1 Supplemental Material 2 Plant traits and plant-soil feedback (PSF) 3 The magnitude and the sign of PSF tells us little about a plant's ability to change soil values 4 (construct a niche) or how adapted it is to unconditioned soil. Therefore, to illustrate this 5 somewhat complex dependence of PSF on the model parameters, Figures S1 and S2 show the 6 values and magnitude of PSF for two cases, when a plant is and is not adapted to soil 7 conditions. In Fig. S1 a plant performs best in unconditioned soil (x1 = θ0), and in Fig. S2 a 8 plant performs poorly in unconditioned soil to indicate the importance of niche construction. 9 To visualize positive and negative feedbacks, we chose a “home” plant with a particular 10 ecological trait (x1) living in a particular unconditioned soil (θ0) and plotted how the sign and 11 magnitude of PSF vary with respect to different values of the ability of the “home” and 12 “away” genotype to change the soil ( f(y1) and f(y2) ). We chose f(y1) and f(y2) rather than y1 13 and y2 because f(y1) and f(y2) represent how much a soil changes due to niche construction 14 and can be measured in empirical studies, while y1 and y2 are the values of niche construction 15 traits themselves and are usually hard to measure. 16 Fig. S1 shows a plot in a case when x1 = θ0. In that case a “home” plant's ecological 17 trait perfectly matches the unconditioned soil value. Figure S1A shows combinations of f(y1) 18 and f(y2) for which PSF will be negative (grey areas) or positive (white areas), while figures 19 S1B, S1C and S1D show the magnitude of PSF for a specific f(y1) value. If f(y1) = 0, (Figure 20 S1C) the “home” plant does not change the soil and the germination probability of the “home” 21 plant in “home” soil will be the highest possible. If the “away” genotype can change an 22 unconditioned soil value ( f(y2) ≠ 0), the “away” soil will differ from “home” soil and there 23 will be mismatch between the “home” genotype and the “away” soil which will cause lower 24 germination probability of the“home” seed in “away” soils. As a result the sign of PSF will 25 be positive. The magnitude of PSF will depend on how much the “away” genotype changed 1 26 the soil. If the “home” plant can change the soil ( f(y1) ≠ 0 ), PSF can be positive or negative 27 (Figure S1B and S1D). Because f(y1) ≠ 0, the “home” plant will mismatch the “home” soil (x1 28 ≠ f(y1) + θ 0). If the mismatch between the “home” plant and the “home” soil is bigger than 29 between the “home” plant and the “away” soil, the “home” plant will have a higher 30 germination probability in the “away” soil, which will result in negative PSF. A smaller 31 mismatch between “home” plant and “home” soil than “home” plant and “away” soil will 32 create positive PSF. 33 Fig. S2 shows an example when x1 = 0.1 and θ0 = 0.7. Since x1 ≠ θ0, this case 34 corresponds to a situation when the “home” plant is not adapted to the (unchanged) soil value 35 in the cell. Having the ability to change a soil value might increase the germination rate of 36 seed if the soil change is in the right direction, and when tested, such a plant would display 37 strong PSF, most of which would be positive (Figure S2D). Having the ability to change a 38 soil might also further decrease the germination rate (Figure S2B), and such a plant would 39 also display strong PSF, most of which would be negative. 40 2 41 42 Figure S1. Strength of plant-soil feedback (PSF) when the x trait of the “home” plant (with 43 genotype x1, y1) perfectly matches the unconditioned soil (x1 = 0.5, 0 = 0.5, = 1.0, =0.05). 44 The sign of PSF is shown on S1A. For some combinations of f(y1) and f(y2) the sign of PSF 45 is negative (grey areas), while for others it is positive (white areas). Other plots show the 46 magnitude of PSF when the “away” plant can change soils by 0.4 (B), 0 (C) or -0.2 (D). 3 47 48 Figure S2. Strength of plant-soil feedback (PSF) when the x trait of the “home” genotype 49 does not match an unconditioned soil (x1 = 0.1, 0 = 0.7, = 1.0, =0.05). As is the case 50 with Figure S1, for some combinations of f(y1) and f(y2), PSF are negative (grey areas), while 51 for others they are positive (white areas). The other panels show the magnitude of PSF when 52 the “away” plant can change soils by 0.4 (B), 0 (C) or -0.2 (D). 4 53 Individual-based Simulations 54 We built a spatially explicit individual-based model to examine the roles of niche 55 construction, selection and seed dispersal on the divergence of plant traits. In particular, we 56 examined if and under what circumstances plant-soil feedbacks would lead to divergence of 57 both the niche construction and ecological traits. Equations and model descriptions (not in 58 main text) describing each parameter are described below. Plant-soil feedback in this model 59 appears because soil change affects the fitness of a genotype that will be present in the cell 60 map in the next generation. 61 Detailed Model Description 62 Space 63 We modelled evolution on a 2D square grid (map). The map consists of L by L cells, where 64 each cell can contain at most one adult plant. Each cell is characterized by a “soil value” (θ0) 65 which creates a linear gradient ranging from 0 to 1 with respect to width of the map. Different 66 values of θ0 represent one or multiple soil features that are related to seed germination (e.g., 67 soil pH, moisture, concentration of limiting nutrients, etc.). 68 Plants 69 Each plant is characterized by two independently evolving quantitative traits, an ecological 70 trait (x) and niche construction trait (y). Trait x is a trait responsible for seed germination 71 probability (e.g., pH sensitivity), and is under stabilizing selection, while trait y is the trait 72 related to the ability of plant to modify its soil (construct a niche). 73 Life cycle 74 Generations are overlapping (effectively, this is same as simulating annual plants). During 75 one generation: 1) each seed disperses to a neighbouring cell with probability ms or lands in 5 76 the cell with a parent plant with probability 1 – ms; 2) seed germinates with the probability 77 that depends on how well their ecological trait matches the soil value; 3) if multiple seed 78 germinates in one cell, seedlings within the cell compete for survival, and only one (randomly 79 chosen) survives to adulthood; 4) adult plants produce hermaphroditic flowers; 5) flowers are 80 pollinated by either neighbouring plants or by self-pollination; 7) adult plant produces seed; 81 and 8) adult plants change the soil based on the strength of its niche construction (y) trait. 82 Germination probability 83 For each cell, there is an optimal value of x, for which a seed will have the highest chance to 84 germinate in that cell, and departing from optimal value will result in a lower germination 85 probability (Figure S3). 86 The probability to germinate (pg) in a cell with soil type θ (which depends on 0 and y trait, 87 see Niche construction section below) is given by a bell-shape curve: 88 89 pg = exp( - (x – θ)2/) 90 (1) 91 By changing the parameter we can change how severely departures from optimum reduce 92 the pg. When is small, small differences in the value of x will have a large effect on the 93 probability of germination, while if is large, even large differences in the value of x will 94 have small differences on the probability of germination. 6 95 96 Figure S3. Seed germination probability (pg) for seeds with different x trait values in cells 97 where soil value () is 0.5 (top) and 0.3 (bottom). When selection on x is strong (full line, = 98 0.05) germination probability decreases more as x departs from the optimal value (which is 99 equal to ) than when selection is weak (dashed line, = 1). 100 101 Niche construction 102 The soil value for each generation in each cell can change due to the fact that the effect of 103 niche construction depends both on the plant's y trait and the unconditioned soil value. The 104 soil value after niche construction () is equal to: 105 θ = θ0 + f(y) 106 (2) 107 Where θ0 is the unconditioned soil value (soil value of cell that had no plants growing in it) 108 and f(y) is the effect of niche construction. We use a linear function to model the dependence 109 of f(y) on y: 7 110 111 f(y) = β (y – ½) 112 (3) 113 Parameter β determines how much the soil is changed by changing y ( is the slope of the 114 line), as well as the maximum amount the soil can be changed due to y. Since y ranges from 0 115 to 1, values of f(y) range from - β /2 to β /2. Table S1 shows f(y) for several values of β and 116 y. 117 118 Table S1. Change of soil due to niche construction (function f(y) = β (y – ½)) for different 119 values of y and β . If β is small, even extreme genotypes (genotypes close to 0 or 1) won't be 120 able to significantly change a soil. Values of β 121 122 Values of y 0.1 0.5 1 2 123 0.1 -0.04 -0.2 -0.4 -0.8 124 0.5 0 0 0 0 125 1 -0.05 0.25 0.5 1 126 127 Flower production, pollen dispersal, pollination and seed production 128 Each plant produces a number of flowers according to a Poisson distribution with mean . To 129 model pollination, we assume that pollen can land on a flower stigma in home and adjacent 130 cells with probability pp. If pollen from multiple plants land on the same stigma, only pollen 131 from one (randomly chosen) plant fertilizes the flower. Self-pollination is allowed. A flower 8 132 can produce seed only if it has been pollinated. The values of the x and y traits of each seed 133 are equal to the mean value of their parent's x and y plus a small number taken from a normal 134 distribution with mean 0 and standard deviation 0.01. Adding this small number represents 135 the effect of mutation. We also assume that trait x or y can experience a mutation of large 136 effect with probability . If such mutation happens, a trait value of a seed is equal to the 137 mean parent's value plus a number taken from a normal distribution with mean 0 and standard 138 deviation 0.1. 139 Parameter choice 140 We manipulated selection on trait x (), parameter in f(y) that controls how the 141 value of the y trait affects the change in soil ( Eq. 3) and seed dispersal probability (ms). The 142 following parameters were constant in the simulations: the grid size was 100 x 100 cells, pp = 143 1.0, = 10-4 and = 4. The soil gradient was linear with respect to the x axis (all cells with 144 the same x coordinate had the same θo value), and the difference between environmental 145 values of two adjacent cells was 1/99. With this design the first column had θ0 = 0 and the 146 last θ0 = 1, while all the other columns had θ0 ranged between 0 and 1. 147 For each manipulated parameter we ran 100 individual based simulations for three 148 different values (low, intermediate, high): = { 0.05, 0.1, 0.2}, = { 0.0, 1.0, 2.0} and ms = 149 {0.1, 0.5, 1.0}. The value = 0.05 represents strong selection, 0.1 intermediate and 0.2 weak. 150 β = 0.0 corresponds to the case of no niche construction ability, while β = 2.0 corresponds to 151 strong niche construction ability. We begin the simulation by filling the right most column of 152 the grid (θ 0 = 1.0) with plants that have x = 1.0 and y = 0.5. Such plants are perfectly adapted 153 to the right of the cell and they cannot modify the soil value. With this initial condition, we 154 are simulating introduction of a small population to a new geographical region. By making 155 individuals perfectly adapted to the right of most cells, we start the simulation under 9 156 conditions that do not necessarily favour the evolution of a niche construction trait, since 157 plants are perfectly adapted to their soil value (x1 = θ 0) and having a y value different from 158 0.5 decreases germination rate. The results reported in the paper are based on the mean values 159 of 100 runs for each parameter set collected at generation 1,000 and 10,000. 160 Population dynamics. 161 During the course of simulation with niche construction, plants spread from the point of 162 introduction (right part of the map) across the map, and by generation 1,000 the whole map is 163 populated by plants that differ with respect to their x and y traits (Figure S4). Plants can 164 spread across the map only if they are able to germinate in soils of different values (local 165 adaptation). Adaptation can be accomplished by a change in x only, by changing the soil 166 value to look more similar to the soil where the seed originated or by a combination of those 167 two. Simulations suggest that adaptation is driven by a combination of both changing y and x 168 traits. Figure S5 shows a single simulation run when there is no niche construction (β = 0.0), 169 in which case plants' spread across the map is accomplished only by adapting an ecological 170 trait to local soils. 171 10 172 173 174 175 176 177 178 179 180 181 182 183 Figure S4. Snapshots of 100 x 100 map showing the spatial distributions of x, y and soil θ 184 traits at three time steps during the simulation (at generations 1, 500 and 1,000), with niche 185 construction. Top row: ecological trait (x). Middle row: niche construction trait (y). Bottom 186 row: soil value (θ). Different colours represent different mean soil gradient values, ranging 187 from 0 (dark grey) to 1 (white). Divergence on an ecological trait is larger than divergence on 188 a niche construction trait. Plant-soil feedbacks are more likely to be observed between plants 189 with more different y values than between genotypes with similar y values. 11 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 Figure S5. Snapshots of 100 x 100 map showing the spatial distributions of x, y and soil θ 218 traits at three time steps during the simulation (at generations 1, 500 and 1,000), with niche 219 construction. Top row: ecological trait (x). Middle row: niche construction trait (y). Bottom 220 row: soil value (θ). Different colours represent different mean soil gradient values, ranging 221 from 0 (dark grey) to 1 (white). Labels and parameters are the same as on Figure S4, except 222 = 0. The distribution of x follows the soil gradient () more closely than on Figure S4, when 223 there is niche construction. Trait y is less diverged and it evolves neutrally. 224 225 12 226 227 228 Fig S6. The mean trait x value across the soil gradient after 10,000 generations is very similar 229 to the situation after 1,000 generations (Fig. 2 in text) suggesting that differences in 230 divergence between cases with and without niche construction can persist over time. See 231 Figure 2 for complete description. 13 232 233 Figure S7. The mean trait y value across the soil gradient after 10,000 generations is very 234 similar to the situation after 1,000 generations (Fig. 3 in text) suggesting that differences in 235 divergence between cases with and without niche construction can persist over time. See 236 Figure 3 for complete description. 237 238 14