Temporal environmental variation imposes differential selection on genomic and ecological traits of virtual plant communities

The reaction of species to changing conditions determines how community composition will change functionally — not only by (temporal) species turnover, but also by trait shifts within species. For the latter, selection from standing variation has been suggested to be more efficient than acquiring new mutations. Yet, studies on community trait composition and trait selection largely focus on phenotypic variation in ecological traits, whereas the underlying genomic traits remain relatively understudied despite evidence of their role to standing variation. Using a genome-explicit, niche- and individual-based model, we address the potential interactions between genomic and ecological traits shaping communities under an environmental selective forcing, namely temporal variation. In this model, all ecological traits are explicitly coded by the genome. For our experiments, we initialized 90 replicate communities, each with ca. 350 initial species, characterized by random genomic and ecological trait combinations, on a 2D spatially-explicit landscape with two orthogonal gradients (temperature and resource use). We exposed each community to two contrasting scenarios: without (i.e. static environments) and with temporal variation. We then analyzed emerging compositions of both genomic and ecological traits at the community, population and genomic levels. Communities in variable environments were species poorer than in static environments, populations more abundant and genomes had a higher numbers of genes. The surviving genomes (i.e. those selected by variable environments) coded for enhanced environmental tolerance and smaller biomass, which resulted in faster life cycles and thus also in increased potential for evolutionary rescue. Even under the constant environmental filtering presented by temporal environmental variation, larger, more linked genomes allowed selection of increased variation in dispersal abilities. Our results provide clues to how sexually-reproducing diploid plant communities might react to increased environmental variation and highlights the importance of genomic traits and their interaction with ecological traits for eco-evolutionary responses to changing climates.

the potential interactions between genomic and ecological traits shaping com-10 munities under an environmental selective forcing, namely temporal variation. 11 In this model, all ecological traits are explicitly coded by the genome. For our 12 experiments, we initialized 90 replicate communities, each with ca. 350 initial 13 species, characterized by random genomic and ecological trait combinations, on 14 a 2D spatially-explicit landscape with two orthogonal gradients (temperature 15 and resource use). We exposed each community to two contrasting scenarios: 16 without (i.e. static environments) and with temporal variation. We then ana-17 lyzed emerging compositions of both genomic and ecological traits at the com-18 munity, population and genomic levels. Communities in variable environments 19 were species poorer than in static environments, populations more abundant 20 and genomes had a higher numbers of genes. The surviving genomes (i.e. those 21 selected by variable environments) coded for enhanced environmental tolerance 22 and smaller biomass, which resulted in faster life cycles and thus also in increased 23 potential for evolutionary rescue. Even under the constant environmental filter-24 ing presented by temporal environmental variation, larger, more linked genomes 25 allowed selection of increased variation in dispersal abilities. Our results pro-26 vide clues to how sexually-reproducing diploid plant communities might react 27 to increased environmental variation and highlights the importance of genomic 28 traits and their interaction with ecological traits for eco-evolutionary responses 29 to changing climates. Communities of plant species are the result of different abiotic and biotic con-36 ditions (Huntley 1991). Changes in those conditions will therefore also reflect 37 on communities and their trait composition. Response strategies that enable 38 species survival under changing conditions may vary across species. They can, 39 for instance, select for survival (Holt 1990), for lower body mass (Parmesan 40 2006), for dispersal (Berg et al. 2010), or for adaptation to new conditions (Joshi 41 et al. 2001, Jump and Peñuelas 2005, Bell and Gonzalez 2009. Given enough 42 time, this will result in the communities passing through ecological species suc-43 cessions (Huston and Smith 1987) and evolutionary taxon cycles (Ricklefs and 44 Bermingham 2002). Even in short periods, populations within communities can 45 change their traits in response to environmental variation via rapid evolution 46 (Maron et al. 2004). In this case, selection on standing variation can be more 47 efficient than aquiring novel mutations (Barrett andSchluter 2008, Bolnick et al. 48 2011). This standing variation can be both intraspecific and intra-individual, 49 i.e., within-genome variation. A high standing variation thus provides a re-50 source for populations to quickly respond to changing environments (Cochrane 51 et al. 2015). However, the genomic traits which enable and maintain standing 52 variation remain largely understudied in ecological and eco-evolutionary studies 53 (but see Schiffers et al. 2012, Matuszewski et al. 2015. 54 Many functional species traits are quantitative and subject to genetic inter-55 actions, such as epistasis, pleiotropy and genetic linkage. To infer a direct con-56 nection between phenotype and genotype is therefore complex (Korte and Farlow 57 2013). Still, all this genomic background determines standing genetic variation, 58 which in turn will constrain which individual phenotypes are possible and thus 59 a population's evolutionary potential. With the increasing availability of ex-60 haustive genetic data, considering detailed genetic factors in eco-evolutionary 61 models has become more feasible, especially for model species (Frachon et al. 62 2019, Exposito-Alonso et al. 2019). Indeed, there is an increasing amount of ge-63 netic data at the population or even at the individual level (e.g. Domingues et al. 64 2012, Alonso-Blanco et al. 2016. Nevertheless, manipulating real-world systems 65 to conduct meaningful experiments to isolate factors on both functional and ge-66 netic levels is difficult (but see Booth and Grime 2003). Thus, although the 67 importance of genetic factors for ecological processes has long been recognised 68 (Holt 1990), investigating its effects in real-world systems remains challenging 69 (Hughes et al. 2008). 70 Simulation models provide a powerful alternative to overcome the empirical 71 challenges of investigating and manipulating genetic traits and all the trait-72 mediated ecological functions they control. Modeling studies can cover any 73 organisational level in biology, from genomes over species to communities (Ma-74 tuszewski et al. 2015, Kubisch et al. 2014, Münkemüller et al. 2012, Saupe 75 et al. 2019, and thus are suitable tools to explore potential eco-evolutionary 76 regulations of species traits. Therefore, we developed a Genome-explicit Meta-77 community Model (GeMM, Fig. 1) to address the interplay of genomic and 78 ecological traits in species communities under an environmental selective force, 79 namely temporal environmental variation. Specifically, we address the follow-80 ing questions. (a) Which ecological and genomic traits enable survival in tem-81 porally variable environments? (b) How do temporally variable environments 82 shape standing variation (phenotypic and genetic)? We designed a simulation 83 experiment under two different environmental scenarios, namely with versus 84 without temporal environmental variation (variable and static environments, 85 respectively) and analyzed genomic and ecological trait characteristics of sur-86 viving communities. We expected communities in variable environments to se-87 lect for higher tolerances (Holt 1990), higher dispersal abilites (Berg et al. 2010) 88 and lower biomass (Parmesan 2006) and to exhibit increased standing variation, 89 both genetic and phenotypic (Cochrane et al. 2015). While our expectations on 90 trait responses were largely confirmed, we find that standing variation is de-91 creased for most traits except dispersal. Our findings on virtual communities 92 suggest how eco-evolutionary dynamics of real plant communities might unfold 93 under changing environments.

Sequence
Allele(s), one or more of:  Figure 1. Schematic of the model. (a) Individuals represent the base agents in the model. They are comprised of a phenotype which interacts with other individuals and the environment, and a genome. The genome is diploid and consists of maternal and paternal sets of linkage units, which combine genes as one hereditary unit. Each gene may code for one or more alleles of functional traits. The expressed trait in the phenotype results as the average of all associated alleles in the genome. The expression of some of the traits ("variable traits") additionally depends on the local current environment and may change over time. (b) Flow of processes each individual passes every year. Some of the processes are dependent on the local temperature and individual biomass (marked "metabolic"), while all processes depend on an individual's phenotypic traits (see (a)). Dashed arrows represent influences, solid arrows represent sequence of events.

95
The model 96 General structure. We use GeMM (version 1.0.0) -a genome-and spatially-97 explicit, niche-and individual-based model for plant metacommunities written 98 in julia (Bezanson et al. 2017, Fig. 1). The model generates metacommunity 99 dynamics (Hanski 2001, Leibold et al. 2004) and it considers explicit local pop-100 ulation and community assembly dynamics emerging from genomic and indi-101 vidual level processes. The model simulates discrete time steps, which can be 102 translated to one year. In the model, individuals belong to species, which are 103 characterized by individuals with identical genetic architecture (i.e. genome size 104 and linkage) and ecological traits (dispersal ability, environmental niche and 105 size) falling within a species-specific Gaussian trait distributions ( Fig. 1 (a)). 106 Thus, individuals of the same species are not functionally identical, depicting 107 intraspecific phenotypic variation. Dispersal of individuals (i.e. seeds) intercon-108 nects grid cells, while the position of individuals is characterized by the grid cell 109 coordinates, i.e., all individuals are concentrated in the center of the respective 110 grid cell.

111
Eco-evolutionary processes. Like some previous ecosystem models ( Har-112 foot et al. 2014, Cabral et al. 2019, see Cabral et al. 2017 for a review), yearly 113 vegetative growth in biomass, fertility and mortality rates in the model are con-114 trolled following the metabolic theory of ecology (MTE, Brown et al. (2004), 115 Price et al. (2010)). Accordingly, the model considers discrete yearly time steps. 116 In MTE, a biological rate b depends on the temperature T and individual mass 117 M , scaling a base rate b 0 as: where E A is the activation energy and k B the Boltzman constant. The expo-119 nent c is 3 4 for biomass growth and reproduction, and − 1 4 for mortality (Brown 120 et al. 2004). This results in smaller individuals having a higher mortality than 121 bigger ones, while individuals in cooler conditions have a lower mortality than 122 those in warmer conditions. Using the MTE means reduced parameterization 123 effort, since b 0 values for the different processes are global constants and thus 124 identical for every species. Additionally, the emerging longevity-fecundity trade-125 off that comes with mass-regulated rates has been shown to inherently supress 126 the evolution of "super-species" (Cabral et al. 2019).

127
Over the course of a simulation, individuals thus grow in size, passing three 128 life stages: (1) seed, (2) juvenile, and (3) adult. Individuals disperse as seeds, 129 establish, grow and become reproductive adults ( Fig. 1 (b)). Both seed biomass 130 and adult biomass, i.e., the threshold biomass where individuals become repro-131 ductive, are two of the central, genetically-coded traits that define individuals 132 ( Fig. 1 (a), Table 1). Adults are monoaecious and reproduce sexually with a 133 random adult of the same species whithin the same grid cell to produce new 134 seeds. Seed dispersal follows a logistic dispersal kernel with genetically-coded 135 mean dispersal distance and shape parameter µ and s, respectively (see Bullock 136 et al. 2017). In our discrete landscapes, dispersal is modeled as centroid-to-area, 137 with expected mean dispersal distances usually around equal to the length of 138 the grid cells (cf. Chipperfield et al. 2011). Furthermore, all individuals have 139 encoded preferences concerning two different environmental measures: the first, 140 temperature, has a direct effect on biological rates, as described by the MTE 141 (Brown et al. 2004) and affects density-independent mortality, while the sec-142 ond is a surrogate for environmental resources, e.g., water. Thus, from here on 143 this second axis is called precipitation for simplicity. Individuals' adaptation 144 to precipitation conditions determine their competitive abilities. Both these 145 preferences are characterized by an optimum and a tolerance, which are rep-146 resented as mean and standard deviation of a Gauss curve, respectively. The 147 degree of mismatch between an individual's preference optimum with the lo-148 cal environment (i.e. within the grid cell) determines its adaptation value (i.e. 149 environmental fitness). Near their optimum, individuals with higher niche tol-150 erance have lower adaptation values than individuals with narrower breadth 151 (i.e. specialists, Griffith and Sultan (2012)). During establishment, the adapta-152 tion values toward temperature and precipitation are calculated for each new 153 seed based on the local conditions and phenotypic traits ( Fig. 1 (b)). Further-154 more, each time environmental conditions change, all individuals in the affected 155 grid cell pass establishment again to re-calculate their adaption values. These 156 adaptation values are functional for two different subsequent processes. First, 157 individuals experience a metabolic, density-independent mortality (Brown et al. 158 2004). This mortality further scales with individual temperature adaptation, so 159 that mortality is higher for individuals which are poorly adapted to the sur-160 rounding temperature (Cook 1979). Second, all individuals in a cell compete 161 for the limited available space in the grid cell, i.e., total sustainable biomass. If 162 the combined biomass of all individuals in a cell exceeds the grid cell's carrying 163 capacity biomass, individuals are removed from the community until biomass 164 is within grid cell limits. The choice which individuals to remove is based on 165 pair-wise comparisons of random pairs of individuals. From any of such two 166 individuals, the individual less adapted to local precipitation conditions is re-167 moved.

168
Genetic architecture. All of the aforementioned traits (see Table 1) are 169 coded by one or more genes in an individual's diploid genome (polygenes ). Sin-170 gle genes can also be associated to several traits at the same time (pleiotropy, 171 Solovieff et al. (2013)). Thus, each trait can be represented more than once 172 in the genome (i.e. through different genes at different loci). Since trait repre-173 sentations are subject to species-specific variation, they can constitute different 174 alleles -both within the haploid genome at different loci, but also between the 175 maternal and paternal haploid genomes or between individuals (cf. Nevo 1978). 176 Realized ecological traits y, i.e., an individual's phenotype, are then determined 177 quantitatively by considering all respective loci y l within an individual's genome 178 and taking their average. This results in a random degree of species-specific phe-179 notypic and genetic, i.e., intra-individual or intra-genomic, trait variation (cf. 180 Mackay 2001). Lastly, genes may be combined to form a linkage unit, which 181 represent a set of spatially close genes within the same chromosome arm. Link-182 age units thus comprise the smallest hereditary entities (Hermann et al. 2013, 183 Lande 1984. Haploid gametes receive a complete random set of those linkage 184 units following a recombination process, where each linkage unit can originate 185 from either the paternal or maternal chromosomal complement of the individual 186 producing the gamete. During reproduction, the gametes of two mating individ-187 uals thus form an offspring's (i.e. seed) genome. The phenotypic characteristics 188 of each offspring are then calculated on the basis of its recombined genome and 189 local environmental conditions ( Fig. 1 (a)).

190
A detailed model description with justification for assumptions, equations and 191 parameter values can be found in Supplementary material Appendix 1 (Grimm 192 et al. 2006(Grimm 192 et al. , 2010. Model parameters are summarized in Table 1.  Biomass at seed stage e −2 g to e 10 g genome µ Dispersal kernel mean 0 to 1 grid cells genome s Dispersal kernel shape 0 to 1 grid cells genome P Precipitation optimum 0 to 10 genome σ P Precipitation tolerance 0 to 1 genome T Temperature optimum 10 • C to 40 • C genome σ T Temperature tolerance 0 • C to 1 • C genome capacity of 100 kg of total biomass, which approximately relates to 100 m 2 of 197 grassland (Deshmukh 1984, Bernhardt-Römermann et al. 2011. Landscape 198 size and carrying capacity was arbitrary but ensured computational feasibility. 199 Two perpendicular environmental gradients (temperature and precipitation) ran 200 along the long and short axis of the landscape, respectively. The rectangular 201 shape of our simulation arena provided a longer gradient in the physiologically 202 important temperature direction.  Table 1). To obtain the ecological characteristics of a species, 211 first an average phenotype was defined by randomly selecting a value for each 212 phenotypic trait. These traits, more specifically, the adult biomass trait, were 213 then used to calculate the number of offspring a single individual of this species 214 would have. Given an already determined genetic architecture (i.e. n l , n u , and 215 σ l ), each individual of a species was then initialized as follows. For each trait 216 representation (i.e. gene) within the genome, the associated trait value was 217 chosen randomly following a Normal distribution with the trait value of the 218 average phenotype as mean and standard deviation the product of σ l and the 219 trait value (Table 1). Afterwards, the initial phenotype for each individual was 220 calculated based on all genes in the genome. This resulted in varying degrees 221 of intragenomic and intraspecific standing variation. We disabled mutations in 222 our experimental design so that this standing variation was the only resource 223 for selection. Grid cells were then filled with populations of several species until 224 carrying capacity was reached. Because species vary randomly in their traits, 225 including biomass, initial grid cell communities varied in richness. This resulted 226 in initial communities with on average 10 species per grid cell and a total of 350 227 species in the landscape.

228
Values for simulation, global and species-specific parameters that were not 229 varied in the different experimental scenarios were chosen to ensure plausible 230 patterns, most importantly to achieve species co-existence by adjusting the 231 mortality-to-fecundity ratio. Species-specific parameter values were drawn at 232 random from a range that extended beyond what would be realisable in sim-233 ulations to reduce geometric artifacts within the parameter space (Table 1). 234 This also kept the need for additional assumptions at a minimum, since viable 235 species emerged via environmental filtering and ecological interactions. Global 236 parameter values were either adapted from the literature (Brown et al. 2004, 237 Fournier-Level et al. 2011) or fine-tuned via trying out a range of realistic values. 238 Scenarios. For investigating our general study question about the interplay of 239 environmental variation and ecological and genomic traits, we designed two sce-240 narios. In the first, temperature and precipitation gradients arbitrarily ranged 241 through constant values of 16.85 • C to 22.85 • C (290 K to 296 K) and 3 to 7 242 (arbitrary quantity), respectively, during the entire simulation run ("static en-243 vironment"). In the second, initial temperature and precipitation values were 244 the same as in static environments, but could change at each year ("variable 245 environment"). The change followed a gaussian random-walk trajectory to yield 246 positive auto-correlation (Fung et al. 2018). The amount of change (δ P and 247 δ T , Table 1) was drawn randomly from a Normal distribution with a standard 248 deviation of 0.2. This value corresponds to a moderate rate of change of no 249 more than 0.5 degrees per year in the majority (ca. 99 %) of cases, which 250 we found by trying different values to produce noteable environmental change 251 that did not kill all individuals in a short amount of time. Since our simu-252 lation arena represents a small spatial scale, all grid cells changed always by 253 the same value at each timestep. The change of temperature was independent 254 from that of precipitation and vice-versa. Confounding effects, such as land-255 scape configuration, different temporal dynamics, complex dispersal behavior 256 and macro-evolutionary processes (e.g. clade diversification) have been studied 257 elsewhere and were thus not included in the present study (Münkemüller et al. 258 2012, Kubisch et al. 2014, Aguilée et al. 2018. Table 1 contains the parameters 259 which were varied for the scenarios, their meaning and their values. We sim-260 ulated 90 different replicates. Each replicate terminated after 1000 simulated 261 years. This duration was adequate to allow quasi-equilibrium (see Results) and 262 short enough to warrant our selection-on-standing-variation rationale (Hermis-263 son and Pennings 2005). Each replicate, i.e., each unique initial community, 264 was subjected to both scenarios. This yielded 180 simulations in total. 265 We recorded the complete state of the individuals in our simulation world 266 at the start and every 50 years of a simulation run. This data encompassed 267 individual phenotypic and genotypic values. Thus, for every year, we tracked the 268 state of local species populations including location, abundance, demographics, 269 median adaptation, and trait values for all ecological and genomic traits. To make the individual information more accessible, we calculated summary 272 statistics at the population level. We defined a population as a group of conspe-273 cific individuals co-occurring in the same grid cell. For each population, we then 274 calculated median values of each phenotypic trait, the variance of each pheno-275 typic trait (phenotypic intraspecific variation), and medians of the individual 276 genetic variation in each trait. We scaled all variance values by the respec-277 tive population-specific medians to get coefficients of variation of the median 278 (CV median). In order to compare emerging ecological patterns and identify 279 when equilibrium is reached, we calculated a set of ecological metrics, namely 280 species-richness, i.e., the average number of species per grid cell, α (α-diversity), 281 the total number of species across the landscape, S (γ-diversity), β-diversity, 282 β = S/α − 1 (Whittaker 1960), population demographic structure (i.e. number 283 of juveniles and number of adults) and range-filling from the data on surviv-284 ing communities. For diversity indices, we converted our data to community 285 matrices and analyzed them using vegan (Oksanen et al. 2018) in R (R Core 286 Team 2019). To assess demographic structure within communities, we analyzed 287 average numbers of juveniles and adults. Range-filling was calculated as the 288 fraction of grid cells that was occupied by a species over all the grid cells that 289 were potentially suitable for the given species. Suitability was asserted as an ar-290 bitrary cut-off where environmental parameters (temperature and precipitation) 291 fell within a species' tolerance (optimum ± tolerance).

292
For our study questions, we analyzed the trait composition of surviving 293 communities genomic trait composition (study question (a)), and differences in 294 phenotypic and genetic standing variation (study question (b)) between envi-295 ronments. Since we were interested in general patterns of the effect of envi-296 ronmental variability, rather than the effects of warming or cooling trends, we 297 excluded precipitation and temperature optimum traits from our analyses. We 298 transformed trait and variation distributions before analysis and visualization 299 using a log (x + 1) transformation, because they were strongly left-skewed and 300 contained values < 1. Additionally, we calculated the degree of genetic linkage 301 as n l nu , because due to our method of initializing species, n u directly depended 302 on n l .

303
To ascertain whether and how trait composition differs between environ-304 mental conditions (study question (a)), we first compared species numbers and 305 identities. Because each community is subjected to both environments, we an-306 alyzed what proportion of species was shared by both environments, and which 307 were unique to one of the environments. To assess how ecological and genomic 308 traits respond to variable environments, we compared trait characteristics be-309 tween scenarios by performing principal component analyses on the population 310 trait data. This way, we were able to describe general patterns in trait space 311 between scenarios by relating the total trait space shift to the principal com-312 ponents and correlated trait axes. Additionally, we compared community trait 313 distributions pairwise between environments to identify trends in traits specific 314 to the environments. For this, we calculated linear mixed models using the R 315 package lme4 (Bates et al. 2015) with trait as response, environment as fixed 316 effect and replicate as random effect.

317
To find out whether there is a selective force on standing variation (both phe-318 notypic and genetic) specific to environmental conditions (study question (b)), 319 we compared the phenotypic and genomic trait variances of surviving commu-320 nities between scenarios for all traits in separate. We again calculated linear 321 mixed models, with trait variances as response, environment as fixed effect and 322 replicate as random effect.

323
The model code, experiment definition files and analysis scripts are available 324 at https://github.com/lleiding/gemm. Albeit reporting of significance values 325 is generally inappropriate for simulation models (White et al. 2014), we use 326 significance here to identify which responses are stronger than others.

328
Differences of ecological patterns between environments 329 Surviving communities in our simulation experiments ( Fig. 1) differed in a num-330 ber of ecological characteristics. Compared to communities in static environ-331 ments, communities in variable environments were only about half as species-332 rich on a local level (α-diversity, Fig. 2(a)) and exhibited less β-diversity ( Fig. 333  2(b)), which resulted in decreased species richness on a regional scale (γ-diversity, 334 Fig. 2(c)). Summing over all replicates, a total of 108 species survived in 335 both enviroments, while 256 and 64 surviving species were unique to static 336 and variable environments, respectively. Emerging functional differences com-337 prised higher total abundances in all demographic stages ( Fig. 2(d), (e)) and de-338 creased range filling for communities in variable environments ( Fig. 2(f)). While 339 all aforementioned metrics were constantly changing during the entire simula-340 tion course in the variable environments, in static environments they reached a 341 quasi-equilibrium by year 500. For the following trait-based analyses, we thus 342 used data from that year.

Response of ecological and genomic traits 344
Surviving communities showed subtle differences in their trait syndromes com-345 bining all traits in a PCA. In the first two principal components, populations 346 from variable environments occupied for the most part a subset of the trait 347 space of populations from static environments (mostly overlapping ellipses in 348 Fig. 3). Nevertheless, the trait space of variable environment communities was 349 shifted towards increased environmental tolerances and dispersal abilities and 350 decreased mean genetic variation (negative direction of second principal com-351 ponent - Fig. 3). With the exception of the first, all principal components and 352 thus correlated traits, contributed similarly to the overall explained variance 353

354
Focusing on single traits, communities showed several differences between 355 the two types of environments ( Fig. 4(a), Supplementary material Appendix 1 356 Table A3). Compared to static environments, surviving communities in variable 357 environments showed on average an increased number of genes (n l ), increased 358 precipitation and temperature tolerances (σ P and σ T , respectively), increased 359 long distance dispersal (s), decreased adult biomass (M r ), and decreased ge-360 netic variation (σ l ). Seed biomass, mean genetic variation and genetic linkage 361 exhibited no significant differences (Supplementary material Appendix 1 Table 362 A3).

Differences in standing variation (phenotypic and genetic) 364
Additionally to differences in the phenotypic characteristics, we found distinct 365 patterns between environments in both phenotypic and genetic trait variation 366 (Fig. 4(b)). While the phenotypic variation of mean dispersal distance and both 367 phenotypic and genetic variation of long distance dispersal was increased in 368 variable environments, all other trait variations (phenotypic and genetic) were 369 decreased in variable environments. The trend towards a decrease in genetic 370 , where only those species survive that are able to adapt to or track envi-378 ronmental changes. As a result, communities are species poorer (see also Menge 379 and Sutherland 1976). The decreased β-diversity furthermore suggests that 380 these fewer species in variable environments are rather generalistic, in compari-381 son to static environments where species seem more specialized to local environ-382 mental conditions, as evidenced by the higher β-diversity (cf. Gilchrist 1995). 383 The fact that, furthermore, range-filling is reduced in the variable environments 384 is likely a mid-domain-like effect (cf. Colwell and Lees 2000), where due to the 385 ongoing temporal variability, the margins of a potential range will often become 386  Red and blue colors indicate negative and positive differences, respectively. Note the different axis scales. The abbreviation "n.s." denotes differences that are not significant (p > 0.05). "N.A." marks trait differences that are not available at the given level.
unsuitable quickly, impeding establishment and survival. Moreover, because the 387 environmental change in our simulations was random rather than periodical or 388 directional, the probability for species to find alternating suitable conditions is 389 low. This alternating suitability, however, is the prerequisite for temporal envi-390 ronmental variability to favor species co-existence and increased species richness 391 (cf. Pacala 1993, Descamps-Julien andGonzalez 2005). In contrast, 392 most communities in static environments passed environmental filtering already 393 after the first 200 years, after which species were distributed according to their 394 environmental preferences and ecological patterns became stable.

395
Study question (a): Which ecological and genomic traits 396 enable survival in temporally variable environments?

397
The trait characteristics of communities in the respective environments repre-398 sent successful strategies in surviving random environmental variation. The de-399 creased values of precipitation tolerance in communities in static environments 400 indicate increased environmental specialization. This is in contrast to commu-401 nities in variable environments, where the variability in precipitation conditions 402 favors species with higher tolerance values (i.e. specialization to local condi-403 tions are detrimental in variable environments, Gilchrist 1995, Kassen 2002. 404 Additionally, temperature tolerance directly affects individual survival due to 405 metabolic constraints (Fig. 2(d)). Since a high temperature tolerance decreases 406 fitness, species are forced to keep tolerances low if they occur at their respective 407 environmental optimum. In variable environments, this environmental optimum 408 is hardly met. As a consequence, selection acts rather on enhancing temper-409 ature tolerance to gain long-term fitness. Therefore, our experimental design 410 captures the evolution towards bet-hedging strategies in terms of adaptation to 411 variable environments (Slatkin 1974).

412
The second aspect of survival strategies lies in the biomass patterns. In gen-413 eral, species in variable environments were smaller than in static environments. 414 Since growth rates and fecundity follow MTE, smaller species are more fecund 415 than bigger species at the expense of survival. A higher and more frequent num-416 ber of offspring will spread the risk over time in variable environments (McGinley 417 et al. 1987, Philippi andSeger 1989). Additionally, the larger range of different 418 biomasses in static environments can be interpreted as temporal partitioning 419 (Pronk et al. 2007), because it means that co-occurring species will reproduce 420 at different times and intervals. This allows species to alternate dominance and 421 thus produce temporally variable biotic conditions (cf. Olff et al. 2000, Wilson 422 and Abrams 2005). Furthermore, both biomass and tolerance patterns suggest 423 that specialization to avoid competitive exclusion plays a larger role in shaping 424 communities in static environments, while communities in variable environments 425 are primarily shaped by generalism and environmental filtering (cf. Menge and426 Sutherland 1976, Hulshof et al. 2013).

427
In order to track suitable conditions, dispersal abilities are of crucial im-428 portance in changing environments (Bourne et al. 2014). While mean dispersal 429 distances in our simulations showed little differences between environments, long 430 distance dispersal indeed increased in variable environments. Besides primary 431 dispersal traits, the dispersal rate also increased in variable environments via 432 the indirect effect of metabolic rates: the high demographic turnover that comes 433 with higher fecundity due to decreased biomass leads to more frequent dispersal. 434 This further explains why there was little change in mean dispersal distance be-435 tween environments. With the rate of change in our simulations and the small 436 spatial extent of our landscape, dispersal distance (which is what is controlled 437 by dispersal traits) is less important than dispersal rate (cf. Johst et al. 2002). 438 However, this might change in fragmented landscapes, where dispersal distance 439 is critical to maintain connection between habitable patches (Bacles et al. 2006, 440 Boeye et al. 2013, Bonte et al. 2010.

441
Lastly, species may survive by adapting to changing conditions (Jump and 442 Peñuelas 2005). This constant adaptation requires an appropriate genetic ar-443 chitecture: we expected genomes to contain a high variation of trait alleles 444 (Holt 1990) which can be recombined easily for a species to quickly respond to 445 novel conditions (Schiffers et al. 2012, Matuszewski et al. 2015 by producing 446 new phenotypes. Indeed, we found increased gene numbers in variable environ-447 ments, which allow potentially larger range of possible expressed trait values, 448 and thus more recombination potential. Since genetic linkage did not differ be-449 tween environments, the genome size increase is due to an increased number 450 of linkage units. Species with these larger genomes can be thought of having 451 undergone polyploidisation or ascedent dysploidy. In fact, polyploidisation cor-452 relates with latitude and, arguably, with environmental stress (Rice et al. 2019), 453 but direct tests of this are difficult due to feasibility (Van de Peer et al. 2017). 454 Moreover, increased fecundity also increases adaptation potential as it leads to 455 more recombination in a given time interval. According to our results, the adap-456 tation response to variable environments is mainly characterised by increasing 457 environmental tolerances. However, the changes in genomic traits did not pre-458 vent the general decrease of mean genetic variation in variable conditions, which 459 contradicts results from a previous modeling study (Matuszewski et al. 2015). 460 With more detailed data on the levels of variation, we will attempt to offer an 461 explanation to this in the following section.

462
Study question (b): How do temporally variable environ-463 ments shape phenotypic and genetic standing variation?

464
Having identified survival strategies on a population phenotypic level, we wanted 465 to know whether there are selection patterns on the standing variation within 466 the populations -both at the phenotypic intraspecific and genetic levels. Our 467 results enable us to identify which traits are under increased selection pressure 468 and in which traits species benefit from variation in the different environments. 469 Most traits, such as tolerance for environmental conditions and biomass, were 470 more specialized, i.e., had lower variation, in variable environments at both in-471 traspecific and genetic levels. However, it appears to be beneficial for species to 472 maintain plasticity in dispersal distances when coping with temporal environ-473 mental variation, as evidenced by the fact that dispersal traits, especially long 474 distance dispersal, maintained similar to higher levels of variation.

475
Since variation in our experiments could be increased neither by mutation 476 (Josephs et al. 2017), nor by external gene flow (Cornetti et al. 2016), selection 477 could only act on standing variation. Under these conditions, a higher linkage 478 of genes preserves variation in the associated trait (cf. Teotónio et al. 2009), 479 while low linkage genes allows faster specialization. This differentiated selec-480 tion pressure might explain why we don't see a net change in genetic linkage, 481 because variation and specialization benefit from contrasting degrees of genetic 482 linkage. Specialization in any given ecological function and thus the emergence 483 of different phenotypes could also be facilitated by a low number of loci for as-484 sociated traits . In contrast, phenotypic uniformity might 485 arise from increased number of loci which stabilize variation (Fraser and Schadt 486 2010). Thus, the number of loci represents a potential trade-off between special-487 ization and phenotypic robustness, which might warrant further investigation. 488 These findings suggest that increasing the number of loci could act as a stabi-489 lizing coexistence mechanism by promoting intraspecific competition caused by 490 phenotypic uniformity and thus greater intraspecific niche overlap. In light of 491 this, further experiments should focus on whether phenotypic variation impacts 492 species coexistence negatively (Hart et al. 2016) or if low species numbers first 493 allow higher phenotypic variation (Hulshof et al. 2013).

494
Our results furthermore exemplify that intraspecific and genetic variation do 495 not need to be correlated. In case of mean dispersal distance, phenotypic varia-496 tion increased in variable environments. However, the genetic variation of mean 497 dispersal distance decreased. Thus, the phenotypic variation in mean dispersal 498 distance is due to very different phenotypes, which, in turn, exhibit relatively 499 specialized genotypes. This further stresses the essential role of ecotypes to 500 ensure species survival under changing environments.

501
Limitations and perspectives 502 The fact that our simulations produced low coexistence in terms of the total 503 number of species across the landscape might be a result of too large a trait 504 space in the initial species pool, most of which would be filtered by the relatively 505 narrow environmental conditions. Since the initial species pool was on average 506 350 species large, the probability is also high for it to contain a few strong 507 generalist species, which outcompete other species. On the other hand, an 508 average initial number of 10 species per grid cell means a low probability for one 509 or more species to be sufficiently adapted to the local conditions. Nevertheless, 510 the coexistence level obtained is also in accordance with theoretical expectations, 511 considering that a niche partitioning along the two gradients would explain the 512 average of four species we count in static environments (i.e. one specialized 513 species per environmental gradient combination, see Armstrong and McGehee 514 1980). The filtering is also evidenced by the reduction of trait value ranges over 515 all traits after simulation initialisation (not shown). In fact, additional post-hoc 516 simulations with more constrained initial communities in terms of species traits 517 resulted in a two-fold increase of surviving species numbers (not shown). This 518 did not, however, change the general results. Small-scale disturbance or trophic 519 interactions, e.g., herbivory could further increase coexistence, as theoretical and 520 empirical studies suggest , Chesson and 521 Kuang 2008. But since these processes likely produce additional confounding 522 effects, we chose not to include them in our model at this stage, albeit we identify 523 them as potential directions for further model development. Trophic and other 524 interactions such as mutualism, can have important effects on species survival 525 under climate change (Berg et al. 2010) and even lead to extinction cascades 526 if keystone species get lost (Brook et al. 2008). Since keystone species would 527 be affected by genetic factors in the same way as any other species, our model 528 likely underestimates net species loss effects mediated by genetic factors.
529 Furthermore, our model simplifies complex genetic factors and dynamics 530 which could potentially have confounding effects on resulting patterns. For in-531 stance, linkage between genes in reality is not a binary decision, but rather a 532 consequence of the physical distance between those genes. The larger the dis-533 tance, the higher the probability of crossing over during meiosis. Additionally, 534 genetic architecture is dynamic, especially in plants. Genomes can grow, e.g., 535 by polyploidisation ( Van de Peer et al. 2017), and shrink in size, both of which 536 affects genetic linkage and potentially genetic variation. Since polyploidisation 537 is often a stress response in plants it will arguably affect survival (Rice et al. 538 2019). Subsequent gene loss may then even initiate speciation, therefore provid-539 ing new opportunities for emerging species (Albalat and Cañestro 2016). Our 540 model hence represent the effects of genetic linkage and genome sizes without 541 explicitly considering their respective genetic origins. Nevertheless, our findings 542 on the interaction betweeen genetic and ecological traits call for empirical works 543 identifying the factors that trigger these genomic processes and assessing their 544 evolutionary relevance ( Van de Peer et al. 2017).

545
To make our model and the findings on genomic and ecological traits un-546 der temporal evironmental variation more applicable and relevant to real-world 547 systems, the model could be constrained by real data in further studies. For in-548 stance, the model could be initiated with simulation arenas which can be directly 549 derived from actual landscapes, including environmental conditions (e.g. from 550 Karger et al. 2017). Species-specific parameters could be taken from databases 551 for phenotypic traits (Kattge et al. 2011) and occurence records (GBI) and en-552 riched by genomic information (Dong et al. 2004, Howe et al. 2020 to constrain 553 initial parameter space for the creation of random communities. Thus, our 554 model represents an oppurtunity to integrate different datasets from genomes 555 over traits and occurrences to environmental in a single mechanistic framework. 556 Even in the current state, our model addresses a number of eco-evo-environ-557 mental phenomena (cf. Govaert et al. 2019). The emerging patterns additionally 558 inspire new hypotheses which can be used to guide fieldwork and experimental 559 studies. The consideration of genomic traits, for example, implicates the explicit 560 consideration of new perspectives on biodiversity dynamics during impending 561 climate change (Fig. 5). For scenarios of short-term change of environmental 562 conditions, i.e., warming, lower or increased precipitation and more frequent 563 extreme events, adaptation can only exploit standing intraspecific or genetic 564 variation, rather than novel mutations. Species with high phenotypic variation 565 will likely have good adaptation potential, regardless of genetic characteristics. 566 For species with low phenotypic variation, adaptation potential depends on ge-567 nomic traits. Species that have highly specialized, i.e., uniform, phenotypes, and 568 show little or no genetic variation will only be able to survive rapidly changing 569 conditions by tracking their specific favourable conditions. Fragmented envi-570 ronments or poor dispersal abilities therefore will likely lead to the extinction 571 of those species. Even if species have high genetic variation, genetic architec-572 ture is crucial for their performance. With a high degree of genetic linkage, 573 species might not be able to adapt critical traits in time to react to changing 574 conditions, since a beneficial trait allele might likely be linked to other disadvan-575 tageous trait alleles. Thus, net fitness is unlikely to increase. Low linkage, on 576 the other hand, might lead to species who quickly adapt to new environments 577 as they are not impeded by genetic hitchhiking. However, if linkage is too low, 578 species will also quickly lose genetic variation, rendering them unfit to react to 579 subsequent change. Any conservation measures targeted at particular species 580 should thus consider population structure and genomic traits of species. Hence, 581 while the importance of genetic diversity is already acknowledged in conserva-582 tion biology (Ramanatha Rao and Hodgkin 2002) -additional to functional 583 diversity (Dıáz and Cabido 2001), it is genetic architecture that will determine 584 adaptation success.

586
In this study we have demonstrated complex interactions between genetic and 587 ecological realms by using a simulation model that explicitly considers ge-588 netic architecture of plant communities in changing environments. These eco-589 evolutionary feedbacks broaden our understanding of the role of trait-specific 590 standing variation in species survival and adaptation (Fig. 5). This enabled 591 identifying ecological strategies of species to survive variable environmental con-592 ditions. Variable environments select species with higher tolerances and faster 593 life cycles while species maintain variable dispersal abilities that facilitate track-594 ing favorable environmental conditions. These adaptations are, however, mostly 595 enabled by large genomes, which allow maintaining a high degree of genetic vari-596 ation. Furthermore, we could show that selection pressure differs between traits 597 and that there might even be positive selection pressure to maintain higher 598 genetic variation for dispersal traits.

599
Our findings suggest that genomes are subject to opposing forces -espe-600 cially under changing conditions. While constant environmental filtering impov-601 erishes genomes, there is a selective force to maintain variation in the genome 602 to adapt for future change. This conflict can be mediated to a certain degree 603 by genetic architecture, namely a higher number of genes which allows more 604 genetic variation and a high linkage of loci which impedes the loss of variation. 605 However, traits that need quick specialization require a low number of weakly 606 linked loci. These complex interdependencies of genomic traits may thus further 607 promote the high diversity in genetic architecture and ecological strategies in 608 real-world species.

609
Additionally, our theoretical approach provided potential mechanisms re-610 sponsible for the incongruence of phenotypic and genetic variation, which is 611 sometimes found in nature. A mechanistic link between negative correlation 612 in those types of variation means that special care is called for when inferring 613 genetic variation from phenotypic variation and vice-versa.

614
In summary, this study highlights the importance of genomic traits for the 615 functional assessment of local populations, species and metacommunities. We 616 hope that conservation studies make more use of these characteristics to pri-617 oritize conservation efforts and expect future studies to investigate the genetic 618 architecture of specific traits in natural populations.  The results should be reproducible by using provided simulation codes and con-633 figuration files (see Code availability). Any data and codes associated with this 634 study will be made accessible upon request.