Invasive Impatiens glandulifera: A driver of changes in native vegetation?

Abstract Biological invasions are one of the major threats to biodiversity worldwide and contribute to changing community patterns and ecosystem processes. However, it is often not obvious whether an invader is the “driver” causing ecosystem changes or a “passenger” which is facilitated by previous ecosystem changes. Causality of the impact can be demonstrated by experimental removal of the invader or introduction into a native community. Using such an experimental approach, we tested whether the impact of the invasive plant Impatiens glandulifera on native vegetation is causal, and whether the impact is habitat‐dependent. We conducted a field study comparing invaded and uninvaded plots with plots from which I. glandulifera was removed and plots where I. glandulifera was planted within two riparian habitats, alder forests and meadows. A negative impact of planting I. glandulifera and a concurrent positive effect of removal on the native vegetation indicated a causal effect of I. glandulifera on total native biomass and growth of Urtica dioica. Species α‐diversity and composition were not affected by I. glandulifera manipulations. Thus, I. glandulifera had a causal but low effect on the native vegetation. The impact depended slightly on habitat as only the effect of I. glandulifera planting on total biomass was slightly stronger in alder forests than meadows. We suggest that I. glandulifera is a “back‐seat driver” of changes, which is facilitated by previous ecosystem changes but is also a driver of further changes. Small restrictions of growth of the planted I. glandulifera and general association of I. glandulifera with disturbances indicate characteristics of a back‐seat driver. For management of I. glandulifera populations, this requires habitat restoration along with removal of the invader.


| INTRODUC TI ON
Biological invasions are an important aspect of anthropogenic global change and are considered to be one of the major threats to biodiversity worldwide (Sala et al., 2000). A well-documented impact of species invasions is to reduce native biodiversity, species abundances, change community patterns, and ecosystem processes such as nutrient cycling in invaded communities (Dogra et al., 2010;Ehrenfeld, 2010;Vilà et al., 2011). However, it is difficult to disentangle cause and effect of an invasion. An alien species can invade an intact ecosystem and cause changes there, thus be the "driver" of the changes (Bauer, 2012;Didham et al., 2005;MacDougall & Turkington, 2005). Alternatively, invasion may be facilitated by earlier ecosystem changes, such as global warming, land use change, or disturbances. Then the invasion is only a symptom, and the invader a "passenger" of the underlying change (Bauer, 2012;Didham et al., 2005;MacDougall & Turkington, 2005). Drivers and passengers are the extreme positions of a continuum, and several invasive species rather fall in-between those categories (Bauer, 2012). Such "back-seat drivers" benefit from previous changes, but once established they become drivers of further changes (Bauer, 2012). Another challenge in assessing the impact of an invader are context-dependencies. Invasion can, for example, depend on ecosystem, invasion stage, or species traits (Kueffer et al., 2013). The more an invader is a passenger of changes, characteristics of the native ecosystem such as habitat conditions and species composition of the receiving community should influence the outcome of invasion and lead to differences between habitats. Invasion of a passenger is rather unlikely the more it relies on previous ecosystem changes. Knowledge about driver and passenger characteristics of an invader and context-dependencies is important to understand invasion processes and to develop more targeted management plans.
Impatiens glandulifera originating from the Himalaya mountains is a very common invasive species in Central Europe. Rapid spread and population growth of this annual species are enabled by a large number of seeds and their effective dispersal. Seeds are catapulted over several meters due to an explosion mechanism of the capsule and subsequently often transported by water flows (Beerling & Perrins, 1993). I. glandulifera invaded various wet habitats such as mesotrophic grass-and woodlands but increasingly also forests and ruderal sites outside of the riparian zone (Beerling & Perrins, 1993;Čuda et al., 2020;Pyšek & Prach, 1993, 1995. I. glandulifera is capable of suppressing native plants because of a high competitive effect along with a vigorous growth and the release of allelopathic substances such as 2-methoxy-1,4-naphthoquinone as shown in experimental studies (Bieberich et al., 2018;Gruntman et al., 2014;Loydi et al., 2015;Power & Sánchez Vilas, 2020;Ruckli et al., 2014;Vrchotová et al., 2011). Another factor benefiting I. glandulifera is, for example, release from natural enemies such as insect herbivores and parasitic rust fungi (Tanner et al., 2014).
Under field conditions, it can form dominant stands with a height of up to three meters (Beerling & Perrins, 1993;Bieberich et al., 2020). Nonetheless, the impact of I. glandulifera on native plant communities is rated ambiguously, and it is not clear whether the impact is causal, thus I. glandulifera being a driver of ecosystem changes. Comparing invaded and uninvaded sites Hejda and Pyšek (2006), Hejda et al., (2009), andDiekmann et al., (2016) found only weak, but Kiełtyk and Delimat (2019) found strong differences of plant diversity and composition. From a previous study, we know that I. glandulifera and native vegetation cover correlated negatively, and the correlation depended on environmental conditions at a particular site (Bieberich et al., 2020). However, with these observational approaches, causality of impact is difficult to address (Hejda & Pyšek, 2006;Kumschick et al., 2015;Stricker et al., 2015). Some studies-also with ambiguous results-experimentally removed the invader I. glandulifera (Cockel et al., 2014;Hejda & Pyšek, 2006;Hulme & Bremner, 2006). Such removal experiments can help to identify whether an effect is causal (Kumschick et al., 2015;MacDougall & Turkington, 2005). If the invader is a driver of changes, removal should rescue the state prior to invasion.
Response of the native community could also be caused by the disturbance of the treatment itself. Removal of any other, even native, species could have the same effect, for example, because this may lead to higher resource availability. The process of native community recovery could also take longer time than the study, and thus effects may not become visible yet, especially if there are legacy effects of the invasion. An effective method to study causal effects is to add the invader to the native community (Stricker et al., 2015). However, this is rarely implemented under field conditions because then, a careful handling of the invader is required.
The aim of this study was to investigate whether I. glandulifera has a causal negative impact on the native vegetation and whether this impact depends on the habitat. Due to its uneven distribution within one field site, I. glandulifera can be transplanted from an invaded patch into an uninvaded patch, without introducing the species to a new site. To disentangle cause and effect of invasion, we combined the classical approaches to compare invaded and uninvaded patches, and to remove I. glandulifera from invaded patches, with transplanting I. glandulifera into uninvaded patches. Thus, the transplanting represents a control for removal and vice versa. To test for habitat-dependence, we replicated this experimental approach in two different riverside habitat types, alder forests and meadows. We expect that I. glandulifera has a negative impact on the native vegetation, specifically on α-diversity, biomass and species composition of the resident vegetation, and on individual plant growth of resident species. For the latter, Urtica dioica was chosen as target species because it is one of the most frequent native co-occurring species of I. glandulifera in both habitats. Because of the high competitive and allelopathic effect of I. glandulifera on neighboring plants, especially native plant growth should be affected even within a short time leading to changed species abundances and plant performance at the spatial scale of the experimental plots. If I. glandulifera is a driver of changes having a causal impact, (a) removal of I. glandulifera is expected to have a positive (recovery) effect on the native vegetation, and (b) planting I. glandulifera into formerly uninvaded plots should have a negative impact on the native vegetation. Additionally, (c) establishment of planted I. glandulifera and impact of planting and removal are expected to depend on the habitat because plant growth and species interactions are shaped by environmental conditions. If I. glandulifera has no causal impact on the resident vegetation, its removal should have no recovery effect, and its planting should have no negative impact on the resident vegetation. The native vegetation could still differ between invaded and uninvaded patches if I. glandulifera has no causal impact but is only a passenger of changes.

| Implementation of the field experiment
Field studies were conducted at four riverside sites around Bayreuth, Germany, also used in a previous study (Bieberich et al., 2020 To choose positions for the plots, a grid of 20 m × 20 m was laid over each study site (Figure 1a), ten meters shifted to the grid of our previous study (Bieberich et al., 2020). In March to April 2016, all grid intersection points were checked for suitability to conduct either removal or planting of I. glandulifera there (Figure 1a). Suitability was predefined as an area of 2 m × 4 m homogeneous herbaceous vegetation, in spring either invaded by I. glandulifera with 5%-40% cover for the removal trial or uninvaded with a maximum of five After about 10 days, we checked whether the transplanted individuals had grown and replaced failed individuals once. We wanted to achieve that the uninvaded plots and plots where I. glandulifera was removed were free of I. glandulifera over summer, while naturally growing and planted I. glandulifera developed 15%-75% cover. This  (Eggenberg & Möhl, 2013;Jäger, 2017;Jäger et al., 2013;Schmeil et al., 2011), and total dry weight was recorded per species. To measure dry weight, all plant material was dried at 90°C for 2 days and weighed to the nearest 0.01 g (weighing scale Mettler PM 4,600). Thus, all biomass data, hereafter, are given as dry mass.
Because of these properties, RII enables further analysis with classical statistical methods (Armas et al., 2004

| Dependence of I. glandulifera performance on treatment and habitat
In the uninvaded control as in the removal treatment I. glandulifera remained mostly absent or occurred at very low abundances only (I. glandulifera dry biomass median 0.00 g, max. 0.87 g, cover less than 5%). On average 47 of the 63 planted I. glandulifera plants, corresponding to 74%, established. However, survival was lower in alder forests than in meadows (51% versus 85%, p = .012, Wilcoxontest). The planted I. glandulifera added up to a biomass of 7-186 g per plot (median 75 g, Figure 2). In natural occurrences in contrast, a higher I. glandulifera biomass was recorded (39-433 g, median 137 g, Figure 2). Cover of I. glandulifera ranged from 10% to 90% (Braun-Blanquet classes 2a to 5) and correlated strongly with biomass (combining planted and natural occurrences, Pearson correlation coefficient r = 0.797, p < .001, Figure A1). Planted I. glandulifera plants reached similar, but slightly smaller sizes as those naturally grown ( Figure 2): with 0.1-61 g biomass (median 4.8 g) plants did not differ significantly in biomass but planted ones had shorter stems than the naturally grown ones (median 126 versus 153 cm). Abundance and plant growth of both, planted and naturally grown I. glandulifera was lower in alder forests than in meadows (Figure 2).

| Habitat-dependent impact of I. glandulifera on the resident vegetation
In total 71 resident species were recorded (Table A1) Figure 4).

F I G U R E 3 Resident vegetation characteristics in the control treatments (a) and impact intensity of
Impatiens glandulifera planting and removal (b) depending on the habitat. With linear mixed-effect models using site as random factor, it was tested whether the shown parameters differed between control plots invaded and uninvaded by I. glandulifera and between habitats (p-values given). Impact intensity of I. glandulifera manipulation on each parameter is expressed by relative interaction index (RII) among manipulation and appropriate control per pair of plots. RII of −1 shows most negative impact, 0 no impact, and + 1 most positive impact. For planting and removal in both habitats separately, it was tested with a one-sample Wilcoxon test whether RII differs from zero (result indicated by asterisks). Sample sizes are given at the bottom of the graphs Note: The PERMANOVA was separately conducted for 1) the invaded and uninvaded control treatments, 2) planting trial, and 3) removal trial. Study sites were used as groups within which permutations were constrained.

| Impact of I. glandulifera on Urtica dioica and other frequent species
Urtica dioica grew significantly better in uninvaded than in invaded control plots regarding total biomass, cover, individual stem length, and individual vegetative biomass (Figure 5a). U. dioica total biomass was not changed by I. glandulifera manipulations while cover was slightly, but not significantly, decreased by I. glandulifera planting and increased by removal (Figure 5b). Individual plants of U. dioica, however, were affected by the manipulations regarding all considered parameters (Figure 4). Impact intensity on stem length was low but significant for planting. Impact on individual plant biomass of U. dioica was slightly higher. Median RII through planting was −0.11 with a maximum decrease from 6.2 to 2.6 g (RII −0.41), median RII through removal was 0.23 with a maximum increase from 1.2 to 4.8 g (RII 0.59). Impact intensity on infructescence biomass was very high but only significant in the removal trial (Figure 5b).
Besides U. dioica, the most frequent resident species were Galium aparine, Filipendula ulmaria, Stellaria nemorum, and Phalaris arundinacea. Total biomass of P. arundinacea was higher in invaded plots, but total biomass of the other species was independent of invaded or uninvaded situations ( Figure A2a). RII of I. glandulifera planting and removal on each of those frequent species was highly variable and never significantly different from zero ( Figure A2b). However, median total biomass of G. aparine decreased by planting and median total biomass of F. ulmaria, G. aparine, and S. nemorum increased by removal.

| D ISCUSS I ON
In this field study, we experimentally removed Impatiens glandulifera from invaded plots, and planted I. glandulifera in formerly uninvaded plots in order to test whether I. glandulifera has a negative impact on the native vegetation in riparian meadows and alder forests, and whether the impact is causal or not. We found  Note: With linear models, it was tested whether the impact of I. glandulifera depended on trial (planting and removal of I. glandulifera), habitat (meadows and alder forests), and their interaction term. Study site was used as random factor (lmer) unless its variance was estimated zero, thus no random factor was used (lm). p-values <.05 are given in bold.

| Impatiens glandulifera had low but causal impact on native vegetation
Removal of Impatiens glandulifera had a positive and planting a negative effect on total resident plant biomass and growth of Urtica dioica individual plants. This indicates that I. glandulifera is a driver of ecosystem changes having a causal negative impact on the resident vegetation. A causal impact of I. glandulifera on native vegetation is also indicated by Hejda and Pyšek (2006), Hulme and Bremner (2006), and Cockel et al., (2014) who all found positive, but often only slight effects of I. glandulifera removal on riparian plant species diversity and composition, which were, however, not affected in the present study. A causal impact on U. dioica plants as found in the present study is underpinned by experimental studies on competitive and allelopathic interactions of both species (Bieberich et al., 2018;Gruntman et al., 2014;Tickner et al., 2001). However, the impact of I. glandulifera on U. dioica competing in a pot experiment was much stronger (relative interaction index RII about −0.7, in Gruntman et al., (2014) and Bieberich et al., (2018)) than under the field conditions in the present study (median RII planting −0.09).
Taken together the impact of I. glandulifera can be rated as low. Total resident biomass and individual plant growth of U. dioica were affected by planting and removal indeed, but only to a small extend, and α-diversity, species composition, vegetation height, and total biomass of the most frequent co-occurring species were not affected by the manipulations at all.
Criteria of a clear driver of changes were only partially met for I. glandulifera in the present study. If the species was a clear driver, planted I. glandulifera should establish and clearly suppress natives, while removal would lead to recovery of the native vegetation (Bauer, 2012;Didham et al., 2005;MacDougall & Turkington, 2005). In the present study, planted I. glandulifera reached similar, but slightly smaller sizes than naturally growing ones. Establishment and growth of I. glandulifera were lower in alder forests than in meadows. Thus, I. glandulifera growth was slightly restricted by resident vegetation and native plant species α-diversity was not affected at all. Species composition, vegetation height, U. dioica total biomass, and U. dioica cover differed among invaded and uninvaded plots. However, they were not subsequently affected by removal and planting of I. glandulifera. On the one hand, this can indicate that differences between invaded and uninvaded plots were not caused by I. glandulifera but are due to F I G U R E 5 Urtica dioica in the control treatments (a) and impact intensity of Impatiens glandulifera planting and removal (b). With linear models, it was tested whether the shown parameters differed between control plots invaded and uninvaded by I. glandulifera (pvalues given). Study site was used as random factor (lmer) unless its variance was estimated zero, thus no random factor was used (lm). Impact intensity of I. glandulifera manipulation on each parameter is expressed by relative interaction index (RII) among manipulation and appropriate control per pair of plots. RII of −1 shows most negative impact, 0 no impact, and + 1 most positive impact. For planting and removal separately, it was tested with a one-sample Wilcoxon test whether RII differs from zero (result indicated by asterisks). Sample sizes are given at the bottom of the graphs. Only pairs of plots are considered in which U. dioica occurred in both plots. Stem length, vegetative, and infructescence biomass of U. dioica represent mean values of 6-8 plants per plot other factors, such as habitat conditions or disturbances. If these factors already differed between plots before I. glandulifera invasion, they themselves could be one reason for the invasion success at a particular patch. In this case, only comparing invaded and uninvaded patches observationally could lead to the false conclusion that I. glandulifera has a negative impact on native vegetation. On the other hand, a response of the native vegetation to the I. glandulifera manipulations indicating a causal effect could take longer time than the study duration of one season (Cockel et al., 2014;Rusterholz et al., 2017). Also between-year variations could obscure long-term effects. However, the manipulations affected total native biomass and performance of U. dioica, the response of which is faster and more sensitive in comparison with diversity measures. This indicates a fast competitive and allelopathic effect on the growth of neighboring plants as known for the annual I. glandulifera from the seedling stage onwards (Bieberich et al., 2020;Gruntman et al., 2014). Another limitation of this experimental study design is that the removal and planting of any other plant species could have the same effect as the removal and planting of I. glandulifera, and thus the results might not be specific to I. glandulifera. However, results of the present study are corroborated by a previous observational study within the same sites, which underpins that I. glandulifera has no impact on α-diversity, species composition, and vegetation height, but on abundance of U. dioica (Bieberich et al., 2020). We suggest that continuing the manipulations for more than one season may lead to a change of total abundance of U. dioica as a consequence of the reduced growth of individual plants.
If I. glandulifera is not a strict driver of changes, it could be a backseat driver, whose invasion is favored by previous ecosystem changes until it becomes a driver of further changes itself (Bauer, 2012).
Affinity of I. glandulifera to habitats with natural and anthropogenic disturbances and changed land use (Ammer et al., 2011;Beerling & Perrins, 1993;Čuda, Rumlerová, et al., 2017;Čuda et al., 2020;Pyšek & Prach, 1993, 1995 also indicates characteristics of a backseat driver. However, to clearly distinguish a back-seat driver from a driver is not possible with the present study. To this end, it would be necessary to test whether removal of the invader would result in recovery of the initial state of an ecosystem only in combination with habitat restoration (Bauer, 2012).

| Causal impact of I. glandulifera depended only slightly on the habitat
We found a consistent effect of I. glandulifera manipulations on native vegetation in alder forests and meadows: In both habitats, I. glandulifera caused a reduction of total resident biomass but had no causal impact on species composition, α-diversity, and vegetation height. According to a linear model, RII on total biomass did not differ between the two habitats, alder forests and meadows. However, there was a small difference between habitats, as the RII on total biomass was significantly different from zero in alder forests but not in meadows in the I. glandulifera planting trial. This indicates a higher impact in elder forests, where both, the biomass of I. glandulifera and the resident vegetation was lower than in meadows. In contrast, in our previous study within the same study sites, we found negative correlations between cover of I. glandulifera and cover of U. dioica, F. ulmaria and total cover, which were stronger under bright conditions with higher I. glandulifera cover than under dark site conditions (Bieberich et al., 2020). Comparing invaded and uninvaded sites, also Diekmann et al., (2016) suggested a higher impact of I. glandulifera in open than in more shady habitats. Thus, the correlative impact seems to be stronger habitat-dependent than the short-time causal impact and more pronounced in bright habitats.

| Implications for assessment of impact and for nature conservation
We found that the impact of I. glandulifera on native vegetation was causal but low. The response of the native vegetation to the I. glandulifera manipulations was quite fast within one vegetation period, even if only some parameters were affected within the study duration. Also other field studies on I. glandulifera using a removal approach found effects on native vegetation within one season (Cockel et al., 2014;Hejda & Pyšek, 2006;Hulme & Bremner, 2006), whereas only in Hulme and Bremner (2006), the effect was quite high. This means that invasion can have a negative impact after a short period of time, but also removal as management measure could have a fast effect. However, the impact of I. glandulifera could also increase over time after invasion (Rusterholz et al., 2017), and longer lasting removal can also enhance a management effect (Cockel et al., 2014;Rusterholz et al., 2017).
We suggest that I. glandulifera is not a clear driver of changes, but it has some characteristics of a back-seat driver benefiting from previous changes such as disturbances or changed land use. This is relevant for nature conservation because drivers and back-seat drivers require a different management strategy. In the case of a driver, removal of the invader, which induced the changes, is ideally sufficient (Bauer, 2012). In contrast, in the case of a back-seat driver, habitat restoration is necessary in addition to removal of the invader (Bauer, 2012). Thus, management of a back-seat driver is more complicated because the previous changes that facilitated invasion have to be known and countered. Such previous changes can be all kinds of alterations of ecosystem properties such as land use change, pollution, nutrient input, or altered disturbance regimes (Bauer, 2012;Didham et al., 2005). Unfortunately, there is often no reliable information on the original community and ecosystem processes available (Parker et al., 1999).  2016)). In this case, it can be recommended to prevent the potential invasion of a back-seat driver while planning and conducting the disturbance (D'Antonio & Meyerson, 2002;Lapin et al., 2016). It is also possible that I. glandulifera invasions are favored by anthropogenic nutrient input as I. glandulifera has an affinity to nutrient-rich patches (Bieberich et al., 2020;Čuda et al., 2014). Thus, reducing the nutrient input into water bodies as a general aim of nature conservation may also reduce invasion of I. glandulifera. In the case of already established populations of I. glandulifera, it can be discussed if a management is reasonable, considering the rather low impact of I. glandulifera in combination with its high abundance and frequency in Central Europe. Since a population control can be very expensive (Leblanc & Lavoie, 2017), it should be reserved for sites which are particularly valuable in terms of nature conservation.

| CON CLUS ION
Impatiens glandulifera had a causal but low impact on the resident vegetation in both riparian habitats, alder forests and meadows. The effect could be seen already after one season, but may also intensify over longer time. Impatiens glandulifera had some characteristics of a back-seat driver, which is facilitated by previous ecosystem changes but is also a driver of further changes having causal impact on the invaded ecosystem. If I. glandulifera has to be managed for nature conservation, this involves the need of ecosystem restoration along with removal of the invader.

ACK N OWLED G M ENT
We would like to thank the Bavarian state water authority

CO N FLI C T O F I NTE R E S T
The authors declare that there are no conflicts of interest.

F I G U R E A 2
Total biomass per 1 m 2 of the most frequent resident species in the control treatments (a) and impact intensity of Impatiens glandulifera planting and removal (b). With a Mann-Whitney-U test, it was tested whether the shown parameters differed between control plots invaded and uninvaded by I. glandulifera (p-values given). Impact intensity of I. glandulifera manipulation on each parameter is expressed by relative interaction index (RII) among manipulation and appropriate control per pair of plots. RII of −1 shows most negative impact, 0 no impact, and + 1 most positive impact. For planting and removal separately, it was tested with a one-sample Wilcoxon test whether RII differs from zero (result indicated by asterics). Sample sizes are given at the bottom of the graphs. Only pairs of plots are considered in which the respective species occurred in both plots