Statistical analyses
All statistical analyses associated with microbial community richness and composition in roots and shoots of A. speciosa were conducted in R (R Core Team 2018, v. 3.5.1) using the vegan package (Oksanenet al . 2017) except where otherwise noted. All analyses were based on rarefied data (160, 7820, 634 and 1400 sequences for bacteria, AMF, non-AM root fungi and FFE, respectively) using the ‘rrarefy’ function. Samples with few sequences were removed from further analyses to allow for greater sequencing depth, which resulted in 36, 36, 34 and 36 samples for bacteria, AMF, non-AM root fungi and FFE, respectively. Analyses of plant ITS2 data when extracting DNA from all remaining foliar tissue was based on a rarefaction level of 8475 sequences. These sampling depths were chosen based on saturation of sequencing effort curves (Figure 2) and in effort to retain the most samples (n = 18 per sampling strategy for bacteria, AMF and foliar fungi and 17 for non-AM root fungi). Sequencing effort curves were produced using the iNext package (Hsieh and Chao 2016). To test how sampling strategy influenced microbial alpha diversity metrics, we calculated richness as the number of SVs in each sample as well as Pielou’s ‘J’ evenness, which describes the similarity of species frequencies. To compare diversity metrics (based on SVs) between the two sampling strategies we performed a Wilcoxon signed-rank test on all paired values. To ensure that potential differences in sampling strategies were not due to artifactual SVs, we also compared results when implementing LULU, an algorithm for post-clustering curation that clustered SVs at 98.5% similarity and a minimum relative co-occurrence of 0.9 (Frøslev et al ., 2017) .
We performed non-metric multidimensional scaling (NMDS) to evaluate community structures for each sampling strategy and target region, individually. Each NMDS analysis was performed using the ‘metaMDS’ function and stress for all plots was between 0.04 and 0.16. These analyses were performed on Bray-Curtis distances of Hellinger transformed sequence abundances as well as Raup-Crick distances of presence/absence data. The ‘Procrustes’ function in vegan was used to assess similarity of patterns produced in the NMDS analyses for the paired sampling strategies and congruency was visualized in Procrustes plots. The ‘protest’ function was used with 1000 permutations to estimate the significance of the Procrustes statistic. We performed analyses on both presence/absence and abundance data to determine how low-abundant SVs influenced the differences between sampling strategies.
To determine variation in microbial communities among individual plants when we extracted from multiple subsamples, we performed NMDS analyses as well as a PERMANOVA using the ‘adonis’ function. We also performed a PERMANOVA to detect seasonal differences between May and September leaf-disc collections. Figures were generated using ggplot2 (Wickham 2016).