Biogeochemical Effects of a Forest Understory Plant Invasion Depend More On Dissimilar Nutrient Economies Than Invader Biomass


 There is an increasing need to better understand how and why invasion impacts differ across heterogeneous landscapes. One hypothesis predicts invader impacts are greatest where the invader is most abundant (the mass ratio hypothesis; MRH). Alternatively, invader impacts may be greatest in communities where the nutrient acquisition strategies of the invader are most dissimilar from those of native species (the nutrient economy dissimilarity hypothesis; NEDH). We tested whether the effects of an invasive grass, Microstegium vimineum, on soil biogeochemistry were best explained by MRH, NEDH, or both. At three locations (Indiana, North Carolina, and Georgia), invaded and reference plots were established across a nutrient economy gradient. Plots varied in the relative abundance of arbuscular mycorrhizal (AM) vs. ectomycorrhizal (ECM) associated overstory trees, reflecting gradients in biotic nutrient acquisition strategies and edaphic factors. At two locations, we found NEDH predicted invader effects on soil conditions. The net effect of M. vimineum homogenized soil properties across the nutrient economy gradient towards conditions consistent with AM-dominated stands; as such, the nutrient economy gradients observed in uninvaded plots were mostly absent in invaded plots. At one location with high N availability and intermediate acidity, both ECM-dominance (NEDH) and invader abundance (MRH) predicted differences in soil moisture, pH, and nitrification rates. Collectively, these results suggest the biogeochemical consequences of M. vimineum depend, in part, on pre-invasion soil nutrient economies. Where pre-invasion conditions are known, we provide a scalable and predictive approach to determine where impacts on biogeochemical cycling of C and N may be greatest.


Introduction
Determining the role of individual species in ecosystem processes is a primary need in ecology given the widespread changes in the distribution and abundance of species resulting from human activities Wright 2003, Hooper et al. 2005). Whereas considerable evidence suggests the addition or loss of species can signi cantly alter processes such as nutrient cycling or productivity, species effects in ecosystems are often highly contingent, which hinders our ability to predict the ecosystem consequences of species gains and losses (Chapin III et al. 1996 Historically, the magnitude of a species' effect on ecosystem processes has been predicted by its relative dominance within the community, assuming its traits will exert effects proportional to its abundance: the mass-ratio hypothesis (MRH) (Grime 1998, Mokany et al. 2008). While there is some evidence that supports MRH (Kramer et al. 2012, Lee et al. 2014), impacts do not always correlate with invader biomass (Peltzer et al. 2009, Sokol et al. 2017, casting doubt on the universality of this hypothesis. As an alternative, and perhaps not mutually exclusive hypothesis, a species could have its greatest effect when its nutrient acquisition strategy is most dissimilar from those of the existing species in the community: the nutrient economy dissimilarity hypothesis (NEDH; Fig. 1c). While the MRH and NEDH are assumed to work in consort (Strayer et al. 2006), direct tests of their relative importance remain limited (Platt 1964, Quinn andDunham 1983) and none have explored the hypotheses at the ecosystem scale. While there is evidence of these mechanisms acting as non-alternative hypotheses in lab microcosms (Kuebbing and Bradford 2019), laboratory, greenhouse, and common garden experiments often do not scale to whole ecosystems (e.g. a forest stand; (Kumschick et al. 2014, Stricker et al. 2015). This knowledge gap has hindered development of models to predict how gains and losses of species in uence ecosystem processes across a range of ecosystem types (Cardinale et al. 2012, Pyšek et al. 2012).
An abundance of research suggests that trees alter soil properties to confer an advantage in effectively obtaining limiting resources (Finzi et al. 1998 vimineum has nutrient acquisition strategies most similar with those observed in AM-dominated stands. Previous work suggests M. vimineum may become dominate in both inorganic (Adams and Engelhardt 2009) and organic nutrient economies (Kourtev et al. 2002); though invasive plants might be more likely to invade AM-dominated forest stands (Jo et al. 2018). Therefore, we expect greater M. vimineum biomass in inorganic nutrient economies (Fig. 1b). However, evidence suggests both abundance and nutrient availability can predict M. vimineum effects on soil independently and collectively (Lee et al. 2014, Craig et al. 2015, Kuebbing and Bradford 2019), making it ideal for exploring MRH and NEDH as alternative and non-alternative hypotheses.
In this study, we evaluated the degree to which impacts of M. vimineum on various soil properties were explained by the ecosystem characteristics of the forest (i.e., supporting the NEDH), its biomass (i.e., supporting the MRH), or both (supporting NEDH and MRH as non-alternative hypotheses). We measured soil characteristics in reference and M. vimineum invaded plots across a gradient of AM-to ECMdominated forest types (Fig. 1a). We determined whether M. vimineum abundance differed in AM-and ECM-dominated forest stands. Since invasion impacts differed across the ECM-dominance gradient, we use reference models to estimate the invader effect as the difference between invaded and reference conditions (Fig. 1b). We use these measures to directly ask whether ECM-dominance, M. vimineum, or both as proxies for MRH and NEDH best explain invasion effects on soil nutrient cycling (Fig. 1c). Finally, to evaluate if patterns are generalizable, we replicated the study at three locations (Indiana, North Carolina, and Georgia) across the invaded range of M. vimineum in the eastern U.S..

Study Locations
We selected three study locations across the distribution of invasive M. vimineum in the eastern U.S. (SI Fig. 1 vimineum invaded plots at each location. Each plot consisted of three adjacent subplots (10m radius) within a single contiguous patch of M. vimineum and three subplots (10m radius) located nearby (i.e., within 40m) that lacked M. vimineum but was otherwise similar in slope, canopy coverage, and overstory and understory plant communities (SI Fig. 2). Within each subplot, we recorded the diameter at breast height and species (or genus) of all trees (≥ 5cm DBH, 4.5ft above average ground level) to attribute a previously known mycorrhizal association to the tree community (Wang andQiu 2006, Phillips et al. 2013). Overstory ECM tree dominance was calculated as a percentage of the total basal area within a subplot (Phillips et al. 2013) and then averaged by invasion status across subplots to estimate ECM dominance for invaded and reference at each plot. Although 'time-since-invasion' can mediate invasion effects (Strayer et al. 2006, Flory et al. 2013, Stricker et al. 2016), our primary goal was to investigate invasion impacts across the forest nutrient economy gradient. We captured a range of 'time-sinceinvasion' at each location, while within location opted to capture the nutrient economy gradient rather than controlling for invasion age. More detailed information regarding 'time-since-invasion' for these locations is published in Craig et al. (2019).

Field sampling
We collected soil and understory vegetation samples from invaded and reference subplots for three growing seasons from 2014-2016. All samples were collected during the time of peak biomass for M. vimineum (August in IN, and September in NC and GA). We collected soil cores using a slide hammer soil corer (5 cm in diameter) from each subplot, removing litter layers and collecting topsoil (0-5cm depth) including O horizons where present (e.g., in ECM-dominated plots) and (5-15cm depths). Since we saw no differences in 2014 and 2015 between the 0-5 and 5-15cm depths, we present the 0-5cm depths for simplicity. Cores collected from the three invaded and three reference subplots were composited at the plot level, totaling 30 invaded and 30 reference samples for each sampling year.
Samples were stored on ice, transported back to the lab, and sieved to 2mm to homogenate soils and remove ne roots. Sieved subsamples were reserved at 4°C for chemical assays and air-dried for total elemental analyses. At peak biomass, we also collected understory vegetation in subplots adjacent the soil samples by hand-clipping samples using 0.0625m 2 quadrats to quantify aboveground M. vimineum biomass on a grams per square meter basis (SI Fig. 3). Vegetation samples were sorted into M. vimineum and non-M. vimineum biomass, oven dried at 55C for at least 48hrs before being weighed. Aboveground biomass was averaged across subplots to obtain for comparison to the soil samples.

Laboratory analysis
Soil moisture content was determined gravimetrically. Soil pH was determined by measuring a 1:8 wet soil to 0.01M calcium chloride (CaCl 2 ) extract on a bench-top meter (VWR symphony pH meter; VWR international, West Chester, PA, USA). To quantify total soil C and N content, air-dried soils were ground to a homogenous powder and combusted on a Costech Elemental Analyzer 4010 (Costech Analytical Technologies Inc.; Valencia, CA, USA). Net nitri cation potential (nitri cation) and inorganic N concentrations were extracted in a 1:2.5 wet soil to a 2M potassium chloride (KCl) solution. Soil moisture was used to control for water content of soils in the extraction. Nitri cation was quanti ed as the change in nitrate concentrations after 14-day aerobic incubations at 23°C. Inorganic N (KCl extractable N) was quanti ed as the initial concentration of ammonium and nitrate. Extracts were quanti ed for inorganic N, ammonium and nitrate + nitrite (henceforth described as nitrate), using salicylate and cadmium column reduction methods, respectively, on a Lachat QuikChem 8500 series 2 autoanalyzer (Lachat; Hach, Loveland, CO, USA).

Data analysis
To determine the role of invasion status, location, and position along the nutrient economy gradient on soil properties and processes, we used univariate linear models with the following xed effects: invasion status (invaded, reference), location (IN, GA, NC), ECM%, and their interactions. Soil response variables were averaged across the three sampling time points (years 2014, 2015, 2016) and included soil pH, moisture (%), nitri cation (ug per g*day), and soil C:N. To determine if M. vimineum differentially established across the ECM-dominance gradient, we test their correlation with Pearson's R test.
To estimate the impact of invasion on soil properties and processes, we originally intended to use a paired plot approach. However, we found that the ECM-dominance of invaded areas did not strongly correlate with those in nearby reference areas (SI Fig. 2), making the paired plot assumption impossible for these data. Instead, we used data from the reference plots to (1) establish a baseline relationship between ECM% and soil response variables at each location, (2) predict the soil response for all invaded areas based on measured ECM%, and (3) estimate the invasion impact by subtracting the predicted from the expected values at each plot (Fig. 1b). Each impact estimate is also accompanied by 95% prediction intervals to show our level of uncertainty in these estimates (SI Fig. 6). Though we acknowledge that we did not directly measure invader effects (e.g., a common garden) and did not manipulate the system, our observational data and modeling approach provide an critical insight into putative drivers. Using a wellestablished natural gradient as a hypothesis for expected soil conditions provides an informed prediction against which we can compare conditions under invasion, offering an alternative approach to estimate invader impacts that incorporates quantitative estimates of the ecological context.
We t univariate linear regression models that used ECM-dominance, M. vimineum biomass and their interaction to explain Δ in soil conditions (i.e., the difference in soil characteristics between invaded and uninvaded reference plots across the mycorrhizal gradient; Fig. 1c). Results from the previous model showed that soil conditions and the baseline relationship between soil conditions and ECM% differed by location. For this reason, we t separate models for each location. For each Δ soil property (Δ soil moisture, Δ pH, Δ soil C:N, Δ nitri cation), we t a linear regression with ECM-dominance, M. vimineum biomass, and their interaction as predictors. We standardized M. vimineum biomass values by dividing by 100, so values were in a comparable range as ECM-dominance. We partition the variation in our main effects (ECM-dominance and M. vimineum abundance) and their interaction by calculating eta-squared (η 2 ) to determine whether NEDH (described by ECM-dominance), MRH (described by M. vimineum abundance), or both explained the most variation in Δ soil property. All statistical analyses were performed in R version 3.4.1" (R Core Team 2017) and visualization of regression models were made in R Package visreg (Breheny and Burchett 2017).

Site characteristics
Whitehall Forest had the highest N availability and more intermediate acidity, in comparison to Moore's Creek and Duke Forest. Locations differed in mean bulk density, texture, inorganic N, and soil pH.
Whitehall Forest had the highest mean KCl extractable nitrate and ammonium concentrations, as well as highest mean pH. Mean bulk density, percent clay, percent sand, and inorganic N were similar between reference and invaded conditions within location (SI Table 2).

Invasions across nutrient economies
We con rmed the presence of the nutrient economy gradient in reference plots at all three locations. ECMdominance was a signi cant predictor of soil conditions at all locations. We observed signi cant negative relationships between ECM-dominance ( Table 2, 'ECM') and pH (p < 0.01, Fig. 2a), soil moisture (p < 0.001, Fig. 2b), and nitri cation (p < 0.001, Fig. 2c); soil C:N ratios increased with greater ECMdominance (p < 0.001, Fig. 2d). Locations differed in their pre-invasion soil conditions, however, the magnitude of the relationship between ECM-dominance soil pH, moisture, and soil CN was consistent across varying soil types at all three locations (Fig. 2a,b,d). Location was a signi cant predictor in linear models, but ECM-dominance and location were not signi cant predictors of soil pH, moisture, or soil CN (Table 2).

Invasion promoted inorganic nutrient forms across the nutrient economy gradient
The effects of invasion status varied across the ECM-dominance gradient. The interaction between ECMdominance and location was signi cant in all the models (Table 2), most often altering ECM-dominated soil conditions more than AM-dominated soil conditions (Fig. 2e-h). Under M. vimineum invasion the relationship between ECM-dominance (Table 2, 'ECM: invaded') and soil pH, soil C:N, and nitri cation rates attened (Fig. 2e-h); yet, the relative differences among locations were still evident in M. vimineum invaded plots ( Table 2, 'invaded'). For example, in Whitehall forest, where reference pH (Fig. 2a), soil moisture (Fig. 2b), and nitri cation (Fig. 2c) were greatest, we observed the highest soil pH (Fig. 2e), soil moisture (Fig. 2f), and nitri cation (Fig. 2g) in the invaded locations.

Nutrient economy dissimilarity hypothesis
Of the two hypotheses, ECM-dominance (NEDH) explained the most variation in models predicting the difference in pH and soil C:N and at two sites for soil moisture and nitri cation (Fig. 3). At Moore's Creek and Duke Forest, ECM-dominance consistently accounted for over 50% of the variation in the difference in soil moisture, pH, and nitri cation under M. vimineum invasion, as well as soil CN at Moore's Creek. ECM-dominance explained a plurality of the variation in the change in soil CN at Duke Forest. At Whitehall Forest, ECM-dominance explained a plurality of the variation in difference in soil C:N. The interaction between ECM-dominance and M. vimineum biomass was signi cant for the difference in soil moisture at Moore's Creek and nitri cation at Duke Forest. Where ECM-dominance positively predicts the change in soil moisture and nitri cation, M. vimineum biomass is not correlated with the difference between reference and invaded soil moisture or nitri cation at both locations. This represents a signi cant difference between the slopes (SI Fig. 6), but not a synergistic or additive effect of both predictors.

Mass ratio hypothesis
Invader abundance, often along with ECM-dominance and the interaction, predicted the differences in soil conditions at one location, Whitehall Forest (Fig. 3). M. vimineum density positively predicted soil moisture, pH, and nitri cation at Whitehall Forest (SI Fig. 7). M. vimineum biomass explained the most variation in models of change soil moisture (eta-squared = 0.37; Fig. 3) and net nitri cation (eta-squared = 0.60; Fig. 3). M. vimineum density interacted with ECM-dominance to explain changes in soil pH at Whitehall Forest (Fig. 3), with both ECM-dominance and invader biomass as positive predictors (SI Fig. 6).

Discussion
We investigated whether invader biomass and ECM-dominance, as a proxies for MRH and NEDH, operate as alterative and non-alternative predictors of invaded soil conditions. Overall, NEDH more often predicted differences between reference and invaded conditions, indicating that knowledge of the pre-invasion nutrient economy gradient may be a useful and scalable tool for predicting where a plant invasion may have the greatest impact on soil biogeochemistry. Notably, M. vimineum invasion across the mycorrhizal gradient was associated with homogenization of the ECM-dominance gradient, altering slow cycling organic soil conditions more than inorganic nutrient economies. To the extent that loss of heterogeneity in nutrient economies may promote a reduction in species richness (Scott and Baer 2019), which has known effects on ecosystem functioning (Hooper et al. 2005), the longer-term consequences of invasion may be greater than demonstrated here.
While it has been widely assumed that MRH and NEDH should operate additively to predict the effects of invasion (Strayer et al. 2006), we found more evidence that NEDH alone was a better predictor of invaded soil conditions. The MRH assumes the traits of the most dominant plant exert the greatest effects on the community. Studies often use measures of native and invasive plant height as a metric of dissimilarly, which are highly correlated with biomass (Martin et al. 2017). While it is true that MRH by de nition incorporates some aspects of plant traits, which here we assume to exert a measurable effect on soil nutrient economies, it is surprising that nutrient economy dissimilarity (NEDH) alone, rather than M. vimineum abundance more often predicted invaded soil conditions. In contrast to our hypothesis, M. vimineum did not achieve higher biomass invade stands with nutrient economies it promotes, suggesting it is equally likely to form dense stands across forest types.
Our ndings suggest that in forested communities, understanding the pre-invasion nutrient economy of soils may be a useful tool for predicting if and where invasion alters soil conditions. In direct tests using experimental microcosms, Kuebbing and Bradford (2019) found that biomass effects interacted with trait dissimilarity to alter C mineralization rates. The authors manipulated invader abundance by altering the percent of invasive plant litter in mesocosms containing maple-poplar and oak-hickory litter mixes.
While this manipulation provides clear evidence that these processes can and do operate together, our observations in the eld are an interesting contrast. We only found evidence at one site for one variable of a possibly synergistic interaction between both mechanisms. There were no live plants in the Kuebbing and Bradford (2019) lab mesocosms, while in the eld in our study, plant interactions with soils and thus rhizosphere processes were present. Roots can alter soil microbial communities in many ways, for example, by altering the quantity (Phillips et al. 2011) and quality (de Vries et al. 2019) of root exudates that, in turn, can affect soil microbial respiration and rates of nutrient mineralization. Some of these processes may mask the direct effects of aboveground biomass on belowground processes in our observational study-perhaps making NEDH a better predictor of the differences between references and invaded conditions, as it inherently integrates both above and below ground plant-soil interactions. In a study of a northeastern forest in Connecticut, Ward et al. (2021) reported that understory shrubs associating with ericoid mycorrhizae weakened the negative relationship between soil CN and ECMdominance in forests. While it's unlikely ericoid associating shrubs had similar effects in our plots (since only one site -Whitehall Forest, GA -has an abundant ericoid shrub layer), the ndings of Ward et al. Pre-invasion soil conditions may help explain when MRH and NEDH drive invasion effects. While we observed nutrient economy gradients that were consistent with the predictions based on ECM% of the overstory trees across all three locations, soil pH and inorganic N at the GA location were higher in both reference and invaded plots than at the other two locations. This result may explain, in part, why Whitehall Forest, GA, was most sensitive to invader-induced changes in nitri cation. In reference plots in GA, for example, we observed soil pH consistently above 4.75, which are conditions favorable to nitrifying bacteria (Vitousek et al. 1982, Mushinski et al. 2019. Given that high M. vimineum abundance can increase pH (Ehrenfeld et al. 2001, Kourtev et al. 2002, Lee et al. 2014, inorganic N availability (Craig and Fraterrigo 2017), and moisture (Fraterrigo et al. 2014), all of which are drivers of nitri cation, the conditions in invaded plots may have become even more favorable for nitri cation. This result could be due to increases in other limiting substrates such as organic matter, which was higher in invaded plots in GA (SI Table 2). Conversely, in landscapes with lower available N and greater soil acidity, strong limiters of nitri cation rates and thus N cycling, the nutrient economy dissimilarity gradient was of greater importance for determining where the invader had the greatest effects on biogeochemical cycling. This result is in line with previous ndings where invader effects on soil conditions were greatest in sites with low soil fertility and high acidity prior to invasion (Dassonville et al. 2008). Nitri cation and soil moisture change on a faster temporal scale (days or months) in comparison to pH and soil CN (years or decades). The temporal scale may make moisture and nitri cation more responsive to inputs from an annual plant.
The outcome of M. vimineum invading AM-and ECM-dominated stands in equal densities and NEDH driving effects, appears to result more uniform soil conditions across the nutrient economy gradient. The ultimate effect homogenizes a once biogeochemically diverse landscape, making pre-invaded organic nutrient economies indistinguishable from an inorganic nutrient economy within each location. Invasive    Figure 1 a) Soil nutrient economies vary across an AM to ECM tree dominance gradient. AM associating trees promote fast-cycling, nutrient acquisitive conditions (inorganic nutrient economies), while ECM associating trees promote slow cycling conditions (organic nutrient economies). This gradient provides a well-established predictable backdrop upon which we can compare the differences between invaded and uninvaded (reference) conditions. b) Invasion can occur across the gradient their impacts can vary across the gradient. To quantify invader effects, we use reference models to estimate an invader effect as the difference between invaded and predicted reference conditions. c) Where mass ratio hypothesis (MRH) drives invader impacts, we expect invader effects to increase with invader abundance. Where nutrient economy dissimilarity hypothesis (NEHH) is driving the effects of invasion, we expect the greatest invasion effects in ECM-dominated soils. Where both operate in consort, we expect the effects to be greatest in ECM-dominated plots with greater M. vimineum biomass.

Figure 2
Forest ECM-dominance (aboveground community) predicts (a) pH, (b) soil moisture, (c) nitri cation, and (d) soil C:N in reference plots. M. vimineum-invasion changes the relationships between ECM-dominance and (e) soil pH, (f) soil moisture, (g) nitri cation, and (h) soil CN. IN as blue squares, NC as red triangles, and GA is depicted as green circles. Figure 3