Environmental modulation of plant mycorrhizal traits in the global flora

1862 | wileyonlinelibrary.com/journal/ele Ecology Letters. 2023;26:1862–1876. INTRODUCTION Functional traits define organism responses to the environment and their effects on ecosystem properties (Dıaz & Cabido,2001), and traitbased functional ecology has emerged as a burgeoning research area in recent decades (Fontana et al., 2021; Garnier & Navas, 2012; Kattge et al., 2011; McGill et al.,2006; Nock et al.,2016; Violle et al.,2007; Zanne et al.,2020). While an important axis of variation in plant form and function is known to reflect whether or not plants form symbiotic relationships with soil microorganisms (Bergmann et al.,2020; Weigelt et al.,2021), we still lack a detailed understanding of belowground traits, including those describing LETTER Environmental modulation of plant mycorrhizal traits in the global flora YimingMeng (孟益明)1 | JohnDavison1 | John T.Clarke2,3,4,5 | MartinZobel1 | MaretGerz1 | MariMoora1 | MaarjaÖpik1 | C. GuillermoBueno1,6 Received: 16 February 2023 | Revised: 15 August 2023 | Accepted: 21 August 2023 DOI: 10.1111/ele.14309 1Institute of Ecology and Earth Sciences, University of Tartu, Tartu, Estonia 2GeoBioCenter, LudwigMaximiliansUniversität München, Munich, Germany 3Department of Earth and Environmental Sciences, Paleontology & Geobiology, LudwigMaximiliansUniversität München, Munich, Germany 4Department of Ecology and Biogeography, Nicolaus Copernicus University in Toruń, Toruń, Poland 5Department of Zoology, Institute of Ecology and Earth Sciences, University of Tartu, Tartu, Estonia 6Pyrenean Institute of Ecology, IPECSIC, Jaca, Spain Correspondence Yiming Meng and C. Guillermo Bueno, Institute of Ecology and Earth Sciences, University of Tartu, Tartu, Estonia. Email: yiming.me[email protected]e and gbueno@ ipe.csic.es Funding information Alexander von HumboldtStiftung; Eesti Teadusagentuur, Grant/Award Number: PRG1065, PRG1789 and PRG741; European Regional Development Fund, Grant/Award Number: Centre of Excellence EcolChange; Horizon Europe, Grant/Award Number: ALFAwetlands; Narodowa Agencja Wymiany Akademickiej, Grant/Award Number: PPN/ULM/2019/1/00248/U/00001; Ramon y Cajal fellowship, Grant/Award Number: Project ALFAwetlands Editor: Josep Penuelas Abstract Mycorrhizal symbioses are known to strongly influence plant performance, structure plant communities and shape ecosystem dynamics. Plant mycorrhizal traits, such as those characterising mycorrhizal type (arbuscular (AM), ecto- , ericoid or orchid mycorrhiza) and status (obligately (OM), facultatively (FM) or nonmycorrhizal) offer valuable insight into plant belowground functionality. Here, we compile available plant mycorrhizal trait information and global occurrence data ( ∼ 100 million records) for 11,770 vascular plant species. Using a plant phylogenetic megatree and highresolution climatic and edaphic data layers, we assess phylogenetic and environmental correlates of plant mycorrhizal traits. We find that plant mycorrhizal type is more phylogenetically conserved than plant mycorrhizal status, while environmental variables (both climatic and edaphic; notably soil texture) explain more variation in mycorrhizal status, especially FM. The previously underestimated role of environmental conditions has farreaching implications for our understanding of ecosystem functioning under changing climatic and soil conditions. KEYWORDS arbuscular mycorrhiza, ectomycorrhiza, ericoid mycorrhiza, facultative mycorrhiza, GBIF, mycorrhizal status, mycorrhizal type, nonmycorrhiza, obligate mycorrhiza, phylogenetic conservatism This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. © 2023 The Authors. Ecology Letters published by John Wiley & Sons Ltd. | 1863 MENG et al. symbioses, as well as their evolutionary and ecological drivers (Carmona et al.,2021; Laliberté,2017). Mycorrhizal symbiosis emerged approximately 400 Mya (StrulluDerrien et al., 2018), and among the 358,000– 435,000 extant plant species (Enquist et al.,2019), a high proportion are believed to form mycorrhizas (about 80%; Smith & Read,2008) with >50,000 fungal partners belonging the phyla Ascomycota, Basidiomycota and Glomeromycota (Tedersoo, Bahram, & Zobel,2020; van der Heijden et al.,2015). As mycorrhizal associations are widespread and known to strongly influence plant performance, structure plant communities and ultimately shape ecosystem dynamics (van der Heijden et al., 2015), traits describing the characteristics of different mycorrhizal associations and the capacity of plants to form them (hereafter plant mycorrhizal traits, PMTs) are among the most promising traits in terms of capturing plant belowground functionality (Chaudhary et al.,2022). Plant mycorrhizal traits include mycorrhizal type, which distinguishes mycorrhizas that emerged at different times during plant evolution and feature specific morphological and functional characteristics (i.e., arbuscular (AM), ecto- (ECM), ericoid (ERM) or orchid (ORM) mycorrhiza), and mycorrhizal status, which classifies plant species into those that are always (obligately; OM), sometimes (facultatively; FM) or never (nonmycorrhizal; NM) mycorrhizal (Moora,2014; Smith & Read,2008). Changing environmental conditions probably provided the stimulus for multiple evolutionary waves of plant innovation via development and shifts in plant mycorrhizal symbioses (Brundrett & Tedersoo,2018; Werner et al.,2018). Based on fossil evidence, the first mycorrhizal partners of terrestrial plants are thought to have been AM fungi in the Glomeromycota and Mucoromycota (Feijen et al.,2018; van der Heijden et al.,2015). During the course of plant evolution, alternative mycorrhizal types and NM status may have been favoured by natural selection, as innovations emerged that allowed plants to withstand environmental stresses or to acquire nutrients via specialised root structures or associations with other fungal groups (in the phyla Ascomycota and Basidiomycota; Wang & Qiu,2006; StrulluDerrien et al.,2018). While the ancestral AM state still prevails among plant species (Smith & Read,2008; Werner et al.,2018), it is proposed that the most recent mycorrhizal evolutionary wave is ongoing and responsive to current and future environmental changes (Brundrett & Tedersoo,2018). A common way to examine variation in PMTs between species involves using taxonomic relationships. This method extrapolates the PMTs of plant taxa, such as genera or families, by leveraging information about the mycorrhizal type observed among closely related species (Tedersoo,2017; Tedersoo et al.,2019). While this information contributes to our understanding of PMTs within floras, it has limitations. Notably, qualitative extrapolation of PMTs to plant orders, families and genera has been shown to overlook key adaptive processes (Albornoz et al.,2021; Brundrett,2009; Bueno, AldrichWolfe, et al.,2019; Osborne et al.,2018; Sun et al.,2019; Tedersoo,2017), leading to discrepancies between plant phylogenybased extrapolations and empirical observations (Bueno, Davison, et al., 2021; Bueno, Gerz, et al.,2019; Moora,2014). To address this, it is important to gain an indepth understanding of the degree to which PMTs are conserved among related species (Brundrett & Tedersoo,2018; Bueno, Gerz, et al.,2019). Studying the relationship between phylogenetic and phenotypic similarity (by means of estimating phylogenetic signal) provides insight into the evolutionary constraints of traits (Blomberg et al.,2003). This could enhance our ability to interpret and predict variation in PMTs, but has yet to be attempted. One likely explanation for switching between plant mycorrhizal types or statuses is the selective pressures imposed by different environmental conditions (Brundrett & Tedersoo,2018; Werner et al.,2018). The theoretical model proposed by Read (1991) and Read and PerezMoreno(2003) suggested that the abundance of plants of different mycorrhizal type changes along latitudinal and altitudinal gradients under the influence of climatic conditions that mediate accumulation of soil organic carbon and availability of different nutrient forms. According to this model, AM, ECM and ERM plants dominate in grassland and tropical forests, temperate and boreal forests, and heathlands, respectively. More than 10% of terrestrial plants are considered NM and 15% are considered FM (Brundrett,2009; Brundrett & Tedersoo,2018). While NM plants may frequently be at a survival disadvantage compared to mycorrhizal species (Richardson et al.,2000), NM status may be favourable in areas lacking fungal symbionts or experiencing disturbance (Delavaux et al.,2019; Duchicela et al.,2020). FM plants, on the contrary, can succeed both in the absence and presence of mycorrhizal fungi (Smith & Read,2008), which may confer an ability to exploit a wider range of environmental conditions (Gerz et al.,2018). Certain aspects of these hypotheses have been explored in previous research characterising the distribution of AM and ECM tree species (Steidinger et al.,2019), extrapolating global mycorrhizal biomass estimates based on dominant species (Barceló et al.,2019) and examining the proportion of mycorrhizal versus nonmycorrhizal species numbers (Delavaux et al.,2019). However, these studies have primarily relied on vegetationlevel data and made strong assumptions for taxonomic extrapolation. Species represent fundamental units of vegetation, and disentangling the evolutionary and ecological processes determining specieslevel PMT expression could importantly improve our understanding of the mechanisms underlying spatial and temporal variation of mycorrhiza. Yet, such empirical specieslevel information is lacking. Characterising the range of environmental conditions where particular PMTs occur can inform predictions about plant belowground functioning in changing environments. For instance, the abundance of different 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1864 | PLANT MYCORRHIZAL TRAITS IN THE GLOBAL FLORA PMTs is expected to strongly influence biogeochemical processes related to carbon cycling and nutrient retention (Phillips et al.,2013; Terrer et al.,2016). It has been found that variations in the share of different mycorrhizal types in the European flora correlate with temperature, soil pH and net primary productivity, while the share of FM species varies with temperature (Bueno et al., 2017). Similarly, recent studies using relatively coarse data (10– 60 arc minutes) suggest that climatic variables are the main drivers of mycorrhizal type distribution across terrestrial ecosystems (Barceló et al.,2019; Steidinger et al.,2019). However, drivers may be scaledependent, and coarseresolution data may not detect the significance of edaphic factors (Barceló et al.,2019; Bueno, Gerz, et al.,2019). Indeed, a finerresolution approach (30 arc seconds) showed that PMTs are driven by a combination of climatic and soil factors in the Pyrenees (Bueno, Gerz, et al.,2021). Here, we compile available mycorrhizal trait information for plant species worldwide. Using information from a plant phylogenetic megatree and climatic and edaphic data layers at 1 × 1 km2, we assess phylogenetic and environmental correlates of PMTs. Specifically, we aim to answer (1) to what degree are PMTs phylogenetically conserved? (2) how does the share of different PMTs vary globally? (3) to what extent do environmental and plant phylogenetic variables alone or in combination explain the mycorrhizal traits exhibited by plant species in the global flora? and (4) which environmental drivers are associated with the different trait states? We hypothesise that (i) the evolutionary younger ECM, ERM and ORM types ( ∼ 100– 180 Mya, Brundrett & Tedersoo,2018) are more phylogenetically conserved than AM type; (ii) environmental conditions explain variation in the prevalence of all PMTs, but especially in the prevalence of mycorrhizal statuses; and (iii) reflecting the findings of Bueno, Gerz, et al.(2021), soil properties estimated at fine spatial scales can be as important drivers of PMTs as climatic conditions. We found that plant mycorrhizal type was more phylogenetically conserved than mycorrhizal status, with plant phylogenetic vectors particularly strong predictors ECM, ERM and ORM type among plant species. Both climatic and edaphic variables explained variation in types and statuses, in particular FM status. These findings inform our understanding of plant mycorrhizal trait expression and can be used to predict plant belowground functioning in changing environments. METHODS Data collection We collated the most uptodate literature available on PMTs (AppendixS1) and distinguished four plant mycorrhizal types (arbuscular (AM), ecto- (ECM), orchid (ORM) and ericoid (ERM) mycorrhiza, as defined by Smith and Read (2008)) and three plant mycorrhizal statuses (obligately (OM), facultatively (FM) and nonmycorrhizal (NM)) for 14,722 species (AppendixS2). To identify phylogenetic and environmental correlates of PMTs, we combined these data with plant species occurrence data, plant phylogenetic information and global climatic and soil environmental data (in 1 km grid cells) (AppendixS1). The plant occurrence and environmental data were used to estimate plant species environmental associations for 11,770 species with sufficient occurrence data. The environmental association data and phylogenetic information were then used to model variations in plant mycorrhizal trait expression. Figure1 presents the process of data preparation and modelling. Phylogenetic signal in PMTs was examined using the 𝛿 statistic (Borges et al.,2019) (further information in AppendixS1). Seven hundred ten dual mycorrhizal plant species (AM + ECM) were distinguished (AppendixS2), of which 665 yielded sufficient occurrence data. Except where stated otherwise, these species were grouped with ECM plant species in further analyses, reflecting ongoing controversy concerning the definition of dual mycorrhizal plant species (Brundrett,2021a; Teste et al.,2020) and the fact that the niches of dual mycorrhizal plants most closely resemble those of ECM plants (Gerz et al.,2018). Modelling approach We aimed to understand how well plant phylogenetic conservatism and plantenvironment associations explain the mycorrhizal traits of extant plant species. To do this, we modelled each plant mycorrhizal type and status as a binary response variable, for example AM vs nonAM and FM vs nonFM. We also modelled the four plant mycorrhizal types and three mycorrhizal statuses as multilevel classifications to assess the overall response of these two traits to environmental and phylogenetic factors. We used the random forest (RF) algorithm to build classification models (Breiman,2001; Liaw et al.,2002). Selection of phylogenetic and environmental eigenvectors We first prepared a set of orthogonal vectors representing phylogenetic relationships between plant species, using the PVRdecomp function from the R package ‘PVR’ (Santos et al.,2013) to decompose the phylogenetic tree into eigenvectors using principal coordinates analysis (PCoA). We next prepared a set of orthogonal vectors representing environmental gradients by performing principal components analysis (PCA) on the plantenvironment association dataset with 137 variables (53 means, 53 standard deviations and 31 soil orders, Appendix S1). We used BorutaPy codebase (https:// github.com/sciki tlearn - contr ib/boruta_py), which uses 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License | 1865 MENG et al. the RF algorithm based on methods from Kursa and Rudnicki (2010), to identify relevant phylogenetic and environmental features (‘feature selection’ of predictors). Variance partitioning RF modelling was performed using the Python machine learning library ‘scikitlearn’ implementation (Pedregosa et al., 2011). We first performed separate RF analyses using the selected phylogenetic and environmental eigenvectors to classify each plant mycorrhizal trait and trait state (RF models 1 and 2, Figure1). We then performed another RF analysis for each (RF model 3; Figure1) using all of the features included in RF models 1 and 2 (i.e., combining selected phylogenetic and environmental predictors). Data points not included in the RF bootstrap sample (out of the bag; OOB) were used to test the predictive error rate of the model (OOB error rate). We obtained OOB errors for each of the three RF models, which respectively represented errors using (1) environmental predictors only, (2) phylogenetic predictors only, and (3) environmental and phylogenetic predictors. We then partitioned variation into portions uniquely explained by each predictor set and that explained by both. Variation explained by both might, for example reflect biogeographic processes or niche conservatism, such that related plants exploit similar environmental conditions. Variance partitioning is generally used with linear models, with model performance described by R2 values. Here, we used an analogous approach based on the OOB error, which is commonly interpreted as a measure of predictive quality (prediction R2) (Acharjee et al.,2016; Povak et al.,2020; Sperling et al.,2016). We calculated the R2 OOB using an approach which sets as a null baseline the frequency of the most frequent category, that is where 11,770 and OOB% represent the total number of species in the study and the OOB error produced by a RF model, respectively, and m represents the frequency of the most frequent trait category (e.g., there were 8501 AM species and 3269 nonAM species in the data; thus, m = 8501 was used when calculating R2 OOB for AM). R 2 OOB = 11770 ×OOB%−m 11770 −m FIGURE 1 Schematic representation of data preparation and modelling, including feature selection and dataset combinations in different random forest (RF) models. Data preparation started with compilation of a plant mycorrhizal trait database, comprising the response variables, and phylogenetic and environmental information as predictors. The predictor matrices were orthogonally decomposed and included in the form of eigenvectors into RF models. Feature selection was conducted using the Boruta algorithm, and five RF models were built to determine the importance of phylogeny and environmental conditions in explaining plant mycorrhizal traits. 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1866 | PLANT MYCORRHIZAL TRAITS IN THE GLOBAL FLORA Importance of environmental variables We included all 137 raw environmental predictors as variables in a fourth RF analysis (RF model 4; Figure1). Feature importance (normalised so that total importance sums to 1), was calculated from the mean and standard deviation of accumulation of the impurity decrease within each tree for each response variable. To check whether the relative importance of different environmental predictors changed in the presence of plant phylogenetic vectors, we incorporated both raw environmental variables and phylogenetic vectors (the selected phylogenetic features from RF model 1) into a final model (RF model 5; Figure1) and recalculated feature importance. RESULTS Mycorrhizal trait dataset Of the 11,770 studied plant species, 72.2% were AM type, while ECM, ERM and ORM types represented 11.4%, 1.4% and 1.2% of species, respectively. 67.7% of species had OM status, 18.5% had FM status, and 13.8% were NM (Table1). AM type and OM and FM statuses were found in each of the 87 most dominant plant families, and NM species were most commonly found in Brassicaceae, Cyperaceae and Amaranthaceae (Figure2, AppendixS3). Phylogenetic conservatism in mycorrhizal traits The strength of phylogenetic signal differed between and within plant mycorrhizal type and status. Among plant mycorrhizal types, relatively informative inferences were obtained, with node entropies close to 0 (i.e. indicating phylogenetic conservatism); conversely, among mycorrhizal statuses, most ancestral inferences were uninformative, producing node entropies close to 1. Thus, phylogenetic signal was higher for plant mycorrhizal types ( 𝛿 = 6.7) than for statuses ( 𝛿 = 3.2) (Figure2). Signal was highest for ORM and ERM types, intermediate for ECM type and lowest for AM type; FM status had lower signal than OM or NM status (Figure2). Trait environment associations The share of AM plant species peaked at low latitudes and exhibited a consistent trend in relation to climatic (climate PC1) and, to a lesser extent, soil conditions (edaphic PC1, Figure3a). AM plants were more likely to occur in warm climates (low climate PC1, Figure3a; high annual mean temperature (BIO1) and solar radiation, low wind speed, FigureS1) with moderate precipitation (BIO12, FigureS1) and alkaline soils with intermediate moisture and nutrient content (soil PC1, Figure3a; available soil water, exchangeable acidity, organic carbon (OC), nitrogen (N) and phosphorus (P), FigureS1). ECM species peaked at high latitudes across the globe, in forest ecosystems of the eastern United States, eastern and southern Asia, and in parts of central and eastern Africa, Australia and, to a lesser extent, the southern cone of South America (TableS1). Compared with AM plants, ECM plants were associated with lower temperatures (climate PC1, Figure3b) and acidic soils with high moisture, cation exchange capacity (CEC), OC, N and P, and low cation content and bulk density (soil PC1 and PC2, Figure3b; also see FigureS1). All ERM species belonged to the Ericaceae family besides two Diapensiaceae and one Escalloniaceae species (AppendixS3; TableS1). ERM plants occurred more frequently in cold and humid climates and in moderately fertile soils, that is climates with low temperature, solar radiation, evapotranspiration and acidic soils with more available water, C, N and higher CEC (climate and soil PC1, Figure3c; also see FigureS1). These conditions are mostly found in the high latitudes of the northern hemisphere and the southern Andes. ORM plant species distribution reflected temperate, boreal and tundra species, as we lacked data on the mycorrhizal associations of tropical orchids. The environmental drivers for these highlatitude orchid species were not clearly dependent on climatic or edaphic conditions (AppendixS4). NM plants peaked at high latitudes of the northern hemisphere (> 50◦ N), but also at high altitudes (Figure3d; TableS1). The share of NM species showed no clear trend with respect to edaphic conditions, but increased in cold and arid conditions (climate PC1 and PC2, Figure3d; BIO4 (temperature seasonality), BIO8 (mean temperature in the wettest quarter) and BIO9 (mean temperature in the driest quarter), FigureS1). TABLE 1 Sample sizes (i.e., number of species representing each trait), outofbag R2 and cumulative feature importance for random forest models classifying plant mycorrhizal traits according to phylogenetic and environmental variables. Trait N R2 OOB% Feature importance Env_only Phy_only Both Env Phy AM 8501 8.41 38.97 00.394 0.606 ECM 1347 2.08 58.80 1.11 0.221 0.779 ERM 152 1.97 68.42 00.163 0.837 ORM 146 3.42 82.88 00.149 0.851 NM 1624 10.59 18.97 00.530 0.470 FM 2177 23.84 0.23 00.865 0.135 OM 7969 11.81 22.44 2.74 0.545 0.455 Type 7.56 41.15 00.362 0.638 Status 13.63 14.39 00.590 0.410 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License | 1867 MENG et al. FIGURE 2 Phylogeny and mycorrhizal traits of 11,770 vascular plant species, and phylogenetic signal for each plant mycorrhizal trait. In the phylogenetic tree, the first ring represents mycorrhizal trait characteristic for each tip (species), the second ring (purple strips) highlights the range of the 15 largest families in our dataset, and the third ring represents the abundance (as a stacked bar) of each trait within each of the largest families. In the violin plot, node entropies were calculated based on the linear version of Shannon entropy. 𝛿 , measuring the degree of phylogenetic signal between a trait vector and a phylogeny, is higher in types than statuses. Note that some species (e.g., among Pinaceae) displayed as ECM exhibited dual mycorrhiza (AM + ECM). AM, arbuscular mycorrhizal; ECM, ectomycorrhizal; ERM, ericoid mycorrhizal; FM, facultatively mycorrhizal; NM, nonmycorrhizal; OM, obligately mycorrhizal; ORM, orchid mycorrhizal. In order to include all species in both trait analyses, NM is presented as a type and a status. FIGURE 3 Global distribution of mycorrhizal traits among studied plant species: Arbuscular mycorrhizal (AM), ectomycorrhizal (ECM) and ericoid mycorrhizal (ERM) types, and nonmycorrhizal (NM), facultatively mycorrhizal (FM) and obligately mycorrhizal (OM) statuses. In order to visualise global patterns of mycorrhizal trait distribution, data for 50 × 50 km grid cells were calculated by aggregating 1 × 1 km plant occurrence data and upscaling environmental variables (see AppendixS1 for details); cells with only one species were excluded. Note that the scales vary between panels to emphasise relative trait variations; figures with a uniform 0%– 100% scale are shown in AppendixS6. Inset (bottomleft): Share of species with particular mycorrhizal traits in relation to edaphic and climatic gradients. For grid cells with complete environmental data (46,957 grids), edaphic and climatic factors were described using their first two principal components. The first two principal components explained 46.1% and 25.1% of variation in climatic variables, and 28.4% and 15.4% of variation in edaphic variables. Climate PC1 and PC2 mainly characterised temperature and precipitation, respectively. Soil PC1 characterised a range of soil properties including soil available water, cation exchange capacity, bulk density, organic carbon, nitrogen and phosphorus content; soil PC2 mainly reflected cation content and base saturation (AppendixS7). Lines represent fitted curves from LOESS. Numbers between brackets in the ×- axes correspond to the variance explained by each principal component. Relationships with individual environmental variables are shown in FigureS1. 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1868 | PLANT MYCORRHIZAL TRAITS IN THE GLOBAL FLORA The share of FM plant species was greater towards high latitudes of the northern hemisphere, while OM plant hotspots were often close to the equator (TableS1). FM and OM plants showed contrasting environmental trends in relation to soil and climate conditions (Figure3e,f). OM plants occurred more in poor soils with high bulk density, low CEC, low C, N and P (soil PC1 and PC2, Figure3f) and in tropical rainy climates (climate 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License | 1869 MENG et al. FIGURE 3 (Continued) 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1870 | PLANT MYCORRHIZAL TRAITS IN THE GLOBAL FLORA PC1 and PC2, Figure3f; BIO1, BIO4, BIO12, low wind speed, high vapour pressure, FigureS1). FM plants were correspondingly less represented in such conditions (FigureS1). Variance partitioning Random forest classification (RF models 1, 2 and 3; Figure1) and variance partitioning revealed that plant phylogeny explained a relatively large amount of variation in mycorrhizal type (41.2% explained by phylogeny; 7.6% explained by environmental conditions), while phylogenetic and environmental variables explained similar amounts of variation in mycorrhizal status (14.4% explained by phylogeny; 13.6% explained by environmental conditions) (Table1). Among plant mycorrhizal types, environmental variables explained 8.4% of variation in AM type. ERM and ORM types were highly phylogenetically conserved (phylogeny explaining 68.4%– 82.3%), while environmental variables explained almost none. For NM and OM species, environmental variables explained approximately 11% of variation while phylogeny explained approximately 19%– 22%. By contrast, for FM species, environmental variables explained 23.8% of variation, while phylogeny explained <1%. Our findings also revealed little variation explained by the joint influence of environmental and plant phylogenetic variables (column ‘Both’ in Table1), indicating that each predictor set explained different aspects of PMTs. We also performed variance partitioning for the dual mycorrhizal type separately (AM + ECM, TableS2), finding that environmental variables explained about 6% of variation, while plant phylogeny explained almost none. Estimates of feature importance generally mirrored the variance partitioning estimates but the relative importance of phylogenetic and environmental variables tended to be more equal than suggested by variance partitioning. A combination of edaphic and climatic variables were important predictors in RF models classifying plant mycorrhizal type (constituting 11 and 9 out of the 20 most important variables, respectively), with soil texture, particularly sand content, along with variability (standard deviation) in soil properties, such as pH, the most important variables (Figure4). Predictors reflecting climate variability, such as the standard deviation of isothermality or precipitation of the warmest month, were also relatively important. For plant mycorrhizal status, climatic variables, specifically solar radiation and evapotranspiration, and to a lesser degree, variability in temperature, wind and evapotranspiration, were consistently strong predictors. After climate, Histosols (soils consisting primarily of organic materials) and variability in soil texture and pH were important (Figure4). Among plant mycorrhizal types, AM type was explained by variability in climate, for instance the standard deviation of isothermality and temperature of the wettest quarter, and edaphic factors including soil pH. Variables reflecting soil texture were also important. For ECM type, soil texture and soil depth were important, along with climate variability, particularly the coldest month and quarter of the year. For ERM type, edaphic features, including Cambisol (largely representing young soils in boreal areas), base saturation, total P, organic carbon and soil texture characteristics, were most important; precipitation seasonality was the most important climatic factor. Among plant mycorrhizal statuses, NM was driven by both edaphic and climatic factors. Soil development characteristics, such as soil depth or the extent of nonsoil (e.g. rocky areas), were most important. Somewhat less important were variation in temperature and precipitation along with variation in certain soil nutrients, including sodium (Na) and N. By contrast, for OM and FM, climatic variables were more important than edaphic factors, with solar radiation and, to a lesser extent, evapotranspiration best predicting both FM and OM statuses. For FM plants, different soil categories, in particular, Histosols and Retisols (soils poor in clay), were also important factors (Figure4). Important environmental predictors obtained from RF model 5 (the combination of raw environmental variables and plant phylogenetic vectors) were broadly similar to those identified by RF model 4 (AppendixS5). DISCUSSION We found that evolutionarily recent plant mycorrhizal types (i.e. ericoid (ERM), ecto- (ECM) and orchid mycorrhizal (ORM) plants) exhibited stronger conservatism than the ancestral arbuscular mycorrhizal (AM) type, and that plant mycorrhizal type overall was more conserved than status. Variance partitioning using random forest classification indicated that plant phylogeny explained 5– 35 times more variation in mycorrhizal types compared with the fraction explained by the environmental characteristics of plant species distribution areas, while feature importance suggested more balanced importance of phylogenetic and environmental drivers (Table1). Thus, the effects of environmental associations were relatively minor but not negligible and may explain some of the deviation from perfect phylogenetic conservatism observed within many plant families (Appendix S3). For mycorrhizal status, phylogenetic and environmental predictors were of approximately equal significance. Among environmental predictors, both climatic and soil niche variables importantly distinguished mycorrhizal types, while climatic variables predominantly explained mycorrhizal status. Thus, our study illustrates the underappreciated importance of environmental characteristics, in particular edaphic factors, in driving plant mycorrhizal trait (PMT) 14610248, 2023, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ele.14309 by Csic Organización Central Om (Oficialia Mayor) (Urici), Wiley Online Library on [17/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License