Statistical analysis
Statistical analyses were done within R version 4.2.2 (Team R
Development Core, 2022). First, we performed a principal component
analysis (PCA) on above- and below-ground plant traits where values of
individual traits were standardized using z-transformation by using the
function PCA from the FactoMinerR package(Csardi & Nepusz 2006). The
explained variance from the first two PCs and individual plant trait
loading weightage on them were extracted. To test the contribution of
PCA loadings, a combination of a threshold selected using the number of
dimensions (Richman 1988) and a bootstrapped eigenvector method
(Peres-Neto, Jackson & Somers 2003) were used. For plant traits and
bacterial diversity indices (ASV Richness, Shannon diversity, and
Pielou’s evenness), we performed linear models to test the main effects
of domestication (wild versus modern barley), soil microbiome (DSM
versus NSM), and their interactions. The step-wise data exploration
protocol from (Zuur, Ieno & Elphick 2010) was followed to avoid common
statistical errors in which, the mean-variance relationship from
residual plots was visually checked. Measured plant and bacterial
variables are presented as means with 95% confidence intervals that
were computed by using a non-parametric bootstrap resampling with 10,000
iterations. The 95% confidence intervals are referred to as
compatibility intervals (95% CI), henceforth (Amrhein & Greenland
2018; Berner & Amrhein 2022). Given the practice of usingp -values (α = 0.05) as dichotomous to test the null hypothesis
and to favor ‘statistically significant’ over ‘non-significant’ results,
we refrain from using the above-mentioned terms and mostly mentioned the
mean differences between treatments and effect sizes wherever possible
while interpreting our results (Halsey 2019; Berner & Amrhein 2022).
We performed network analysis to investigate how multiple plant traits
are interacting. In the network analysis, traits are assigned as nodes
and their connections as edges. We extracted network parameters such as
edge density, diameter, distance, and modularity which have clear
ecological significance (Flores-Moreno et al. 2019). For
instance, edge density is the ratio of present to total possible
connections, ranging from 0 to 1 and traits with higher edge density are
considered more efficient. Modularity determines connectivity among
trait modules where trait networks with higher modularity have tighter
traits within than between modules. Trait networks with shorter diameter
and mean distance imply stronger coordination among various traits. For
the trait-trait relationship, data was log-transformed followed by the
calculation of correlation coefficients for both wild and modern barley
separately for DSM and NSM. Trait network analysis was described by
significant correlation coefficients and illustrated by using the igraph
package (Csardi & Nepusz 2006). Network properties such as edge
density, network diameter, mean path length, and modularity were
extracted (He et al. 2020; Xie, Wang & Li 2022). We excluded
root AMF colonization data from correlation and trait networks as this
dataset was comprised of many zero values.
The effect of DSM verus NSM and wild versus modern barley on overall
bacterial community composition was analyzed via NMDS ordination of
weighted uniFrac distance and PERMANOVA using the ordinate and adonis2
function of the vegan package (Oksanen et al. 2022), whereas, for
the identification of biomarker taxa, a generalized linear model was
used from MASS package (Ripley 2023). Additionally, differences between
log2fold changes were calculated and both LEfSe (Linear discriminant
analysis Effect Size) and ANCOMBC (Analysis of Compositions of
Microbiomes with Bias Correction) were used to validate the results.
Multiple test correction was performed by p-value adjustment via the
Benjamini-Hochberg method. The core microbiome of barley species under
DSM and NSM was defined as genera occurring with 0.1% in at least 75%
of replicates. Plots were created in R using ggplot2 (Wickham et
al. 2023) and metacoder (Zachary, Grunwald & Gilmore 2023) packages.