Positive associations with native shrubs are intense and important for an exotic invader but not the native annual community across an aridity gradient

Positive interactions influence the assembly of plant communities globally, particularly in stressful environments such as deserts. However, few studies have measured the intensity and relative importance of positive interactions involving native and invasive species along aridity gradients. These measures are essential for predicting how dryland communities will respond to biological invasions and environmental change. Here, we measured the intensity and importance of positive associations formed between native shrubs and the annual plant community, which included highly invasive Bromus madritensis ssp. rubens (“B. rubens”) and native neighbours, along an aridity gradient across the Mojave and San Joaquin Deserts.

Facilitation occurs when a foundation species (i.e. the facilitator) offsets biotic or abiotic stresses that would otherwise inhibit the performance, abundance or species richness of beneficiary species (mechanisms reviewed by Callaway, 2007;Filazzola & Lortie, 2014;Michalet & Pugnaire, 2016). Importantly, foundation plants are not always interchangeable-some foundation species are better facilitators than others (Callaway, 1998), and large plants can be better facilitators than small ones (Tewksbury & Lloyd, 2001). The strength and relative importance of facilitation can also depend upon environmental severity. The stress-gradient hypothesis (SGH) predicts that the frequency (Bertness & Callaway, 1994), intensity (le Roux & McGeoch, 2010) and importance (Bertness & Callaway, 1994;Callaway & Walker, 1997) of facilitation should increase with environmental stress such that positive interactions are most intense and most important in the most extreme environments. Here, intensity refers to the absolute impacts of biotic interactions, and importance refers to the impacts of biotic interactions relative to all other factors (Brooker et al., 2005). Positive interactions do occur even in mild environments (Holmgren & Scheffer, 2010), but there is relatively consistent empirical support for the SGH across taxa and biomes (see meta-analyses by Lortie & Callaway, 2006;He, Bertness, & Altieri, 2013;Romero, Goncalves-Souza, Vieira, & Koricheva, 2015;Dangles, Herrera, Caprio, & Lortie, 2018; but see Butterfield, Bradford, Armas, Prieto, & Pugnaire, 2016).
Most empirical studies of plant invasions have focused on negative interactions, that is competition and predation (reviewed by Jeschke et al., 2012;Maron & Vila, 2001;Mitchell et al., 2006;Roy, Lawson Handley, Schonrogge, Poland, & Purse, 2011), but positive interactions can also influence invasion trajectories (reviewed by Simberloff, 2006;Travaset & Richardson, 2014). In this context, native species in deserts can exacerbate plant invasions by strongly facilitating the abundance (Lucero et al., 2019;Schafer et al., 2012), performance (Holzapfel & Mahall, 1999) and population growth (Griffith, 2010) of invasive plant species, or by indirectly increasing the competitive effects of invasive species on native neighbours (Llambi, Hupp, Saez, & Callaway, 2018;Reisner, Doescher, & Pyke, 2015). There is some evidence that the intensity of positive interactions between native and invasive species can vary along environmental gradients (Badano, Villarroel, Bustamante, Marquet, & Cavieres, 2007;Saccone, Pages, Griel, & Michalet, 2010), but very few dryland studies have measured the intensity and importance of such interactions along an aridity gradient. This knowledge gap is significant because dryland ecosystems are predicted to become hotter and drier in the future (Abatzoglou & Kolden, 2011;Archer & Predick, 2008), which could favour the expansion of exotic plant species (Bradley, Blumenthal, Wilcove, & Ziska, 2010) and shift the frequency and importance of biotic interactions away from competition and towards facilitation (He et al., 2013).
In changing drylands, positive interactions can benefit exotic plant species more than their native competitors (Abella & Chiquione, 2018). For instance, Lucero et al. (2019) monitored associations between native shrubs and the annual plant communityincluding native and exotic taxa-over three years in a California desert and found that shrubs facilitated the abundance of exotic annual species 2.75 times stronger than native annual species.
Interestingly, shrub-annual associations were least positive in the wettest years, which is consistent with the SGH. However, Lucero et al. (2019) explored a limited spatial scale that did not incorporate geographic variation in aridity and did not consider the importance of facilitation relative to other factors. Understanding variation in the intensity and importance of positive interactions involving native and invasive species along aridity gradients is essential for predicting how dryland communities will respond to biological invasions and environmental change (Badano et al., 2016;He et al., 2013).
The objective of this study was to investigate the extent that the highly invasive annual species Bromus madritensis ssp. rubens associations increase with relative aridity. To better understand interactions between B. rubens and the native annual community, we correlated the abundance of B. rubens with that of native annuals near and away from shrubs.

| Study area and species
We surveyed annual plant communities at peak flowering in April 2019 at six sites that spanned an aridity gradient (see Table A1 for site names, locations and aridity values) across the Mojave (n = 3) and San Joaquin (n = 3) Deserts, USA (Germano et al., 2011 where P was the total precipitation (mm) from 1 May 2018 to 30 April 2019, and T was the mean annual temperature (°C) during the same interval. Thus, low A dM values indicated high aridity. We chose this index because its components are recorded at practically all weather stations, which facilitates direct calculations of A dM at fine spatio-temporal scales relevant to biotic interactions in local communities. We also calculated a 20-year A dM value for each site by averaging yearly A dM values over the last 20 years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019). Importantly, our statistical analyses used the A dM of the study year (2018-2019), not the 20-year average, because long-term climatic trends are less relevant to the establishment and performance of B. rubens than current trends, as B. rubens seed banks persist less than two years in the field (Jurad, Abella, & Suazo, 2013).
Sites in the Mojave Desert were dominated primarily by the native shrub Larrea tridentata, but the native perennials Ambrosia dumosa, Lycium andersonii, Yucca brevifolia and Y. utahensis were also present at relatively low densities. Sites in the San Joaquin Desert were dominated almost exclusively by the native shrub Ephedra californica, but the native perennial Agave americana was present at low densities at one site (Cuyama). Here, we focused on the potential for native shrub species to act as facilitators because they are the dominant physiognomic class across our study area (Pan et al., 2015). All sites were invaded by the exotic annual species B. rubens, Schismus spp. and Erodium cicutarium. Bromus diandrus was present at low densities at the Cuyama and Carrizo sites. Among these exotic species, we chose to focus on B. rubens because it is reported as one of the region's most problematic invasive species (Hunter, 1991;Salo, 2004) due to strong negative impacts on community-level biodiversity (Brooks, 2000;Salo, 2005) and historic fire cycles (Abatzoglou & Kolden, 2011;Brooks et al., 2004;Fusco, Finn, Balch, Nagy, & Bradley, 2019 At each site, the A dM for the study year fell within the 95% CI of the 20-year A dM , except at the Las Vegas and Carrizo sites, which were more and less arid, respectively, than usual (Table A1)

| Sampling
We sampled the annual plant community using a paired shrubopen microsite contrast with a 0.5 × 0.5 m quadrat subdivided into 100, 5-cm 2 frames (Pescador, Chacon-Labella, de la Cruz, & Escudero, 2014). Shrub microsites were defined as the area immediately beneath the canopy of a shrub, and open microsites were defined as interstitial spaces at least 1 m from any shrub canopy. For shrub microsites, sampling quadrats were placed midway between the shrub centre and dripline. We did not sample areas more than In sampling quadrats, we recorded the abundance (no. of plants rooted inside the quadrat) and percentage cover (percentage of quadrat frames with a plant rooted inside) of B. rubens and native species (pooled), as well as the richness of native species, exotic species, and all species combined. Relationships among these particular measures are used to describe the invasiveness and impacts of exotic species in non-native communities (Pearson, Ortega, Ozkan, & Hierro, 2016). For all species, individual plants were easy to distinguish because asexual reproduction is absent. In addition, we haphazardly collected a single B. rubens individual from each quadrat and counted the number of spikelets produced on the longest inflorescence, as a proxy for fitness. We transported collected B. rubens plants back to the laboratory in individual paper sacks and measured the aboveground biomass (g) of each after drying to constant mass at 70°C for 72 hr, as a proxy for plant performance (Holzapfel & Mahall, 1999).

| Statistical analyses
Relative interaction indices (RIIs; Armas, Ordinales, & Pugnaire, 2004) were used to estimate the intensity of shrub-mediated effects on the annual plant community. We calculated RIIs as follows: where M s was a vegetation measure (e.g. B. rubens abundance, species We estimated the importance of shrub-mediated effects on the annual plant community using the I imp index (Seifan, Seifan, Ariza, & Tielbörger, 2010). We calculated I imp as follows: where N imp was the contribution of shrub-mediated interactions to a particular vegetation measure (e.g. B. rubens abundance, species richness), and E imp was the environmental contribution to the same measure. These components and their calculation are fully explained by Seifan et al. (2010). Like RII, I imp values range from −1 to + 1. Negative  (Seifan & Seifan, 2015).
We used t tests and linear mixed-effects models to characterize associations between native shrubs and the annual plant community. To evaluate the direction and magnitude of the intensity and importance of shrub-annual associations across all study sites, we performed independent one-sample t tests with RII or I imp (averaged at the site level) as the response variable. We evaluated the effects of aridity and shrub traits on the intensity and importance of shrubannual associations using independent linear mixed-effects models with RII or I imp as the response variable: A dM , shrub species and shrub height as fixed factors; and study site as a random factor. To better understand the ability of E. californica, L. tridentata and A. dumosa to act as facilitators, we used independent linear mixed-effects models for each shrub species with RII as the response variable; A dM and shrub height as fixed factors; and study site as a random factor. We could not explicitly consider L. andersonii because this shrub was surveyed at only one study site (Las Vegas). We contrasted the absolute mag- All analyses were performed in R, version 3.5.1 (R Development Core Team, 2018). All linear mixed-effects models used the lmer function of the "lmerTest" package (Kuznetsova, Brockhoff, & Christensen, 2017), which uses Satterthwaite's method to calculate denominator degrees of freedom (df hereafter). Post hoc Tukey contrasts were estimated using the emmeans function of the "emmeans" package (Lenth, Singmann, Love, Buerkner, & Herve, 2018). We did t tests using the t.test function (R Development Core Team, 2018).

| RE SULTS
Bromus rubens formed exceptionally intense (based on RII) and important (based on I imp ) positive associations with native shrubs. At each site, B. rubens abundance, cover, biomass and spikelet production (i.e. fitness) were at least 2.52 (and up to 70.47) times greater in shrub microsites than in open microsites (Table A2) In contrast, the native annual community did not associate positively with native shrubs. At the site level, native abundance, cover and species richness did not always differ by microsite but were never greater in shrub microsites (Table A2) Shrub traits and relative aridity influenced the intensity of some shrub-annual associations (Table 1), but never importance (Table 2).
When L. tridentata was the focal shrub (Table 3), RII values for B. rubens abundance and cover were least positive (though never negative) at the most arid sites when shrubs were shortest ( Figure 3). No other RII (Table 1) or I imp (Table 2) measurement was influenced by relative aridity, shrub species or shrub height.
We found no spatially based evidence of biotic interactions between B. rubens and the native annual community ( Figure A2). Bromus rubens abundance had no relationship with native abundance in any microsite or aridity context (Table 4). However, we reemphasize that microsite had a strong effect on the absolute abundance of B. rubens across the entire aridity gradient (Table A3). Interestingly, when L. tridentata was the focal shrub, the intensity (RII) of shrub-related effects on B. rubens abundance and cover was least positive (though never negative) under the shortest shrubs at the most arid sites. Otherwise, aridity did not predict the intensity or importance of B. rubens-native shrub associations. Hence, positive associations between B. rubens and native shrubs did not follow predictions derived from the SGH. In stark contrast to B. rubens, the native annual community generally formed negative associations with shrubs, regardless of aridity or shrub traits. Taken together, these findings suggest that native shrubs mediated biotic interactions that generally benefitted B. rubens but not the native annual community.

| D ISCUSS I ON
F I G U R E 1 Mean intensity (RII ± 95% CI) of shrub-mediated effects on the annual plant community at each of six study sites spanning an aridity gradient across the Mojave and San Joaquin Deserts, and averaged across all sites ("All"), according to independent one-sample t tests with RII as the response variable. RII > 0 suggests positive (i.e. facilitative) effects, and RII < 0 suggests negative (i.e. antagonistic) effects. Study sites are arranged from the least arid (Panoche) to the most arid (Carrizo). See Table A6 for complete statistics Our findings coincide with a growing number of studies reporting strong facilitation of exotic plant species by native species.
Positive interactions among exotic species are common (reviewed by Simberloff, 2006), and such "invasional meltdown" (Simberloff & Von Holle, 1999) is a key concept in invasion biology (Jeschke et al., 2012). The potential for native plant species to facilitate invasive species has received surprisingly little attention (Gallien & Carboni, 2017), but there are striking examples in the genus Bromus.
Griffith (2010) (2019) showed that native shrubs generally facilitated the abundance of exotic annual species, including B. rubens, much more than native annual species, and Abella and Chiquione (2018) reported a similar pattern in a long-term experimental study in the Mojave Desert.
The present study provides further evidence that exotic invaders can capitalize on positive interactions to a greater extent than native competitors, and extends this evidence to a regional scale. In addition to describing the intensity of such positive interactions, we also evaluated their relative importance. Measuring the intensity and importance of biotic interactions is essential for understanding the capacity of competition and facilitation to influence community assembly in general (Brooker et al., 2005) and the trajectory of biological invasions in particular. In this context, we suggest F I G U R E 2 Mean importance (I imp ± 95% CI) ("Imp") of shrub-mediated effects on the annual plant community at each of six study sites spanning an aridity gradient across the Mojave and San Joaquin Deserts, and averaged across all sites ("All"), according to independent onesample t tests with I imp as the response variable. I imp > 0 suggests that positive (i.e. facilitative) shrub-mediated interactions are important, and I imp < 0 suggests that negative (i.e. antagonistic) shrub-mediated interactions are important. Panel arrangement follows Figure 1. See Table A6 for complete statistics TA B L E 1 Results of independent linear mixed-effects models testing the influence of relative aridity and shrub traits on the intensity of shrub-annual associations along an aridity gradient spanning the Mojave and San Joaquin Deserts Note: RII was the response variable; de Martonne aridity (A dM ), shrub species ("Species") and shrub height ("Height") were fixed factors; and study site was a random factor (not shown). Significant (i.e. p < .05) effects appear in bold.

TA B L E 2
Results of independent linear mixed-effects models testing the influence of relative aridity and shrub traits on the importance of shrub-annual associations along an aridity gradient spanning the Mojave and San Joaquin Deserts  Accordingly, positive interactions mediated by trees, shrubs and cacti have been touted as an "insurance" (Michalet, 2006) for dryland biodiversity (see also Cavieres et al., 2015). This may often be the case, but we found that shrubs did not facilitate any community-level measure of biodiversity considered here (i.e. native species richness, exotic species richness, whole-community species richness) and ac- Note: RII for B. rubens abundance (log-transformed) or cover was the response variable; de Martonne aridity (A dM ) and shrub height ("Height") were fixed factors; and study site was a random factor (not shown). Significant (i.e. p < .05) effects appear in bold. RII for other vegetation measures did not vary with respect to shrub species (Table 1), and are thus not shown. Results specific to L. tridentata are displayed in Figure 3.

TA B L E 3
Results of independent linear mixed-effects models testing the influence of relative aridity and shrub height on the intensity of associations between Bromus rubens and the native shrubs Ephedra californica, Larrea tridentata or Ambrosia dumosa along an aridity gradient that spanned the Mojave Desert portion of the study F I G U R E 3 Results of independent linear mixed-effects models testing the influence of shrub height and relative aridity on the intensity of associations between Bromus rubens ("Bromus") and the native shrub Larrea tridentata at three sites spanning an aridity gradient across the Mojave Desert portion of our study. RII for Bromus abundance (log-transformed) or cover was the response variable; shrub height (cm) and de Martonne aridity ("Aridity"; low values indicate high aridity) were fixed factors; and study site was a random factor. See Table 3 for complete statistics. Regressions show ± 95% CI.  Travaset & Richardson, 2014).
To this point, over twenty years ago, Holzapfel and Mahall (1999) quantified associations between the native shrub L. tridentata and the annual plant community in the Mojave Desert and, contrary to our findings, reported that the annual plant community, including B. rubens and native species, generally formed positive associations with this shrub species. Importantly, the relative abundance of B.
rubens was much lower in the study of Holzapfel and Mahall (1999) than in our study, hinting that high levels of B. rubens invasion might be necessary to disrupt positive shrub-native annual associations.
If so, we might expect strong competitive interactions between B.
rubens and the native annual community under shrubs (Salo, 2005), as experimentally demonstrated by Brooks (2000). However, we found no evidence for this, regardless of relative aridity. There are several potential explanations for this. First, we observed relatively little variation in native abundance under shrubs, which may have reduced our ability to detect evidence for competitive interactions via spatial abundance relationships. Alternatively, our study may not have been conducted at the appropriate temporal stage of invasion to find evidence for competition in action. Said differently, the damage of B. rubens competition under shrubs-the depletion of the native annual community-may have already been done (note that there were no shrubs without B. rubens beneath them; Figure A2).

Furthermore, the effects of exotic annuals on native neighbours
can fluctuate year to year, ranging from negative in some years to positive in others (Lucero et al., 2019; see also Brooks, 2000), and we may have simply missed strong competition. Finally, B. rubens is not the only invasive annual that could disrupt positive associations between shrubs and native annuals. All study sites were invaded by exotic Schismus spp. and E. cicutarium. Both can be facilitated by native shrubs (Holzapfel & Mahall, 1999;Lucero et al., 2019; but see Brooks & Berry, 2006), and both can impose competitive effects on native annuals (Bishop et al., 2019;Schutzenhofer & Valone, 2006 studies have examined facilitation using coarse biodiversity metrics, especially species richness at the community level (Vega-Alvarez, Garcia-Rodriguez, & Cayuela, 2019), a very conservative approach.
We found little evidence for the SGH, but like most studies, our surveys included coarse biodiversity metrics, with the exception of more-detailed surveys of B. rubens performance. Furthermore, our study area sampled a modest range of aridity values potentially experienced by B. rubens populations across the non-native range, and our results may have differed had we included more arid or mesic locations in our surveys. Finally, RII values for B. rubens were exceptionally high and varied relatively little along our aridity gradient (see Cavieres et al., 2014, for an example of wider-ranging RII values), which may have reduced the power of our regressions (but see results specific to B. rubens-L. tridentata associations; Table 3). Given these considerations, it may not be particularly surprising that the SGH did not "hold water" here (see also Metz & Tielbörger, 2016).
It is unclear why RII values for B. rubens abundance and cover became less positive with aridity when L. tridentata shrubs were shortest. Compared to other shrub species, L. tridentata can be a poor facilitator (Hutto, McAuliffe, & Hogan, 1986; reviewed by Callaway, 2007) due to relatively strong competitive effects and allelopathy (Mahall & Callaway, 1992), but we found limited evidence for this. Across all sites, RII values for B. rubens abundance and cover (the only vegetation measures with RII values affected by shrub species; Table 1) were no different under L. tridentata than any other shrub species (Table A4). That said, L. tridentata was the only shrub species whose positive effects on B. rubens appeared to decline with aridity (especially when shrubs were short)-the opposite pattern predicted by the SGH. This pattern could arise if B. rubens became relatively less abundant under L. tridentata or relatively more abundant in the open as aridity increased, but it is unclear which occurred (Table A5; note the lack of a significant microsite × aridity interaction). The former could occur if the quantity, quality or availability of soil resources concentrated under L. tridentata canopies (Schlesinger, Raikes, Hartley, & Cross, 1996) declined with aridity, or if the competitive/allelopathic effects of L. tridentata increased with aridity.
Regardless, our findings underscore the potential for shrub traits (species identity and height in this case) to mediate the effects of aridity on shrub-annual associations (reviewed by Callaway, 2007), though not necessarily as predicted by the SGH. However, we emphasize that L. tridentata canopies were surveyed at only three study sites, all in the Mojave Desert. Thus, RII-aridity relationships under L. tridentata canopies were based on a small sample size (n = 3 sites) that spanned a narrow aridity gradient. Accordingly, we urge caution in interpreting these patterns.
This observational study did not test for mechanisms of facilita- exotics. However, no differences in the importance of positive effects among shrub species suggest a relatively simple and consistent mechanism such as shade or soil fertility (Schlesinger et al., 1996).
However, Abella and Chiquione (2018) recently showed that efforts to use positive interactions to restore native biodiversity benefitted exotic species more than native species. Similarly, we found that shrub-mediated interactions greatly benefitted B. rubens but not the native annual community, underscoring the potential for strong facilitation of invasive species to confound restoration efforts.

| CON CLUS IONS
We found evidence that spatial association with native shrubs strongly and consistently increased the abundance, cover, biomass

ACK N OWLED G EM ENTS
This work was supported by a York Science Fellowship to JEL. Nissa B. provided outstanding logistical support. We thank the BLM and Nature Conservancy for supporting our fieldwork on public lands.  (Kahle & Wickham, 2013). Site names are abbreviated by their first three letters. Table A1 provides full site names TA B L E A 1 Location, total annual precipitation (TAP) (mm), mean annual temperature (MAT ± SE) (°C) and the de Martonne aridity index (A dM ; formula given in main manuscript) for each study site during the study year  and over the past 20 years (± SE   (Lenth et al., 2018) applied to the linear mixed-effects models described in

TA B L E A 6
Mean (± 95% CI), intensity (RII) and importance (I imp ) of shrub-mediated effects on the annual plant community at each of six study sites (n = 20 shrub-open pairs at each site) that spanned an aridity gradient across the Mojave and San Joaquin Deserts, and averaged across all sites ("All"; n = 6). Whether means differed from zero (i.e. p < .05) was tested with independent one-sample t tests with RII or I imp as the response variable. Means (± 95% CI) are plotted in Figures 1 and 2