Supporting Information for

0 downloads 0 Views 1MB Size Report
(at 0 h) or the invading species (at 48 h) by plating 50 µl of the initial ... (t=48 h) or when invading (t=96 h), we plated 1 µl of nectar from each well (diluted 1:50) on. YMA plates, incubated plates at 27 ºC for one week, and counted CFU. ... enough time for nectar yeasts to reach equilibrium by assessing growth curves for a ...

Supporting Information for Applying modern coexistence theory to priority effects Tess Nahanni Grainger, Andrew D. Letten, Benjamin Gilbert and Tadashi Fukami

Corresponding author: Tess Grainger Email: [email protected]

This PDF file includes: Supplementary methods Main experiment Supplementary experiments Supplementary analyses Supplementary results Tables S1 and S2 Figures S1 to S9 References for SI citations

1 www.pnas.org/cgi/doi/10.1073/pnas.1803122116

Supplementary methods Main experiment Smooth and bumpy strains of Metschnikowia reukaufii We have identified two morphologically, functionally, and genetically distinct strains of M. reukaufii (1), which we call Smooth and Bumpy based on the appearance of their colonies (Fig. S9). The distinct colony morphologies of these two strains arise from differences in the prevalence of chlymadospores (2) relative to vegetative cells (1). The function of chlymadospores is unknown, but may include protection from harsh environments, and these two strains have shown consistent functional differences in growth rates and responses to sugar and amino acid concentrations (3). Initial inoculations Four days prior to the start of each experiment, we streaked each species on yeast malt agar (YMA) plates from stock solutions stored at -80 ºC. On day 1 of the experiment, we used a hemocytometer to create suspensions containing each yeast species at known total cell density. As this total cell count contains both viable and nonviable cells, and the proportion of viable cells varies across species, we created higher density solutions to account for nonviable cells, and based these densities on the ratio of viable:nonviable cells for each species estimated from previous work. We then determined the population size of viable yeast cells of the initial species (at 0 h) or the invading species (at 48 h) by plating 50 µl of the initial microcosms (diluted 1:10) on YMA plates, incubating plates at 27 ºC for one week, and counting the number of colony forming units (CFU). We used this CFU to calculate an initial population sizes of viable cells at introduction (in CFU/µl) for each species in each of our four temporal replicates (mean 78 viable

2

cells ± 6 SE). We used these cell densities as the starting population sizes in our growth rate calculations. Population estimates To determine the population sizes of each yeast species after 48 h of growth in monoculture (t=48 h) or when invading (t=96 h), we plated 1 µl of nectar from each well (diluted 1:50) on YMA plates, incubated plates at 27 ºC for one week, and counted CFU. To verify species identification based on colony shape on plates containing two species (i.e., plates used to estimate population sizes at 96 h), we examined cell morphology from a subset of colonies under 40 × magnification. Our four species had distinguishable colony shapes (Fig. S9), and previous research has shown that colony morphology and cell structure can be reliably used to distinguish yeast species (4). Model assumptions Two related assumptions of the models that we used to calculate fitness and stabilizing differences are that the first species is at equilibrium when the second (invader) species is introduced, and that the complete elimination of shared resources (‘resource extinction’) by the first species does not prevent the invasion of the second species (5). These assumptions necessitate a careful choice of time interval between the introduction of the first and second species, as well as repeated resource input into the system (6). We ensured that 48 h is generally enough time for nectar yeasts to reach equilibrium by assessing growth curves for a subset of our treatments that we created using instantaneous growth rates from a recent experiment (described below in Supplementary Analyses, Fig. S6) (7). In some of our treatments, populations did increase in abundance after 48 h (Fig. S5), so we tested the sensitivity of our results by reanalyzing our data with these treatments removed (described below in Supplementary Analyses,

3

Figs. S7 and S8). To ensure that our results would not be affected by resource extinction, we replenished resources throughout the experiment (Fig. S2), confirmed that population sizes of the first species were not decreasing between 48 and 96 h (Fig. S5), and conducted an amino acid addition experiment to test whether the priority effects we observed were robust to additional resources supplied during the invasion of the second species (described below in Supplementary Experiments, Figs. S3 and S4). We also wanted to ensure that if our estimates of fitness and stabilizing differences were sensitive in any way to the timing of introductions, we would capture the most biologically realistic competitive outcomes. We selected 48 h as a realistic time between yeast introductions because in Diplacus aurantiacus, the hummingbird-pollinated plants from which we isolated our yeast, this is a realistic time between pollination events (8) and has been used previously as the time lag in competition experiments in this system (4, 9). This design also created a biologically plausible total length of our experiment (four days), given that these species’ habitat (D. aurantiacus flowers) only lasts an average of seven to ten days (8). One challenge in estimating intrinsic growth rates is that the time elapsed between our initial and final population size counts may have allowed for some intraspecific densitydependence. This would cause an underestimation of growth rates, and would be most likely to affect monoculture growth rates of the most rapidly growing species in the most favorable conditions (i.e. Bumpy, Fig. S6a,b). Slower-growing species (e.g., Gruessii, Fig. S6c,d) and species growing in less favorable environmental conditions are less likely to have reached carrying capacity quickly enough to bias growth rate estimates. If Bumpy’s monoculture growth rate were underestimated, this would cause an underestimation of Bumpy’s sensitivity to competition, and thus an overestimation of both fitness differences and stabilizing differences for

4

species pairs containing Bumpy (Eqns. 3 and 4). However, there was no indication that species pairs that included Bumpy were biased toward the high FD, high SD region of Fig. 2a. More importantly, any underestimation of monoculture growth rates would not alter the competitive outcomes that we observed. Finally, we note that the models we used cannot be used to assess coexistence in the special case in which one species is unaffected by its competitor and so has a sensitivity of zero (e.g., 𝑆𝑖,𝑗 = 0). This could occur if a species releases a chemical that suppresses heterospecifics but not conspecifics. In this case, 𝐹𝐷 = 𝑖𝑛𝑓𝑖𝑛𝑖𝑡𝑦 and 𝑆𝐷 = 1, and the mathematical criterion for coexistence is 𝐹𝐷 > 1/(1 − 𝑆𝐷), which in the above scenario becomes infinity > undefined. While this situation did not occur in our experiment, it could pose a challenge to future research evaluating coexistence in systems mediated by interactions other than resource competition (see Discussion). Data analysis To determine the effect of our pH and sugar treatments on the performance of our focal species, we ran a linear model with pH, sugar, and species identity as predictors and monoculture growth rates as the response. The method that we used to calculate fitness and stabilizing differences includes a calculation of each species’ sensitivity to competition (Eqn. 2). This metric, as defined by Carroll et al. (5), indicates the degree to which a species’ per capita growth is reduced when invading a population of its competitor, compared to when it is grown in monoculture. We note that this definition of sensitivity differs from its use by Meszéna et al. (10) to indicate the degree to which a change in a regulating variable (e.g., resources or predation) alters a species’ intrinsic growth rate. We note also that this metric includes a ratio, which biases estimates calculated from relatively small sample sizes (11). We therefore used bootstrapping to determine the bias in our 5

sensitivity estimates caused by the ratio. In short, we sampled (with replacement) the same number of observations in our experiment (8 for most treatments) to generate 1000 datasets from which estimates of sensitivity were calculated. We used the distribution of sensitivity values to determine the percent bias on each of our original sensitivity estimates. We found that ratio estimation caused minimal bias: the average bias was 0.77%, most (86%) bias values were under 1%, and none were above 5%. The models that we used to calculate fitness and stabilizing differences assume positive growth of both species in monoculture (5). Whenever a species’ growth in monoculture is negative it cannot persist even in the absence of competition, and fitness and stabilizing differences do not apply and cannot be calculated because one species’ sensitivity is negative and imaginary numbers are produced. We therefore excluded all replicates for which growth in monoculture was negative for at least one species, which left a minimum of four remaining replicates for all treatments. In addition, the four species pair-by-environment combinations for which the average monoculture growth for one species across all replicates was negative were excluded from all analyses (the four NA values in Table S2). Our analyses of the effect of pH and sugar treatments on competitive outcomes, the distance to neutrality, and fitness and stabilizing differences had N = 6 species pairs per environment, except for two environments for which two species pairs were removed due to negative growth in monoculture (Table S2). For all analyses, we started with the most complex model, tested for interactions using a likelihood ratio test (‘drop1’ function), dropped all non-significant higher order interactions, and re-tested the simplified model. We present the highest order significant interaction or significant main effects from each model. Where necessary, we included the ‘weights’ function in the

6

‘nlme’ package to account for heteroscedasticity of variance among treatments. Data were logtransformed as needed to meet model assumptions.

Supplementary experiments Spent nectar experiment To determine the impact of our focal species on the nectar environment, we grew each of our four species in monoculture (plus a no-yeast control) in each of six nectar environments for 48 h and measured the change in nectar pH and total sugar concentration (sucrose + glucose + fructose) after 48 h (N=3 replicates per species). We used pH strips to measure acidity due to the small volume of liquid available for testing (EMD Millipore ColorPhast Strips, MA, USA) and a handheld refractometer to estimate total sugar content (Bellingham and Stanley Ltd., GA, USA). To determine the effect of our focal species on nectar pH and sugar after 48 h of growth in monoculture, and whether this effect varied across nectar environments, we ran two separate linear models. The model for pH had species identity (control, Bumpy, Smooth, Gruessii, or Rancensis) and nectar environment (3 pH levels × 2 sugar levels = 6 environments) as predictors and final pH as the response. The model for sugar had species identity and nectar environment as predictors and final sugar concentration as the response. Amino acid addition experiment To determine the robustness of our results to variation in resource availability, we conducted a separate amino acid addition experiment. This experiment followed the same experimental protocol as the main experiment except that we added 1 µl of one of 12 concentrations of additional amino acids (see x axis of Fig. S3) at 48 and 72 h. We did this by replacing 1 µl of nectar in each well with 1 µl of fresh nectar that had the appropriate concentration of amino acids

7

and the same sugar content as that well. We conducted this experiment on a subset of our species and environment combinations: Bumpy invading Gruessii in the three high sugar environments, and Gruessii invading Rancensis and all six nectar environments. We only tested one-way invasions, in which the dominant species was invading the weaker species in each pair, because amino acid additions were most likely to facilitate successful invasions in this scenario. Our highest amino acid addition level brought nectar in microcosms to 50 mM, which is likely >15 × higher than the amino acid concentration of amino acid-rich natural nectar. We replicated each treatment twice, so that each species pair-by-environment combination had a total of 24 possible invasions (12 amino acid addition treatments × 2 replicates). To determine whether the strength of our observed priority effects predicted their stability (robustness to resource addition), we ran a linear model with the strength of priority effects in the main experiment as the predictor and the stability of priority effects in the resource addition experiment as the response. We quantified the strength of priority effects in the main experiment as the average distance of each species pair-by-environment combination to the line separating priority effects from competitive exclusion, divided by the average standard error on SD and FD measurements (so that larger errors were quantified as weaker priority effects). We quantified the stability of priority effects in the resource addition experiment as the number of unsuccessful invasions (cases where µi < 0), out of a possible 24, when amino acids were added at 48 and 72 hours. Unsuccessful invasions indicate that amino acid addition did not change the competitive outcome from priority effects (neither species can invade) to competitive exclusion (one species can invade).

8

Supplementary analyses Growth curves To ensure that 48 h is sufficient for nectar yeasts to reach equilibrium, we estimated instantaneous growth rates (r) for a subset of our species-by-environment combinations (Bumpy and Gruessii at high pH low sugar and high pH high sugar) using data from a previous study by Letten et al. (7) in which population sizes were measured at 0, 12, 24 and 48 h. We used these estimates of r, along with initial population sizes and carrying capacity (K) from the current study, to generate logistic growth curves (Fig. S6). We modeled population growth as logistic because this provides a good representation of yeast population dynamics and is a widely used approach in microbial ecology (Fig. S6) (12-14). These curves predict that 48 h is sufficient time for these species to reach carrying capacity (Fig. S6). Re-analysis excluding treatments that had not reached carrying capacity by 48 h To ensure that our main results were robust to the exclusion of treatments in which yeast populations were still growing between 48 and 96 h, we re-analyzed our data with these treatments excluded. We first determined which treatments (species-by-environment combinations) had yeast populations that were increasing between 48 and 96 h. We did this by calculating the log ratio of population sizes at these two times for each replicate (log(N_96h+1)/(N_48h+1)) and running a one-tailed t-test for each treatment to determine whether the log ratio was significantly different from zero. This amounted to 24 t-tests (4 species X 6 environments), so we used a Bonferroni correction that adjusted our cut-off for significance to P > 0.002. Out of 24 species by environment combinations, eight had significantly higher population sizes at 96 h, two had significantly lower population sizes at 96 h, and 14 did not have significantly different population sizes at 48 vs. 96 h. We then removed the 14 species pair by

9

environment treatments (out of the 32 included in our original analyses) in which either resident species was increasing between 48 h and 96 h and used the remaining 18 treatments to re-run our main analyses.

Supplementary results Monoculture growth rates The effect of pH on yeast performance in monoculture varied across sugar treatments and species (significant pH × sugar × species interaction, P < 0.001; Fig. S1), but for Bumpy, Smooth, and Gruessii, growth rates tended to increase with increasing pH (Fig. S1). Sugar had a more variable effect, with high sugar levels benefitting Gruessii, benefitting Bumpy and Smooth when pH was low, and harming Rancensis when pH was low or medium (Fig. S1). Spent nectar results After 48 h of monoculture growth, all yeast species lowered pH slightly compared to control nectar containing no yeast (mean pH reduction of 0.3). There was an interactive effect of species and nectar environment on the lowering of pH, as Smooth lowered pH more than other species, but only in the high pH, high sugar environment (species X treatment interaction P