Temporal fitness fluctuations in experimental Arabidopsis thaliana populations

Understanding the genetics of lifetime fitness is crucial to understand a species’ ecological preferences and ultimately predict its ability to cope with novel environmental conditions. Yet, there is a dearth of information regarding the impact of the ecological variance experienced by natural populations on expressed phenotypic and fitness differences. Here, we follow the natural dynamics of experimental A. thaliana populations over 5 successive plantings whose timing was determined by the natural progression of the plant’s life cycle and disentangle the environmental and genetic factors that drive plant ecological performance at a given locality. We show that, at the temperate latitude where the experiment was conducted, a given genotype can experience winter-, spring- or summer-annual life cycles across successive seasons. Lifetime fitness across these seasons varied strongly, with a fall planting yielding 36-fold higher fitness compared to a spring planting. In addition, the actual life-stage at which plant overwinter oscillated across years, depending on the timing of the end of the summer season. We observed a rare but severe fitness differential coinciding with inadequate early flowering in one of the five planting. Substrate variation played a comparatively minor role, but also contributed to modulate the magnitude of fitness differentials between genotypes. Finally, reciprocal introgressions on chromosome 4 demonstrated that the fitness effect of a specific chromosomal region is strongly contingent on micro-geographic and seasonal fluctuations. Our study contributes to emphasize the extent to which the fitness impact of phenotypic traits and the genes that encode them in the genome can fluctuate. Experiments aiming at dissecting the molecular basis of local adaptation must apprehend the complexity introduced by temporal fluctuations because they massively affect the expression of phenotype and fitness differences.


Introduction
There is overwhelming evidence that local selection pressures have shaped the genetic composition and ecological performance of natural populations [1][2][3]. But in the context of rapid climate change, greater attention must be devoted to the detailed ecological events to which a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 species have had to evolve to best adjust to their ecological niche [4][5][6]. The phenotypes that plants manifest in nature, i.e. their timing of germination, their growth rate, and their timing of reproduction, are plastically regulated by environmental conditions that vary both temporally and geographically [7]. Much is known about the molecular mechanisms of many plastic plant traits [8][9][10]. This plasticity may buffer the effect of environmental fluctuations on fitness and thereby decrease year-to-year fluctuations in selective pressures [5]. Yet, we know little about how environmental fluctuations influence the amount of phenotypic variation that is actually expressed in natural conditions, nor do we know how it impacts population levels of fitness variance, an elemental requirement for adaptive evolution [11].
Here we document seasonal variation in expressed plant life cycles and measurements of fitness in Arabidopsis thaliana, a plant species that stands as a model for the dissection of the molecular underpinnings of local adaptation [12]. In this species, analyses of genetic variation along altitudinal transects often revealed patterns of local adaptation [13][14][15]. Patterns of nucleotide variation uncovered the population genetics signatures of local adaptation on diverse traits [16][17][18][19][20]. Reciprocal transplants of a Swedish and an Italian population suggested phenotypic trade-offs evolved in Scandinavian populations [21,22] and common garden experiments in four different climatic zones highlighted the hundreds of nucleotide variants that associate with local measures of fitness [23]. While these studies reveal the pervasive impact of local selective pressures, we asked ourselves what really these selective pressures are in nature. Because both the environmental conditions and the genetics of specific individuals can vary, the prediction of which plastic traits will be displayed by any one individual and whether it will impact fitness is far from straightforward [24][25][26]. This question is especially relevant in monocarpic species that target their seasonal growth window with an array of lifehistory traits (e.g. germination, flowering) that are themselves greatly plastic [27]. In most field experiments, the seasonal window was imposed by the experimenter [21,27,28]. Yet, in climatic regions were summers are sufficiently mild and wet to allow the completion of multiple generations on some years, it is unclear whether seasonal fluctuations can cause fluctuations in the seasonal window a genotype might choose. This in turn can dramatically alter the selective pressures exerted and thus the type of adaptive traits required. Does seasonal variance have cascading effects on the choice of successive seasonal windows? Does it impact the magnitude of phenotypic and fitness differences manifested by diverse genotypes? Does life history plasticity tend to buffer differences? How does the effect of temporal variance compare with the magnitude of microgeographic or genetic variance?
To answer these questions, we designed an experiment that dissects how environmental factors affect fitness profiles in natural populations at temperate latitude. We report the monitoring of lifetime fitness in experimental populations over 5 natural successions of plant cycles in conditions differing in soil substrate and water availability. We monitored 4 genotypes, including two genomic backgrounds, one of local and one of Southern origin (Col-0 and C24), and the reciprocal chromosome 4 (Chr 4)-introgressions of several polymorphisms previously reported to associate with differential fitness in various locations [23]. We expected that the Southern genotype would perform comparatively better in summer conditions and that these differences would be modulated by the Chr4-introgression and the substrate. We show that at the temperate latitude of our experiment, genotypes can complete multiple cycles per year, alternating summer, winter and spring cycles. Our experiment documents how overall fitness results from complex interactions between seasonal oscillations, soil composition, genomic background and genotypes. As a consequence, both the choice of successive growth seasons and the expressed phenotypic and fitness differences between genotypes fluctuate markedly at a given locality under natural conditions.

Genetic plant material
Arabidopsis thaliana is an annual weed growing across a broad geographic range [29]. The fruits it produces are called siliques, a character of the Brassicaceae family. In this study, we monitored field performance of two wild type genotypes, Col-0 and C24. Genomic information for these two lines suggest that the Col-0 genotype is typical of Western European populations whereas C24 is more related to Southern European genotypes [30]. We further included in our experiment two reciprocal near-isogenic lines (NILs). The two sets of NILs contained either an introgression of chromosome 4, between 11.8 to 13.8Mb, of C24 accession in Col-0 genomic background or a genomic fragment between 5.1 and 15.3Mb of Col-0 accession in C24 background [31]. Seeds were provided by T. Altmann (IPK, Gatersleben, Germany). Our experiment thus includes genomic background and chromosome-4 introgression as two separate genetic factors. In total, Col-0 and C24 differed for 284 of the 867 SNPs reported to associate with survival and seed production in at least one of 4 locations throughout the range of the species [23]. Of those, 25 (8%) and 38 (13%) SNPs mapped to the introgressed fragments in Col-0 and C24 backgrounds, respectively. They associated with enhanced survival in Finland and silique number in Spain for the Col-0 genotype [23].

Field plantings
In this study, we monitored germination, flowering time and fitness of the 2 wild type genotypes, Col-0 and C24 and their reciprocal Chr4 introgressions in the field over five consecutive generations that followed a quasi-natural population establishment dynamics. Fifty field-collected seeds of the two NILs and their wild-type parents were placed in 12 replicate pots (e.g. the experimental unit of this study), for each of the two soil compositions and 4 genotypes. Seeds were taken from a pool of 500 seeds collected from 5 plants per pot in the previous planting. There were two exceptions to this scheme: planting 1 was started with 200 seeds and conducted in only four replicates (32 pots), and planting 5 was performed with only Col-0 and Col NIL due to limitation in personnel (48 pots). The timing of the different plantings is given in Fig 1. The field was located in the middle of a flat grass meadow and fenced with iron nets to prevent grazing by rabbits. No vegetation protected it from sun or wind. We filled large flat pots (30cmx40cmx12cm) with two soil types: Arabidopsis turf soil (Einheitserde Typ Minitray, plus 1kg Osmocote Start per m 3 , Einheitserde Werkverband, Sinntal-Jossa) or a soil mixture enriched in sand (a 1:3 mixture of the turf described above and sand). Pots thus dried faster in the sand mixture compared to turf. Pots were randomized and placed on a large plastic cover that prevented germination from the soil below (S1 Fig). Each replicate population was prepared in a pot by sowing seeds mixed with 5 g silicon dioxide (Sigma-Aldrich, S5631). Pots had been watered to saturation to facilitate seed adhesion to the soil. After sowing, seeds were left exposed to natural conditions in the field and no watering other than natural rain was done. The first planting was started with seeds fully ripened in the greenhouse. All other plantings were started with seeds collected in the field in the previous planting, with the exception of C24 seeds in planting 4, which were taken from planting 2 (the C24 genotype did not survive winter in planting 3). At complete senescence, plants were collected and dried in paper bags, seeds were harvested so that the following planting could be initiated within 8 days after harvest.
For all plantings, seedlings other than A. thaliana were immediately removed when they appeared. Our experiment included empty pots to evaluate whether migration occurred between pots. After detecting migrants in control pots, we verified the genotype of each seedling. Individuals with the C24 background display glabrous leaves whereas those of individuals with the Col-0 background are hairy. The genomic background could thus be verified visually. A leaf was collected from each single plant to extract DNA and genotype the individual at Chr4 by PCR with primers mi05 (5'-CTCTTGCTGCGTAGGGTTCCC-3') and mi06 (5'-AT CTCCCCACTCCCCAATTTT-3'), which generate a length polymorphism at AT4G24415 within the chromosome 4 introgression [32]. We detected 26 heterozygotes in 453 individuals in planting 4, suggesting an outcrossing rate in the field of about 5%. These individuals were removed from further analysis.
The field study was conducted on the experimental land area of the Max Planck Institute for Plant Breeding Research (Cologne, Germany). This field study did not require specific permission because it did not include transgenic plant material and did not involve endangered or protected species.

Fitness and phenotype scoring
For the first planting, early, intermediate and late germinants were observed but the exact germination date was not scored. Germination time was scored for plantings 2-5, flowering time (date of the first open flower) was scored for all five plantings. Silique number was counted for a sample of 3-8 randomly chosen plants in each population. In total, fitness was scored for 96, 401, 583, 301 and 691 plants in plantings 1 to 5, respectively (for a total of 2072 plants distributed in a total of 368 pots). In addition, in plantings 2, 3 and 4, the number of seeds was determined for three siliques in each of 3 randomly chosen plants per population. For fruit number flowering time and germination time, an average was computed for each population (e.g. each pot). The average fruit number per pot was taken as an estimate of lifetime fitness. Variation in average seed number per silique did not change the conclusions based on fruit number (not shown). When plants in a pot failed to germinate or survive until reproduction, lifetime fitness was set to zero. To account for variation in germination rates, we included plant density as a covariate in the analyses (see below). Raw phenotypic and fitness data is shown in S1 Table. Statistical analysis All statistical analyses and figures were performed in R. A generalized linear model (GLM) was used to establish the minimal model explaining phenotypic or fitness variation. For this, we followed the procedure described in [33]. A first model with average lifetime fruit production as a dependent variable and including, as explanatory variables, the main factors planting, soil type, genomic background and chromosome 4 region, as well as the 2-way, 3-way and 4-way interactions between these factors was run. To take into account variation in fruit number that may result from competition in densely populated pots, the total number of plant in each population was also included as an explanatory covariate. Non-significant interaction terms were removed one by one, from the most complex to the least complex, using the F-test implemented in the R function drop1. To control for the non-Gaussian error distribution, we used a negative binomial distribution of error (implemented in R Package MASS) and tuned its free parameter to adjust the residual deviance to the degrees of freedom of the model. This adjustment was performed at each step of the procedure, until the minimum model was determined. The resulting "minimal" model included the four main factors and the significant interaction terms. Interpreting the effect of a single experimental factor is not straightforward in the presence of interactions between factors, especially when the 4-way interaction is significant. We thus broke down the factors and conducted separate ad-hoc GLM for each genomic background or for subsets of plantings. The fold-change attributed to experimental factor was calculated by taking the exponential of the log estimates given by the model (GLM with negative binomial distribution of error used the natural log as a link function). This measure corresponds to the odds ratio associated with the factor. The proportion of variance explained by each factor or interaction term, was calculated as the ratio of the deviance calculated for a given factor against the total deviance. To compute the fitness fold-change attributed to the chromosome 4 introgression within each combination of experimental factor (planting, soil composition and genomic background), we ran a separate GLM analysis with introgression as a factor nested within a factor called "setting", which combined soil condition, planting and genomic background in 16 levels. Fitness fold-change attributed to variation in the chromosome 4 region was calculated as described above.
The product of day length (h) by mean temperature (˚C) was calculated for each day and summed for the time windows covering date of sowing to average germination (or flowering) date to compute the cumulated Photo-Thermal Units (PTU) for each planting, following Brachi et al. 2010 [19].

Effect of experimental factors on lifetime fitness
We monitored average silique (fruit) production in experimental A. thaliana populations over five naturally consecutive plantings by counting the average number of siliques produced per pot (Fig 1). Climatic differences between successive plantings are summarized in Table 1. A generalized linear model with a negative binomial distribution of error adjusted to minimize overdispersion revealed that all factors of the experiment interacted with each other to alter this measure of fitness ( Table 2).
The average number of siliques (fruits) produced in each population was significantly affected by the number of germinants (or survivors) in each replicate population, i.e plant density (p = 1.03e-11, Table 2 The C24 background, instead, did not yield high density because of a lower germination rate, especially in sand for the planting initiated in the summer (not shown). In the remaining of the analysis, plant density was included as a cofactor, so that the genetic and environmental effect reported below are independent of population density.
The largest part of variance was explained by variation across plantings (36% of the variance, p <2.2 e-16, Table 2). Indeed, both highest and lowest values for average silique number per population were observed in planting 3, where most of the experimental populations with C24 genomic background did not survive (Fig 2A). As a consequence, the second largest effect was controlled by the interaction between genomic background and planting (18%, p<2.2e-16, Table 2). To characterize the details of average fitness variation across plantings for a single genotype reaching reproduction, we focused on the Col-0 background, which completed its cycle in all 5 plantings (Table 3). The log estimate of this analysis revealed that average silique production per plant in planting 3 was 14 times higher than in planting 1, which was itself 2.6 times higher than in planting 5 ( Fig 2B). We thus estimate that, in this experiment, average fruit number per population varied by up to 2.6 Ã 14~36-fold across plantings for the sole Col-0 genomic background.
As expected soil quality also had a significant effect on silique production (p = 6.07e-06, Table 2), with fitness being lower in sand across all plantings, except in planting 3 (Fig 3A). Although a relatively small proportion of fitness variance could be attributed to the main effect of genomic background (3%, p = 5.22e-11, Table 2), this factor accounted for a large part of the significant interaction effects with planting and soil. In the three plantings (1, 2 and 4) where the two backgrounds could be compared for fruit production, Col-0 displayed 2-fold greater fitness than C24 in sand, an effect that was not seen in turf (main effect, log estimate 0.85, p = 1.8e-07, interaction with soil, log estimate -0.89, p = 9.8e-06, Fig 3B, Table 4).
The effect of the chr 4 introgression (Tables 2 and 4) was modulated by all factors, i.e. planting, soil and genomic background (F = 10.42, p = 1.67e-13, Table 2). The interaction between planting conditions and background also changed significantly the effect of the introgression  Tables 2 and 3). To investigate which factorial combination revealed the effect of the chr4 introgression, we ran a separate GLM analysis with introgression as a factor nested within a factor called "setting", which combined soil condition, planting and genomic background in 16 levels (Fig 4). This analysis revealed that the effect of the introgression was significant for 3 out of the 16 settings and its effect was estimated to reverse from a 50% decrease to a 2-fold increase in silique production for the C24 Chr4 allele. Although the introgression of Col-0 into the C24 background was larger than the reciprocal introgression, we detected a significant effect either in both backgrounds or in the Col-0 background only (Fig  4) suggesting that it is the overlapping portion of the reciprocal introgression that displays fluctuating fitness effects.
The effect of the introgression, when it is significant, is comparable in magnitude to the 2.3fold increase in silique production per plant of Col-0, compared to the C24 genomic background (Table 4). Thus, in some conditions, the Chr4 introgressed region can reach an effect comparable in magnitude to the average effect caused by the whole genomic background.

Effect of experimental factors on life-history variation
We observed that plant lifespan varied drastically across the 5 successive plantings (Fig 1,  Table 1). In the second planting, the life cycle was completed in less than two months. In planting 3, the life cycle was completed after 8 months and in planting 4, in approximately 4 months. Flowering in planting 4 was terminated in late September (a month later than in planting 2 the previous year), which resulted in a late-fall distribution of seeds for planting 5 and an overwintering at the seed stage.
The duration of vegetative growth (e.g. the time from germination to flowering) is believed to have an important impact on final fitness [27,34]. Yet, in our experiment, we observed no obvious correlation between silique production and the duration of the life cycle (Fig 5). Furthermore, using photo-thermal units (PTU, expressed in˚C.hour) to control for temperature and day-length differences across plantings, we observed that Col-0 in planting 3 yielded higher fitness than planting 4, despite a lower cumulative PTU (Table 1). We nevertheless observed a marginally significant correlation between days to flowering and silique number within the experimental populations of planting 3 (Spearman r = 0.29, p = 0.07).
We observed in the 3 overwintering seasons a greater spread in the timing of life-history transition. Planting 1 was started in early December and no germination occurred before

Discussion
In this study, we designed a field experiment to quantify the fitness outcome of climate, substrate and genotypic variation on phenotypic and fitness variation in the natural setting of successive seasons. Our results reveal that environmental variance experienced over successive growth seasons at a given locality of temperate latitude with permissive summer season, is not buffered by phenotypic plasticity. Genotypic differences in phenotype and fitness fluctuate in magnitude across soil conditions and seasons. Seasonal fluctuations experienced at temperate

Seasonal fluctuations in the life cycles expressed by a single genotype
The performance of various A. thaliana genotypes in the field has already been reported in several studies [20,21,23,27], but our experiment is the first to monitor the life cycle dynamics naturally unfolded by single genotypes. This experiment was conducted along 5 consecutive seasons and included two summer plantings and three fall plantings. From the second season onward, the date of planting was determined by the date of fruit maturation in the field, and the date of germination was determined by the level of dormancy imposed in the field during and after seed maturation. The genotypes were thus allowed to express their natural life cycle. At the temperate latitude of our experimental site, we show that A. thaliana completes two cycles per year while the third one is suspended during winter, either at the rosette (planting 3) or at the seed stage (planting 5). Our experiment shows that a given genotype may manifest diverse life cycles in the course of successive seasons, confirming theoretical predictions based on a combination of hydrothermal and photothermal models of germination and flowering decisions [26][27][28][29][30][31][32][33][34]. The life stage at which the winter interruption occurred seemed to depend on the timing at which the second generation was finished, either because of variation in the maternally imposed dormancy, the adverse conditions for germination, or possibly the onset of secondary dormancy triggered by low temperatures [34][35][36][37][38]. We further observed that life time fruit production also varied by up to 36-fold across successive seasons. This is likely to be due to the dramatic environmental differences to which genotypes are exposed throughout the successive seasons. Indeed, level of fluctuation for a winter annual cycle replicated over 5 distinct years reported up to 7-fold fitness fluctuations [21]. Because the overwintering generation tended to reach higher fitness than summer or spring generations, we can further conclude that, at the latitude where this experiment was conducted, variation in characters important for rosette survival during winter may be comparatively more crucial for adaptation than those important for completing a life cycle in summer. Nevertheless, since we also observed that the life-history stage of overwintering could change across years, the traits exposed to winter selective pressures also likely fluctuate.

Seasonal fluctuation in expressed life history differences between genotypes
Based on previous laboratory studies, we expected that the two backgrounds would display markedly different life histories in the field. The genotype C24 was reported to flower 11 days after Col-0 in long day conditions but is a relatively early genotype under short day conditions because of a weak FLC allele [31]. In addition, high ambient temperature (27-30˚C) triggers faster flowering in Col-0, but delays flowering in C24, whereas C24 seeds matured under long day conditions at 20˚C showed 2 weeks longer primary dormancy than Col-0 (JDM, pers. com.).
Life history transitions such as germination and flowering time are known to be dependent on the environment [27,35,[39][40][41]. Here, we show that phenotypic differences between genotypes can be either masked or exposed across the successive life cycles of a natural setting. Indeed, a major difference in the timing of flowering of Col-0 and C24 was observed only when seeds were sown at the end of August. The late summer time window has been demonstrated to impact the manifestation of flowering time differences controlled by the vernalization pathway [39].
The Col-0 genotype appears better adapted to the experimental site The relative fitness of the Col-0 and C24 backgrounds were reported to change markedly across geographical sites [23]. The Col-0 genotype produced 50% more siliques in Halle, Germany, but its fitness was 12% of that of C24 in Norwich, UK, while both genotypes showed comparable fitness in Valencia, Spain [23]. In our experiment, the Col-0 background generally displayed higher fitness than C24 despite temporal seasonal fluctuations. This suggests that it is generally well adapted to the climatic conditions of our field. Its genome is assigned to a Western European A. thaliana clade of genotypes originating from natural stands in Germany [30]. The magnitude of the fitness difference nevertheless changed markedly across soil conditions and seasons. A recent study reported that A. thaliana genotypes originating from Southern latitudes, which is the case for C24, perform comparatively well at intermediate temperate latitudes [41]. C24, however, did not perform better than Col-0 even in sand where water limitations might have provided it with an advantage. In fact, the Col-0 genotype performed better than C24 in sand, even though plants suffered a 3-week long period of high temperature without precipitation in planting 2 (Fig 1, Fig 3). Yet, we also observed that micro-geographic heterogeneity in substrate, which is commonly found in natural A. thaliana stands, can magnify or mask differences observed between Col-0 and the Southern genomic background C24 (Fig 3B).

Rare but radical events counter select inappropriate flowering time
The evolution of gene frequencies in natural populations depends on differential fitness, and therefore on the environmental factors that promote or buffer the expression of this variation [11]. Our experiment shows that although phenotypic differences in the timing of flowering are masked in 3 of 4 seasons, their expression coincided with a severe selection against the C24 genotype. This severe frost event caused equal damages on the rosettes of C24 and Col-0. Only Col-0, which had not flowered, and a couple of populations (pots) in which a handful C24 individuals had not flowered, were able to regenerate their rosettes and complete their cycle. It is thus possible that it is the transition to flowering that might have caused the plants' vulnerability to low temperatures, although we cannot formally exclude that Col-0 and C24 may differ in frost tolerance. This experiment therefore depicts relatively rare but radical selective events by which a genotype can be counter-selected in regions with sometime severe winter conditions. Such an ecological scenario may explain why local adaptation remains pervasive in this species despite the seasonal fluctuations affecting the expression of adaptive phenotypes that we document [20,23].

Temporal fluctuations of allelic effects on fitness
Our experiment further shows that seasonal and environmental fluctuations must be taken into account when studying the genomic underpinnings of local adaptation. Extrapolating the evolutionary trajectory of mutations based on their impact on ecological performance in a reductive experimental setting is not directly possible. Whereas the Col-0 background generally performed better over the whole experiment, the Chromosome 4 fragment had effects of changing signs in a small subset of conditions. This small region, which only contains about 10% of previously reported fitness-relevant differences across the geographical range of the species, can have, in some conditions, an effect comparable to that of the genomic background. In fact, interactions between genetic and temporal variation are probably more the rule than the exception. As an example, variation in both plant density and water supply, two environmental factors that are likely to fluctuate over time, has been recently shown to change the effect of a polymorphism impacting water-use efficiency in A. thaliana [42].

Conclusion
The current development of faster and cheaper sequencing methods boosts our ability to investigate the basis of local adaptation in a number of plant species, including rare endemics or aggressive invasive species [43,44]. Annotating how functions encoded in the genome impact ecological performance and fitness seems within reach [45]. Our findings, however, have important bearings on the challenges we are facing for plant (or animal) species that can shift their temporal niches across generations. Our experiment indeed clearly shows that adaptation to local conditions requires the optimization of fitness in both summer and winter conditions. The likelihood that given allele meets the seasonal window where its selective advantage can be exposed will have to be characterized. For this, the genotypic specificities of plastic life history regulation (as e.g. in [46]), the temporal environmental variance, and the scale and impact of micro-geographic substrate heterogeneity will have to be jointly evaluated. A. The effect of plant density in the pot was only strong in planting 2, 3 and 5. For 3 and 5, Co-0 was the only genomic background that survived. B. For planting 2, the C24 background did not perform well, so that density was too low to reveal an impact of competition on total life time fitness. (TIFF) S3 Fig. Relationship between days to flowering (DTF) and days to germination (DTG) in each planting for each genomic background. Plantings 2 and 4 had similar germination and flowering time. By contrast, generation 3 showed spread flowering time and generation 5 rather spread germination (spring germination), suggesting that the buffering of germination and flowering differed across planting (and across genotypes). Note that the Chr4 introgression did not impact these differences. In generation 1, the date of germination was not scored, but three cohorts were observed, reflecting a spread of germination over March/April, whereas flowering time occurred within 1-2 weeks. (TIFF) S1