**4. Discussion**

We addressed the effect of connectivity, fragment size, edge effect, and their interactions on insect and seed predation of arthropods. The predation rates of forest-steppes and kurgans responded differently to the effects of fragmentation depending on the fragmented grassland ecosystem. Predation pressure was higher in the edges of small fragments of forest-steppes and in the edges of more isolated landscapes of kurgans. Thereby, our results emphasize the differences of the two fragmented grassland ecosystems.

#### *4.1. Predation in Forest-Steppe*

Connectivity did not affect predation (Hypothesis 1); however, our second and third hypotheses about edge effect and fragment size, respectively, were partly supported, as the difference between edges and centres was more pronounced in small than in large fragments. In the case of insect predation, we found higher predation rates in the edges in the centres of small fragments, but we found the opposite pattern in the case of large fragments. Seed predation was higher in the edges of both small and large fragments, but the difference was more pronounced in small fragments.

Edge effects were stronger in small than in large fragments, which was in line with the findings of Laurance and Yensen [58]. This was explained partly by the higher edge/area ratio, i.e., smaller fragments had longer edges related to their area [59]. The edge effect on arthropod predation, and on other interspecific interactions and functional trait composition of arthropods, is relatively well-known [60–62]. Some of them measured with the very same sentinel prey method [40]. It is expected that resource consumption in terrestrial arthropod communities is higher at fragment edges [63]. The abundance of generalist predators increases near edges, which may alter food webs [64]. However, the response of insect predation by arthropods to edges is highly variable. Although there is a general positive correlation between the abundance of predators and predation rates [65,66], predation rates were found to be even higher in the habitat interior than in the edges in some cases [40]. There are contradictory results for seed removal ratios at edges vs. centres, with negative [67], or neutral [53] responses to edges. Fragmentation may shape many attributes of edges [10,11]; landscape and fragment-scale factors may interact with edge effects [63,67]. Our results emphasize that several effects may modify seed predation at edges, as we found higher rates at edges than in centres, but this effect was modified by fragment size. The interaction of these factors may explain the variability of predation responses to edges.

The main difference between forest-steppe and kurgan systems is the size of the fragments. Forest-steppe fragments (0.16–6.88 ha) were an order of magnitude larger kurgan fragments (0.01–0.44 ha), presumably maintained by abundant predator and seed predator assemblages. This might have had an effect on predation rates, which caused fragment size to override the effect of connectivity that we expected in the hypothesis (1).

#### *4.2. Predation in Kurgan*

Our first hypothesis that predation pressure increases with decreasing connectivity was supported for insect and seed predation in edges of kurgans. We found that increasing connectivity had a negative effect on predation rates in the edges but had a minor effect in the centres of kurgans. In addition, predation rates of edges mainly stayed under the rates of centres, in contrast to our presumptions during the hypothesis (2). Presumably, spillover from arable lands to grasslands increased predation rates in the edges.

We assumed that predators of arable fields dominated the predator assemblages in the edges of kurgans, therefore the decreasing amount of agricultural fields associated with increasing connectivity in the matrix had a negative effect on predation in the edges. Spillover between agricultural fields and natural habitats is recognised as an important mechanism shaping biodiversity, biotic interactions in communities, and ecosystem functions [19,68,69]. In order to gain insight into these mechanisms, not only should the number of species and individuals that move between adjacent habitats be observed, but also the functional responses should be addressed directly [70]. Native herbivores in natural habitats may experience increased predation pressure in landscapes with increasing habitat loss, due to spillover of generalist predators from surrounding cropland habitats [71]. Agricultural landscapes appear to augmen<sup>t</sup> generalist predators resulting in dramatically higher predator–prey ratios and reduced herbivore abundance [72].

Seed predation also decreased in the edges of kurgans when connectivity increased. Granivorous arthropods from the agricultural matrix may contribute greatly to seed predation rates in the edges of grasslands. When the proportion of arable fields decreased in the surrounding matrix; i.e., connectivity of habitat fragments increased, it led to a decrease in seed predation by arthropods. Our presumption about the tendency of movement from arable field to the adjacent habitat is recognised for some seed consumer carabids [73]. Furthermore, our results align with those of the study of Craig et al. [67], who found that seed predator arthropods cause higher predation rates with decreasing connectivity in the landscape. Taken together, connectivity had similar effects on insect and seed predation, presumably due to the same phenomenon: spillover.

Connectivity moderated the edge effect on predation; however, we did not find an effect of fragment size (Hypothesis 3), in contrast to in the forest-steppes. All kurgan fragments were relatively small.

Presumably, the e ffect of connectivity overrode the e ffect of fragment size. These results are in line with those of Rösch et al. [23], emphasising that connectivity is more important for small, isolated fragments, whereas fragment size is more relevant for large fragments.
