Impacts of detritivore diversity loss on instream decomposition are greatest in the tropics

The relationship between detritivore diversity and decomposition can provide information on how biogeochemical cycles are affected by ongoing rates of extinction, but such evidence has come mostly from local studies and microcosm experiments. We conducted a globally distributed experiment (38 streams across 23 countries in 6 continents) using standardised methods to test the hypothesis that detritivore diversity enhances litter decomposition in streams, to establish the role of other characteristics of detritivore assemblages (abundance, biomass and body size), and to determine how patterns vary across realms, biomes and climates. We observed a positive relationship between diversity and decomposition, strongest in tropical areas, and a key role of abundance and biomass at higher latitudes. Our results suggest that litter decomposition might be altered by detritivore extinctions, particularly in tropical areas, where detritivore diversity is already relatively low and some environmental stressors particularly prevalent.

A key question in contemporary ecology is whether changes in biodiversity lead to alterations in the functioning of ecosystems and associated biogeochemical cycles 1,2 . Interest in this topic emerged in the 1990s, motivated in part by the remarkable increase in global biodiversity loss 3 , and led to hundreds of experiments that manipulated biodiversity at different levels (species, genes or functional traits) in different groups of terrestrial and aquatic organisms, to examine possible effects on ecosystem processes 4,5 . While this large body of primary research and subsequent syntheses have demonstrated a strong, positive role of diversity of primary producers on biomass production [6][7][8] , the patterns for decomposition have proven to be weaker and less consistent 6,9 . This contrast may occur because decomposition can be simultaneously affected by the diversities of plant litter, microbial decomposers and animal consumers, with consequently more complex relationships 10 .
Plant litter decomposition is a key process in the biosphere, as 90% of the annual plant production escapes herbivory 11 and eventually becomes litter, which is ultimately decomposed or sequestered in terrestrial or aquatic ecosystems 10 . Streams play a particularly important role in receiving and processing litter from their catchments 12 , contributing significantly to global carbon and nutrient fluxes [13][14][15] . Litter enters streams mainly in the form of leaves, and it is decomposed by microorganisms (mostly aquatic hyphomycetes) and specialised invertebrates (litter-consuming detritivores) that can obtain carbon and nutrients from the litter and associated fungi 16,17 .
Multiple studies have manipulated detritivore diversity and assessed its effect on decomposition locally in streams or in laboratory microcosms, with inconsistent results 10 . These inconsistencies have been attributed to the existence of different species interactions driving either positive 18,19 or negative effects 20,21 , which can compensate for each other and sometimes result in overall neutral effects 22 . However, there has been no global assessment of the relationship between detritivore diversity and decomposition in streams, which would help account for local and regional environmental contingencies in the diversity-decomposition relationship 23 . A metaanalysis of terrestrial and aquatic studies revealed strong effects of detritivore diversity on decomposition, but there was no separate assessment of instream decomposition 9 . Several stream studies have suggested a direct link between faster decomposition 24 and greater detritivore diversity 25,26 in temperate streams, but did not explore the relationship explicitly. A large-scale study demonstrated that decomposition in streams was enhanced when detritivore assemblages were more complex (large-and medium-sized organisms as opposed to medium-sized only), although it did not examine detritivore diversity 27 .
Here, we describe results from a global-scale decomposition experiment conducted by partners of the GLoBE collaborative research network (www.globenetwork.es) in 38 streams distributed across 23 countries in all inhabited continents. We use a standardised design and methodology to examine global-scale ecological questions, which reduces the number of confounding factors that need to be statistically controlled for in a meta-analysis 28,29 . Our main working hypothesis is that detritivore diversity has a major positive effect on decomposition 9 , although we also expect an influence of other detritivore assemblage characteristics such as abundance, biomass, and body size 18,22,27 . Moreover, we predict that biotic drivers of decomposition vary across sites at different latitudes, possibly because of the varying interplay between positive and negative species interactions 22 . We also explore detritivore variation across latitudes, biogeographic realms, biomes and climates, to further explain their global distribution and the potential consequences of reduced diversity for decomposition in different areas of the world. Unlike previous large-scale decomposition studies using 1 or 2 litter types 24,30 , we use several mixtures representing a variety of litter traits to maximise the generality of our results. Our global experiment supports the expected positive relationship between detritivore diversity and decomposition, and reveals that detritivore species loss may have its greatest consequences on stream ecosystem functioning in the tropics.

Results
The model that best explained global variation in total decomposition explained 73% of the variation and revealed a significant influence of detritivore diversity, abundance, biomass, latitude, and interactions between diversity and latitude, abundance and latitude, and biomass and latitude (Table 1 and Supplementary  Table 1). The model that best explained global variation in detritivore-mediated decomposition explained 82% of variation in the data, and showed that the interactions between diversity and latitude, abundance and latitude, and biomass and latitude were significant (Table 1 and Supplementary Table 1). As these results indicated that the three detritivore variables were important predictors of decomposition, but their influence varied with latitude, we explored the interactions with a second type of model where latitude was a categorical variable (Supplementary Table 2). These models revealed that the relationship between detritivore diversity and decomposition was stronger in tropical areas than in temperate areas and absent in boreal areas; and that abundance and biomass were important in temperate and boreal areas, but not in tropical areas ( Fig. 1 and Supplementary Table 2).
All detritivore variables varied significantly among realms, biomes and climates, and so did assemblage composition (Figs. 2-4, Table 2 and Supplementary Table 3). Diversity and abundance were highest in the Palearctic realm, tundra and temperate broadleaf and coniferous forests, and warm temperate and snow climates; and lowest in Neotropical, Afrotropical and Indomalayan realms, tropical wet forests and savannas and xeric shrublands, and equatorial climates. Biomass and mean body size were highest in Palearctic and Nearctic realms, temperate broadleaf and coniferous forests, and again warm temperate and snow climates, with the lowest values in the Indomalayan realm, tropical savannas and xeric shrublands, and Table 1 Results of the best additive models explaining variation in total and detritivore-mediated litter decomposition based on detritivore diversity, abundance, biomass, mean body size, latitude, and interactions between detritivore variables and latitude. All predictors were fitted as tensor product interaction smooths. We show effective degrees of freedom (edf) and values of F and p for each factor. Models explained 69% and 78% of variation in the data, respectively.

Total decomposition
Detritivore-mediated decomposition Detritivore mean body size (mm) meta-analysis of controlled experiments performed in terrestrial and aquatic ecosystems 9 . Our result also agrees with results of controlled experiments that found average increases in decomposition of 12-30% per detritivore species added 18,19,31 , suggesting that positive interactions (i.e. resource partitioning and facilitation) are prevalent in detritivore assemblages. Clearly, our field study does not demonstrate causality among these variables or the suggested mechanisms, but the finding of a consistent relationship across 113°of latitude indicates that detritivore diversity, at least at the family level, is indeed a driver of decomposition. Whether this relationship would change by considering species diversity cannot be currently ascertained due to limited taxonomic knowledge in many regions 32 .
The relationship between detritivore diversity and decomposition, when data were grouped according to latitudinal zone, was most evident in tropical areas, less important in temperate areas and unimportant in boreal areas (although the latter were underrepresented in our dataset). Others have demonstrated a positive relationship between detritivore diversity and decomposition in some streams of boreal areas 33 , but our global dataset indicates a relatively weak relationship when compared to other latitudinal zones. Importantly, the stronger relationship between detritivore diversity and decomposition in the tropics suggests that species losses in these areas, where detritivore diversity is already lower than at higher latitudes as shown here and elsewhere 25,26 , may cause the greatest impact on decomposition. Detritivores in tropical areas are particularly vulnerable, because of the prevalence of multiple environmental stressors. For example, concentrations of agricultural pesticides have limited regulation in many tropical countries 34 and are known to cause mortality in many detritivores [35][36][37] . Climate warming is also likely to cause more extinctions in the tropics because more detritivore species are closer to their thermal maxima than elsewhere 25 and are likely to suffer greater physiological changes, despite the smaller changes in temperature occurring in this latitudinal zone 38 . Nevertheless, other climatic changes such as increased droughts can be more important at higher latitudes 39 .
We found that the influence of detritivore abundance and biomass on decomposition also varied with latitude, but with negligible effect in the tropics and more important at higher latitudes. These variables have previously been found to be important predictors of decomposition in some tropical streams 40 , but here their importance was lower in the tropics than elsewhere. In temperate areas, both relationships were non-linear and complex (with decomposition first decreasing and then increasing with higher abundance or biomass), which impedes predictions about how decomposition might be altered by changes in these variables. Moreover, responses of abundance and biomass to environmental stressors are not as straightforward as diversity loss, because lost species can be replaced by more tolerant ones that thrive under stressful conditions and can cause an overall increase in numbers 41,42 . Smaller detritivores are often more sensitive to stressors than larger ones 42 , although this variation could be due to taxonomic differences rather than to size. Our results suggest that species replacements under environmental stress could result in an overall increase in biomass, but this possibility needs confirmation.
The distribution of most detritivore families corresponded to broad realms (Fig. 5), with 26 families showing a Laurasian distribution (i.e. being present in the Palearctic and/or Nearctic realms) and 14 families a Gondwanan distribution (Neotropical, Afrotropical, Australasian, and/or Indomalayan realms). Although we did not perform phylogenetic analyses, this dichotomy, together with the observation that diversity and abundance of detritivores were higher in the Palearctic and Nearctic (and their predominant biomes and climates), suggests that patterns of variation in diversity and abundance were at least partly determined by biogeography. Our findings contrast with those for angiosperms, current distributions of which do not correspond to tectonic history, possibly because of the existence of high transoceanic dispersal 43 ; however, they support patterns for organisms with lower dispersal, such as liverworts and conifers 44 , which show clear Laurasian-Gondwanan disjunctions 45 .
The strong influence of biogeography on detritivore diversity and abundance, and the fact that these two variables are key Table 2 Results of linear mixed effects models exploring variation in detritivore and total invertebrate diversity, abundance, biomass and mean body size, and PERMANOVAs exploring variation in assemblage composition, among realms, biomes and climates.  Fig. 4 Global distribution of study sites in different climates [A, equatorial (Af, fully humid; Am, monsoon; As, with dry summer; Aw, with dry winter); C, warm temperate (Cfa, fully humid with hot summer; Cfb, fully humid with warm summer; Csa, with dry and hot summer; Csb, with dry and warm summer); D, snow (Dfb, fully humid with warm summer; Dfc, fully humid with cold summer)]; n = 38. Box plots show the median, interquartile range and minimum-maximum range of litter-consuming detritivore diversity (number of families per litterbag), abundance (number of individuals per litterbag), biomass (mg per litterbag) and mean body size (mm) in each climate (ordered from highest to lowest diversity); different letters indicate significant differences. The NMDS ordination of litter-consuming detritivores with biomes is represented by polygons of different colours as in maps and box plots. Significant differences in assemblage structure were: Aw vs. Cfb, Cfa, Dfb; Af vs. Cfa, Cfb, Dfb. drivers of decomposition, suggest that the split of Pangea in the Late Jurassic (≈200 Ma ago) had a crucial legacy effect on the current functioning of stream ecosystems and the influence of ongoing environmental change. The lower detritivore diversity of tropical streams 25 and the higher susceptibility of their fauna to extinction 38 make these streams more vulnerable to reductions in decomposition rates that are associated with impaired ecosystem functioning 46,47 . This observation, together with the overexploitation of natural resources that severely affects tropical stream ecosystems 48 , indicates that tropical detritivore species should be of high conservation concern globally. Methods Study sites. We conducted our study in 38 headwater streams located in different regions in 23 countries (Figs. 2-4). A random distribution of sites was unfeasible, so some regions were underrepresented (mostly Africa and northern Asia), which is usually the case for globally distributed experiments 28,49,50 . Streams were similar in size (mean ± SE: wetted channel width, 3.9 ± 0.1 m; water depth, 28.7 ± 0.4 cm; 1st-3rd order) and physical habitat (alternating riffles and pools). Most had rocky substrate and were shaded by a dense riparian vegetation (64 ± 1%) representative of the region. They were located in 6 realms, 7 biomes, and 10 Köppen climate classes 51 . In each stream we selected a ca. 100-m long reach with 5 consecutive pool habitats in which to conduct the experiment. Further information on site physicochemical characteristics is given in Supplementary Table 4.
Field and laboratory work. At each site, we incubated 6 different 3-species litter mixtures, which included 9 species in total (Supplementary Table 5). The species and mixtures were chosen to represent different levels of functional diversity for a companion study 52 , but here our interest was to use a variety of mixtures and thus increase the generality of our results (as opposed to working with a single or a few species). The 9 species were collected at different locations around the world and distributed among partners 52 ; we considered the possible home-field-advantage effect of using litter from different origins negligible based on available literature 53,54 . Litter mixtures were enclosed within paired coarse-mesh (5 mm) and fine-mesh (0.4 mm) litterbags containing the same amount and type of litter. The two types of litterbag respectively quantified total and microbial decomposition, and allowed the calculation of detritivore-mediated decomposition (see below). There were 60 litterbags per stream (n = 5 per litter mixture and mesh size), each containing 3 g of senescent litter (1 g per species), which had been collected freshly fallen from the forest floor, air-dried and distributed among research partners 52 . Litterbags were deployed in each stream (one litterbag per litter mixture type and mesh size in a different stream pool, with all 5 pools consecutive) in 2017-2019 at the local time of the year with the greatest litter input and were retrieved after 23-46 d, depending on water temperature in each stream, thereby halting the decomposition process at a comparable stage (mean ± SD: 32 ± 17% litter mass loss on average for all the litter mixtures, 41 ± 18% for the fastest decomposing mixture 52 ; mean values for each biome are given in Supplementary Fig. 1). Litterbags were transported to the laboratory on ice enclosed individually in zip-lock bags and rinsed with filtered stream water to remove attached sediment and invertebrates. Litter was oven-dried (70°C, 72 h) and a subsample weighed, incinerated (500°C, 4 h) and re-weighed to calculate the final ash-free dry mass (AFDM). Invertebrates were sorted, and litterconsuming detritivores were counted and identified under a binocular microscope to the highest taxonomic level possible (mostly species or genus, and family in some cases), using available literature and local expert knowledge.
Calculation of variables. We quantified litter decomposition in each litterbag as the proportion of litter mass loss (LML) per degree day (dd), to account for differences in temperature across sites; LML = [initial AFDM (g) -final AFDM (g)]/initial AFDM (g), where initial AFDM was previously corrected by leaching, drying and ash content, which were estimated in the laboratory 55 . We calculated detritivore-mediated decomposition as the difference in LML between paired coarse-mesh and fine-mesh litterbags 30 . Total and detritivore-mediated decomposition were strongly correlated (r 2 = 0.90, p < 0.001), but we used both as response variables in the analyses because the former is more relevant at the ecosystem level and the latter reflects patterns mediated solely by detritivores.
We quantified detritivore diversity in each coarse-mesh litterbag as taxon and family richness; as they were strongly correlated (r 2 = 0.90, p < 0.0001), we used family richness for analyses to avoid taxonomic inconsistencies among sites. We quantified abundance as the number of individuals per litterbag. We estimated total biomass based on mean body size using published equations for each family, and mean body size based on abundance and the mean of a body size category (2.5-5.0, 5.0-10.0, 10-20, 20-40 and 40-80 mm) that was assigned to each family using available literature [56][57][58][59][60][61][62][63] .
Data analyses. We examined the influence of detritivore diversity, abundance, biomass, mean body size, latitude and the interactions between detritivore variables and latitude on decomposition, using generalised additive models (GAMs, gam function, 'mgcv' package v. 1.8.31 64 Table 2). Models were fitted using tensor product interaction smooths (ti) with a normal or gamma distribution (depending on model fit and residuals) and the identity-link function 67 . We used this type of model instead of a linear model because preliminary data exploration showed the existence of non-linear patterns 68 . Total or detritivore-mediated decomposition was the response variable, and detritivore diversity, abundance, biomass, mean body size, absolute latitude and the interactions between detritivore variables and latitude were predictors, fitted as smooth terms. Exploring differences among litter mixtures was beyond the scope of this study (but see Boyero et al. 52 , where litter diversity effects on decomposition were examined based on the same experiment described here), so we averaged values of different mixtures rather than including the mixture as a random factor in a generalised additive mixed model, which would be highly complex and would not converge when using interactions and variance functions (see below). Spatial correlation among sites was tested using the autocorrelation function (ACF) with residuals of the final model; all values were <1 as recommended by Zuur et al. 67 . Abundance and biomass data were log (x + 1)-transformed to avoid the disproportionate influence of outlying data observations on model estimates 68 . As interactions of detritivore variables with latitude were significant, we explored the relationships for tropical (≤23°of latitude), temperate (24-60°) and boreal zones (>60°) through a model that was similar to the one described above, but with latitude as a categorical rather than continuous predictor. This was done to facilitate the representation and interpretation of complex nonlinear relationships between two continuous predictors. We explored differences in detritivore variables across realms, biomes and climates with linear mixed-effects models (lme function, 'nlme' package v. 3.1.151 69 ) where realm, biome or climate were fixed factors and litter mixture type was a random factor, followed by pairwise comparisons using adjusted P-values (glht and mcp functions, 'multcomp' package v. 1.4.13 70 ). The variance was allowed to differ among realms and biomes using the VarIdent structure. Normalised residuals of the final model were inspected with plots of residuals vs. each predictor, and no pattern was observed. Variation in assemblage composition was explored with non-metric multidimensional scaling (NMDS, monoMDS function, 'vegan' package v. 2.5.6) 71 calculated on Hellinger transformed abundance data and permutational analysis of variance (PERMANOVA) based on a Bray-Curtis dissimilarity matrix. We compared realms, biomes and climates (adonis function, 'vegan' package), followed by pairwise comparisons (pairwise. adonis function), and determined which were the most representative families in each assemblage (simper function). All analyses were run on R v. 4.0.2.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.