Non-native plant drives the spatial dynamics of its herbivores: the case of black locust (Robinia pseudoacacia) in Europe

Non-native plants typically benefit from enemy release following their naturalization in non-native habitats. However, over time, herbivorous insects specializing on such plants may invade from the native range and thereby diminish the benefits of enemy release that these plants may experience. In this study, we compare rates of invasion spread across Europe of three North American insect folivores: the Lepidoptera leaf miners Macrosaccus robiniella and Parectopa robiniella, and the gall midge Obolodiplosis robiniae, that specialize on Robinia pseudoacacia. This tree species is one of the most widespread non-native trees in Europe. We find that spread rates vary among the three species and that some of this variation can be explained by differences in their life history traits. We also report that geographical variation in spread rates are influenced by distribution of Robinia pseudoacacia, human population and temperature, though Robinia pseudoacacia occurrence had the greatest influence. The importance of host tree occurrence on invasion speed can be explained by the general importance of hosts on the population growth and spread of invading species.


Introduction
Plants introduced to new, non-native habitats may have an advantage over the native flora by escaping herbivore pressure, allowing them to allocate more resources toward vegetative and reproductive growth, as formulated e.g. in the enemy release hypothesis (Keane and Crawley 2002). In such a setting, non-native plants can quickly become widespread and invade various habitats. Black locust, Robinia pseudoacacia (Fabaceae), is a prime example of this, now being one of the most widespread non-native trees in Europe (Vítková et al. 2016). The native range of this species is limited to the central Appalachian and Ouachita mountains and the Ozark Plateau in the Eastern and Central United States (Huntley 1990). Black locust was introduced to Europe during the first half of the 17 th century as an ornamental tree planted in parks and gardens (Wein 1930), and from 1750 on, it was used in forest plantations in Central Europe for purposes of timber and honey production. From these plantings, it spread prolifically and is currently found throughout most of temperate and sub-Mediterranean Europe (Fig. 1A), displacing native vegetation and altering ecosystem properties (Vítková et al. 2016).
Although widely distributed, European populations of black locust were little affected by the few native generalist herbivores feeding on it, with generally marginal impact on the tree (e.g. Bartha et al. 2008). In contrast, five specialist herbivores accidentally introduced to Europe from the native range of black locust were found to have a considerably higher impact on the tree. The first North American insect species discovered feeding on Robinia in Europe was the sawfly Euura tibialis (Newman, 1837) (Hymenoptera: Tenthredinidae), formerly placed in Nematus (Prous et al. 2014). This species was first recorded in Europe in 1825 (Rasplus et al. 2010), twelve years before the species' formal description from the Isle of Wight (Newman 1837). In Europe, this parthenogenetic species is found feeding on Robinia pseudoacacia and R. viscosa, while in its original North American range it also feeds on R. hispida and Gleditsia triacanthos (Darling and Smith 1985;Liston 2011).
More recently, four additional Robinia herbivores were accidentally introduced from North America to Europe: In 1970, Parectopa robiniella Clemens, 1863, a Lepidoptera leaf miner of the Gracillariidae family, was recorded from Northern Italy (Vidano and Marletto 1972). It was followed by Appendiseta robiniae (Gillette, 1907) (Aphididae), an aphid first found in 1978 in Italy (Micieli De Biase and Calambuca 1979). Another Gracillariidae leaf miner, Macrosaccus robiniella (Clemens, 1859), was first found in 1983 in Northern Switzerland (Whitebread 1990). This species was placed in Phyllonorycter Hübner, 1822 until recently, when it was transferred to Macrosaccus Davis &De Prins, 2011 (Davis andDe Prins 2011). Finally, in 2003 the black locust gall midge, Obolodiplosis robiniae (Haldeman, 1847) (Diptera: Cecidomyiidae), was reported from Northeast Italy (Duso and Skuhravá 2002). Upon the arrival of the four most recently introduced Robinia herbivores in Europe, black locust was widely distributed and naturalized on the continent. The four herbivore species thus found their food source in abundance and were subject to little competition from more generalist native European herbivores, so that they could extend their distribution range.
Relatively little is known about how the range expansion of specialized non-native herbivorous insects is affected by the distribution of their native host plant in nonnative regions. European Robinia pseudoacacia and its introduced specialist herbivores are a prime opportunity to study such a setting in more detail. In order to better understand the factors promoting the range expansion of these non-native herbivores and to better predict spread patterns in other parts of black locust's non-native range, we analyze the three most well-documented Robinia herbivores present in Europe (P. robiniella, M. robiniella, and O. robiniae), their patterns of historical spread across the continent, and potential factors facilitating this spread. For this, we investigate and quantify different potential drivers of the spread of these herbivores: Robinia distribution, human population, mean annual temperature and precipitation, and proximity to previously invaded regions. We hypothesize that both the human population and R. pseudoacacia distribution would positively affect herbivore spread via effects on propagule pressure and habitat invasibility.

Methods
In order to avoid confusion among the similar species names, we will refer to the three species by their genus names, i.e., Parectopa for P. robiniella, Macrosaccus for M. robiniella, and Obolodiplosis for O. robiniae. In figures and tables, we state the full species names. We furthermore refer to Robinia pseudoacacia simply as Robinia, unless other Robinia species are mentioned. Country and regional first records of the presence of Parectopa, Macrosaccus and Obolodiplosis across Europe were obtained from the published literature, online databases and in one case from a photographic record. Coordinates for the localities were obtained through Wikipedia's GeoHack (https://www.mediawiki.org/wiki/GeoHack) and Google Maps (https://www.google.com/maps). Suppl. material 1: Table S1 provides a full list of records for the three folivore species. We also obtained georeferenced occurrence records for each of these three species at a global scale. These were sourced from GBIF (https://www.gbif.org), EPPO (https://gd.eppo.int), CABI (https://www. cabi.org/ISC), Davis and De Prins (2011) and Shang et al. (2015). These global records were not used for analysis of spread rates.
Radial rates of spread were estimated for each species from European first records using the distance regression method (Gilbert and Liebhold 2010). According to this method, a linear regression model was fit to the distance from the first discovery point in Europe as a function of year of first discovery. The slope of the estimated regression equation provides an estimate of the radial rate of range expansion. Distances between the distribution records were calculated with the R packages geosphere 1.5-10 (Hijmans et al. 2019) and sp 1.4-2 (Pebesma and Bivand 2005;Bivand et al. 2013), using the 'Vincenty' (ellipsoid) great circle distance function (distVincentyEllipsoid). Linear regressions were performed using the lm function in the R language.
In order to explore factors affecting spread of each species, we applied Cox Proportional Hazard analysis following the approach used by Ward et al. (2020). This model quantifies the probability that each uninvaded location will become invaded at annual time steps as a function of a series of candidate explanatory variables. Five predictors for herbivore spread were considered: human population, Robinia distribution, mean annual precipitation, mean annual temperature (see Fig. 1), and spatial proximity. Human population (expressed as number of inhabitants in the year 2000) was extracted from a human population density raster at a resolution of 30 arc-seconds from the Global Rural-Urban Mapping Project (Balk et al. 2006). Data on Robinia distribution (expressed as total tree area in km 2 ) were extracted from the European Atlas of Forest Tree Species (Sitzia et al. 2016) as a relative probability of presence raster at a resolution of 1 km, based on the C-SMFA model and field observations . Values of total annual precipitation (cm) and annual mean temperature (°C) for the period 1970-2000 were obtained from the WorldClim v2 database (Fick and Hijmans 2017) at a resolution of 30 arc-seconds. No data on Robinia distribution were available for points located in Moldova, Belarus, Ukraine and the European part of Russia. Values for each variable were calculated for areas in a 10 or 50 km buffer radius zones around each of the individual records for each species. Spatial proximity (sp) to previously invaded points (associated with diffusive propagule pressure) was a time-varying predictor and calculated as: where d is the distance (in km) between a given point i and each previously invaded point j. Thus, spatial proximity was estimated for each point in each year, while all other predictors did not change annually. Human population and Robinia distribution were log-transformed to reduce skewness. In addition to locations of individual records for each species, the Cox proportional hazard model was fit using "pseudo-absence" points. These are locations falling outside of the invaded range of each species that were never invaded during the time span of records. Pseudo-absence records were generated in a 50 km grid across a 300 km buffer zone outside of the minimum convex hulls around each set of records for each species (see Fig. 2). The minimum convex hull, individual buffer zones and spatial statistics for the selected variables were created using ESRI ArcMap 10.5.1 (ESRI 2016).
Given uncertainty about the identity of most relevant spatial scales of the predictor variables, all possible combinations of 10 km and 50 km scale predictors were fit in full models. The model with the lowest Akaike Information Criterion (AIC; Akaike 1973) was then further reduced (if applicable) by iteratively removing predictors with the higher p-value until all remaining were p < 0.05. To assess robustness of our models to the missing values of Robinia density for Eastern Europe, the entire model fitting and selection process was redone without considering Robinia distribution as a predictor. Models were fit using the R package survival 3.2-7 (Therneau 2020).
Macrosaccus mainly spread east-and northward in the first two decades after its introduction ( Fig. 2B), as did Parectopa. The latter species was first discovered in Northern Italy, south of the Alps. In order to reach the areas north of the Alps, it spread east-and later northward around this mountain range that acted as a geographical barrier ( Fig. 2A). Obolodiplosis spread more or less equally in all directions from its first occurrence location in Northern Italy. Within Europe, it is the most widespread of the three investigated Robinia herbivores, with distribution records stretching from Portugal to the Caspian Sea and from Sicily to Southern Sweden and the Baltic states ( Fig. 2C). In the 18 years since its first discovery in Europe, it has invaded a larger area than either of the two leaf miners, which had been introduced considerably earlier.
Results of the reduced Cox proportional hazard models are shown in Table 2, correlation matrices of predictors for the best-fitting model for the three species in Suppl. material 2: Tables S4-S6, and Akaike Information Criterion (AIC) values for the three species in Suppl. material 2: Tables S7-S9. Annual mean precipitation is found to have the least predictive power among the five investigated predictors. It is absent in all reduced models (Table 2), and is significant only for Parectopa in the full model (Suppl. material 2: Table S2). In the reduced (Table 2) and full (Suppl. material 2: Table S2) models, colder annual mean temperatures were associated with an increased risk of invasion for Parectopa and Macrosaccus (as indicated by the negative Z-scores), and less so for Obolodiplosis. In the models with Robinia omitted (Suppl. material 2: Table S3), it is significant for Parectopa, and less so for Macrosaccus; no significance is observed for Obolodiplosis. In the full and the reduced models, human population has a highly significant positive influence on the invasion risk for Parectopa and Macrosaccus, and less so for Obolodiplosis. In the models with Robinia omitted (Suppl. material 2: Table S3), it is highly significant for all three species. Robinia distribution is found to be the most   consistent predictor, explaining the spread of all three species with high significance both in the full and the reduced models. In the full and the reduced models, proximity to previously invaded areas is highly significant for Parectopa and Macrosaccus, but much less so for Obolodiplosis. In the models without Robinia, it is highly significant for all three species, along with human population.
The known global distribution of Robinia is shown in Fig. 4A, and the distributions of the three herbivore species are shown in Fig. 4B-D. Robinia is widely distributed in virtually every temperate and subtropical portion of the world. The distributions of the three herbivore species appear to be more limited. Of the three species, Obolodiplosis is the most widely distributed, having established in Europe, East Asia and New Zealand. However, there is no record of its presence in either the Afrotropic or Neotropic regions. The two Lepidoptera species Parectopa and Macrosaccus appear to be slightly less successful invaders, having only established in Europe.

Discussion
The three herbivores show similar patterns of radial range expansion in Europe, although with substantially different annual spread rates. All three species were initially discovered in the same general region of south-central Europe with only ~200-400 km separating their sites of initial discovery. Strikingly, Parectopa, which was the first of the three investigated Robinia herbivores to be recorded from Europe over 50 years ago, has the smallest annual spread rate (about 35 km/year) and is reported from the fewest number of countries (24). Macrosaccus, first reported 13 years later in 1983, exhibits an average spread rate of 73 km/year, but spread much faster in Hungary with its abundant black locust stands, invading the entire country from west to east in two years (Csóka 2001). The species is currently recorded from 25 European countries. The newest invader, Obolodiplosis, has the by far highest spread rate (128 km/year) and has spread to 33 countries since its first report in 2003. Of the three species, Obolodiplosis has also spread the most widely on the global scale (Fig. 4D). While Obolodiplosis has successfully invaded Europe, East Asia and New Zealand, Parectopa and Macrosaccus have only invaded Europe. In North America, all three species have also extended their range beyond the native range of Robinia, with Macrosaccus and Obolodiplosis having spread as far as the west coast of the US and Canada.
Invasion spread is driven by population growth coupled with movement. Thus, any factors that affect either population growth or movement are likely to influence patterns of spread. It is likely that the differences in invasion patterns observed among these species (both within Europe and globally) can be attributed to their biological traits that influence their population growth rates or dispersal, either natural dispersal or accidental long-distance movement by humans. Obolodiplosis develops through three generations per year in the Czech Republic, and in up to four generations in more southern regions such as Italy, Hungary and Serbia (Skuhravá et al. 2007;Mihajlović et al. 2008;Duso et al. 2011). For China, however, up to six generations per year have been reported (Shang et al. 2015). The capacity for this species to develop through multiple generations likely facilitates rapid population growth (Fahrner and Aukema 2018). The small size of adults also probably leads to this species being easily transported in wind though such natural dispersal probably only facilitates local dispersal. Long-distance transport (including inter-continental spread) is most likely to occur via hitch-hiking with cargo, vehicles, etc. Pupation of Obolodiplosis takes place in the galls, except for the last generation of a year, where pupation takes place in the soil (Uechi et al. 2005;Tóth et al. 2009). Because this species overwinters as a diapausing larva (Duso et al. 2011), this probably creates potential for the species to be accidentally transported long distances with vehicles and other objects that might be placed under Robinia trees prior to transport.
Even though both of the two leaf miner species belong to the same Lepidoptera family (Gracillariidae), their biologies exhibit differences that potentially explain differences observed in their success and rate of spreading across Europe. Parectopa produces two to three generations per year, with two in more northern regions such as Belarus, and up to three in more southern regions like Transnistria (Moldova) and Croatia (Maceljski and Igrc 1984;Antyukhova 2010;Sautkin and Evdoshenko 2012). Macrosaccus is reported to produce two to five generations per year: two generations in Southern Germany, Switzerland and Austria (Wipking 1991;Huemer et al. 1992;Huemer 1993;Rietschel 1996), two to three generations in Hungary (Csóka 2001), three generations in Serbia and Belarus (Stojanović and Marković 2005;Sautkin and Evdoshenko 2012), four generations in Slovenia (Seljak 1995), and potentially even a fifth generation in Croatia (Maceljski and Mešić 2001). Furthermore, often two to three (and up to eight) Macrosaccus larvae share a common mine (Huemer 1993;Rietschel 1996;Šefrová 2001), whereas Parectopa caterpillars usually inhabit mines solitarily (Baugnée 2014). In addition to a generally higher reproduction rate, Macrosaccus may thus be able to attain a higher population density.
Pupation takes place in the leaf litter in the case of Parectopa, whereas Macrosaccus larvae pupate on the leaves (Antyukhova 2010; Davis and De Prins 2011). In urban areas, Parectopa pupae might therefore be removed with the leaf litter in the autumn (Antyukhova 2010), whereas Macrosaccus, which overwinters in the adult stage (Deschka 1995), probably remains on or near its host plants, increasing its chances of reoccupying Robinia stands in the following season. However, pupating in the leaf litter, where it is presumably less exposed to parasitoids, might increase the survival of Parectopa as compared to Macrosaccus (Csóka et al. 2009). Given that Parectopa exhibits the slowest rate of spread of all three species, we can hypothesize that their biology of overwintering as pupae in leaf litter does not facilitate their anthropogenic movement to the extent seen in Macrosaccus and Obolodiplosis.
The small adult body size and wing anatomy of the two leaf miners indicate that they likely spread passively with wind, but transport of hibernating or resting adults with trade cannot be excluded (Rietschel 1996;Šefrová 2001. Passive wind transportation might explain the generally stronger eastward spread of the leaf miners with the prevailing west winds in Central Europe. We find a negative correlation between mean annual temperature and the spread of the two leaf miners, meaning that colder temperatures promote the spread of these species. Considering the geographical setting in which the range expansion of these species occurred, this is not surprising: with their first records in Northern Italy resp. Northern Switzerland, range expansion would occur mostly north-and eastward, as expansion southwards is limited by the Mediterranean Sea. The negative correlation between temperature and spread might thus be a result of generally more sampling points in the north-and eastward direction of the points of first record, where annual mean temperatures are generally lower than those in Northern Italy (see Fig. 1D).
Our findings of colder annual mean temperatures promoting the spread of both leaf miners are in contrast to published information at least of Parectopa, which is reported to be "more thermophilous" than Macrosaccus (Baugnée 2014). This is consistent with its slower northward spread and its presence in Southern Italy (i.e., south of the Emilia Romagna region), where Macrosaccus is absent (Stoch 2003). Parectopa was also reported as "massively present" with 50-80% of leaflets infested in the hot and dry, sandy environments of coastal Croatia, whereas habitats in inland Croatia with a more continental climate experienced a low infestation rate of 3% (Maceljski and Igrc 1984;Stojanović and Marković 2005). Parectopa thus seems to have more specialized habitat requirements than Macrosaccus. Parectopa might therefore continue its spread in the more southern parts of Europe and into the Transcaucasian region where its hostplant is present. Fodor and Hâruța (2009) find almost no niche overlap between Parectopa and Macrosaccus in Romania, despite both utilizing Robinia leaves as their food source, where they occupy mostly opposite sides and different parts of the leaflets. The two leaf miners are thus not in direct competition for their food source.
Both leaf miner species are often reported to exhibit high population densities during their initial colonization phase following establishment in a new region, while subsequently becoming much rarer (Seljak 1995;Šefrová 2001;Tomov 2003;Antyukhova 2010;Baugnée 2014). In Poland however, Parectopa was mostly first recorded from single mines in isolated locations, apparently as a result of anemochorous dispersal. The following absence of Parectopa mines in these locations for several years suggests that these founder populations were unable to establish. More successful northward spread of Parectopa occurred along river valleys, e.g. the Vistula valley, where Robinia finds favorable growing conditions on the sunny slopes (JB, pers. obs.). Macrosaccus, on the other hand, quickly spread through Poland over a wide front and in considerable abundance until 2005, when areas of rarer Robinia occurrence (presumably due to less suitable growth conditions) were reached (JB, pers. obs.). There are also records of Obolodiplosis being very abundant in recently invaded regions, particularly in East Asia (Yang et al. 2006). Though lacking quantitative data, it appears that none of the three species is particularly abundant in their native range in North America (AML, pers. obs.). Along these lines, we note that most of the records of Obolodiplosis from North America lie outside of the native range of its host, Robinia, which may be indicative of the low abundance of Obolodiplosis in its native range.
Parasitization might play an important role in the speed of spread. Since their establishment in Europe, the two leaf miners have accumulated a large number of generalist parasitoids (summarized in Serini 1990;De Prins and De Prins 2006-2020, and Csóka et al. 2009), with 20 species recorded for Parectopa, and 37 for Macrosaccus. Parasitization rates vary considerably though, ranging in the case of Macrosaccus from 1-3% in Upper Austria (Deschka 1995), 10-30% in Southern Moravia (Šefrová 2001), <40% in Kraków, Poland (Wojciechowicz-Żytko and Jankowska 2004), 35-50% in Trentino, Italy (Angeli et al. 1996), up to 47.6% in Hungary (Csóka et al. 2009), and >60% in Bosnia-Herzegovina (Dimić et al. 2000), to 30-67.5% in Serbia (Stojanović and Marković 2005). Information on parasitization rates in Parectopa are few, reaching a maximum of 15.3% in Hungary, where Macrosaccus is up to three times more heavily parasitized (Csóka et al. 2009). Obolodiplosis hosts few parasitoids, which likely promotes its rapid spread in Europe and other regions of the world. It is to be expected that Obolodiplosis will have a fairly large impact on Robinia populations wherever it is introduced, which might however be compensated by the fast growth and reproductive abilities of Robinia. On the other hand, Platygaster robiniae, the gall midge's primary parasitoid infesting the host eggs and feeding gregariously on the larvae (Buhl and Duso 2008;Duso et al. 2011;Kim et al. 2011), is reported to cause parasitization rates of 51.6% to 84.8% (Park et al. 2009;Lu et al. 2010), making it a promising candidate as control agent of the locust gall midge (Lu et al. 2010).
Our quantitative analysis indicates local Robinia density to be the single factor having the strongest impact on the spread of Parectopa, Macrosaccus and Obolodiplosis across Europe. Skuhravá et al. (2007) reached a similar conclusion for Obolodiplosis based on a qualitative evaluation of historical European spread. Since feeding of all three insect species is limited to Robinia, it is understandable that its density would strongly affect population growth rates and consequently affect spread. Several other studies have reported that host densities influence rates of invasion spread of invading species (e.g., Meier et al. 2014;Hudgins et al. 2017;Ward et al. 2020). The resource concentration hypothesis posits that more abundant host plant resources promote insect herbivore population growth rates (Hambäck and Englund 2005), and such elevated rates can be expected to translate into increased invasion spread rates.
The fact that Robinia is itself an invasive species has interesting implications regarding the positive effect of Robinia density on spread of these folivore species. It has been noted that at a global scale, plant invasions or widespread planting of non-native plants promote invasions by herbivore species that use these plants as hosts (Liebhold et al. 2018;Branco et al. 2019;Guo et al. 2019). There are many examples in which abundant distributions of non-native plants have promoted invasions by insect herbivores that specialize on those plants (e.g., Hurley et al. 2016). This phenomenon can be regarded as a type of "invasion meltdown" where invasion by one species triggers subsequent invasions of other species (Simberloff and Von Holle 1999). However, less is known about how host insect invasions can mediate invasions of their parasites beyond theoretical studies (e.g., Fagan et al. 2002).
Previous studies have also identified human population density to be related to the spread of invading insect species (Gilbert et al. 2004;Ward et al. 2020). It is logical that humans may accidentally transport insect life stages and therefore promote long-distance dispersal. Population models show that when occasional long-distance dispersal is coupled with frequent short-distance dispersal, this leads to much greater rates of spread than when long-distance spread is lacking (Shigesada et al. 1995;Hastings et al. 2005). Long-distance dispersal is often associated with passive movement by humans and thus high human densities may drive higher rates of longdistance movement and thereby facilitate invasion spread (Gippet et al. 2019). But the significant influence of human population may also be confounded with Robinia occurrence since human-caused disturbance typically promotes this tree species (Vítková et al. 2016).
Similar to human population, annual mean temperature was found to have a significant influence on the spread of the two leaf miners, but less so for the gall midge. This result is in concordance with the wider climate spectrum of invaded regions of Obolodiplosis: in Europe, the gall midge is now distributed from the hot-summer Mediterranean climate of Portugal, Sicily and Greece to the humid continental climate of Southern Sweden and the Baltic states. On the global scale, it has been recorded from Vancouver Island, Canada (Skuhravá et al. 2007), Japan and South Korea (Kodoi et al. 2003;Woo et al. 2003;Uechi et al. 2005;Tokuda et al. 2019), China (Yang et al. 2006;Shang et al. 2015), the Russian Far East (Csóka et al. 2017), and New Zealand (Anonymous 2009;Bain 2009) (Fig. 4D). Parectopa and Macrosaccus, on the other hand, have only been reported outside their native range from Europe and the west coast of North America (Fig. 4B, C). In contrast to mean temperature, we found annual mean precipitation to have no significant influence on the spread of the three herbivore species.
Our results indicate that spatial proximity to previously invaded regions plays an important role for the spread of Parectopa and Macrosaccus, but much less so for the gall midge Obolodiplosis. Obolodiplosis showed an extremely fast spread across most of Europe in the 18 years since its first record in Europe, now occupying a considerably larger area than the much earlier established leaf miners. The results of Roques et al. (2016) indicate that it spread faster than any other insect species invading Europe in their analysis. The spread of this species exhibited several longdistance jumps to form discontinuous populations, the most prominent one being a 2,000 km dispersal from its first record in Northeast Italy to the East Ukraine in just three years. Furthermore, Obolodiplosis successfully invaded the islands of Great Britain, Sicily, Corfu, and the Balearic Islands, none of which have been reached by either of the two leaf miners. Skuhravá et al. (2007) speculate that the gall midge may frequently be transported over long distances with nursery trees, and/or through passive transport by freight traffic. Our finding that spatial proximity to previously invaded areas plays a minor role in the spread of Obolodiplosis concord with its high spread rate of 128.3 km/year, and the observed long-distance dispersal.
None of the scatterplots of the three herbivore species (Fig. 3) show a clear establishment phase preceding the expansion phase. Macrosaccus, however, was closely monitored in the area of its first discovery in 1983 around Basel in Northern Switzerland. There, the distribution range did not exceed an 85 km radius around Basel by 1989, although the species was "already common around Basle in 1983" (Whitebread 1990). The absence of a clear establishment phase in the scatterplots might indicate that the species arrived some years before their first record in Europe, when they had time to establish a sufficiently large population and propagule pressure to expand their ranges. The scatterplots also provide little evidence of geographical "saturation" in any of these three species. As invading species spread to all suitable areas in a region, such plots can be expected to asymptotically stop increasing (Shigesada and Kawasaki 1997). Eventually, all three species can be anticipated to become established in all regions with suitable habitat. That environmental niche is presumably defined both by the presence of a suitable climate and by the presence of Robinia hosts. Comparison of the current distribution of these species (Fig. 2) with the distribution of Robinia suggests that all three species will soon saturate their potential habitat. However, on a global scale these same species are far from saturation and further invasions can be anticipated in the future.
In addition to these three species that utilize Robinia as a host, Hargrove (1986) identified 72 other herbivore species associated with Robinia in its native North American range. Given the three species studied here, along with Euura tibialis and Appendiseta robiniae, it is evident that only five out of 75 North American Robinia specialists have presently invaded Europe. Thus, we can anticipate that additional herbivore species are likely to invade Europe and elsewhere in Robinia's invaded range and that this will contribute to the dilution of enemy release in populations of this invasive plant.

Conclusion
Specialist herbivores are crucially dependent on the presence of their host plant. Our results show that the widespread presence of Robinia in Europe, especially in humaninfluenced environments, greatly facilitated the spread of the introduced North American herbivores. The excessive proliferation of Robinia increases the likelihood of establishment and spread of non-native specialist herbivores, thus creating a negative feedback where the initial beneficial effects of enemy release on Robinia are diminished, and Robinia populations are potentially reduced.
With Robinia having been introduced to most regions of the world with a suitable temperate climate, conditions are thus beneficial for the establishment of these insects, and potentially other specialist herbivores from black locust's native range. Obolodiplosis has already become established in East Asia and New Zealand, where it has exhibited rapid spread similar to that in Europe. Its success can be attributed to the ability for long-distance jumps as well as to life history traits, such as high reproduction rates, and a presumably small guild of parasitoids. For the two leaf miner species, spatial proximity to previously invaded areas is another important factor affecting range expansion, reflecting the ability of these species to disperse into adjacent uninvaded areas following initial colonization. Although the three investigated herbivores invaded Europe under similar conditions, there are pronounced differences in their invasion success, which can be explained with species-specific life history traits. Furthermore, pan-European cargo traffic has increased over the past decades, increasing the likelihood of long-distance spreading.