Analysis
We calculated productivity as the aboveground plant biomass for each plant community. A small proportion (~4%) of plants died during the experiment, of which almost all (~94%) were Asclepias syriaca . To determine if the death of plants was influenced by inoculation treatments, we modeled the survival ofA. syriaca as a function of soil and leaf microbe inoculation. We found no effect of inoculation treatments on survival. We excluded the plant communities with dead individuals, leaving 316 out of 360 plant communities that were included in further analyses. During the sampling, we found a different phenotype of Solidago canadensis in 5 of out the 316 pots, the analysis with and without the 5 pots show consistent results. Therefore, we present here the results of analyses including these phenotypically distinct individuals.
We used linear mixed-effect models and analysis of variance (ANOVA) (packages ‘lme4’ and ‘lmeTest’ in R version 4.2.0; ) to test effects of plant diversity, soil microbial inoculation, leaf microbial inoculation and all interactions among these three explanatory variables on plant productivity. To control the effect of plant community composition on plant productivity, we included the composition as a random term in the model . To specifically test the effect of microbial inoculations on plant productivity and plant diversity-productivity relationships, we contrasted the marginal means of plant productivity between microbial treatments using Tukey’s HSD post hoc test; and compared the slopes of diversity-productivity relationships between microbial treatments using the emtrend function in the ‘emmeans’ package (Lenth et al. 2018). We further calculated net biodiversity effects, also known as overyielding, for all pots with more than two plant species and partitioned the biodiversity effect into selection and complementarity components according to Loreau and Hector (2001). We tested the effect of microbial inoculations on net biodiversity effects, selection effects, and complementarity effects using Tukey’s HSD post hoc test. Since we did not observe any significant effect of low-concentration leaf microbial inoculum on productivity, we only present comparisons regarding high-concentration microbial inoculum when discussing leaf microbial effect although all data were included in our analysis.
We used DADA2 to identify the amplicon sequence variants (ASV) of bacteria and fungi . For bacterial 16S sequences, we calculated the quality scores (Q) of the reads to inspect the sequencing quality, then we removed 30 and 28 nucleotides from the start of reads and kept total sequence length at 180 and 220 for forward and reverse sequences respectively. This trim method removed the primers and low-quality nucleotides resulting in reads of fixed length. However, the ITS2 region of fungi is highly variable in length and this variation reflects the biological differences between fungal taxonomic groups. Thus, we only removed the primers from the ITS amplicon sequences but did not trim them to fixed length. After that, we inferred the error rates of nucleotide substitution and assigned sequences to ASVs following DADA2 default parameters . The paired read ends were merged with a minimum overlap of 12 nucleotides, and non-target sequences and chimeras were removed. Bacterial and fungal taxonomy was assigned by comparison with the SILVA SSU r138 and ITS v8.3 databases respectively .
The total number of reads of both bacterial and fungal ASVs varied greatly among samples which could bias the estimation of sample diversity. Thus, we rarefied the community data using the ‘vegan’ R package (Oksanen, J., et al. 2022). For soil microbial communities collected at the end of the experiment, we firstly used generalized linear models (Poisson family with log link function) to test for effects of plant diversity, soil and leaf treatments on microbial ASV richness (alpha diversity) and then applied distance-based redundancy analysis on the Euclidean distance matrix of Hellinger-transformed community data to test the effects of these treatments on microbial community composition (beta diversity). Additionally, we used Permutational multivariate analysis of variance (PERMANOVA) to quantify the variation in microbial community composition explained by plant diversity, soil, and leaf inoculation respectively. We did not consider the interactions among plant diversity, soil and leaf inoculation in microbial analysis due to a limited number of microbial samples. Finally, we conducted ANCOM-BC analysis to identify the microbial taxa with differential abundance between inoculated and non-inoculated soils using the ‘ANCOMBC’ R package .