Statistical analyses
In order to assess whether gene copy number could directly explain the
performance differences of seedlings, partial regression analysis based
on linear mixed-effects models (LMMs) were used to identify the effects
of gene copy number of each defense response on seedling survival rate
at species level and community level using the lme4 package in R 4.0.2
(Bates et al. 2015). For the model at species level, gene copy number of
four defense responses was set as fixed effect and the year interval as
a random effect to account for interannual change. For the model at
community level, the habitat type of seedling stations was also added as
a random effect to account for habitat heterogeneity. The Akaike’s
Information Criterion (AIC) values were used to determine the optimal
model in stepwise backward multiple regression. The partial effect
(r∂ ) as described in Grace et al. (2016) was
evaluated and shown in multivariate partial plots.
The generalized linear mixed-effects models (GLMMs) were used to
evaluate the effects of functional CNV dissimilarities of seedling and
adult neighbors on the survival of focal seedlings. GLMMs with binomial
errors were established for each defense response separately using the
lme4 package in R 4.0.2. The annual survival (alive or dead in May of
second year) of seedlings in all the six year intervals (2009 to 2015)
was the response variable. Eight candidate models were developed with
different factors as fixed effects for each defense responses (Table 1).
The log-transformation of initial seedling height (logHT) was added to
all the models to exclude its effect on seedling survival. S_con and
A_con were set as fixed effects to evaluate the effects of the density
of the local conspecific neighbors, while relative nearest taxon
functional diversity of seedling (S_NTFd’) and adult (A_NTFd’)
neighbors which were calculated as neighborhood functional CNV
dissimilarity were set as fixed effects to evaluate the effect of
heterospecific neighborhood functional CNV dissimilarity on defense
response. Before being added to models, these four variables (S_con,
A_con, S_NTFd’ and A_NTFd’) were standardized by subtracting the mean
of observed value and dividing by one standard deviation. Species and
year interval were added as the random effects to account for the
inherent difference of each species and interannual change in response
to neighbors. Plot and station were also included in the models as
nested random effects to account for spatial autocorrelation. The AIC
values were used to identify the best-fit models for each defense
responses.