Human-induced decrease of ectomycorrhizal vegetation led to loss in global soil carbon content

Vegetation impacts on ecosystem functioning are strongly controlled by mycorrhiza, a plant-fungal association formed by almost all terrestrial plants. Using global high-resolution maps of mycorrhizal associations, we quantitatively examined contributions of distinct mycorrhizal types to the processes of soil carbon sequestration in the context of land-use change. We show that shifts between mycorrhizal types within biomes modify global soil carbon content. Conversion of natural ecosystems into croplands reduced the global coverage of ectomycorrhizal plants by 19%. This global reduction in activities of ectomycorrhizas has led to 23.5 GT loss of soil carbon, contributing substantially to the total human-induced soil carbon debt. Our work provides a benchmark for global, spatially explicit and quantitative assessments of mycorrhizal impacts on ecosystem functioning and biogeochemical cycles.

has been transformed to AM and non-mycorrhizal (NM) vegetation. However, the question of how such human-induced ecosystem changes have affected global distribution patterns of mycorrhizal types and hence their role in ecosystem functioning remains unanswered.
Quantitative information about the global distribution of mycorrhizas is required to understand connections between mycorrhizal distributions and shifts in ecosystem function.
Despite existence of regional maps of current (14, 15) and past (16) mycorrhizal vegetation, we still lack global data on mycorrhizal vegetation patterns.
Based on a comprehensive quantitative evaluation of plant-mycorrhizal associations and the distribution of vascular plant species across biomes and continents, we assembled the first high-resolution digital maps of the global distribution of AM, EcM, ErM, and NM plants ( Fig.1). Using these maps, we examined (i) how conversion of natural ecosystems to croplands has affected the distribution of mycorrhizal types globally, (ii) the relationship between soil carbon content and the relative abundance of AM, EcM and ErM plants in ecosystems, and (iii) to what extent conversion of natural areas to croplands has impacted soil carbon content via differential activities of mycorrhizal types.
To assemble global maps of mycorrhizal type distribution, we compiled available information about mycorrhizal status of plants and vegetation in relation to habitat type and ecoregion.
Based on data accumulated over the past three decades, we established the largest data set containing information about the presence and type of mycorrhizal colonisation of vascular plant species, which is comprised of 27,736 records for 12,702 plant species obtained from 1,565 sources including published reviews, data compilations and previously neglected or recent case studies (Table S1). Based on information on locality and vegetation composition provided by vegetation surveys obtained from 2,169 sources (Table S2), abundance estimates of mycorrhizal types were assigned to all combinations of continents and 98 ecological regions, as defined by Bailey (17) (Table S3), and to land cover types. The latter was obtained from the most recent high-resolution (300 m) map comprising 38 land cover types (18), Table   S4. We evaluated the accuracy of our maps using four independent datasets: (i) forest biomass structure for Eurasia (19), (ii) global analysis of mycorrhizal impacts on carbon vs nitrogen dynamics (20), (iii) USA-based analysis of mycorrhizal associations using remote sensing (21), and (iv) West Australian map of mycorrhizal root abundance (14) (Fig. S1). This validation revealed that the great majority of the data (87% of the AM data points and 89% of the EcM data points) deviate by <25% from the measurements (19-21) (Fig. S2) when excluding ESA land use classes (18) that are poorly resolved and hence difficult to couple to our classification scheme. Agricultural practices drive the replacement of natural vegetation by mostly non-mycorrhizal and AM crops (1, 22). Using past vegetation estimates, Swaty and co-workers (16) showed that across conterminous USA, agriculture has reduced the relative abundance of ectomycorrhizal plants compared to other mycorrhizal types. However, global quantifications of agricultural impacts on distribution of mycorrhizas have not been possible until now. The current land use data underlying our maps (Table S4) enabled us to assess mycorrhizal distributions on Earth in the absence of croplands. We replaced the data on AM, EcM, ErM, and non-mycorrhizal plant abundances in each of the ecoregion-continent-land cover combinations that contained croplands by estimates of AM, EcM, ErM and non-mycorrhizal plant fractions that were expected at these locations based on natural vegetation types (see   Methods for details, and Tables S7-8 for data). Based on these data, we generated additional maps presenting potential natural distributions of AM, EcM, ErM, and non-mycorrhizal plants in a cropland-free world (Fig. S3).
Comparison of the current and cropland-free distributions of mycorrhizal types revealed that crop cultivation has led to a 50% (5.8 million km 2 ) increase in non-mycorrhizal vegetation, 19% reduction (5.5 million km 2 ) in EcM plant cover, 5.8% reduction (0.15 million km 2 ) in ErM plant cover and 0.6% reduction (0.3 million km 2 ) in AM plant cover (Fig. 2). Current coverage of AM plants is the result of an extensive re-distribution pattern, with large increases in Europe, Asia and North America (45, 27 and 12% of total AM gain, respectively), but also large declines in Africa, Asia (mostly India) and South America (25, 40 and 24% of total AM loss, respectively), where it was mostly replaced by non-mycorrhizal crops ( Fig. 2 A,C). Fig.   3 shows relative changes of each mycorrhizal type per continent.   There is growing empirical evidence that soil biogeochemical cycling processes qualitatively differ between ecosystems dominated by plants hosting distinct types of mycorrhizas, especially between ecosystems dominated by AM and EcM plants (7)(8)(9). For instance, EcMdominated ecosystems often exhibit higher soil carbon-to-nitrogen ratios (proxy for recalcitrance of carbon compounds and low nitrogen availability in soils) (8,12). However, it is not known how decreasing abundances of EcM plants in ecosystems are quantitatively associated with soil carbon content. The relative contribution of mycorrhizas to soil carbon sequestration processes compared to that of biome characteristics, previously suggested to be the central driver of soil carbon stocks (23), is also unknown. To address this question, we examined the global relationships between soil carbon content in the topsoil (uppermost 30 cm soil layer) (24), biome type (25), and abundances of AM and EcM plants. These analyses are justified, because the data sources are independent: the ecoregion classification and hence mycorrhizal type distribution does not account for edaphic parameters, whereas our soil C data (24) is unrelated to that of vegetation. Model comparison was based on the Akaike information criterion (AIC). The relative importance of each predictor was examined using the Lindemann-Merenda-Gold (LMG) metric. As ErM plants contribute to a small proportion of the biomass in the majority of ecosystems, the reliability of their cover estimation is lower than for AM and EcM plants. Therefore, we did not analyze the relationship between ErM plant cover and topsoil C content.
The model comprising biome and areal cover of EcM plants explained 46% of the variance in topsoil carbon (Cragg and Uhler's pseudo R 2 ). Despite the large data spread, the areal cover of EcM plants was an important driver of soil carbon content across biomes (P<0.001), accounting for one third of the explained variation in topsoil carbon (LMG=30%) in addition to that explained by biome (P<0.001, LMG=65%). The interaction between biome and EcM plant cover was only marginally important (LMG=5%, while P=0.043), indicating that the increase in topsoil carbon along with an increase in EcM plant cover is mostly a biomeindependent process. Fig. 4 shows the relationships between topsoil carbon content, biome and EcM plant cover. In contrast, AM plant cover showed a weak and idiosyncratic relationship to soil C, varying across biomes (Fig. S3). In the model comprising biome and AM plant cover as predictors of soil C content, biome was the main driver of soil C (LMG = 94%), while the explanatory power of AM cover was marginal (LMG=3%). This model performed worse than the model based on predicting topsoil C with biome and areal cover of EcM plants (ΔAIC= 1382). These contrasting patterns provide strong evidence for a global relationship between EcM plant cover and soil carbon accumulation. Although it can be argued that high abundance of EcM and ErM plants is a consequence rather than a driver of soil C accumulation, a large body of recent findings provides evidence that EcM are likely to be the drivers of soil carbon accumulation through two interacting mechanisms. First, EcM and ErM fungi produce greater biomass of more recalcitrant mycelium compared to AM fungi (4). Second, while EcM and ErM fungi are more efficient in taking up N in N-poor soils than AM fungi or roots (7), the former guilds immobilize most of the N taken up in their own biomass. This suppresses saprotrophic decomposition (26) and may further aggravate the soil N limitation and reinforce the competitive advantage of the EcM and ErM plants, although the extent to which EcM fungi affect activities of saprotrophic organisms has been suggested to be context-dependent (26). In agreement with these findings, recent paper of Averill and co-workers (27) suggest that at a local scale, decreases in abundance of EcM plants in ecosystems negatively affects carbon accumulation in topsoil.
Building on this hypothesis, we estimated the impact of declining EcM vegetation cover and EcM-mediated processes on global carbon content in topsoil. We based this estimation on the quantitative relationship between the within-biome cover of EcM plants and topsoil carbon content, provided by our glm (Fig. 2). Using the model, we calculated the difference in the amount of total soil carbon associated with mycorrhizal activities for the current distribution The relationship between EcM cover and topsoil C is significant and clearly shows the same trend across biomes, even though it is associated with a high uncertainty (Fig. 4).
Recent estimations (28) suggest that the total soil carbon loss due to agricultural practices accounts for 133 GT, with great acceleration of losses during the past 200 years (28). Our results suggest that soil C losses associated with reduction of ectomycorrhizal impacts due to agricultural practices account for 20% of total lost carbon. The magnitude of soil C losses associated with the reduction of activities of EcM due to loss of ectomycorrhizal vegetation is roughly 28% of the values of aboveground carbon losses due to deforestation in the period between 1700 and 2000 (86 GT C (28)). These estimates collectively suggest that changes in mycorrhizal vegetation types, associated with the conversion of natural habitats to croplands, constitute an important driver in the development of agriculture-induced soil carbon debt.
Our findings provide quantitative estimates for the importance of plant-soil interactions in human-driven alterations of the global carbon balance. Previous research has revealed that ecosystems dominated by EcM plants are typically associated with soils featuring higher carbon content (7,11). However, recent analyses conducted at regional scales suggest that while increased EcM dominance is associated with higher soil carbon-to-nitrogen ratio, it may be that decline in nitrogen rather than increase in carbon is the main factor underpinning the relationship(12, 20, 27). In contrast to these regional scale studies, our analysis indicates that across large geographical scales, higher cover of EcM vegetation is broadly associated with higher soil C content in topsoil.
In this study, we aim to provide a benchmark for global assessments relating ecosystem processes to the functioning of distinct types of mycorrhizas. Considering the virtual absence of quantitative global data about mycorrhizal distribution while there is a major demand for such data (9,10,29), our mycorrhizal distribution maps provide an essential basis for detailed and systematic analyses of mycorrhizal biogeography and the environmental drivers thereof.
Inclusion of these data into vegetation models will allow accounting for impacts of mycorrhizas on ecosystem processes from landscape to global scale as well as for spatially explicit and quantitative assessments on the role of mycorrhizas in the functioning of  Data and materials availability: All data is available in the main text or the supplementary materials.

Global data sets of mycorrhizal types and mycorrhizal vegetation
We integrated information about the plant mycorrhizal types and registered occurrences from  Because medium-sized islands were missing from the ecoregions map of Bailey, we added mycorrhiza information for such islands manually based on the literature references in our dataset (Table S2).
To correct for impacts of anthropogenic activities, water regime and edaphic factors on the areal distribution of mycorrhiza types within each ecoregion, we overlaid the distribution of mycorrhizal types obtained using the ecoregion approach with an up-to-date satellite observation-based land cover map for the year 2015 (38 land cover categories, spatial resolution of 300 m), generated by the European Space Agency (18) (see Table S4 for the list of land cover categories). Using this overlay and the vegetation by ecoregion data (Table S2) In this study we focused on the abundance of mycorrhizal plants and not on species diversity; we therefore did not map the distribution of orchid mycorrhizas. Plants species featuring orchid mycorrhizas exhibit great diversity, exceeding in species number ErM and EcM plant species together (2). However, orchid species are never abundant in ecosystems in terms of biomass, and therefore are unlikely to play an essential role in biogeochemical cycles.

Impact of crop cultivation on areal coverage of distinct mycorrhizal types
Using the data of mycorrhizal type abundance in ecoregions by continent by land cover, we replaced the data on AM, EcM, ErM, and non-mycorrhizal plant fractions in each of the ecoregion-continent-land cover combinations that contained croplands (land-covers of types 10,11,12,20, and 30; see Table S4) Table S2). In cases where forests are not expected within the ecoregion (for instance in tundra) a combination of grasslands and shrublands was assumed, and, similarly, the expected mycorrhizal fractions were assigned following the mycorrhizal types associated with the respective plant types in the current vegetation based on the per ecoregion references in Table S2. This resulted in an extra dataset describing combinations of vegetation as defined by ecoregion-continent-land cover without croplands (Table S7 for continent data and Table S8 for island data). Maps of potential mycorrhizal distribution in a cropland-free world (Fig. S4) were created based on this dataset, using the same R scripts as for the current distribution of mycorrhizas. We calculated the changes in area covered by particular mycorrhizal types for the current situation and the cropland-free world based on the sum of projected areas of grid cells for which land cover differed, multiplied by the fraction of each type of mycorrhizal vegetation in the grid cell.

Map validation and uncertainty analysis
The maps of the current mycorrhizal distributions were validated using the datasets of forest The data of forest biomass structure for Eurasia (19) provide information on per plot tree species abundances for a large number of European sites. As the data contain all records obtained since the 19th century, we used only the data recorded after 1999. Using our database of plant species and associated mycorrhizal types we assigned every tree species with its mycorrhizal type (1344 data points, Fig. S1). This provided us with a per-site data on the relative biomass of AM and EcM trees. We used these data as proxies for AM and EcM cover to compare with the data in our maps. We used the same approach for the data of Lin  Map uncertainty analysis.
To assess the uncertainty sources in the maps, we examined which land use classes represent the data points that deviate from the observed data by more than 25 percent units. Our analysis showed that the large proportion of those deviations (63% for EcM and 40% for AM) falls into those land use classes that represent a poorly described mixture of evergreen or mixed forests and grasslands, i.e. ESA classes described as various forms of "closed to open (>15%) forest" (Table S4). Further improvement of the ESA classification data will provide a possibility to improve precision of our maps.