Diverse phylogenetic neighborhoods enhance community resistance to drought in experimental assemblages

Although the role played by phylogeny in the assembly of plant communities remains as a priority to complete the theory of species coexistence, experimental evidence is lacking. It is still unclear to what extent phylogenetic diversity is a driver or a consequence of species assembly processes. We experimentally explored how phylogenetic diversity can drive the community level responses to drought conditions in annual plant communities. We manipulated the initial phylogenetic diversity of the assemblages and the water availability in a common garden experiment with two irrigation treatments: average natural rainfall and drought, formed with annual plant species of gypsum ecosystems of Central Spain. We recorded plant survival and the numbers of flowering and fruiting plants per species in each assemblage. GLMMs were performed for the proportion of surviving, flowering, fruiting plants per species and for total proportion of surviving species and plants per pot. In water limited conditions, high phylogenetic diversity favored species coexistence over time with higher plant survival and more flowering and fruiting plants per species and more species and plants surviving per pot. Our results agree with the existence of niche complementarity and the convergence of water economy strategies as major mechanisms for promoting species coexistence in plant assemblages in semiarid Mediterranean habitats. Our findings point to high phylogenetic diversity among neighboring plants as a plausible feature underpinning the coexistence of species, because the success of each species in terms of surviving and producing offspring in drought conditions was greater when the initial phylogenetic diversity was higher. Our study is a step forward to understand how phylogenetic relatedness is connected to the mechanisms determining the maintenance of biodiversity.

The current theoretical framework and evidence suggest that both stochastic 1,2 and deterministic mechanisms [3][4][5][6][7] operate simultaneously on the assembly of plant communities [8][9][10][11] . Abiotic and biotic filters-mostly acting at the regional and the fine spatial scales, respectively-are important drivers of species assembly in drylands 12 , together with facilitation that has been described as an important coexistence mechanism in stressful environments (i.e., the Stress Gradient Hypothesis, sensu Bertness and Callaway 13 ). Plant trait-based community ecology is recognized as an invaluable tool to understand these processes because it provides morphological or physiological trait-based indices in order to identify the role played by each species at the community level in a niche complementarity context 14 . Thus, a species will become part of a realized species assemblage only if it possesses suitable traits to pass through the filters imposed by restrictive environmental conditions and it reduces niche overlap with neighbor species 15 . In the last two decades, the toolbox of community ecologists has incorporated analyses of the phylogenetic patterns of plant communities to understand assembly processes 16,17 . It is evident that historical and evolutionary mechanisms related to migration and speciation are critical for the formation of the regional species pool, but it is not clear how the phylogenetic diversity that describes the degree of relatedness among species can provide information about assembly processes that occur at the ecological time scale 5,18 . A phylogeny should summarize the ecological requirements of coexisting species because it synthesizes the morphological, physiological, and phenological changes in each species throughout evolutionary time in a reduced geographical domain [19][20][21] . However, phylogenetic distance among species could indicate not only niche

Materials and methods
The target plant community comprised annual plant communities on gypsum soils in the Tagus valley, central Spain, which has a semiarid Mediterranean climate with mean annual temperatures around 14.5 °C and mean annual precipitation of 400 mm m -2 year -1 . Precipitation events occur mainly in the late autumn and early spring, and there is an intense summer drought (Aranjuez weather station, 40° 4′ 2″ N; 3° 32′ 46″ W, 540 m). The dominant vegetation comprises gypsophilous dwarf shrubs (e.g., Lepidium subulatum L., Centaurea hyssopifolia Vahl, Gypsophila struthium L., Helianthemum squamatum (L.) Dum. Cours., Thymus lacaitae Pau, Herniaria fruticosa L., and Frankenia thymifolia Desf.) scattered in a matrix of bare soil covered mostly with a biological  (1) If phylogenetic relatedness predicts the intensity of niche overlap-differentiation among species, then plants will coexist more readily in high phylogenetic diversity scenarios due to functional/niche complementarity. Conversely, if phylogenetic relatedness predicts the competitive ability of species, then coexistence will be more likely to occur in low phylogenetic diversity scenarios (i.e., competition symmetry will enhance the coexistence among similar competitors). (2) If functional traits related to water economy are phylogenetically conserved, the response of plants to drought would be more heterogeneous in high diversity assemblages, resulting in a faster decline of species richness. In contrast, if water economy traits in the species pool are convergent among distantly related taxa, phylogenetically diverse assemblages will be more resistant to drought than those formed by close relatives. . Seeds were cleaned and submitted to a light hot thermal shock (15 days at 50 °C) to simulate hot summer conditions to break the seed dormancy. We established 6 experimental scenarios, but we finally maintained 4 of them because two of the scenarios did not fulfill the requirements to enter the experiment (seed germination was not enough at each plot), thus, we finally used 28 species to build the species assemblages (see below). We prepared a common garden experiment with 110 experimental assemblages and more than 7000 seedlings.

Scientific Reports
The experimental design consisted of manipulating the phylogenetic diversity of starting experimental assemblages together with water availability treatments. The plant emergence of species in these communities is highly synchronized, so we prepared different phylogenetic combinations at this early demographic stage for our experimental treatments. In order to select the high and low phylogenetic diversity scenarios, we calculated the phylogenetic species variability (PSV) index because it is bounded between 0 and 1 and it is easily interpretable (PSV ~ 1: high phylogenetic diversity; PSV ~ 0: low phylogenetic diversity) 55  These indices were calculated running the R packages "ape" 57 and "picante" 58 based on the phylogenetic tree for the 28 species involved in the experiment built using "V.Phylomaker" package, using phylo.maker function and the "scenario1" option to bind new tips 59 (Fig. 2). To control for the idiosyncratic effect of species identities, we established two different species combinations for each phylogenetic diversity level. Thus, four taxonomic combinations were constructed Figure 2. Distance-based phylogenetic tree for the 28 annual plant species used to prepare the experimental scenarios. Based on "V.PhyloMaker" package in R. The capital letters between brackets next to the names of species indicate the species combinations in which they participated. In nonbold typeface, the high phylogenetic diversity scenarios (A and B combinations); in bold, the low phylogenetic diversity scenarios (C and D combinations). We established water availability treatments with two levels in a fully crossed factorial design: average precipitation vs. drought. The average precipitation treatment simulated the monthly average rainfall recorded between 1981 and 2010 in the study area, and the drought treatment used 33% of the average rainfall for each month. We established two phylogenetic diversity levels × two taxonomic combinations of species × two water availability treatments (eight experimental scenarios). Each scenario was replicated in 10 to 16 units, thereby resulting in 110 experimental assemblages.
A common garden experiment was conducted in a greenhouse at Rey Juan Carlos University (https:// urjcculti ve. webno de. es/ Móstoles, Madrid, Spain: 40° 20′ 2″ N, 3° 52′ 00″ W, 650 m) from October 2017 when the seeds were sown, until June 2018 when the last individuals were collected. We used round pots with a diameter of 30 cm and height of 10 cm, which were filled with seed-free gypsum soil from a gypsum quarry located close to the collection sites. We aimed to establish 10 plants of each seven coexisting species per pot, so we initially sowed 70 seeds per species in each one. Excess emergent seedlings were removed every two days trying to avoid clusters of seedlings to ensure the planned abundance of each species. By this way, we got to reproduce high densities of ephemeral, highly synchronized annual plants (i.e., around one plant per each 10 cm 2 in average) in our experimental assemblages, close to those often observed in the field, thus promoting the full activation of the interaction net among participant plants. We watered pots to the soil water-carrying capacity for the first 20 weeks to ensure the establishment of experimental assemblages at the emergence stage mimicking natural field conditions and then commenced the water availability treatments, which were maintained for 19 weeks. Between February and June, we monitored plant survival per species and per pot (summing 7700 plants) every two weeks, and we recorded the numbers of flowering plants once a week. In addition, for each species and pot we registered the final number of plants that reached the fruiting stage.
Generalized linear mixed models (GLMMs) were employed to analyze the proportion of surviving, flowering, and fruiting plants per species and pot (Table 1; Appendix 2) and to evaluate the overall proportion of species and plants that survived per pot (Table 2; Appendix 2). We used the irrigation treatment (2 levels: average and drought) and the initial phylogenetic diversity (2 levels: high and low PD) as fixed factors and we included the interaction term between both. Pot identity (n = 110) and taxonomic composition (4 combinations of species were set up in the experimental design: 2 species combinations with high PD and 2 with low PD) were included in the models as random factors. Sampling moment (Time) was considered as the number of days since the beginning of the experiment and used as a covariate (n = 10 for surviving plants and n = 19 for flowering plants). We did not consider the sampling moment to model the proportion of fruiting plants, because this variable was the percentage of the total cumulative number of fruiting plants per species in each pot. We used the "glmer" function in the "lme4" package with a binomial error distribution and the logit link function All statistical procedures were performed in R (4.0.3 version) (R Core Team, 2020). Table 1. Generalized linear mixed models (GLMMs) for the proportion of surviving, flowering and fruiting plants per species and pot. Dependent variables were modelled using binomial error distributions and logit link functions. Pot identity (n = 110) and taxonomic composition (n = 4) were included in the model as random factors. Sampling moment (time) was considered as the number of days since the beginning of the experiment and used as a covariate (n = 10 for surviving plants and n = 19 for flowering plants). We did not consider the sampling moment to model the proportion of fruiting plants, because this variable was just the percentage of the total cumulative number of fruiting plants per species in each pot. Phylogenetic diversity (PD) and water availability (W) were used as fixed factors. Type III Wald Chi-square tests were performed to estimate significance. Number of observations are shown for each dependent variable. Coef. Coefficient, Df degrees of freedom, SE standard error, CI confidence interval, ns not significant. *p < 0.05; **p < 0.01; ***p < 0.001.

Results
The annual plant species that formed the experimental assemblages completed their life cycle within 5 months (Fig. 3a). Plant mortality concentrated between the 2nd and the 3rd month of the experiment since plants died shortly after fruit maturation. Flowering started in the first weeks of the experiment and lasted for nearly four months (Fig. 3b). Our results showed that under low water conditions (33% of the average precipitation treatment), phylogenetically more diverse assemblages favored species coexistence. In particular, in low water conditions, we found that the experimental assemblages formed of distantly related species resulted in more surviving plants per species (Table 1, Fig. 3a) and more species coexisting in each pot along time ( Table 2; Fig. 4a) In addition, under dry experimental conditions more plants flowered (Fig. 3b) and fructified (Fig. 5) per species in distantly related assemblages compared with those that were closely related. Furthermore, plant survival (regardless of species identity) was higher in high phylogenetic diversity assemblages under drought conditions (Fig. 4b). Consequently, the experimental assemblages with high phylogenetic diversity were less sensitive to drought than the low phylogenetic diversity assemblages in terms of the plant survival, number of coexisting species, and numbers of flowering and fruiting plants in each experimental unit.

Discussion
As hypothesized, phylogenetic relatedness among coexisting plants drives community level processes such as survival and reproduction. In particular, we demonstrated the higher resistance of phylogenetically diverse assemblages to drought in terms of plant survival and number of coexisting species over time, and even more, plants not only were able to survive more successfully to drought in phylogenetically diverse assemblages, but also more individuals completed the reproductive stage by setting flowers and fruits. Overall, these results support the idea that phylogenetic relatedness predicts niche differences among species (Hypothesis 1a in Fig. 1), and that drought resistance might be convergent along phylogeny (Hypothesis 2b in Fig. 1).
Our results are consistent with the species niche complementarity concept 4,60 , which predicts that species with differences in terms of their resource use are more likely to coexist due to the reduced competitiveness among them 22,[60][61][62] . Distantly related species are more likely to be phenologically and functionally complementary, and thus to suffer less from the effects of competition compared to living among conspecifics or close relatives in the neighborhood [63][64][65] . Closely related species are likely to have ecologically similar requirements 66,67 , so they would share fundamental niches and be more prone to compete strongly for resources. Several mechanisms could promote niche complementarity, such as phenological differences among species 68,69 , different resource use traits [70][71][72] , or different root foraging activities 73 . Maynard et al. 74 found that phenological differences among the species in a community could affect the competitive dynamics to promote coexistence, thereby possibly leading to an increase in species richness at fine spatial scales, which could promote stabilizing dynamics. In our study, the distantly related species probably differed in terms of their phenology, resource uptake, and physiological efficiency, which could have reduced the intensity of the competitive interactions among them 14 , particularly when water availability is limited in low irrigation treatments This situation could have promoted individual plant survival and species richness, as well as higher plant fitness in high phylogenetic diversity assemblages in drought conditions (see also Ref. 75 ).
Furthermore, our results support the idea that drought resistance is a convergent strategy along phylogeny of annual plant species in our study system (Hypothesis 2b in Fig. 1). Our results also agree with García-Camacho et al. 76 who did not detect phylogenetic conservatism in terms of the rainfall preferences of 111 annual plant species from an aridity gradient in Israel. Thus, the annual plant species in dryland areas have evolved over a long period under strong pressure due to drought events and highly unpredictable rainfall events, which might Table 2. Generalized linear mixed models (GLMMs) for the proportion of surviving species and proportion of total plants per pot. Dependent variables were modelled using binomial error distributions and logit link functions. Pot identity (n = 110) and taxonomic composition (n = 4) were included in the model as random factors. Sampling moment (time) was considered as the number of days since the beginning of the experiment and used as a covariate (n = 10) to statistically control for the effect of time. Phylogenetic diversity (PD) and water availability (W) were used as fixed factors. Type III Wald Chi-square tests were performed to estimate significance. Number of observations are shown for each dependent variable. Df degrees of freedom, Coef. Coefficient, SE standard error, CI confidence interval, ns not significant. *p < 0.05; **p < 0.01; ***p < 0.001.

Fixed effects Df
Proportion of surviving species per pot (n = 1100) www.nature.com/scientificreports/ have resulted in the convergent adaptation of distantly related phylogenetic clades to cope with limiting water conditions. Clearly, a powerful abiotic filter such as severe droughts could have shaped the regional species pool over an evolutionary time scale. Consequently, regardless of phylogenetic relatedness, all the species in the community would be able to cope with water limitation, including when it occurs over an ecological time scale 77 . Nevertheless, our results could also concur with the stress gradient hypothesis, which states that facilitation is more relevant than competition for the assembly of species under stressful abiotic conditions 13 . In that case, the clades adapted to withstand drought could have improved the micro-environmental conditions in their close neighborhoods, thus favoring survival and fertility of distantly related less tolerant clades 35 . However, to the best of our knowledge, this has not been tested with annuals at the small spatial and time scales monitored in our experimental setup, where all the species are very small (less than 15 maximum height) and had very synchronous  Table 1). Black lines represent high phylogenetic diversity (PD) scenarios and grey lines denote low phylogenetic diversity scenarios. Solid lines represent the average precipitation treatments based on natural precipitation for 30 years, and spotted lines denote drought treatments (33% of the average precipitation). Vertical bars represent the standard error.  78 , for positive plant-plant interactions with a large annual nurse plant in a field study). Assembly processes that occur at the community level may affect the evolution of species over the long term (see also Refs. 21,79,80 ). Coexistence among closely related species can trigger character displacement to reduce competition intensity 81 or character convergence to reduce competition asymmetry 52 . Community processes seem to exert crucial effects on evolution 79 , and provide a plausible explanation for the intriguing question regarding how so many annual plant species can coexist in the harsh conditions of semiarid gypsum systems. Based on this, we suggest that the species assembly occurring at the finest spatial scales where species interact, could a be major driver of the high taxonomic diversity observed at larger spatial scales in our study system (more than 120 annual species 43 ). In this line, our findings support the "united we stand" framework 82 , which states that rare species might remain in assemblages by establishing diffuse positive interactions. Such "diffuse  Table 2). Black lines represent high phylogenetic diversity (PD) scenarios and grey lines denote low phylogenetic diversity scenarios. Solid lines represent the average precipitation treatments based on natural precipitation for 30 years, and spotted lines denote drought treatments (33% of the average precipitation). Vertical bars represent the standard error. www.nature.com/scientificreports/ positive interactions" could be related to high phylogenetic diversity among neighboring plants at fine spatial scales (see Ref. 83 ).
In conclusion, our results agree with the existence of niche complementarity and the evolutionary convergence of water economy strategies as major mechanisms for organizing annual plant assemblages in semiarid Mediterranean gypsum habitats. Species that grow in assemblages of distantly related species are more likely to survive and fructify in drought conditions than those that grow in closely related ones. Intense droughts occur often in semiarid Mediterranean ecosystems 12,47 and their intensity is expected to increase in the future 84 . Furthermore, the United Nations recently declared the next decade as "The Decade on Ecosystem Restoration" (United Nations Environment Programme, 2019), and they remarked the importance of understanding the drivers of community assembly. In this context, our study is timely since restoration strategies are moving beyond traditional restoration actions and adopting new tools that better describe the characteristics of species assemblies. In this way, we demonstrate that phylogenetic diversity is an excellent measure that can be used to understand species assembly processes. Furthermore, our study highlights that experimental approaches can provide new answers to old questions in community ecology by connecting assembly processes and patterns in a more robust causal framework. We show that species rich annual plant communities are excellent model systems for such investigations, due to the feasibility of manipulating species assemblages and the short time lapses needed to account for a complete generation.  Table 1). Black bars represent high phylogenetic diversity scenarios and grey bars low phylogenetic diversity scenarios. Average precipitation treatments are based on natural precipitation for 30 years in the field and drought treatments represent 33% of the average precipitation. Vertical bars represent the standard error.