https://doi.org/10.5061/dryad.dfn2z3594

In this study, we conducted a meta-analysis testing the effects of climate change-related abiotic factors on plant-associated microbes. To this end, we analyzed studies involving experimental manipulations of climate change-related abiotic factors (e.g., warming, drought, and eCO2) and measuring abundance of microbes (e.g., virus, bacteria, fungus) in the phyllosphere or rhizosphere. We aimed at: (1) Describing the overall magnitude and direction of effects of abiotic factors on plant-associated microbes (2) Testing whether such abiotic effects were contingent on the type of microbe, namely bacteria, fungus or virus, and plant part where microbes are found, namely the phyllosphere or rhizosphere. In doing so, this study furthers our understanding of climate change-related abiotic forcing on plant–associated microbes and its implications for ecosystem responses to global change threats.

Description of the data and file structure

To be included in our analysis, studies had to meet the following criteria: (a) provide a measure of plant-associated microbial abundance (e.g., amount, frequency, disease intensity, transmission rate, virus load) in the phyllosphere or rhizosphere of plants growing under experimental manipulation of climate change-related abiotic conditions (eCO2, warming, drought, etc.), and (b) report treatment level means (abiotic manipulation vs unmanipulated control), variability (i.e., variance, standard error or standard deviation), and the sample size in either the text, figures, tables or appendices. When needed, we extracted data from figures following digitalization using WebPlotDigitizer software. We excluded studies that applied two or more different abiotic manipulations together on the same plants. After applying these criteria, the resulting dataset consisted of 513 case studies from 96 studies (out of the original 5450) from the primary literature published between 1975 and 2021 in 47 scientific journals.

"NA" means empty values.

Code/Software

For each study case, we estimated effect sizes using Hedges’ d metric and a confidence interval using the “metafor” package 3.8-1 version in R 2022.07.2. Hedges’ d is calculated as the standardized mean difference between plants subjected to the abiotic manipulation and control (unmanipulated) plants, such that negative values indicate that microbial abundance had lower mean values on abiotically-manipulated plants compared to control plants, whereas positive values indicate the inverse.

We first estimated the grand mean effect size and 95% confidence interval (CI) across all studies to assess whether there was an overall effect of abiotic factors on microbial abundance. This grand effect size does not separate the effects of different types of climate change-related abiotic factors. Rather, the purpose of this analysis was to estimate the degree of consistency among studies by means of the between-studies heterogeneity (τ² and associated Q statistics), an important overall estimator for our analysis. High heterogeneity can be accounted for by using explanatory variables (referred to as “moderators” in meta-analysis literature). Total heterogeneity is split into among-group heterogeneity (i.e., among abiotic factors) and within-group heterogeneity (i.e., variance of effect sizes within moderator level). The τ² and associated Q statistics for heterogeneity aim at determining whether among-group heterogeneity is large enough as compared to within-group heterogeneity to conclude on the significant effect of the moderator tested. Because τ² is dependent on sample size, we also calculated *I*2 value which is a standardized estimate of total heterogeneity ranging from 0 and 1.

We next evaluated the effects of the type of climate change-related abiotic factor (eCO2, warming, and drought) on the abundance of plant-associated microbes by estimating mean effect sizes and 95% CIs for each abiotic factor and running models with the type of abiotic factor as a moderator. Then, we tested whether effects of the type of climate change-related abiotic factor on the abundance of plant-associated microbes were contingent on the microbial taxonomic group (bacteria, fungus or virus), and the plant part with which microbes were associated (phyllosphere or rhizosphere). For this, we ran models including as moderators: the type of climate change-related abiotic factor, one of the above-mentioned factors (microbial taxonomic group or plant region of colonization), and the two-way interaction between the type of abiotic factor and the microbe grouping factor. We note that there was not enough replication to test for the three-way interaction between abiotic forces, microbial group, and plant region of colonization. We reported results from the omnibus test (i.e., overall effect of all moderators) as well as from the coefficient parameter estimate and associated confidence interval. In all the above models, we performed multi-level error meta-analyses with the rma.mv function of the R package metafor v. 2.0-0 , and included the primary study and study case nested within primary study as random factors in order to account for non-independence among multiple effect sizes drawn from a single primary study. Multiple comparisons of abiotically-manipulated plants with the same control plant were accounted for by computing the variance-covariance matrix among all effect sizes. We considered an effect size as significant if its 95% confidence interval did not overlap with zero.

To ensure that our findings were robust, we conducted a sensitivity analysis in which we sequentially removed one primary study at a time. This analysis was aimed at testing whether the main result could have emerged from the inclusion of any particularly influential study, for instance one providing a large number of study cases. For each of the 95 runs, corresponding to removing each of the 95 primary studies included in the main analysis, we checked that model parameter estimates for each treatment (abiotic manipulated vs unmanipulated control plants) were comparable, regardless of whether each study was later included or not in the analyses. This analysis indicated that our findings were robust and unbiased by non-independence among effect sizes. In addition, we used several approaches to verify that our results were not affected by publication bias: (1) inspection of funnel plots, (2) exploration of the relationship between effect-sizes and journal impact factor, and (3) cumulative meta-analysis. These analyses indicated that our findings were robust to selective reporting and dissemination bias.