R. irregularis phylogeny and phylogenetic signal

All nine isolates used in this study were previously sequenced with ddRAD-seq with a minimum of 3 replicates [23]. The raw sequence reads were retrieved from the NCBI bioproject (accession number PRJNA326895) and were trimmed and analysed following the workflow of Wyss et al., (2016) [32] and Savary et al. (2017) [23]. Genetic distance matrices among the nine isolates were calculated based on the scalar distance method of Wyss et al., 2016 [32]. For this calculation, data used were taken from the replicate with the deepest sequencing of each of the nine isolates and with the highest number of markers available.

The package phylosignal [33] was used to test phylogenetic signals between the phylogeny of the 9 isolates and i) AMF colonization and mycorrhizal responsiveness across the 6 plant species and ii) the mean AMF colonization, mean mycorrhizal responsiveness, evenness, total productivity and mean flower production per mesocosm, 78 and 105 days after inoculation.

Five phylogenetic signal indicators were calculated on variables of each of the plant species and on community metrics of the mesocosms. These were Moran’s I index [34,35], Abouheif’s Cmean index [36], Blomberg’s K and K* [37] and Pagel’s λ [38]. These were then tested against the null hypothesis of a random trait with no significant signal [33]. We wanted to report the results of different metrics of phylogenetic signal since to our knowledge, we are the first to assess phylogenetic signal using within species phylogenetic distances based on thousands of SNPs. A major trend with several significant indices would indicate a true biological meaning. Plant species and mesocosm variables showing significant phylogenetic signal towards AMF isolates were kept for following analyses on co-inoculation treatments. AMF community phylogenetic diversity for treatments involving more than one AMF taxa was calculated using Faith’s PD [39] and relationships between the retained variables and phylogenetic diversity were assessed using quantile polynomial regression on the median, 20th and 80th quantiles of the response variable.