Statistical analysis
In all analyses, we identified and excluded outliers as explained in Supplementary Methodology S3. We used multiple regression analyses to test how phylogenetic isolation aboveground (PIA) and belowground (PIB) conjointly affect i) the decomposition parameters (litter mass loss, C loss, N loss and 13C isotope ratios) and ii) the decomposer biota, i.e. Acari abundance, Collembola abundance, and microbial biomass after 8- and 14-month decomposition, and fungal abundance and fungal diversity after 8 month decomposition. The regression results were illustrated in partial residual plots.
We then explored the role of different trajectories by which PIA or PIB may affect decomposition (Table 1): (i) by changing the abundance or diversity of a given group of decomposers, i.e. PIA or PIB influences significantly a particular group of decomposers (abundance or diversity) which in term influences decomposition (either mass loss or C loss or N loss either at 8 or at 14 months); (ii) by changing the efficiency of a given decomposer group (abundance or diversity),i.e. a significant interaction term between either of the phylogenetic isolations and the decomposer group in their effect on decomposition; or (iii) by affecting decomposition via abiotic conditions, i.e. PIA or PIB maintain a significant effect on decomposition after accounting for abundances and diversities of decomposer groups and the corresponding interaction terms (assuming that all pertinent decomposer groups have been accounted for). Testing the influence of PIA and PIB on decomposers as part of relationship (i) was explained in the before paragraph. The remaining relationships are explained below.
We explained decomposition by including PIA and PIB and decomposer biota as predictors, as well as the interaction terms of decomposers with both PIA and PIB. Due to the large number of decomposer biota as predictors we took a stepwise approach to select the most pertinent interaction terms with decomposer variables. For each of the six decomposition variables to be explained, we first carried out five distinct statistical models corresponding to five biotic predictors: Acari abundance, Collembola abundance, and microbial biomass. Each model tested the effects of PIA, PIB, one biotic predictor and its interactions with PIA and PIB. Finally, in order to decipher the relative contributions of the five biotic predictors to explain litter decomposition, we created a last model in which we included all the significant predictors of the five previous models. The full models were then simplified to determine the most parsimonious models using the ‘stepAIC’ function of the ‘BMASS’ package, an established model selection procedure with both forward and backward selection algorithms, which ranks all candidate models (all initial predictor variables included in the full model) based on the lowest AICs (Crawley 2013). We conducted a separate analysis for fungal abundances and diversities, as these were only available for the 8-month sampling. The procedure of including interaction terms was the same as before.
In the end, we further explored whether the observed direct or interaction effects of PIA or PIB could be explained by litter traits such as leaf phenolics, leaf phytophagy or leaf litter C/N ratios, or by characteristics of the abiotic environment, as suggested by the mechanisms hypothesized in the Introduction and in Table 1. We did so by replacing either PIA or PIB by a given trait or environmental characteristic (as listed in Supplementary Table S1) and repeated the procedure for different traits/environmental characteristics. Statistical analyses were performed with the R software (version 3.3.3).