Towards a predictive framework for biocrust mediation of plant performance: A meta‐analysis

Understanding the importance of biotic interactions in driving the distribution and abundance of species is a central goal of plant ecology. Early vascular plants likely colonized land occupied by biocrusts — photoautotrophic, surface‐dwelling soil communities comprised of cyanobacteria, bryophytes, lichens and fungi — suggesting biotic interactions between biocrusts and plants have been at play for some 2,000 million years. Today, biocrusts coexist with plants in dryland ecosystems worldwide, and have been shown to both facilitate or inhibit plant species performance depending on ecological context. Yet, the factors that drive the direction and magnitude of these effects remain largely unknown. We conducted a meta‐analysis of plant responses to biocrusts using a global dataset encompassing 1,004 studies from six continents. Meta‐analysis revealed there is no simple positive or negative effect of biocrusts on plants. Rather, plant responses differ by biocrust composition and plant species traits and vary across plant ontogeny. Moss‐dominated biocrusts facilitated, while lichen‐dominated biocrusts inhibited overall plant performance. Plant responses also varied among plant functional groups: C4 grasses received greater benefits from biocrusts compared to C3 grasses, and plants without N‐fixing symbionts responded more positively to biocrusts than plants with N‐fixing symbionts. Biocrusts decreased germination but facilitated growth of non‐native plant species. Synthesis. Results suggest that interspecific variation in plant responses to biocrusts, contingent on biocrust type, plant traits, and ontogeny can have strong impacts on plant species performance. These findings have important implications for understanding biocrust contributions to plant productivity and community assembly processes in ecosystems worldwide.

Biological soil crusts (biocrusts) -biotic soil surface communities comprised of varying assemblages of cyanobacteria, algae, bryophytes, lichens and fungi -occupy the top few millimeters of the soil surface in dryland ecosystems globally (Belnap, Weber, & Büdel, 2016). Fossil data suggest early biocrusts began their colonization of Earth's terrestrial surface some 2,500 million years ago (Beraldi-Campesi, 2013), predating the evolution of seed plants by at least 2,000 million years (Kenrick & Crane, 1997). This suggests that, during their colonization of dry land, early vascular plant (hereafter 'plant') communities likely encountered biocrusts, and that biotic interactions between biocrusts and plants may have been playing out for millennia. Today, biocrusts are estimated to cover ∼12% of the Earth's terrestrial surface (Rodriguez-Caballero et al., 2018), and are particularly widespread in dryland ecosystems, which comprise ~45% of global landmass (Prăvălie, 2016). As biocrusts and plants continue to coexist in ecosystems worldwide, we are offered a unique opportunity to study the impacts of biocrusts on plant performance in present-day communities where biocrusts and plants co-occur.
In recent decades, a growing number of individual studies have investigated biocrust effects on plant species performance worldwide Zhang et al., 2016) for understanding biocrust contributions to plant productivity and community assembly processes in ecosystems worldwide.
Biocrust community composition may also determine effects on plant species given biocrust type largely determines the magnitude of biocrust contributions to soil hydrology, and C and N cycling (Barger et al., 2016;Bowker, Mau, Maestre, Escolar, & Castillo-Monroy, 2011;Chamizo, Cantón, Miralles, & Domingo, 2012). Finally, community theory predicts biotic interactions may differentially influence species performance and trait organization along environmental gradients as resource limitations shift (Cornwell & Ackerly, 2009;He et al., 2013;Maestre et al., 2010), and the importance of niche-based processes increases with increasing abiotic stress (Bruno et al., 2003;Gross, Liancourt, Choler, Suding, & Lavorel, 2010;Liancourt, Callaway, & Michalet, 2005). As such, we posit that the magnitude and direction of plant responses to biocrusts may also be mediated by the ecosystem of origin of study organisms and disturbance.
To address knowledge gaps concerning the outcomes and predictors of plant responses to biocrusts, we compiled a global database of biocrust-plant interaction literature and employed meta-analytical techniques to synthesize global patterns in existing data. Our specific research objectives were to assess the overall effects of bio- Results from this meta-analysis are expected to have broad implications for understanding the effects of biocrusts on plant species performance. In turn, this knowledge will allow incorporation of biocrusts into broader plant community theory and ecosystem management practices. Moreover, given that global landcover of biocrust communities is expected to decline 20%-40% within the next 65 years in response to climate change and land use intensification (Rodriguez-Caballero et al., 2018), and local biocrust community structure may also shift in response to climate change (Ferrenberg, Reed, & Belnap, 2015;Reed et al., 2012), we believe it is critical and timely to examine relationships between biocrusts and plant communities to better understand how the ecosystems in which they co-occur will respond to global change.  [germination, survival, growth, cover, nutrient uptake, phenology, reproduction and diversity]) to generate the set of records to be considered. We then employed a systematic screening process to retain or exclude articles for this meta-analysis ( Figure S1). Eligible articles were defined as those including any comparison ('study') of the performance of plants grown in the presence of biocrusts to plants that were grown in biocrust-absent controls (i.e. bare soil, biocrust removal, or biocrust disturbance). We retained articles that quantified the impacts of biocrusts on plant performance variables (i.e. germination, survival, growth, cover, nutrient uptake, phenology and diversity) in observational or experimental settings, omitting studies that considered the effects of plants on biocrust communities. Individual articles often yielded multiple studies: for example, if a study compared multiple responses (e.g. germination and growth) of multiple plant species to biocrust presence, each plant response and species was considered separately, but given a unique numerical identifier to later test for non-independence.  Table 1) used as moderators in our multi-factor meta-analysis. We recorded the mean (X), standard deviation (SD), standard error (SE), and sample size (n) of both the biocrust and biocrust-absent (control) plots for the plant response variables. Data were extracted directly from tables, published supplementary materials, and from digitized figures using "xyscan" version 4.2.1 (http://rhig.physi cs.yale.edu/~ullri ch/softw are/xysca n/). A detailed description of our data extraction protocol is summarized in Appendix S1.

| Effect size
For each biocrust-present and absent comparison, we calculated an effect size for each plant response variable using mean values.
In addition, to investigate biocrust effects on 'overall plant performance', we estimated an overall effect size (and within-study variance; see below) for plant performance by averaging the effect sizes of all plant responses reported for each reported plant species. Specifically, the effect size of biocrust presence was calculated as the log response ratio: ln(X crust /X ctrl ), where X crust is the mean plant response in the biocrust treatment, and X ctrl is the mean plant response in the biocrust-absent control. When positive, this metric indicates that biocrusts have a beneficial influence on the plant response of interest and when negative, a detrimental influence. Log response ratios provide a standardized measure of plant performance with favorable statistical properties for meta-analysis (Hedges, Gurevitch, & Curtis, 1999) and means for comparisons among studies with different plant response metrics.

| Within-study variance
To account for differences in study precision, we weighted our analysis by estimating within-study variance for each study as in Hedges et al. (1999). Specifically, the within-study variance used in our weighted regressions was calculated as follows: where X crust and X ctrl are the mean plant response with and without in biocrust, SD crust and SD ctrl are the standard deviation of treatment and control means, and n crust and n ctrl are the number of replicates with biocrust versus biocrust-absent soil treatments, respectively.
If no measure of variance was reported for a study (SD or SE;20.8% of studies), we used imputation to calculate missing variances in our dataset (Nakagawa, 2015) using Taylors Law, the relationship between mean and variance (for of our dataset (log(SD pooled ) = (log(Xpooled ) * 0.7998) − 0.5236; R 2 = 0.73).

| Boosted regression tree data exploration
To explore the relative importance of the candidate moderators and their potential interactions in explaining variation among plant response to biocrusts, we performed boosted regression tree (BRT) analyses on candidate variables in each of the five plant response models ( We performed BRTs using the 'gbm.step' function in the gbm (Ridgeway, Southworth, & Runit, 2013) and dismo packages (Hijmans, Phillips, Leathwick, & Elith, 2017) as in Elith and Leathwick (2017).
This and all subsequent statistical analyses in this study were conducted in the R open-source software environment (version 3.3.3; R Core Development Team, 2017). In each BRT model, we included only those moderators that had sufficient representation in the dataset and corresponded to meaningful a priori hypotheses ( Figure S1a); we then weighted each analysis according to the within-study variance.
Models were simplified using the 'gbm.simplify' function suggested by Elith and Leathwick (2017). Simplified BRT models for each analysis included the most influential moderators and ranked them according to their relative contributions (which are scaled to sum to 100% within each model-i.e. the moderator explains X % of the variation explained by the fitted BRT) to the explanation of variation in effect size. Relative variable influences were derived as an average of variable influence in all trees in each BRT model (Friedman & Meulman, 2003). Potential interactions between moderators in final BRT models were explored using the 'gbm.interaction' function ).

| Mixed multi-factor meta-analysis
Following the selection of key moderators to be retained in each of the five plant variable response models via BRT, meta-analyses were performed by fitting mixed-effects meta-regression models using the rma.mv function from the metafor package (Viechtbauer, 2010) with restricted maximum likelihood estimation of parameters. We first used pure random effects models to estimate the overall weighted mean effect size for each plant response model (i.e. the weighted, overall log response ratios of the plant response variables to biocrust presence;  the residual between-study variance component ('STUDY_ID') as a random-effect variable. Then, for each of the five separate analyses, we investigated the relative importance of the categorical fixed-effect moderators (Table 1) included in each model (Table S1b, Figure 1) by analyzing a series of mixed-effect multiple meta-regression models, including a global model containing all the fixed factors (moderators) being considered for that dataset and each of the nested subset models containing one more fixed factor. Every model also contained the random effect STUDY_ID to account for residual between-studies variation. When categorical moderators were significant (Q statistic < 0.05), differences in moderator levels were detected using planned contrasts with the 'linearHypothesis' function from the car package (Fox & Weisberg, 2011). To explain residual heterogeneity and understand the potential effect of contextual factors on plant responses to biocrusts, we ran a series of separate univariate metaregression models for each analysis that included single significant moderators. Interaction terms were only fitted in models if found to be influential in simplified BRT models. Parameters associated with moderators with non-significant effects are not depicted graphically.
With these studies, we evaluated the response to biocrusts in a total of 171 plant species occurring in 40 plant families.  were most commonly identified as important moderators in simplified BRT models (Figures 2-4), while PLANT_ROOT_ MORPHOLOGY, PLANT_DURATION, and STUDY_LOCATION were unimportant. Importantly, BRT analyses identified no significant interactions among moderators in any of the plant response models. A lack of influential interaction terms among main effects in simplified BRT models could suggest that interactions were unimportant. However, it could also suggest that our dataset did not contain adequate sample size to assess the importance of these interactions as it can often take a substantially greater sample size to assess interaction terms relative to main effects in mixed-effects regression models (e.g. Leon & Heo, 2009). Following BRT identification, strong moderators identified for the five plant models were included in mixed multi-factor meta-analyses (Table S1b). Results for final simplified BRT models are summarized in Figure 2 and in additional detail in Appendix S3.
F I G U R E 3 Plant performance responses to biocrusts (weighted mean ± SE): (a) overall plant response ("AVG"), and the three important moderators of this model: (b) BIOCRUST_TYPE, (c) PLANT_NATIVNESS, and (d) PLANT_FUNCTIONAL_GROUP. The number of studies in each moderator group level are shown in parentheses. The p-value in the corner of each graph denotes the statistical significance of the explanatory variable in the plant performance model. Lowercase letters denote statistically significant pairwise differences between moderator levels at p < .05, and "*" and "+" denote the effect size of a given moderator level is statistically different from zero at p < .05 or 0.10 > p> .05 respectively [Colour figure can be viewed at wileyonlinelibrary.com] F I G U R E 4 Plant responses to biocrust presence (weighted mean ± SE) for the SOIL_REFERENCE_STATE explanatory variable in the five plant response models: (a) overall plant performance, (b) germination, (c) survival, (d) growth, and (e) cover. The number of studies in each moderator group level are shown in parentheses. Lowercase letters denote statistically significant pairwise differences between moderator levels at p < .05, and "*" and "+" denote the effect size of a given moderator level is statistically different from zero at p < .05 or 0.10 > p > .05 respectively [Colour figure can be viewed at wileyonlinelibrary.com]

| Plant functional group
PLANT_FUNCTIONAL_GROUP was also important for predicting plant responses across all models. Overall plant performance was impacted by plant functional type (p < .001; Table 2 Table 2). Grasses received the most benefit from biocrust presence, with C 4 grasses experiencing a 200% increase (p < .001; Figure 3), and C 3 grasses experiencing a 149% increase, in growth (p < .001; Figure 3) compared to biocrust-absent controls. Growth of non-N-fixing woody plants also increased 56% with biocrust presence (p = .016; Figure 3), while growth of N-fixing woody plants decreased by 38% (p = .010; Figure 3). Biocrust presence decreased the overall growth of plant communities with multiple plant functional types ('Community') by 42% (p = .011; Figure 3).
Plant cover responses to biocrusts were only statistically distinct from zero for N-fixing woody plants, which decreased 70% (p = .011; Figure 3). However, pairwise contrasts between plant functional types revealed among grasses, C 4 cover was 59% greater than that of C 3 species in the presence of biocrusts (p < .001; Figure 3). Among non-grasses, cover of non-N-fixing woody plants was approximately one-fold greater than that of N-fixing woody plant species (p < .001; Figure 3).

| Plant nativeness
PLANT_NATIVENESS was also an important predictor of overall plant performance (p = .011; Table 2), although pairwise differences between native and non-native species in the overall dataset were not statistically significant from zero or each other (Figure 3).
However, this overall neutral effect was likely driven by opposing native and non-native responses to biocrusts during germination and growth stages of the plant life cycle (Figure 3; Figure 5).
The presence of biocrusts reduced germination in non-natives by 10% (p = .100; Figure 3), while native species were unaffected ( Figure 4c). In contrast, while plant growth responses to biocrusts were also influenced by PLANT_NATIVENESS (p < .001; Table 2; Figure 3) the direction of biocrust influences on native and non-native species growth were reversed. Non-native species growth increased 51% in the presence of biocrust relative to biocrust-absent controls (p = .005; Figure 3), whereas the growth of native species was not affected.

| Soil reference state and other important moderators
Plant responses to biocrusts were also moderated by the type of uncrusted soil used to compare to biocrusted soils (SOIL_REFERENCE_ STATE; bare soil, biocrust removal, disturbed biocrust, or filter paper; Table 1). SOIL_REFERENCE_STATE influenced overall plant performance responses to biocrust presence (p < .001; Table 2; Figures 4 and 5), with overall performance 34% greater in the presence of biocrusts when compared to biocrust-removed controls (p = .024; Figure 4). Plant germination responses to biocrusts were mediated by soil reference type (p = .045; Table 2; Figure 4) with seedling germination marginally lower on soils with biocrust relative to disturbed biocrust controls (−12%; p = .097; Figure 4). Survival responses also differed by SOIL_REFERENCE_STATE (p < .001; Table 2). Mean effect sizes of biocrusts were negative for all control types, though SOIL_REFERENCE_STATE levels were not different from one another (Figure 4). Plant growth responses to biocrusts were influenced by SOIL_REFERENCE_STATE (p < .001; Table 2; Figure 4). Among biocrust-absent control surfaces, pairwise contrasts revealed plants benefited most from biocrust presence when compared to biocrust-removed controls (+190%; p < .001; Figure 4) while biocrust impacts on plant growth were slightly negative when compared to biocrust disturbance controls (−27%; p = .094; Figure 4).
Control type also influenced plant cover responses to biocrusts (p < .001; Table 2, Figure 4) with biocrust presence corresponding to a more than two-fold increase in plant cover when compared to biocrust removed controls (p < .001; Figure 4).
Finally, PLANT_DURATION was also an influential explanatory variable in predicting plant survival responses to biocrusts (p < .001;

| Biocrusts community composition determines plant responses
Biocrust community composition was consistently an important explanatory factor for understanding variation in overall plant performance, germination, growth, and cover (Figures 2, 3 and 5). While cyanobacterial biocrusts had few effects on plants at any stage, moss biocrusts increased both overall plant performance and cover, while lichen-dominated biocrusts considerably reduced overall plant performance and germination but lichen-dominated and mixed biocrusts increased plant growth. Potential mechanisms for such contrasts could be differences in water relations and soil fertility driven by differences in biocrust composition. Soil water availability can strongly influence biotic interactions and the structure of plant assemblages in dryland environments (Chesson et al., 2004;Miranda, Armas, Padilla, & Pugnaire, 2011) and has specifically been shown to mediate biocrust effects on plant community structure (Luzuriaga et al., 2012).
Differences in germination responses to biocrusts may be ascribed to differences in physical structure and water relations among biocrust types. Adequate water availability is first critical to seed water absorption during germination and subsequent seed metabolic activity and radical emergence (Fenner & Thompson, 2005).
Therefore, variability in germination responses among biocrust types can likely be ascribed to differences in community physical structure and impacts on soil water balance. Lichen-dominated biocrust surfaces, especially those with crustose, foliose, or squamulose lichens, are often hardened and hydrophobic (Souza-Egipsy, Ascaso, & Sancho, 2002;Tighe, Haling, Flavel, & Young, 2012), and can obstruct seed contact with, or penetration into mineral soil (Zhang & Belnap, 2015), which can expose seeds to drying or predation on the soil surface which may lead to decreased germination (Deines, Rosentreter, Eldridge, & Serpe, 2007;Schupp, 1995;Serpe, Orm, Barkes, & Rosentreter, 2006). In contrast, mosses grow in cushions (sometime loosely) and can capture water, including dew and fog (Pan et al., 2016) and thus often promote water infiltration into the soil (Eldridge et al., 2010) and soil water availability (Concostrina-Zubiri et al., 2017). This would enhance water availability to seeds and seedlings, promoting germination, possibly leading to moss- containing lichens have complex effects on soil hydrology (Chamizo, Belnap, et al., 2016), but can increase soil moisture by reducing runoff (Chamizo, Belnap, et al., 2016) and increasing absorptivity and water holding capacity (Belnap, 2006) which could increase soil water availability to plants. Lichen-dominated and mixed biocrust communities may also increase soil fertility (Barger et al., 2016). Plants grown with lichen and mixed biocrusts have been shown to have greater concentrations of N and phosphorus in their tissues than plants grown in the absence of these biocrust types (Ferrenberg, Faist, Howell, & Reed, 2018). Lichens with N-fixing cyanobacterial photobionts (cyanolichens; e.g. Collema) are associated with high levels of N-fixation (Barger et al., 2016;Rosentreter, Eldridge, Westberg, Williams, & Grube, 2016) and N-fixation may be higher yet in communities containing both cyanolichens and free-living N-fixing cyanobacteria (e.g. Nostoc, Scytonema; Barger et al., 2016).

| Plant functional group: photosynthetic pathway and symbiotic N-fixation influence plant responses to biocrusts
Plant functional traits, particularly those of beneficiaries of biotic interactions (Soliveres & Maestre, 2014), often predict the outcome of biotic interactions that may in turn influence community structure (Ackerly & Cornwell, 2007;Kraft & Ackerly, 2014;Kunstler et al., 2016;Lavorel & Garnier, 2002;Lebrija-Trejos et al., 2010;McGill et al., 2006). In this study, plant functional type, a proxy for multiple key plant functional traits (i.e. life form, photosynthetic pathway, N-fixation, woodiness), mediated plant response to biocrusts across all models (Table 2, Figure 3). Overall, C 4 species performance, survival, and cover responses to biocrusts were greater than that of C 3 species. C 3 grasses were only positively affected by biocrusts during growth (Figure 3). In contrast, C 4 species, despite a significant decrease in germination, showed an increase in both overall performance and growth by biocrusts. This pattern is similar to studies that have shown C 4 species receive greater benefits than C 3 species from the presence of soil microorganisms such as arbuscular mycorrhizal fungi (e.g. Hetrick, Wilson, & Todd, 1990;Hoeksema et al., 2010). Overall, our results conflict with our predictions for C 3 and C 4 grasses. C 3 species have lower water-and N-use efficiency compared to C 4 species (Pearcy & Ehleringer, 1984). Thus, we would expect C 3 species overall would receive greater benefits from biocrusts, which presumably increase soil water and nutrient availability relative to uncrusted soil. One potential explanation for this pattern is that biocrusts that contain darkly pigmented cyanobacteria (e.g. Nostoc, Scytonema, Tolypothrix) are often associated with elevated soil surface temperature (Couradeau et al., 2016), C 4 species may respond more favorably to biocrusts given their greater temperature requirements and tolerances compared to C 3 species (Pearcy & Ehleringer, 1984;Sage & Kubien, 2007).
Among non-grasses, plants species lacking bacterial N-fixing symbionts exhibited a more positive response to biocrusts than Nfixing species (Figure 3). This result suggests the benefits of N-fixing symbionts to plants are precluded in the presence of N-fixing biocrusts. Empirical evidence suggests that when soil nutrient limitations are relaxed, net benefits of maintaining N-fixing symbionts are decreased and may in turn lead to decreased performance of N-fixing plant species (Suding et al., 2005;Vitousek, Menge, Reed, & Cleveland, 2013). This pattern was less defined in survival, growth, and cover analyses, perhaps due to relatively low sample size of Nfixing forbs and woody plant species in these analyses, indicating additional studies are needed that directly compare the responses of plant species with and without N-fixing symbionts.

| Plant nativeness: Biocrust influences on native versus non-native plants shift across plant ontogeny
We might expect that biocrusts, acting as strong facilitators or inhibitors would similarly influence both native and non-native plant species performance in the case of similar traits among native and non-native species. However, since the native plant community has likely coevolved in the presence of biocrusts and may have already experienced historical and ongoing facilitation or filtering, we might expect a divergence in traits of exotics and native plants and a differential response to biocrusts.
Overall, biocrusts inhibited the germination of non-native species.
This negative effect is consistent with past reports that biocrusts pose greater inhibition to non-native versus native seeds (Deines et al., 2007;Hernandez & Sandquist, 2011;Song, Li, & Hui, 2017) and may be partially explained by physical interactions between nonnative seed morphological traits and biocrusts. Nearly half (48.6%) of germination studies included in our database addressed biocrust effects on non-native grasses with seeds with large awns (e.g. Bromus, Schismus spp.). Large awns may decrease or prevent contact between the seed and the mineral soil surface and can prevent the seeds from slipping into small cracks found in the biocrusts leaving seeds on the soil surface vulnerable to predation and lacking sufficient moisture to germinate (Belnap, Phillips, & Troxler, 2006;Deines et al., 2007;Morgan, 2006;Zhang & Belnap, 2015). Seed size may also govern plant germination responses to biocrusts. For instance, a study conducted by Morgan (2006) in grasslands of southwestern Australia found the large-seeded non-native grass species Briza maxima showed stronger inhibition by biocrusts than smaller seeded native species. Together, these morphological mechanisms are thought to play an important role in biocrust suppression of germination in awned, large-seeded Bromus species in the western US (Evans & Young, 1984;Hernandez & Sandquist, 2011;Howell, 1998;Peterson, 2013;Reisner et al., 2013) and Israel (Prasse & Bornkamm, 2000), Salsola species in Australia and the US (West, 1990), and Schismus species in Australia and Israel (Crisp, 1975;Zaady et al., 1997).
In contrast to germination responses, non-native plant species growth increased on average two-fold by biocrusts (Figure 3), indicating potential tradeoffs in non-native plant responses to biocrusts across plant ontogeny. This result is supported by individual studies that have reported increased growth in non-native and invasive plants by biocrusts (Defalco et al., 2001;Ferrenberg et al., 2018;Pendleton et al., 2003). Most existing studies compare responses of exotic annuals to native perennial plants. As annual plants often have greater relative fitness than native perennials when key resources are not limiting, as often found in biocrusted soils, these results are not surprising (Davis, Grime, & Thompson, 2000;Van Kleunen, Weber, & Fischer, 2010). These results also suggest intact biocrust communities can act as a barrier exotic grass species invasion by inhibiting germination. However, once established, the exotic annuals may be more able than the native perennials to utilize the resources available in biocrusted soils leading to heightened competitive ability.

| Soil disturbance mediates biocrust impacts on plant performance
Perhaps the best approach for understanding the importance of biotic interactions in filtering or facilitating plant species is to remove a putative influence and observe the effects. This approach TA B L E 3 Identified knowledge gaps and future research needs

Knowledge gap or needed research Description
Biocrust impacts on plant community assembly and diversity Direct tests of hypotheses pertaining to biocrust mediation of plant community assembly and diversity patterns at multiple spatial scales are needed.

| Biocrusts: biotic filters and facilitators for plant community assemblages?
Biotic interactions can strongly influence plant community assembly outcomes (Boulangeat et al., 2012;HilleRisLambers et al., 2012;Levine et al., 2004;Lortie et al., 2004 grasses responded more positively to biocrusts than C 3 grasses and N-fixing species were more negatively affected by biocrusts than non-N-fixing species.

The effect of biocrusts on plants shifts across plant ontogeny and may
suggest trait-based tradeoffs that may equalize overall performance of functionally diverse competitors. Biocrusts reduce germination in non-native plants and C 4 grasses but subsequently benefit these two groups in later life stages. Such trade-offs in interaction outcomes across plant ontogeny could be a mechanism that allows inferior competitors to coexist with these two groups which otherwise have adaptations that help to buffer them against environmental fluctuations.
4. Biocrusts can facilitate or inhibit potential plant community members, depending on the disturbance level. Our results suggest that, compared to a simulated highly disturbed environment, biocrusts are likely to exert a positive influence on potential plant community members, although the magnitude is contingent on biocrust type and plant traits. This observation aligns with ecological hypotheses that increased disturbance and/or abiotic stress may increase the importance of niche-based processes once stochastic influences of species dispersal dissipate (e.g. Ferrenberg et al., 2013;Jiang & Patel, 2008) and competition and facilitation between interacting species begins structuring communities (Bruno et al., 2003;Gross et al., 2010;Liancourt et al., 2005).
Biotic interactions are increasingly being incorporated into plant community theory (Bruno et al., 2003;Lortie et al., 2004;Maestre, Callaway, Valladares, & Lortie, 2009) and predictions into how communities will respond to accelerating environmental change (Brooker et al., 2008;He et al., 2013;McCluney et al., 2012;Van der Putten et al., 2010). Given the acute vulnerability of biocrusts to ongoing and future climate change and land-use intensification (Ferrenberg et al., 2015;Reed et al., 2012;Rodriguez-Caballero et al., 2018), understanding biocrust contributions to plant community assembly and structure may be particularly important for predicting how communities will respond to global change. We show biocrusts can have strong, context-dependent effects on plant species. Therefore, we suggest their integration in the development of plant community theory is needed, in a manner akin to ongoing efforts to understand the broader influences of soil microbial communities on vegetation community structure (Bever et al., 2010;Kardol, Cornips, Van Kempen, Bakx-Schotman, & Van Der Putten, 2007;Van Der Heijden et al., 2008).

DATA AVA I L A B I L I T Y S TAT E M E N T
The database compiled and used in this meta-analysis (BSC-PLANT Database) is available from the Dryad Digital Repository: https ://doi. org/10.5061/dryad.sr83ph7 (Havrilla et al., 2019).