MA lines are generated from a single nearly homozygous individual founder and cultivated via limited effective population number. In the case of A. thaliana, this occurs through single seed descent, resulting in Ne= 1. Thus typical MA line cultivation results in an unbiased sample of mutation effects ranging from deleterious to advantageous, although lethal mutations are excluded (Lynch and Walsh 1998). Note that we do not expect lethal mutations to contribute to genetic variation associated with selection response and lethal mutations should only minimally contribute to community level dynamics. Each MA line accumulates independent spontaneous mutations. After the propagation of a set of MA lines through multiple generations, the genetic differences among the MA lines and between those lines and the founder reflect the input of mutation. Significant MA line effects for multiple traits, including performance and trait measures, were found under both field and greenhouse conditions (Rutter et al. 2010; Roles et al. 2016, Rutter et al. 2018). Each of the MA lines in our experiment is fixed for an average of 20 different sequence level mutations, single nucleotide mutations (SNMs) and indels combined (Ossowski et al 2010, Rutter et al. 2012, Weng et al. 2019). To better understand interaction effects within and between MA lines and founders we are referring to intra-genetic and inter-genetic effects respectively throughout the manuscript.
MA lines and field experiments
We used survival and seed set data of A. thaliana MA lines and the founder as assessed in field experiments in 2004 and 2005 from Rutter et al. (2010, 2012 and 2018) planted in a randomized design (Fig. 1). Rutter et al. (2010, 2012) planted seedlings of 100 MA lines and the founder at the four-leaf stage, approximately two weeks post germination, into a secondary successional field at Blandy Experimental Farm (BEF) in Virginia (39°N, 78°W). Each of the 100 MA lines was used to found up to five sublines to minimize biases due to maternal effects introduced by the specific location within the greenhouse. We founded six sublines from each of the six lines representing the premutation founder genotype. In 2003, subline plants were used to generate all seed utilized in all field experiments. In each planting, 7504 individuals were planted, 7000 individuals of 100 MA lines (70 replicates per MA line, 14 replicates per subline) and 504 individuals of the founder (14 replicates per subline). The planting environment corresponds to a spring ephemeral life-history, where plants germinate and complete the life-cycle in the spring. At the time of planting, vegetation was scant but present. By harvest, the A. thaliana individuals were dwarfed by naturally occurring vegetation, consisting of a mixture of mostly biennials and perennials typical of old field succession. We did not characterize the surrounding vegetation for our plots. Thus, we assess mutational contribution to intraspecific interactions against a background of a varying plant community.
The plot was arranged in 14 spatial blocks with each containing 12 sub-blocks (Fig. 1B) (total plot area approximately 35 x 25 m). Each block included one seedling from each subline and in total 7504 individuals. We used the spatial information of each individual within the described design and created a raster of all plant individuals with R packages (raster, maps, maptools). We used individual-based maps neighborhood matrices with exact spatial and trait information of each genotype. If all five sublines did not produce enough seedlings to distribute in all blocks, seedlings from other sublines of the same MA line were overrepresented in blocks to maintain the same overall number of plants per MA line. Plants dying within the first 3 days of transplant (about 50 plants) were considered to have died from transplant shock and were replaced with another plant from the same MA line. Plants were censused weekly for survival. Plants were harvested by late May, by which time they had senesced. In 2004 a total of 5915 individuals including 394 founders survived. In 2005, a total of 4506 individuals survived including 302 founders. Plants were oven dried and biomass was measured. All fruits produced by each plant were counted and in combination with seed production used as the measurements of seed set. For our analysis, we used measurements of two response variables representing an important part of the life history of A. thaliana to calculate direct neighborhood effects on: a) the survival rate of plant individuals and b) individual seed set from plants that survived and produced fruits.
Statistical Analyses
We use eco-evolutionary trait-based neighborhood models that include intraspecific genetic variation and phenotypic variation expressed by plant trait biomass to analyze competition and facilitation between individuals of A. thaliana. We analyzed plant survival rate and focal seed set measurements from years 2004 and 2005 separately. For each year, we considered all individuals as focal plants in the analysis based on individual-based neighborhood matrices to analyze 1) rate of plant survival to reproduction and 2) seed set of those plants that survived to reproduction. We analyzed neighboring plants to focal plants in a radius of 80 cm (small-scale) or 200 cm (large-scale) of a given focal plant as two spatial scales in the neighborhood analyses (Fig. 1) to quantify selection, competition or facilitation between plants depending on their genotypic and phenotypic variation.
We used extensions of linear mixed models (package lme4, Bates et al. 2014) in R ver. 3.5.2 (www.r-project.org) to conduct neighborhood analyses of focal seed set and survival. We assumed binomial errors for the analyses of plant survival and Poisson errors for analyses of seed set. The mixed models described interactions among plants by including neighborhood indices as explanatory variables at two spatial scales in separate models. Neighborhood indices are spatial density effects of surrounding neighborhood plants that affect focal seed set and survival. For each plant, we used the Euclidian distance between the focal plant and the neighboring plants to compute response effects of within MA line or between MA line neighbors in a given radius around focal plants (Nottebrock et al. 2017b). Moreover, we used a neighborhood index that accounts for the decline of neighbor effects with distance from the focal plant (Uriarte et al. 2010) and summed the amount of biomass from all individuals in a radius of 80 cm or 200 cm respectively by a Gaussian interaction kernel (Lachmuth et al. 2018, Nottebrock et al. 2017, Damgaard 2004). We used random effects of block and subblock to correct for environmental variation between local heterogeneous conditions. Importantly, we correct for between MA line effects by including MA lines as a random effect. That is, we account for the fact that different MA lines may have different effects on a focal plant by treating these as random effects. In addition, including a random slope of biomass on each random intercept corrects for the intraspecific phenotypic variation depending on local conditions (sub-block and block differences) of plant focal individuals. Moreover, the weighted neighborhood density by plant biomass accounts for environmental variation between neighboring plants. A detailed model description of survival and seed set models can be found in S1 supplementary text (Supplementary Material S1).
Neighborhood matrices of all individuals (individual-based maps) were used to analyze the effect of intra- and inter-genotypic neighbors on survival and focal seed set with spatial interaction kernels of neighborhood (plant biomass) density. By incorporating different genotypes and phenotypic variation, we can quantify how important genetic variation is for neighborhood models and if the phenotypic variation explains spatial interactions between individuals. We assume the consequences of genetic differences to be larger between MA lines than between any MA line with itself, or the founder with itself. This is a valid assumption since each MA line differs from the other by approximately 20 + 20 = 40 mutations, while any two replicates within a MA line or the founder lines will differ by one generation, < 2 mutations. On average founders should differ from the individual MA lines by roughly half the difference of the number of mutations differentiating MA lines. Thus, we simulated line effects from parameters derived from MA lines as random effects with the R package ‘merTools’ and the function ‘plotREsim’ in R 2018. Not surprisingly, because the founder performance was always the average of all MA lines, we found no difference between founder and MA line effects on a focal plant (see results). Consequently, in many of our analyses we refer to the founder line and MA lines as just lines.
We used the trait values of neighbors to calculate trait-based neighborhood indices including plant biomass as a trait (Goldberg & Fleetwood, 1987; Goldberg & Landa, 1991; Cahill et al., 2005). We fitted eco-evolutionary trait-based neighborhood models at two different spatial scales for response variables (survival and seed set) for each of the two and both years. To address our objectives, we first analyzed models with differential effect in which within all MA lines and founder had a different effect on survival (A1, A2, Table 1) and seed set (B1, B2, Table1) than between MA line and founder neighbors at small-scale (s) and large-scale (l). In addition, we analyzed models with neutral effects on survival (A1, A2, Table 1) and seed set (B1, B2, Table 1) that included total neighbor density without the split between within and between MA line and founder neighbors at small-scale (s) and large-scale (l). To this end, all models were fitted with two separate neighborhood indices that were calculated from intra-and inter-genotypic neighbors. To justify the inclusion of individual plant biomass as trait-values for interacting plants in the model, we used AICc to compare the models with and without the trait-proxy (Burnham and Anderson 2002). We found that models perform generally better including biomass as a trait-proxy and calculated trait-based neighborhood indices for similar or different genotypes (MA lines) from biomass density (ΔAICc > 2, Supplementary Material S2). All eco-evolutionary trait-based neighborhood models contained random effects of subblock nested in block at block scale and subblock scale on the intercept, MA line identity on the intercept and the focal trait-value (plant biomass) on the slope. Additionally, because direct environmental variables were not measured during the field experiments, we included in each model the individual’s biomass to correct for environmental conditions for spatial autocorrelation. All variables are scaled and centered to assure comparability between predictor variables. Models of differential and neutral effects for 2004 and 2005 (Table 1, A1-A2, B1-B2) are fitted at small-scale (80cm scale) and at large-scale (200cm scale). Hereafter, the 80 cm scale models are referred to as “small-scale” models and the 200 cm scale models are referred to as “large-scale” models. Neighborhood indices, intra- and intergenotypic variation and total variation of biomass density are included as inverse density variables (1/1+density). We compared models of differential and neutral effects through likelihood ratio tests (LRTs).