Data analysis
We used generalized linear mixed-effects models (GLMMs, with the package
lme4, version 1.1-23 (Bates et al. 2014)) in R, version 3.2.3 (R
Core Team 2019) for all models, unless specified otherwise (see Table 1
for final models). We used the emmeans package (Lenth 2020) in R to
calculate estimated marginal means and conduct post-hoc tests.
Chi-square tests and p-values were calculated using the Anova function
in the package car (Fox & Weisberg 2019).
We calculated species richness (i.e. alpha-diversity) of fungal
pathogens in plant roots and plant communities. First, total fungal
richness was calculated as the sum of OTUs present in a taxa/sample
matrix of all fungi that was rarefied to the minimum size per sample
(1,500 sequences). Rarefaction ensures that all samples can be compared
at the same sequencing depth, but did result in the loss of 15
individual plants (out of 490) from the dataset (those which had
<1,500 sequences). Pathogen richness was calculated by
subsetting the rarefied fungal taxa/sample matrix to include only the
taxa identified by FUNGuild as pathogenic to plants. To calculate
pathogen relative abundance in plant roots, we summed the total number
of fungal pathogen sequences found in each individual plants’ roots and
divided that by the total number of fungal sequences in that root. To
calculate pathogen relative abundance at the community level, we took
the ratio of the sum of the total number of fungal pathogen sequences
and the sum of all fungal sequences found in the roots in each
community.
To study partner sharing, we calculated the species-level interaction
network metrics normalized degree and closeness centrality (weighted
closeness) using the R package bipartite (Dormann et al. 2011).
Normalized degree measures the proportion of all possible partners in
the community that an organism interacts with, which we used as an
indicator of generality. Normalized degree was calculated for each 1)
plant and 2) pathogen inhabiting the plant roots. Closeness centrality
measures the degree to which a plant species mediates sharing/indirect
interactions among the community. It was calculated on a unipartite
(plant only) projection of the bipartite (plant-fungal) network, whereby
plant species are connected if they share a fungal partner. High
closeness centrality indicates that a plant species has a short path to
any other plant species in the unipartite network, such that it tends to
share partners with many species or with particular species that share
with many others. Given this sharing, we used closeness here as an
indicator of how much influence a plant could have in a community in
terms of its ability to indirectly affect other species by spreading
pathogens; a negative equivalent to previous findings that closeness
confers indirect benfits in belowground mutualisms (Tylianakis et
al. 2018). Normalized degree of plants was modeled using a normal error
distribution, pathogen relative abundance with a binomial distribution
and normalized degree of pathogens and closeness centrality with a Gamma
distribution. We used the logit link function in the binomial models and
the log link function in all other models.
To test whether exotic plants accumulate fungal taxa that are known to
cause plant disease (Question 1), we quantified how pathogen richness
and relative abundance changed as a function of the proportion of
exotics planted into the community (ranging from 0-100%), our soil
treatment (home or away) and their interaction as fixed effects. To
account for non-independence of plants in the same mesocosm community,
we added mesocosm (pot 1-80) nested in plant community (community 1-20)
in the models as random effects. To test how the generality of
interactions and the potential for interactions to be shared changed
along the exotic plant gradient (Questions 1 & 2), we modeled
normalized degree and closeness centrality (respectively) as a function
of the same predictors described above. As an additional check, we ran
all models again, replacing the proportion of exotics planted with the
proportion realized at the end of the experiment to be sure both model
results corroborated. We did this for each model that tested the effects
of the proportion of exotics planted, and results were consistent
between these two approaches (Table S2).
To understand differences in the way native vs. exotic plants interact
with pathogens, independently of the extent of invasion, we ran a
separate analysis focusing on only the 40 plant communities containing
100% natives or 100% exotic plants. We tested how pathogen richness
and relative abundance, the generality of interactions (i.e. normalized
degree) and the relative efficiency with which plants could potentially
spread pathogens (i.e. closeness centrality) varied as a function of
plant provenance, our soil treatment and their interaction as fixed
effects. We added mesocosm nested in plant community and plant species
as random effects in these models.
To test whether natives vs. exotics differ in the proportion of fungal
pathogens they share (Question 2), we modeled pathogen sharing between
native and exotic plants in each mesocosm community as a function of the
proportion of exotics planted, the soil treatment and their interaction,
with random factors as described above (and in Table 1). We excluded
communities containing both 100% native and 100% exotic plants from
this analysis (i.e. retained communities with 25, 50 and 75% exotic)
because there could be no sharing in those communities. To calculate the
proportion of pathogens that were shared (or not) between native and
exotic plants in a community, we calculated the number of pathogenic
fungal taxa that were identified from 1) only native plant roots, 2)
only exotic plant roots, or 3) both native and exotic plant roots in
each mesocosm pot. We then calculated the proportion of pathogenic
fungal taxa that were shared (between natives and exotics) in a mesocosm
by dividing the number of shared pathogen OTUs by the total pathogen OTU
richness in each mesocosm.
To determine whether exotics shared pathogens with natives more than
would be expected by chance in each community, we simulated a null model
showing how much sharing (as calculated above) we would expect along the
full realized exotic gradient (i.e. proportion of exotic plants at the
end of the experiment) if native and exotic plants were equivalent in
their potential to share pathogens. We ran that model 1,000 times,
randomizing the provenance of plants and randomly shuffling pathogen
reads across plant individuals, to calculate a distribution of expected
sharing. We then used a Monte Carlo simulation to compare that null
model data to our observed molecular data of shared pathogens along the
realized exotic gradient (more details about the null model can be found
in Supplementary methods). We used polynomial regression and fitted a
curvilinear line to model the relationship between sharing and exotic
dominance and modeled the deviation in sharing from the mean of the null
expectation along the realized exotic gradient for each mesocosm.
To test whether pathogen sharing between native and exotic plants
correlates with exotic plant impact (Question 3), we calculated the
relative interaction intensity index (RII, Armas, et al. 2004)
for native and exotic plants and quantified its relationship with the
proportion of pathogens shared between native and exotic plants in the
community. The relative interaction intensity index quantifies the
relative change in plant biomass from what was initially planted to what
was realized at the end of the experiment, using the formula ((realized
proportional biomass-planted proportional biomass)/(realized
proportional biomass+planted proportional biomass)). We used this
formula to measure changes in total native and total exotic biomass
separately in each mesocosm. RII is bounded by 1 and -1, so it controls
for extreme values, with values above zero reflecting realized
proportional growth that was higher than what was originally planted and
values below zero reflecting realized proportional growth that was lower
than what was originally planted.
To test whether plant-soil feedbacks in monoculture predict feedbacks in
a community (Question 4), we evaluated whether plant-soil feedbacks
(i.e. the difference between biomass of a given species in home vs. away
soils) differed between native vs. exotic plants (i.e. interaction
effect between soil treatment and plant provenance) and whether this
difference was influenced by whether a plant was grown in monoculture or
a community. To test this, we modeled 1) plant biomass from the
monoculture experiment as a function of plant provenance, the soil
treatment and their interaction as fixed effects with plant species as a
random effect and 2) plant biomass from the communities as a function of
plant provenance, the soil treatment and their interaction as fixed
effects with plant species and mesocosm nested in plant community as
random effects. While others have presented response ratios to evaluate
plant-soil feedbacks (Brinkman et al. 2010), we were unable to
calculate a response ratio in communities due to differential plant
mortality across treatments.