATISTICAL ANALYSIS
Quantification of spore bias in chrono-chimeras 24h after plating, spores were harvested by washing the three Petri dishes corresponding to the three technical replicates in 500 mL SorC buffer. Spores suspensions were incubated for 5 min with 0:5% Triton X-100 and then centrifuged for 7 min at 2000 rpm to remove stalk cells or unaggregated cells that would have survived the 24h-starvation-period. Finally, the pellet was re-suspended in 800 mL SorC buffer and the proportion of GFP and/or RFP-spores was scored using a cytometer. Spore bias of the focal population was quantified as the deviation between its proportion in the spores f S and its proportion f in the initial mix.
To quantify the effect of growth phase heterogeneity on spore bias, we measured spore bias in chrono-chimeras composed of cells harvested in EE, ME, LE and ES phases of vegetative growth. Starting from the same batch of frozen aliquots, we tested a range (between 7 and 15 for each binary combination) of proportions f to assess frequency-dependent effects on spore bias. In principle, transformed strains were expected to produce no bias upon co-aggregation if the inserted fluorescent markers were strictly equivalent. However, we realized that chimeras composed of AX3-RFP and AX3-GFP cells grown in co-culture and both harvested in EE phase yielded a reproducible and significant bias ( Figure S2A). The same bias was observed after having repeated the transformation protocol. In order to compensate for such labeling effect on spore bias, we substracted from our measures of the spore bias the bias predicted based on co-culture of differently labeled populations. Such intrinsic bias was computed for every frequency by interpolating with a third degree polynomial constrained with f = 0 and f = 1 ( Figure S2A). In order to validate the use of this correction, we confirmed that spore biases are reversed when the fluorescent labels are swapped in a chrono-chimera where cells in EE and ME phase are mixed ( Figure S2B).

Quantification of loner bias in chrono-chimeras
Loner bias was estimated in chrono-chimeras composed of a comparable number of cells from the two populations (i.e f $ 0:5). Chrono-chimeras were prepared as previously described and plated on a Petri iScience Article dish that was scanned and imaged at regular time intervals (5 min). Images were taken with an automated inverted microscope Zeiss Axio Observer Z1 with a Camera Orca Flash 4.0 LT Hamamatsu, using a 10X objective, which yielded phase contrast and fluorescence images. Cell aggregation was considered complete when the last streams disappear. At that time, the number of unaggregated cells from the two populations was scored. Images corresponding to different areas of the Petri dish were first analyzed using ImageJ software 69 : aggregates were manually contoured and discarded and the ''Find edges'' ImageJ function was applied to highlight the contour of individual unaggregated cells. f L , the fraction of loners produced by the focal population, was then estimated on several images as the number of unaggregated cells from this population divided by the total number of unaggregated cells. Based on this observable we were able to quantify the bias in the fraction of unaggregated cells as the deviation between the proportion of cells from the focal population found in the pool of unaggregated cells (f L ) and f. The chrono-chimeras for which we measured the loner bias in parallel of the spore bias ( Figure 3) were started from a stock of frozen aliquots with higher initial cell density than that used to measure spore bias for a range of f values (Figure 2). If spore bias variation is overall consistent, the EE/ES chimeras showed a more variable and mostly positive spore bias, suggesting possible long-term memory effects of population density.

Motility assay
Cells were first starved as previously described. After the last centrifugation, the pellet was re-suspended in 3 mL of buffer with a density of 10 4 cells mL À1 . This cell density was sufficiently low for cells not to touch with one another during the assay. The cell suspension was then poured in an empty 6 cm Petri dish. After 30 min -the time for the cells to attach to the bottom of the dish-cells trajectories were tracked for 1h (one image per 30 s) under an inverted microscope equipped with a moving stage and a 5X objective (alike to the measure of 'loner bias'). A large area of the Petri dish was scanned to analyze around 600 cells trajectories per sample. Cell trajectories were then automatically extracted from the time lapse movies using the Python package Trackpy. 70 Three biological replicates were imaged for populations harvested in EE, ME, LE and ES phase. Another script was used to analyze trajectories. Mean square displacement (MSD) was computed as a function of time lag for every single cell. The slope of the log(MSD) vs. log(time lag) curve at low Dt values (Dt<150 seconds) was used as a criterion to distinguish slowly moving (slope < threshold) from fast-moving cells (otherwise). The threshold value was set to 0.5 in order to separate the two modes of the slope distribution ( Figure S5), and the proportion of cells belonging to each class was scored. Individual cell total displacement was quantified as the sum of cells displacements between two successive frames along the trajectory.

Cell-substratum adhesion assay
Cell-substratum adhesion was quantified based on cells' ability to attach to the bottom of a TC treated culture flask (CytoOne, CC7682-4325) as in. 51 Cells were first starved as previously described. After the last centrifuging, the pellet was re-suspended in 10 mL buffer and cell density was adjusted to 2; 5:10 5 cells mL À1 . The cell suspension was then incubated in a 25 cm 2 flask (CytoOne CC7682-4825) for 30 min at 22 C, the time for cells to attach to the bottom of the flask. Each culture flask was gently shaken to resuspend cells that were not attached to the bottom of the flask. The supernatant (containing unattached cells) was transferred into a 15 ml tube. Cell density in the supernatant was measured using a hemocytometer. The fraction of adhesive cells was obtained by dividing the density of cells in the supernatant by the total cell density inoculated in the flask and used as a proxy for cell-substrate adhesion level. This assay was performed on three biological replicates for populations harvested in EE, ME, LE and ES phase.

Cell-cell adhesion assay
Cell-cell adhesion was quantified with a modification of the method by Gerrish. 71 Cells were first starved as previously described. After the last centrifugation, the pellet was re-suspended in 0.5 mL buffer at a density of 10 6 cells mL À1 . The cell suspension was rotated at 150 rpm and 22 C for 1 h, allowing cells to form multicellular clumps. The number of unaggregated cells (singlets and doublets) was determined using a hemocytometer. The percentage of cells that had been recruited into aggregates was calculated as the total cell density minus the density of unaggregated cells, divided by the total cell density. This quantity was used as a proxy for cell-cell adhesion level. The assay was performed on three biological replicates for populations harvested in EE, ME, LE and ES phase.

ll
OPEN ACCESS