Acceptable Loss: Fitness Consequences of Salinity-Induced Cell Death in a Halotolerant Microalga

Environmentally induced reductions in fitness components (survival, fecundity) are generally considered as passive, maladaptive responses to stress. However, there is also mounting evidence for active, programmed forms of environmentally induced cell death in unicellular organisms. While conceptual work has questioned how such programmed cell death (PCD) might be maintained by natural selection, few experimental studies have investigated how PCD influences genetic differences in longer-term fitness across environments. Here, we tracked the population dynamics of two closely related strains of the halotolerant microalga Dunaliella salina following transfers across salinities. We showed that after a salinity increase, only one of these strains displayed a massive population decline (−69% in 1 h), largely attenuated by exposure to a PCD inhibitor. However, this decline was followed by a rapid demographic rebound, characterized by faster growth than the nondeclining strain, such that sharper decline was correlated with faster subsequent growth across experiments and conditions. Strikingly, the decline was more pronounced in conditions more favorable to growth (more light, more nutrients, less competition), further suggesting that it was not simply passive. We explored several hypotheses that could explain this decline-rebound pattern, which suggests that successive stresses could select for higher environmentally induced death in this system.


Introduction
Living organisms are pervasively exposed to stressful environments that may reduce their demographic vital rates (survival, fecundity), in turn decreasing their population size and making them vulnerable to extinction. A widespread mechanism by which organisms can cope with such environmental stresses is phenotypic plasticity, whereby a given genotype produces different phenotypes in response to the environment (Levins 1963;Bradshaw 1965). When the expressed phenotypes increase fitness across environments, plasticity is said to be adaptive, while plasticity that reduces fitness across environments is described as maladaptive (Ghalambor et al. 2007). Adaptive plasticity and its evolution can reduce extinction risk in fluctuating environments that are sufficiently predictable (Reed et al. 2010;Ashander et al. 2016) and may also allow tolerance of higher rates of directional environmental change (Chevin et al. 2010).
However, beyond these simple definitions, demonstrating whether a plastic response is adaptive, and why, may be challenging in practice. This is especially true for traits whose most immediate effect is to decrease a component of fitness. For instance, increased cell death rate in response to environmental stress would seem like the archetypical maladaptive, passive trait, which the organism can neither prevent nor control. Yet many unicellular organisms can undergo forms of cell death that are genetically controlled and involve complex cellular cascades-from caspase-like activity to nuclear DNA degradation (Jiménez et al. 2009;Kroemer et al. 2009;Bidle 2015)-that share features with programmed cell death (PCD) in multicellular organisms (Ameisen 2002;Bidle and Falkowski 2004;Deponte 2008;Kasuba et al. 2015). Because such PCD appears as an active process, it is generally considered as adaptive, in effect representing a form of adaptive plasticity triggered by a variety of environmental stimuli. However, how and why PCD may be favored by natural selection in unicellular organisms (Franklin et al. 2006;Nedelcu et al. 2011;Durand and Ramsey 2019;Vostinar et al. 2019), where it amounts to cell suicide, remains difficult to understand in most cases, except for a few wellstudied examples Berngruber et al. 2013). In general, the main argument for considering PCD as adaptive in unicells is that it is an active process, but this is not satisfying because (i) the line between PCD and less active forms of death is somewhat blurred, with a range of intermediate situations that are difficult to classify (Kroemer et al. 2009;Barreto Filho et al. 2022a, 2022b; and (ii) even plastic responses that are not evidently active or programmed could still have been favored by natural selection. Therefore, understanding the causes of selection on PCD-or any form of environmentally induced rapid death, regardless of where it stands along the PCD continuumrequires demonstrating that this trait varies genetically and that it is associated with differential fitness (as highlighted in, e.g., Reece et al. 2011). Even though PCD is unlikely to be favored directly, since it causes an immediate demographic decline of the corresponding genotype, it may still be selected indirectly if it is associated with higher fitness in the long run, as any other trait that is selected through its covariance with fitness (Price 1970;Durand and Ramsey 2019). Therefore, understanding selection on PCD requires establishing its covariance with longer-term fitness across genotypes as well as experimentally testing specific hypotheses for how its fitness costs and benefits arise.
Here, we undertake such an approach with the unicellular microalga Dunaliella salina. This halotolerant microalga is the main primary producer of hypersaline biomes such as continental saline lakes, coastal lagoons, and industrial salt ponds, where salinity fluctuates along the year (Ben-Amotz et al. 2009). Dunaliella salina can tolerate a broad range of salinities, from below seawater to saturated brine ([NaCl] ∼ 6:2 M), through a number of wellcharacterized, phenotypically plastic mechanisms (Oren 2005;Ben-Amotz et al. 2009;Chen and Jiang 2009). These include rapid morphological flexibility in response to sudden osmotic shock-allowed by the absence of a cell wallfollowed by slower physiological adjustments, most notably the production of glycerol as an osmoprotectant (Zidan et al. 1987). Overall, phenotypic plasticity is thus a key adaptive feature in this species, and it was also recently shown to readily evolve in the laboratory (Leung et al. 2020).
Species of the genus Dunaliella can undergo PCD under various environmental triggers, such as darkness (Berges and Falkowski 1998;Segovia et al. 2003;Orellana et al. 2013), UV radiation (Jiménez et al. 2009), or hyperosmotic shock (Jiménez et al. 2009), the latter being directly relevant for adaptation to salinity. But here again, the potential benefits of PCD remain unclear. A study in D. salina has suggested an advantage of PCD via resource sharing, since dying cells can release the large amounts of intracellular glycerol they produce for osmoregulation, which may then be remineralized by prokaryotes or directly consumed via a heterotrophic pathway (Orellana et al. 2013). However, this study did not compare genotypes that differ in their rate of PCD, to investigate how these differences covary with fitness in the longer run. Recently, Leung et al. (2022) have identified two closely related strains of D. salina that vary in their rates of salinityinduced death. Interestingly, it has also been shown that strain-specific demographic response to salinity, including a rapid decline followed by rebound under hyperosmotic stress for one strain, is also related to the gene expression response involved in chloroplast functions . This lends support to the hypothesis of chloroplastmediated PCD, as described in plants and different unicellular organisms (Zuppini et al. 2009;Thamatrakoln et al. 2013;Murik et al. 2014;Ambastha et al. 2015;Bidle 2016). We here use these strains to explore hypotheses about what drives the intensity of cell death and, more crucially, how this decline relates to later population growth. One of our most puzzling findings is that more cell death is associated with faster subsequent population growth across treatments.

Strains
We focused on two closely related strains of Dunaliella salina, CCAP 19/12 (hereafter, strain A) and CCAP 19/ 15 (hereafter, strain C; Emami et al. 2015), which were previously found to vary in their death rates in response to salinity . We received these strains in 2017 from the Culture Collection of Algae and Protozoa (United Kingdom) and maintained them at constant salinity ([NaCl] p 2:4 M) for more than 3 years prior to this experiment, transplanting them into fresh media twice a week for 5 months (as described in Rescan et al. 2020, constant treatment) and then about once per month thereafter. In all experiments described below, we used 10 (nearly) isogenic lines (five for each strain), founded from a single haploid cell isolated using cell-sorting flow cytometry (BD FACSAria III; BD Biosciences) in May 2019.

Common Experimental Design
Culture Conditions. All lines were cultured in 50-mL flasks (Cellstar, VWR 392-0016; Greiner Bio-One) using custom-made artificial saline water (table S1) that had a modified NaCl concentration and was enriched with 2% of Guillard's F/2 nutritive medium (G0154-500 ML; Sigma). The target culture salinity was obtained by mixing the required volumes of hyposaline ([NaCl] p 0 M) and hypersaline ([NaCl] p 4:8 M) solutions, taking into account the dilution of the inoculum for a final volume of 25 mL. Cultures were maintained in a growth chamber under 12L∶12D photoperiods, with light intensity of 100 mmol/m 2 /s (unless otherwise stated; table 1), temperature of 247C, and position randomized with respect to treatment.
All assays followed the same general experimental design. We first acclimated the lines at intermediate salinity ([NaCl] p 2:4 M) for several days (between 4 and 13 days; see table 1). We then induced either hyperosmotic shock or isoosmotic transfer by inoculating a volume of the acclimated cultures into fresh medium at the required expected initial density and then followed the demographic dynamics over 10 days. The initial densities varied between 5,000 and 50,000 cells/mL across experiments for practical reasons: we needed the minimal density (after putative population decline) to be large enough to be detectable with our flow cytometer but still low enough for substantial exponential growth to occur following decline. More details on each experiment appear in table 1 and the corresponding sections below.
Population Density Measures. Population density was assayed using a Guava EasyCyte HT cytometer (Luminex), with a laser emitting at 488 nm. We checked the cytometer performance before each measure with the Guava Easy-Check kit. The acquisition settings were 30 s or until the number of counted events reached 200,000 (never reached). Band pass filters based on forward scatter, side scatter, and fluorescence emissions in natural red (695/50 nm [wavelength/bandwidth]) and yellow (583/26 nm) enabled the discrimination of live Dunaliella cells from other particles, owing to chlorophyll a natural fluorescence . Note that some dead (or dying) cells had reduced red fluorescence relative to live cells and were thus not taken into account for population density ), but these apparently dead cells do not account for the large population decline we focus on here (fig. S1). We also isolated single cells from doublets (i.e., when two cells were detected as a single event) on the basis of the area-toheight ratio of the electronic pulse induced in the flow cytometer (Wersto et al. 2001). To determine population density, we sampled 200 mL of each population to count the number of live cells at several time points: first at the Note: In the "Acclimation [NaCl]" column, M stands for the molar concentration of NaCl. All variables in the "Fixed effects for GLM" column were treated as categorical except for inhibitor concentration and time (in the generalized linear model [GLM] for maximum growth during rebound), which were treated as continuous. PCD p programmed cell death. end of the acclimation step for the calculation of initial population density before each experiment, then 1 h after the osmotic stress (day 0), and finally once per day from day 1 to 10.
When the cytometer was unavailable (nutrient limitation experiment below, days 0 and 1), we measured wholewell fluorescence (excitation at 390/80 nm and emission at 685/40 nm) of cultures in 96-well plates, using a BMG ClarioStar spectrometer (BMG Labtech). To convert these fluorescence measurements to population density, we fitted a linear model (LM) between cytometer count and fluorescence for days 2-4 of the nutrient limitation experiment (see fig. S2), for which both measurements were available and the population had not yet reached stationarity ( fig. S6).

Experiments and Treatments
We performed a series of experiments to test different hypotheses about the drivers and population consequences of rapid cell death, as summarized in table 1. to 4 M of NaCl) to assess whether the response depends on salinity change or salinity value (three replicates per condition).
Experiment 2: PCD Inhibitor. A plausible explanation for the observed rapid population decline ( fig. 1) is PCD, known to be common in D. salina (Orellana et al. 2013) and related species (Berges and Falkowski 1998;Segovia et al. 2003;Jiménez et al. 2009), notably in response to salt. To investigate this, we used Z-VAD(Ome)-FMK (Cell Signaling Technology), a cell-permeable irreversible caspase inhibitor that prevents caspase-3-like from adopting its active form and that has been used to inhibit cell death in a diversity of unicellulars (Bidle et al. 2007;Zuppini et al. 2007;Bidle and Bender 2008;Segovia and Berges 2009;Yordanova et al. 2013), including Dunaliella tertiolecta (Segovia et al. 2003). Following a preliminary test ( fig. S3), we compared the rate of decline of cells treated with four concentrations of this inhibitor: 0, 1.6, 3, and 4 mM. Treated cells were incubated in the dark for 30 min before isoosmotic and hyperosmotic transfers were performed for both strains, with three replicates per treatment. The final volume in the flask was 10 mL, with an initial total number of cells being 200,000. Population densities were then assessed through cytometer measurements 1 and 4 h after the transfer on day 0 and then on days 1-3 (see fig. S4 for supplementary days 6-8 and for strain C dynamics). : Demographic dynamics of Dunaliella salina following salinity transfers. Each panel represents a transfer condition from the acclimation salinity (S1 p salinity 1) to the assay salinity (S2 p salinity 2). Transfer occurs on day 0, and the first measures were made 1 h after the transfer; the cross represents the expected initial density, predicted from dilution of the acclimated population. The number of live D. salina cells per milliliter on each day is represented on a log scale, for strains A (circles) and C (triangles). Symbols are averages over three replicates, error bars indicate the standard error, and lighter points are the raw density measurements.

Experiment 3: Initial Density and Population Growth
Phase. The rapid cell death that we focus on strongly impacts population dynamics, which is indeed how it is detected ( fig. 1; Berges and Falkowski 1998;Leung et al. 2022). It is thus likely to interact with population density and competition occurring in the population (Ameisen et al. 1995;Christensen et al. 1995). In particular, the initial decline may influence the postdecline growth rate because fewer live individuals means reduced competition for resources (relaxed density dependence). These effects may also depend on the physiological state of the population prior to the transfer, which is itself influenced by competition and resource limitation in the recent past.
To test for these effects, we acclimated lines of strain A and strain C for 4 days (midexponential phase), 13 days (early stationary phase), and 41 days (late stationary phase), representing three different regimes influencing the current physiological state, and then exposed them to hyperosmotic stress starting at three different initial densities (5,000, 20,000, and 30,000 cells/mL; fig. S5). We used four different lines per strain as replicates, resulting in 72 flasks (2 strains#3 initial densities#3 growth phases#4 lines). For the control treatment (isoosmotic transfer), populations were started at the acclimation salinity 2.4 M, with only one line per strain (2#3#3 p 18 flasks). The 90 flasks were treated in two temporal blocs, 1 day apart.
Experiment 4: Nutrient Limitation. It has been hypothesized that the population rebound observed following a population decline could result from a beneficial effect of nutrients released by the dying cells (Orellana et al. 2013), which may compensate for a lack of nutrients in the medium. To test this hypothesis, we compared a treatment with nutrient limitation (concentration of F/2 nutritive solution divided by four relative to our standard growth conditions) with a standard nutrient treatment, using five replicates per treatment. Population measures on days 0 and 1 were made with the spectrometer (fluorescence and optical density), while the cytometer and spectrometer were used for measures on the following days ( fig. S6).
Experiment 5: Cultures on Population Filtrates. To further investigate whether the postdecline rebound is related to substances that dying cells might release into the culture medium (nutrients or informative molecules about environmental conditions), we used population filtrates as culture medium. Indeed, it has been shown in another microalgal species of the same order (Chlamydomonas reinhardtii) that the supernatants of cells that have experienced PCD is more favorable to growth than the filtrates of cells that have undergone other types of death  and that this effect depends on the recipient species (Durand et al. 2014). The latter is important, as it determines to what extent PCD may be favored by kin selection (Durand et al. 2014), through so-called private goods that are preferentially used by relatives rather than public goods that may also be used by cheaters (Estrela et al. 2016(Estrela et al. , 2019. We thus tested for any strainspecific potential benefits of cell death in terms of demographic rebound in D. salina, more specifically by contrasting a declining and nondeclining strain. To prepare the filtrates, cultures of strains A and C (two and three lines per strain, respectively) were grown in a large volume (70 mL) for 4 days at intermediate salinity ([NaCl] p 2:4 M) and then transferred to high salinity ([NaCl] p 4 M) at an initial cell density of 100,000 cells/ mL, in V p 110 mL. On the next day, when the decline was expected to be maximal for strain A (according to fig. 1), we verified that strain A actually showed a decline (∼50% in 24 h) while strain C did not. We centrifuged and filtered these large volumes so as to get our two filtrates as culture media (filtrate A and C, respectively, for strain A and C). We then inoculated three replicate lines per strain in three different culture media (filtrate A, filtrate C, and standard artificial saline water as a control), mixed at a 1∶1 ratio with our artificial saline water to insure the presence of nutritive solution. We applied three salinity transfers-isoosmotic at 2.4 and 4 M and hyperosmotic from 2.4 to 4 M of NaCl-for a total of 54 flasks: 3 culture media#3 salinity transfers#2 strains#3 replicates (minus three flasks because one acclimated population of strain A did not reach a sufficient density; fig. S7).
Experiment 6: Light Intensity. Different studies have underlined the potential role of chloroplast activities in the PCD phenomenon (Murik et al. 2014;Ambastha et al. 2015;Bidle 2016), and recent gene expression analysis on our focal strains revealed that a strain-specific gene expression response to salinity changes mostly involved genes related to chloroplast structures and activities ). In addition, light represents a resource that is not modified by putative elements released in their medium by dying cells (as long as transparency is not affected). We thus tested whether light intensity influences the decline and rebound induced by an increase in salinity through an effect of photosynthetic activity on the physiological state of cells. To do so, we repeated experiment 1 but at a light intensity of 200 instead of 100 mmol/m 2 /s during both the acclimation phase and the assay phase ( fig. S8). These two light levels are commonly considered in the literature as not stressful (Berges and Falkowski 1998;Sui and Harvey 2021), but 100 mmol/m 2 /s is slightly limiting compared with 200.

Fitness of Cell Death 829
Experiment 7: Successive Osmotic Shocks. The pattern of population decline and rebound that we observed in response to osmotic stress may be explained by the elimination of cells that were initially damaged or of low condition (e.g., old); a rapid population rebound would then be the result of faster multiplication of cells that were initially in a better physiological state. This hypothesis implies a process akin to evolutionary rescue (Gomulkiewicz and Holt 1995) but where selection acts on a nonheritable phenotypic variation (as we used nearly isogenic lines). We would thus expect that the decline would be drastically reduced after a second osmotic shock because most cells in bad condition would already have been eliminated.
To test this hypothesis, we assessed the population decline for populations that were subjected to two successive hyperosmotic shocks separated by a return to intermediate salinity, which we compared with populations subjected to a single hyperosmotic shock. We acclimated populations of strains A and C at intermediate salinity ([NaCl] S9). In the control condition, cultures were subjected to isoosmotic transfer first (constant salinity at 2.4 M) and exposed to hyperosmotic shock only at the second transfer. We used five different lines per strain as replicates for each treatment (except for four replicates for strain C in the successive shock treatment because of one broken flask).

Statistical Analysis
To investigate the effects of our treatments on population dynamics, we performed generalized linear models (GLMs) using cell count data as the response variable (except for the nutrient limitation experiment, based on whole-well fluorescence) with a log-link function and a negative binomial error structure, as this has proved to be more accurate than Poisson in our setting . First, to test for population decline following the transfer, we applied such a GLM to population size on day 1, using the logarithm of expected initial population density (based on population size at the end of the acclimation step) as offset. The linear predictor in this model thus estimated the logarithm of the relative change in population density from day 0 to 1. For the light experiment, we compared the experiment with brighter light against all other experiments under hyperosmotic stress (except under nutrient limitation, where cytometer counts were not available for all days, and with the PCD inhibitor). For the nutrient limitation experiment, we applied a LM to the logarithm of population density estimated by the whole-well fluorescence values (day 1) because fluorescence is lognormally distributed . The offset was the logarithm of the initial population density estimated by fluorescence of strain C (1 h after the transfer), as this strain does not display population decline ( fig. 1).
We then applied a second GLM for population growth in days corresponding to a phase of exponential growth (see table 1 for details), visually detected as a linear trend on population dynamics on a log scale (this phase is denoted as gray backgrounds in figs. 1, S4-S9). In this model, no offset was used for initial population size, but day was included as a continuous explanatory variable. The day effect in this model thus estimated the rate of exponential growth per day (linear trend on the log scale), and any interaction of day with other factors (treated as covariates in the regression) was an effect of these factors on maximum exponential growth. The fixed effects were experiment dependent (table 1). For population growth under nutrient limitation, we applied the same approach but using a LM on the logarithm of population densities based on whole-well fluorescence. For the light experiment, we applied the GLM on the subset of days corresponding to the exponential phase in each experiment (determined as specified above).
Our results suggested that across experiments, a larger initial decline was associated with faster maximal growth under hyperosmotic stress. To formally test for this effect, we estimated the decline rate (on the log scale) predicted by the GLM as where e 1 is the linear predictor of the GLM (i.e., the estimator of population density on the log scale, no longer corrected by the offset) on day 1 after the shock and N 0,exp is the initial population density expected from the acclimation culture. A positive value of D means that the population has declined after the transfer. The standard error for the decline rate D is directly that of the linear predictor on the log scale. In the results, we also express the decline as proportional reduction in population size: population reduction (%) = (1 2 e 2D )#100: We also estimated the maximal daily growth rate r in the exponential phase as where e x 1 and e x 2 are the linear predictors of the GLM (log population size) on days x 1 and x 2 after the shock, corresponding to the phase of exponential growth (gray backgrounds in figs. 1, S4-S9). We calculated the standard error of each estimated growth rate as We then tested the significance of the relationship between the estimated decline rate and the rebound growth rate using a bootstrap method. For each D and r pair (corresponding to a specific condition in a given experiment), we drew random values from a normal distribution with the mean given by the estimate and the standard deviation given by the standard error of the estimate, to account for the uncertainty in each data point. A LM was then fitted for the relationship between D and r across conditions in each of these simulations, and the corresponding regression slope and intercept were drawn from normal distributions centered on their estimates and with the standard deviation given by their standard error, to account for uncertainty in each regression. We repeated this process 10,000 times, from which the proportion of positive slopes was used as a P value. Finally, the mean slope b and the intercept a over these 10,000 simulations were used to compute the predicted population density on day t for a nondeclining population as or for a declining population starting to rebound on day c (corresponding to x 1 in eqq. [3], [4]) as From this, we could then estimate the day on which a declining and a nondeclining population would reach equal population density, assuming exponential rebound: All statistical analyses were performed in Rstudio (R ver. 3.6) using the MASS version 7.3.53 package (Venables and Ripley 2002).

Results
We wished to understand the proximal determinants and fitness consequences of PCD in a unicellular organism. Our aim was to find whether, and under which conditions, such an apparently detrimental phenotype may be associated with genotype-dependent benefits for population growth and thus potentially be favored by natural selection.

Decline in Response to Osmotic Shock Is Strain Specific
As salinity is a major component of Dunaliella salina's niche, we first investigated its influence on demographic dynamics in this species. We observed that an isoosmotic transfer into fresh medium without change in salinity (either at intermediate [2.4 M] or high [4 M] NaCl concentration) did not affect demography, since population densities measured 1 h after transfer matched the expected initial densities, accounting for the dilution into fresh medium ( fig. 1; table S2). In contrast, transfer from intermediate to high salinity induced a sharp demographic decline in strain A in the hours following the transfer, with the population size reduced by 69% in 1 h (table S2) and reaching a 77% decline at its minimum, 24 h after transfer. This decline was not observed for strain C ( fig. 1; table S2), leading to a significant effect of strain identity on population density on day 1 and a significant interaction between strain and type of salinity transfer (table 2). This indicates that the rate of rapid cell death varies among closely related strains of D. salina (as shown by Leung et al. 2022) and does so in a manner that depends on the type of salinity transfer, being triggered in strain A by a salinity increase rather than by high salinity per se. Note that the osmotic shock also led to an increase in the number of apparently dead cells (detected by their lower red fluorescence; Rescan et al. 2020), but it was orders of magnitude lower than the observed decline in live cells and of similar magnitude between both strains ( fig. S1). Note: We used a GLM with a log-link function and a negative binomial error structure for population density on day 1, using the expected initial (log) density as offset. Tested effects are strain, transfer, and the strain # transfer interaction. R 2 was calculated as a deviance ratio (Nakagawa and Schielzeth 2013). *** p ! .001.
Interestingly, the population decline of strain A was associated with a higher maximum growth rate in the following days, compared with its growth rate without initial decline in isoosmotic transfers ( fig. 1, left and right panels; table S3; P ! 3e28 and P ! 3e215, respectively, for comparison of growth rate in hyperosmotic transfer to isoosmotic transfers at 2.4 and 4 M). In contrast, strain C, which did not decline following hyperosmotic shock, had a weaker maximum growth rate compared with strain A under this salinity condition (table S3; P ! 2e216). As a result, even though strain A had a much lower population density than strain C after its initial decline under hyperosmotic stress, it eventually compensated for its lag in population size, reaching a density similar to strain C after 6 days.

The Decline Is Reduced by a PCD Inhibitor
To assess whether the rapid population decline may result from PCD, as commonly documented in this species and genus (Berges and Falkowski 1998;Segovia et al. 2003;Jiménez et al. 2009;Orellana et al. 2013), we performed an experiment with a caspase-3-like PCD inhibitor at four different concentrations. The PCD inhibitor had a clear quantitative effect on the intensity of the decline, with a higher inhibitor concentration leading to less severe decline for strain A under hyperosmotic shock ( fig. 2). Accordingly, the GLM treating inhibitor concentration as a continuous variable had a slighter better Akaike information criterion (AIC) than the one treating it as a categorical factor (AIC p 220:78 vs. 223.33). We found a signif-icantly positive relationship (table S4, pt. A; P ! 1e23) between (log) population size on day 1 (postdecline) and inhibitor concentration, with a slope of 0.150 (SD p 0:045). The model thus predicts that the population size 1 day after hyperosmotic shock is reduced by 70.8% without inhibitor (consistent with our previous experiment) but is proportionally increased by~16% with each additional micromolar concentration of inhibitor. In a preliminary experiment (with only two replicates), we even found that strain A already started to decline 15 min after the osmotic shock (43% reduction in population size), while it did not in the three treatments with inhibitor ( fig. S3). In contrast, we found no effect of inhibitor concentration on demographic dynamics in other conditions (strain A under isoosmotic transfer and strain C under any salinity transfer; fig. S4). There was no significant effect of inhibitor concentration on maximum growth rate between days 1 and 6 (table S4, pt. B; P p :107), despite the observed differences in decline intensity. These results therefore indicate that the population decline observed for strain A under hyperosmotic shock is at least partly explained by PCD.

Initial Decline Is Greater under Weaker Competition
To understand what benefit such salinity-induced PCD may confer, we investigated hypotheses about reduced competition. A possible explanation for the higher growth rate of strain A following its initial decline under hyperosmotic stress may be relaxed density dependence (competition release), whereby the resources unused by dead cells become available to those cells that survived. The influence of initial decline on later growth would thus be mediated by densitydependent processes. More specifically, we may expect the postdecline population to have a similar growth rate to a population that did not decline but started at a similarly low density. In addition, the rate of death (and thus the population decline) could itself be modified by the initial population density if PCD is a plastic trait that responds to cues of the density of conspecifics or amount of resources. However, these influences of competition-mediated resource limitation on population growth are likely to be modulated by the metabolic state of individual cells (Delong and Hanson 2009), which in turn not only depends on the immediate density of competitors but also reflects the previous growth conditions during the acclimation phase . This occurs because the influence of competition and resource limitation on the physiological states of individual cells (and hence population growth) depends on the rates of nutrient uptake and metabolic pathways involved and may involve some delays (Droop 1973). We thus tested for an influence of not only the initial density of the inoculum but also the growth phase of the populations in the acclimation phase, as an indicator of the physiological state of the population. We report results about rates of initial decline and maximal growth rate in figure 3, while the full population dynamics appear in figure S5. Overall, strain A showed greater decline in conditions where population growth is expected to be more rapid, that is, when cultures were transferred during their expo-nential or early stationary phase and/or started at low initial density ( fig. 3A). When the populations experienced osmotic shock in the midexponential or early stationary phases (after 4 or 13 days of acclimation, respectively), the initial density did not impact the intensity of the observed decline in strain A ( fig. 3A; table S5, pt. A). In contrast, populations transferred in their late stationary phase (41 days) showed no decline when starting at a high density (30,000 cells/mL) but a decline comparable to that in earlier phases when starting at a low density (5,000 cells/mL; fig. 3A). Overall, across initial densities, populations transferred in their late stationary phase declined less than when transferred in earlier phases (table S5, pt. B). Contrary to strain A, strain C never declined under any demographic conditions ( fig. 3A).
Following the decline, strain A displayed a higher maximum growth rate than strain C for the midexponential and early stationary phase populations ( fig. 3B). Furthermore, the growth phase ("age" of the cultures before the transfer) influenced the maximal growth rate for strain A, but with an effect that also depended on density. The maximum growth rate was highest for populations transferred in their early stationary phase, when starting at N 0 p 30,000 cells/mL (table S6; P p :037 for the comparison with exponential phase starting at N 0 p 5,000), while when starting at N 0 p 5,000 cells/mL they did not show a significantly different rebound than those transferred in their exponential phase (table S6; P p :839). In particular, the rebound growth rate of strain A inoculated at N 0 p 30,000 cells/mL was higher than the growth rate  Figure 3: Impact of competitive conditions on initial decline and subsequent maximal growth following hyperosmotic shock. Estimates and standard errors (across four replicates) from the negative binomial generalized linear models for decline (days 0 and 1 after the osmotic shock; A) and maximal growth rate (corresponding to the shaded area in fig. S5; B) are shown for different conditions of initial density (gray shading) and population growth phase (x-axis), for strains A (circles) and C (triangles). Population growth phase corresponds to the midexponential phase (4 days of acclimation; "Mid-exp"), the early stationary phase (13 days; "Early-stat"), and the late stationary phase (41 days; "Late-stat"). In A, larger positive values indicate faster population decline, while negative values correspond to growing populations. of the nondeclining strain C that started at a density similar to A postdecline (N 0 p 5,000 cells/mL; see figs. S5, fig. 3). This confirmed that density-dependent effects on population growth are largely mediated by persistent responses to past conditions of competition and resource limitation (as shown for salinity in Rescan et al. 2020). The weak and inconsistent effect of initial density on the rate of population rebound further suggests that this rebound is unlikely to be due to relaxed density dependence from many cells having died. In fact, our focus on the rate of exponential growth limits the possibility for density to have a large effect, since exponential growth is density independent by definition.

Decline Intensity Depends on Resources or Cues from the Culture Medium
To more directly assess the effect of resource competition on the observed pattern of decline and rebound, we manipulated nutrient abundance. When fewer nutrients were available, strain A declined significantly less than populations in the control medium ( fig. 4A; table S7; P p 1e24).
(Note that the estimated decline and maximal growth rate were predicted on the basis of spectrometer measures [ fig. S6A] instead of flow cytometry counts for this experiment.) The estimated maximal growth rate in the exponential phase (from days 1 to 3 in this experiment) was clearly lower for strain A when nutrients were missing ( fig. 4B; table S8; P p :003). As in the previous experiment, under standard nutrient conditions strain A displayed a higher maximal growth rate than strain C (table S8; P p 3e27).
Since the medium content affects the intensity of decline for strain A, we tested the hypothesis that the dying cells causing population decline release nutrients that may be used by the remaining cells to grow faster or informative cues about the population demographic status. We also investigated whether the ability to use this resource or information was genotype specific. To do so, we grew populations of each strain in medium filtered from previous cultures that had undergone a salinity increase, either from strain A (filtrate A) or strain C (filtrate C; fig. 4C, 4D). Populations of strain A growing on filtrate A had a slightly but significantly lesser decline than control populations A growing in standard medium (table S9; population reduction of 73% vs. 77%; P p :026). Those growing on filtrate C instead showed a greater decline than the control (table S9; population reduction of 81% vs. 77%; P p :021). In contrast, we did not detect any effect of culture medium on the rebound growth rate for strain A (control: r p 0:64; strain A in filtrate A: r p 0:59, P p :129; strain A in filtrate C: r p 0:68, P p :153; table S10). We also did not find any effect on the population growth of strain C, either on the first day (P p :058 and P p :219, respectively, in filtrate A and filtrate C compared with control culture; see table S11, pt. A) or after day 1 (under all conditions r ∈ [0:26; 0:28]; P p :092 and P p :783; see table S11, pt. B).

Decline and Rebound Depend on Resources
Outside the Culture Medium We also studied the effect of light intensity, an important resource for photosynthetic organisms, which unlike the medium should not be directly affected by the dying cells.
We compared the control conditions from the experiments described above (plus one described below with successive osmotic shocks), where populations faced a salinity increase at a light intensity of 100 mmol/m 2 /s, to populations undergoing a salinity increase at a higher light intensity, 200 mmol/m 2 /s ( fig. 4E, 4F). The decline rate of strain A was significantly greater under brighter light (population reduced by 82% at 200 mmol/m 2 /s vs. 57% at 100 mmol/m 2 /s; P ! :001; table S12, pt. A). Similarly, strain A grew significantly faster in the exponential phase (shaded area on demographic dynamics) under brighter light (r p 0:87 vs. 0.43 for 200 vs. 100 mmol/ m 2 /s; P p :007; table S12, pt. B). This confirmed that resource conditions more favorable to growth led to a faster decline followed by a greater growth rate.

A Consequence of Heterogeneity in Cell Condition?
Initial heterogeneity in the metabolic state or condition of cells could explain the decline-rebound pattern observed in response to hyperosmotic shock ( fig. 1). Indeed, older or more damaged cells within the population may be rapidly eliminated when facing hyperosmotic shock, while those that were originally in good condition may survive and reproduce faster.
To investigate this hypothesis, we applied successive salinity increases to the two strains, separated by a short stay of 17 days at intermediate salinity (corresponding to~17 generations). Our prediction under the hypothesis of an effect of cell condition was that a second osmotic shock imposed shortly after the first one should lead to a more moderate decline (if any) because damaged cells have already been eliminated. Contrary to this expectation, we found that the decline of strain A was greater after a second salinity increase than after a single salinity increase (population reduction of 75% vs. 62%; fig. 4G; table S13, pt. A; P ! 2e216), while the number of salinity increases did not impact the initial growth rate of strain C (no decline; fig. 4G). For strains A and C, the maximum growth rate did not differ between the two treatments ( fig. 4H; table S13, pt. B; P p :345).

Sharper Decline Is Associated with Faster
Rebound across All Treatments The results of these experiments suggest that faster initial decline was generally followed by a higher maximum growth rate (e.g., fig. 4). To formally test for this pattern, we gathered all estimated pairs of decline intensity D (log reduction in population size between days 0 and 1, with larger positive values denoting sharper decline) and rebound rate r (daily rate of maximum exponential growth) for strain A facing a salinity increase across experiments ( fig. 5). We found a highly significant positive relationship (P p 8e24) between D and r. Decline intensity D ranged from 0.22 (19.8% reduction in population size for late stationary phase cultures transferred at high density) to 1.6 (79.8% reduction in population size under bright light). The slope of the regression of r against D is 0.268 (SD p 0:074). This means that relative to a putative population that would not decline, a population that initially declines by D but starts rebounding on day 1 would compensate for its initial demographic deficit by day 5 if D p 1:47 (77.0% initial reduction in population Fitness of Cell Death 835 size) and by day 10 if D p 0:22 (19.7% initial reduction) under exponential growth. In other words, a population that initially declines faster is predicted to match the size of a nondeclining population more rapidly during exponential growth and/or may reach its carrying capacity first, thus potentially exerting competitive exclusion on the nondeclining strain (Kot 2001).

Discussion
We have investigated the factors influencing rapid PCD in two closely related strains of the halotolerant microalga Dunaliella salina and its demographic implications beyond the initial decline it produces. Our aim was to decipher how variation among strains in these demographic responses may impact selection on a trait that seems at first like an archetype of maladaptive plasticity.

Implications of the Decline-Rebound Pattern
We established that the demographic response to high salinity characterized by a rapid decline followed by rebound (shown in Leung et al. 2022) is induced by a salinity increase rather than high salinity per se ( fig. 1). This fast response (population size reduced by 69% in 1 h [  Zuppini et al. 2010). The decline-rebound pattern that we observed further matched the previous finding in D. salina that darkness (rather than salinity increase, as here) causes the death of ∼65% of cells but later enhances population growth (Orellana et al. 2013). Interestingly, this phenotypic response varied drastically between two genetically close Dunaliella strains (Emami et al. 2015;Leung et al. 2022), one of which never displayed the characteristic decline-rebound pattern after a salinity increase, regardless of other experimental factors. While the specific genetic differences behind these contrasted responses to salinity are unknown, it is noteworthy that gene expression and DNA methylation responses to salinity vary between these strains . Orellana et al. (2013) Assunção et al. 2013;Emami et al. 2015) that did decline, suggesting that there is genetic variation for this trait within D. salina, unlike in other species where such demographic responses have been investigated (Nedelcu et al. 2011;Durand 2020). The putative benefits of this decline also appeared to be strain specific, as the growth of strain C was not influenced by filtrate of declining strain A ( fig. 4D; table S11), mirroring the lack of significant effect of induced-death filtrate on two  Figure 5: Faster decline is associated with faster maximal growth across experiments. All estimates and standard errors for decline and rebound rates are shown for strain A following a salinity increase. Gray lines represent 10,000 linear regressions of growth rate against decline rate, based on simulated samples where each pair of values (for decline and rebound) was randomly drawn from a normal distribution with the mean set to the predicted value and the standard deviation set to the standard error of estimators for each condition. Eight of the 10,000 replicates had a negative slope (P p 8e24). Estimates from the nutrient experiment (blue empty circles) are not included in the linear regression, as they were based on whole-well fluorescence instead of cytometer counts. strains in Chlamydomonas reinhardtii . The markedly distinct demographic responses of these closely related strains imply that natural selection can act on rapid, environmentally induced cell death, favoring it in conditions where it is associated with higher accrued fitness during the demographic rebound. Such a positive correlation between initial decline and later demographic rebound following osmotic shock was one of our most striking results ( fig. 5). This allowed the declining genotype to compensate for its initial lag, eventually reaching a density similar to-or even higher thanthe nondeclining genotype (figs. 1, S6, S7) before attaining the stationary phase where population growth stops (commonly described as carrying capacity). Assuming that the dynamics in monocultures predict what would happen in cocultures (i.e., neglecting genotype-by-genotype demographic interactions causing frequency-dependent selection; Chevin 2011), these results suggest that the lethal and at first glance disadvantageous phenotype of rapid cell death is likely to be favored in an environment with fluctuating salinity (including hyperosmotic shock).

Causes of Selection
Which mechanism could explain the positive correlation we found between decline and rebound rates for strain A? Our main hypothesis was that the dying cells release material that is beneficial to the remaining cells, thus representing a form of cooperation Orellana et al. 2013). The importance of environmental dynamics for social evolution, especially in microbes, was recently emphasized (Estrela et al. 2016(Estrela et al. , 2019 and could involve different mechanisms (Kojic and Milisavljevic 2020). For example, dying cells might provide resources by liberating nutrients (public goods) in the culture medium. This organic material may include intracellular glycerol (Zidan et al. 1987), which D. salina stores massively in its cytoplasm as an osmoprotectant at high salinity and could be used directly by the surviving cells (D. salina being capable of heterotrophy; Chavoshi and Shariati 2019) or metabolized by a halophilic archaea, as proposed by Orellana et al. (2013). However, we did not find a faster rebound for strain A growing on filtrate A, (and no "cheating" effect on the growth of strain C) and hence found no support for the altruistic nutrient-release hypothesis. This effect either does not exist in this system or exists but could not be detected because death-induced molecules were too rare in the diluted filtrate or because naturally co-occurring archaea that remineralize glycerol were absent. Another altruistic mechanism could be the detoxification of the environment by dying cells (Estrela et al. 2019), but we are not aware of any mechanism through which D. salina might reduce salinity, which is the major stressful condition applied here. Signaling molecules could also be released, providing information about the environmental conditions, such as quorum-sensing molecules (Christensen et al. 1995;Durand 2020). This hypothesis cannot be discarded since strain A growing on filtrate A showed a more moderate decline ( fig. 4C), suggesting that live cells may perceive that a demographic decline already occurred, therefore modulating the proportion of cells that will die. To sum up, our results did highlight that the declining strain is sensitive to the medium content, but we do not have clear evidence that this is an altruistic trait.
Another less adaptive hypothesis explaining the declinerebound pattern could be a trade-off between reproduction and the rate of plastic change in a trait involved in salinity tolerance. Genotypes that invest more in salinity tolerance by having more rapid plastic change in dedicated traits should have higher survival probability when facing osmotic stress, although possibly at the expense of a lower growth rate if salinity tolerance mechanisms are costly for reproduction. Such a trade-off could involve the metabolism of glycerol, which is known to divert resources produced by photosynthesis or stored in starch (Ben-Amotz and Avron 1973) and is thus likely to impact population growth (Jones and Galloway 1979). A related concept in ecology is the trade-off between resistance and recovery with respect to resilience, whereby a strain that is more susceptible to stress (less resistant) is able to recover faster from it (Hodgson et al. 2015). However, this hypothesis would not explain why for a given genotype (strain A), conditions that cause a sharper decline also lead to faster growth during rebound.
A third hypothesis explaining the decline-rebound demography could be cell heterogeneity in the population, whereby damaged cells die when facing osmotic shock while cells in good shape survive and have a higher growth rate. However, we found that a second osmotic shock in fact induces more (rather than less) cell death than the first ( fig. 4G), so this explanation is unlikely to hold unless a very high proportion of damaged cells is produced during the relatively short stay at intermediate salinity, which seems unlikely. These proposed hypotheses are not mutually exclusive and may partially explain the observed results depending on the ecological context.

Passive versus Active Plasticity and PCD
A strong-albeit indirect-argument in favor of adaptive plasticity is when the plastic response can be established to involve an active mechanism (Pigliucci 1996). For the rapid decline we observed, a natural candidate of active mechanism would be PCD, more commonly named cell suicide in unicellular organisms (Ameisen 2002;Durand 2020). PCD has been reported in unicellular chlorophytes under a range of environmental stresses (Bidle and Falkowski 2004;Zuppini et al. 2010;Durand et al. 2014). This includes Dunaliella species, such as D. tertiolecta (Berges and Falkowski 1998;Segovia et al. 2003), D. viridis (Jiménez et al. 2009), and our focal species, D. salina (Orellana et al. 2013). Signatures of active cell suicide via PCD include the early externalization of phosphatidylserine in the outer membrane, caspase-like activities in mitochondria, and DNA fragmentation (Barreto Filho et al. 2022a). Our assays with caspase-like inhibition showed a clear quantitative reduction in the population decline with increasing inhibitor concentration ( fig. 2). Although caspaselike activity is not specific to PCD (Abraham and Shaham 2004;Barreto Filho et al. 2022a) and PCD can occur without caspase activity (Leist and Jäättelä 2001;Abraham and Shaham 2004), our finding that a caspase inhibitor routinely used to detect PCD distinctly reduces population decline is a clear indicator that the cell death causing this decline is not an entirely passive process. Furthermore, the tested inhibitor was also used in the related species D. tertiolecta, where PCD has been firmly established using several assays and morphological criteria (Segovia et al. 2003).
In addition to this direct test for PCD, one consistent line of evidence for an active process in our experiments was that the decline was more pronounced when the growing conditions were closer to optimal: brighter light ( fig. 4E; table S12, pt. A), cells in better condition and/or with less interindividual competition ( fig. 3A; table S5, pt. A), and nonlimiting nutrients ( fig. 4A; table S7). All of these effects point toward this death being an active and energy-demanding process (Sathe et al. 2019).
In conclusion, our results indicate that the strain that experiences more PCD in response to hyperosmotic stress eventually reaches a higher population size under at least some conditions (figs. S6, S7). Future work should investigate whether these demographic dynamics in monoculture accurately predict what happens in competition between these strains, to better understand whether and how natural selection can favor a lethal phenotype as a form of adaptive plasticity in a stressful environment. The fact that salinity-induced death and its consequences for longer-term fitness varied drastically between closely related strains isolated in the same location suggests that PCD is a very evolvable trait, even on short evolutionary timescales.