Geology and climate influence rhizobiome composition of the phenotypically diverse tropical tree Tabebuia heterophylla

contaminants using Duk). Sequence clustering was done with USEARCH [21] including sequence cleaning and identity assignment. Clustering was done iteratively starting at 99% identity, and decreasing by 1% identity until the level described in the config file is reached (97%). Finally, we used Quantitative Insights Into Microbial Ecology (QIIME) pipeline ver- sions 1.8 and 1.9 [22] for downstream sequence processing such as chimera removal taxo- nomic classification to each cluster using RDP Classifier. From the 144 subsamples, only 137 samples were analyzed considering only samples with more than 1,000 reads. Lastly, sequences corresponding to chloroplast DNA (plant DNA) were removed from the OTU table through a taxa filtering function. For some of the analyses, we collapsed 137 subsamples through the col- lapse function of QIIME v1.9 [22], by summing the sequences from all subsamples per tree, which is our sampling unit (n = 36). We selected the minimum number of sequences of col- lapsed samples, (30, 170 sequences; to rarefy all samples and subsequently perform the statisti- cal analyses. Statistical analyses For each site we extracted the total annual precipitation (bio12) and mean annual temperature (bio01) from BIOCLIM data set using ArcGIS 10.7 [23]. The two climatic variables were con- verted into categorical variables each with four levels (very high, high, medium, and low; Table 1), which together with the parent material (volcanic, serpentine and karst) were used to build a mapping file. Our sequence data and this mapping file were inputted to QIIME v1.9

pipeline [22] and the Phyloseq R package [24]. We examined variation in species’ richness (OTUs) across the parent material and climate matrix based on the Chao 1 index estimated by QIIME [22] and visualized through R [25]. The error bars represent confidence intervals (95%, df = 9, two tail T values) that were calcu- lated using 10 iterations of the Chao1 index of the subsamples, depending on the case. We also examined variation in species diversity through Shannon index and visualized through Phylo- seq [24]. Analyses of Variance tests, implemented in R (50),were used to find significant differ- ences in alpha diversity related to a given category. Additionally, differences among categories, i.e combinations of microhabitat/ sample type (rhizosphere and bulk soil), geology and cli- mate, were tested with an ANOVA and Tukey HSD posthoc test for all variables and were con- sidered significant at an FDR-corrected p < 0.01. We ran two complementary analyses to examine variation in community structure (beta diversity). The first used the subsamples to run a non-metric multidimensional scaling (nMDS) ordinations based on a Bray Curtis dis- tance matrix using ggplot2 [26]. The second, used the collapsed samples into sampling units in a Principal Coordinate Analysis (PCoA) ordination based on a Unifrac distance matrix and visualized through Phyloseq [24]. Measures of beta diversity were assessed for differences between sample groups using two different complementary non-parametric analyses for multivariate data analysis of similarity

(ANOSIM) [27] and nonparametric multivariate analysis of variance (Adonis) using distance matrices [28]. We used the Bray–Curtis similarity index to calculate a distance matrix for ANOSIM and Adonis, and performed the analyses with the Vegan package [29] in the R pack- age using R [25]. We also generated taxa summaries with QIIME [22], filtering only OTUs present in at least 40% of the samples using the script compute_core_microbiome.py, and visualized using R [25]. Biomarker discriminant taxa analyses were performed with the LEfSe algorithm [30] for multi-class comparison. In order to reduce false discovery rates, FDR p-values <0.01 were considered for significant differences—these analyses was performed using the MicrobiomeA- nalyst pipeline [31]. PLOS ONE Rhizobiota of the tropical tree Tabebuia heterophylla PLOS ONE | https://doi.org/10.1371/journal.pone.0231083 April 7, 2020 5 / 16