A systematic review of the literature using the Web of Science was conducted in September 2011 using following keywords: “invas* plant* (insect OR arthropod OR herbivor* OR natural OR phytophag*) and (diversit* OR abundance OR richness OR herbivory OR removal OR enem*)”. A total of 1746 studies were retrieved. However, studies were retained for this review only if they explicitly included arthropods, i.e., studies on mammals were excluded. Aquatic systems and secondary studies not based on experimental data directly collected by the authors (i.e. review or idea articles) were also excluded. Finally, all references cited within these articles were also inspected and included to further extend scope.

A total of 53 relevant articles published in 31 different journals were selected for inclusion in this review. The first study was published in 1982, and only three studies were published prior to 2000. The majority of studies (38%) were published in 2009 and 2010. These articles have been cited a total of 759 times as of December 2011. The number of citations/article ranged from 0 to 104 (0 to 14.86 citations/year; mean = 2.23), with most articles (70%) being cited less than 10 times, indicating that perhaps literature corresponding to arthropod community dynamics on invasive plants is not highly visible. Journals contributing the highest number of articles were Biological Invasions (17%), Biological Conservation (9%), and Environmental Entomology (7%).

To characterize the literature on native arthropod communities associated with non-native plant invasions, the following parameters of each study were recorded: ecosystem type (e.g. grassland, experimental field, waste area); the country in which it took place and whether or not it was biogeographical (i.e. data on arthropod communities in association with the invasive recorded in more than one region); native plant species community richness; invasive plant species taxonomy; the plant functional group (PFG) of each invader (tree, shrub, graminoid, or herb); native arthropod community characteristics on invasive host plants/within invaded habitats (i.e. abundance, order, family, and species level richness) and; the class and number of arthropod trophic levels examined (i.e. herbivores, predators, detritivores). Studies were permitted more than one database entry if they examined more than one non-native plant species or geographic region. As this study is strictly a systematic review and not a meta-analysis, effect sizes were not calculated.

Descriptive statistics were used to explore the first two broad patterns associated with the literature including Chi-square tests for differences in relative proportion of studies where appropriate. Generalized linear models (GLMs) were used to explore the latter two patterns that diversity of arthropods is affected by native versus invasive plants and then by PFG and phylogenetic measures of these plants (firstly, we used the entire dataset and then did a second more direct test via paired t-tests of only the studies that used coupled contrasts). Alpha was set at p < 0.05, and post hoc contrasts were applied when significant to identify specific differences if more than two levels (Nonparametric Wilcoxon signed-rank tests were used as highly conservative between-level tests). Studies were included in these analyses if more than a single trophic group was examined, arthropod richness estimates were provided, and contrasts between target (i.e. on the invasive plant) and native plants or within the community were reported in some form. A total of 4 studies reported only order-level arthropod richness whilst all others reported species-level estimates. The order-level values fell within one standard error of the mean of species-levels estimates so were not excluded. The log response ratio (LRR) was also calculated to summarize the strength of the relative difference between arthropods associated with native versus invasive plants (Hedges et al. 1999) within each study (using only the studies that used a target invasive-native paired plant design directly).

Finally, phylogenetic relationships among all 1045 plant species reported were constructed by grafting published phylogenies onto a family-level backbone based on the APG3 derived megatree produced with Phylomatic (Webb and Donoghue 2005). Polytomies were present below the family level and were resolved from published clade-specific phylogenies to the genus level (see Supplementary File 1 for references and Supplementary File 2 for the Newick file). Polytomies among species within genera were randomly broken as species-level phylogenetic information was rarely available or consistent across studies. The lack of resolution at terminal nodes is likely to make subsequent tests more conservative (Swenson 2009). Dated nodes from Wikstrom et al. (2001) and TimeTree (Hedges et al. 2006) were used to restrict branch-lengths based on estimated divergence dates with undated descendant nodes evenly spaced using the bladj algorithm in Phylocom (Webb et al. 2008). Within a study, the mean and maximum phylogenetic distance was calculated between the invader and all other species within the community. Regression analyses were used to test whether these two phylogenetic measures impact arthropod richness. All statistics were performed with JMP 9 ver. 9.0.2 (SAS).

From the 53 studies included in this review 11 ecosystem types were censused for arthropods. In decreasing order of prevalence these were: grassland, mixed, forest, experimental field, marsh/wetland, shrubland, riparian, waste area, desert, dune, and floodplain ecosystems. Two studies did not detail the ecosystem from which data was collected. Geographically, arthropod communities were censused in 27 countries (Figure 2). Fifty three percent of all studies were conducted in North America, while 28% were conducted in Europe (Figure 2). Less than 8% of all studies (4/53) used biogeographical contrasts to record arthropod dynamics in the native and introduced ranges of invasive plant species.

A total of 38% of studies measured only one trophic level whilst 30% of studies evaluated 4 trophic levels. Fewer studies evaluated only two or three trophic levels (Figure 3, 11% and 21%, respectively). These proportionate differences were significantly different (Chi-square, c² = 8, p = 0.039, n = 53). A breakdown of studies based on which feeding guilds were examined indicated that the majority (92%) targeted at least herbivorous arthropods. Predators were measured in 64% of the studies, detritivores in 53%, and parasites/parasitoids in 34% (Figure 4, Chi-square, c² = 16, p = 0.0013, n = 129).

Arthropod richness estimates associated with invasive plants were significantly depressed relative to native plants or monoculture/community estimates using all cases (Figure 5, GLM, chi-square = 385, p = <0.0001, n = 124). Using only paired contrasts within studies, the strength of the relative depression in arthropod richness between invasive and native plants was -0.18 +/- 0.06 (mean LRR with s.e.), and this estimate was significantly different from a null of 0, i.e. no difference (t-test for mean diff from 0, t = -2.5, p= 0.01, n = 62 cases).

Both mean and maximum phylogenetic distances significantly predicted arthropod richness on invasive and on native plants (Figure 6, Regression analyses, all p < 0.0001, r² values listed on plots). Increasing phylogenetic distances reduced arthropod diversity (Figure 6). Plant functional group significantly influenced arthropod richness at the species level (GLM, Chi-square = 33.8, p = 0.001, df = 1) but only for arthropods associated with invasive plant hosts - not native plants (GLM, Chi-square PFG*host= 80.3, p = 0.0001, df = 3 with Wilcoxon post hoc paired contrasts, p > 0.05 for all natives). Specifically, arthropod species richness differed between invasive trees and herbs (Figure 7, Wilcoxon paired contrasts, p = 0.02),  and the abundance of arthropods associated with invasive trees differed from graminoids (Wilcoxon paired contrast, p = 0.05). Given the exploratory nature of this review, corrections for multiple comparisons were not made (Rothman 1990; Saville 1990), and importantly, inflated type I error is controlled by the overarching GLMs used to determine which pair-wise comparisons to make. In a strict two-way comparison, PFG for woody plants significantly influenced arthropod richness at the species level (Wilcoxon rank sum test, Z=2.27, p = 0.023). Specifically, arthropod species richness differed between invasive trees and shrubs (Figure 7).

Studying invasive species from a biogeographical perspective is a powerful yet underappreciated tool in invasion ecology (Hierro et al. 2005). At different spatial scales, biogeographical contrasts can provide a direct way to infer the overall extent of invasion as well as to directly compare community dynamics between ranges. For instance, Lamarque et al. (2012) contrasted two congeneric maple species (density, relative abundance, age structure, effects on native species) between France and Canada locally and regionally and demonstrated that density is a viable and useful proxy for invasibility. A study from this review by Cripps et al. (2006) contrasted herbivore community dynamics (diversity, evenness, richness, host utilization) on Lepidium draba in its native, expanded, and invasive range effectively testing whether biotic restraint can be mediated through biogeography (it was). However, studies such as these remain scarce in the literature probably due to high financial costs and logistical constraints when sampling in both regions. Regardless, wider scales in sampling of arthropods will dramatically improve our understanding of the full community consequences of invasion.

Sampling regimes focusing on only the herbivore feeding guild comprise a large proportion of the literature (e.g. two of the four biogeographic contrasts in this review). While informative, these studies are not adequate to fully explain the mechanisms by which plant invaders are successful and may introduce uncertainty and thus false conclusions regarding observed declines in herbivores within invaded regions (i.e. Figure 1B). Herbivore declines are often attributed directly to invasive plants but they may be the product of an indirect interaction whereby an invader facilitates predacious or parasitoid species that in turn depress herbivore communities. Specialized enemies such as parasitoids use both visual and volatile cues from plant hosts and their prey items when hunting. In invaded habitats novel plants may initially mask prey presence, although novel cues can be learned after successful foraging (Vet and Dicke 1992). Pearson (2009) found that native spiders were enhanced in grasslands invaded by Centaurea maculosa due to altered vegetation architecture for building webs, whereby formerly simple vegetation was replaced by more complex stands. This resulted in a substantial increase in invertebrate predation rates. Similarly, female condition and reproductive output of the endangered wandering spider Arctosa fulvolineata were enhanced in salt marsh habitats invaded by Elymus athericus (Pétillion 2005; 2009). These positive effects were attributed to a more complex litter layer in invaded habitats compared to uninvaded ones and food limitation was not considered a factor (Pétillion 2005). In other instances where invaders have altered litter inputs within novel habitats, micro and macrodetritivores have responded both positively and negatively to changes in detritus microclimates and food resources (Gratton and Denno 2005; Mayer et al. 2005; Kappes et al. 2007; Wolkovich et al. 2009). Within invaded systems, how shifts in detritivore communities influence predacious or parasitic arthropods, and in turn, how changes in consumer guilds may impact arthropod herbivores is not well understood. Whilst it is intuitive and convenient to limit arthropod studies to herbivores, it would be imprudent to regard invaded systems so simplistically because plant-herbivore interactions have evolved through selection pressures acting in both bottom-up and top-down directions (Hairston et al. 1960; Price et al. 1980; Bernays and Graham 1988; Agrawal 2000; Dicke 2000). Predators and parasitoids have tremendous potential to mold arthropods community structure in the context of plant architecture. In general, plants with greater architectural complexity (e.g. shrubs vs. grasses) provide more places for arthropods to hide from natural enemies (Lawton 1983). arthropods can gain spatially mediated “enemy-free space” on architecturally complex plants by modifying their distribution or behaviour in a way that eliminates or reduces their vulnerability to natural enemies (Jeffries and Lawton 1984). For example, the polyphagous tansy leaf beetle Galeruca tanaceti is hypothesized to gain spatial enemy-free space by ovipositing in structurally diverse habitats over simpler ones, which reduced the searching efficiency of its specialized egg parasitoid (Meiners and Obermaier 2004). Conversely, plants can also provide shelter and alternative food to predators, resulting in depauperate herbivore communities in the presence of predators (Dicke 2000).

Phylogenetic tools are rapidly being applied to the study of plant interactions, community dynamics, and invasion. Phylogenetic similarity between host plants can be associated with similarity in herbivory levels (Hill and Kotanen 2009; Ness et al. 2011), mutualistic interactions (Rezende et al. 2007), and overall arthropod community composition (Weiblen et al. 2006). The link between phylogenetic and ecological similarity can be attributed to the high degree of phylogenetic conservatism in relevant traits (Wiens et al. 2010); invasive plant species may often be both phylogenetically (Gerhold et al. 2011) and functionally (Godoy et al. 2011) unique from the invaded native community. We propose, and show for the first time, that these tools can be an effective means to infer or even predict relative impact on arthropods communities by novel plant species. We must point out however that this approach was tested via a systematic review using data compiled across studies and not from single, controlled experiments. Nonetheless, this broad test showed a clear correlation between phylogenetic distance estimates and richness.

Diversity is an important response variable in ecology, a major ecosystem service, and sometimes a predictor of relative sensitivity to perturbation at larger scales. Plant invasions in general have been shown to reduce diversity of native plant species (Alvarez and Cushman 2002; Flory and Clay 2010). The impact of plant invasions on arthropods is predicted to also negatively impact diversity (Simao et al. 2010), and this finding was confirmed here in the first thorough systematic review of the topic. In general, more complex and productive habitats increase arthropod species richness - particularly that of herbivores (Murdoch et al. 1972; Root 1973; Allan et al. 1975; Siemann et al. 1998; Agrawal et al. 2006; de Groot et al. 2007; Simao et al. 2010). Aggressive plant invaders thus have the potential to drastically alter native plant communities both directly through plant competition with natives and indirectly through introduction of poor host plant material for native arthropods (Lau and Strauss 2005). The most parsimonious explanation for the depressed diversity detected here is the lack of suitable host plants. Nonetheless, it is also reasonable to extend this implication to much larger ripples including eventual collapse of arthropod communities through reduced trophic complexity or even melt-downs and additional invasions with less potential arthropods controls persisting within regions. Arthropod diversity must be incorporated into the study of plant invasion so as to effectively assess impact and resilience more broadly. Like most correlative approaches however, it is difficult to infer cause and effect or decouple drivers from passengers in the invasion literature (MacDougall and Turkington 2005). Consequently, it is important to note that whilst documentation of background arthropod community patterns in plant invasion studies effectively enhances our capacity to infer larger scale impacts of invasion, delineation of mechanism and tests of top-down control should also be nested within studies of arthropod community dynamics.