Short-term effects of horse grazing on spider assemblages of a dry meadow ( Western France )

Short–term effects of horse grazing on spider assemblages of a dry meadow (Western France). In this study, the biodiversity impacts of a little studied herbivore, the horse, were assessed in a high conservation value habitat of dry meadows in Brittany (Western France). Spiders, a diversified and abundant group of predators, were used as bioindicators. Three complementary sampling techniques were used to assess changes in spider assemblages in both soil and vegetation strata, over time (diachronic comparison of managed unit before vs. after management) and space (synchronic comparison of managed vs. control units). Few effects of grazing, i.e. only one significantly indicative species, were found on assemblage composition (ANOSIM), and none on abundances, α– and β– diversities (GLM on pitfall trap data). On the contrary, important differences were found between units before management took place. The main effects of grazing management were revealed over time (after one year), and not between managed and control units (CCA on pitfall trap data and x2–tests on guilds from each sampling method), showing the relevance of a diachronic approach more than a synchronic approach in such management monitoring. Grazing by horses could be relevant to manage meadows because it creates a high spatial heterogeneity, but further (long–term) studies including other model groups are required.


Introduction
Management is frequently carried out in grasslands to simulate past, naturally occurring disturbances such as grazing by large herbivores (Bakker, 1989) or fire (Valkó et al., 2014).The impact of management practices is usually monitored using plant diversity (e.g.Kahmen et al., 2002).However, plant species diversity seems to be a poor predictor for the diversity of other groups, such as arthropods (Kirby, 1992;Morris, 2000;Van Klink et al., 2015).It is therefore necessary to use various groups of organisms to evaluate the potential of a management practice to conserve overall biodiversity.Grazing affects arthropods both directly and indirectly.Direct effects of grazing concern trampling and accidental predation of insects (as well as scavengers and dung feeders: Lumaret et al., 1992), whereas indirect effects are more complex and mainly encompass vegetation and soil-mediated changes (Van Klink et al., 2015).
In general, grazing by large herbivores has mostly a negative impact on species richness and abundances of arthropods as it reduces plant cover and biomass.Many large herbivores produce similar effects on the diversity of phytophagous (including flower-visiting) arthropods (Van Klink et al., 2015).Some studies linked changes in arthropod diversity with changes in plant diversity (e.g.Foote and Hornung, 2005), height (Ausden et al., 2005;Ryder et al., 2005) or biomass (Kruess and Tscharntke, 2002), but shifts in functional plant groups are believed to have a higher explanative power (Van Klink et al., 2015).
On the contrary, measured indirect effects are likely more dependent on the target taxa or metrics considered, and not consistently reported for all herbivores (Read and Andersen, 2000).As an example, arthropods have been shown to be more diverse and more abundant under ungulate grazing, but arthropod biomass overall is reduced (González-Megías et al., 2004).The effects on predatory arthropods depends on the intensity of grazing (e.g.Dennis et al., 1998;Dennis et al., 2002;Pétillon et al., 2007;Rosa García et al., 2009;Van Klink et al., 2013).Although several studies did not find any effect on arthropod diversity, some found effects on their species composition (Gardner et al., 1997;Woodcock et al., 2005;Fadda et al., 2008).
Reported effects of management on arthropod diversity are overall negative, but they differ depending on the types of herbivores and grazing intensity, and are likely context-dependent.Conclusions are also dependent on the taxa and metrics used, and thus on the objectives of management (e.g.Leroy et al., 2014;Török et al., 2016).In this study, we assessed the impact of horse grazing on spiders in a high conservation value habitat in Brittany, dry meadows.Despite its high potential for conservation management (mainly because of an increased plant consummation, for example, higher than similar numbers of cows grazing (Ménard et al., 2002), the effects of horses on arthropod diversity remains poorly documented (Bell et al., 2001;Joern and Laws 2013).Spiders were used because of their bioindicative values (Maelfait and Baert, 1988;Marc et al., 1999), such as to monitor the effects of biological invasions (Pétillon et al., 2005;Mgobozi et al., 2008), success in habitat restoration (Cristofoli et al., 2010;Pétillon et al., 2014), and changes in land use (Schmidt et al., 2005;Prieto-Benítez and Méndez, 2011).We conducted a pre-management inventory so both synchronic and diachronic approaches could be used.Changes in assemblages of both epigeic and vegetation-dwelling spider assemblages were assessed using (three) complementary sampling methods.
Taxonomic and functional changes in community structure were evaluated in the short-term to test the following hypotheses: i) grazing alters the functional composition of a spider community as it will select aeronautic and disturbance-resistant species (Bell et al., 2001) as well as hunting spider species that are less dependent on the plant physiognomy (e.g.Churchill and Ludwig, 2004); ii) abundance and alpha diversity are both expected to be lower in the grazed unit than in the non-grazed unit because of reduced plant biomass and cascading effects (e.g.Kruess and Tscharntke, 2002) whereas β-diversity is expected to be higher in the grazed unit than in the non-grazed unit due to an increased heterogeneity of vegetation (Loucougaray et al., 2004).

Study site
The study site is located in Brittany (Western France), 20 km south-west of Rennes, near the city of Guichen (47° 97' 16 N-1° 89' 20 W).The natural reserve, Vallée du Canut, is a public land of 147 ha, part of a larger Natura 2000 site (total 427 ha) that encompasses meadows, heathlands, forests and a dense network of hedgerows.The dry meadows are located on Cambrian outcrops of red shale on the slopes of a small valley, Le Canut River.The meadows have a N-S slope, with the lower part dominated by bracken (Pteridium aquilinum).Before horses were introduced, the upper part, where the sampling took place, had a mean vegetation height of 1.3 m (visual estimation), and the following dominant plants: Dactylus glomerata, Rumex spp., Stellaria holostea, Plantago lanceolata, Holcus mollis, and different species of Apiaceae, Geranium, Ranunculus, Centaurea and Trifolium.The sampled area was 4 ha in total, subdivided into two experimental units (hence referred to as 'units'), grazed and non-grazed (during the second year of the study).Two horses were then introduced in October 2003 after the first year of sampling.Although the impact of grazers on vegetation cover and diversity was obvious (see fig. 1s in supplementary material), data on habitat changes were not used here because our hypotheses relate to changes in spider assemblages using both diachronic and synchronic approaches (effects of time vs. management).

Sampling design and methods
In order to sample spiders both at ground level and in the different strata of vegetation, we conducted three complementary sampling methods over the two years of the study: pitfall trapping, hand-collecting and sweep-netting.
Traps to catch ground-active spiders consisted of polypropylene cups (diameter: 12 cm, height: 15 cm) filled with ethanol 70° and covered with a wooden roof to prevent overflow by rainfall.Four traps arranged in a square grid were set up in each unit, and spaced 10 m apart (Topping and Sunderland, 1992;Churchill and Arthur, 1999;Ward et al., 2001), each trap representing one sample (for a total of four replicates per treatment, 16 samples in total).Traps were active on 15 days in June and 15 days in September (i.e. the most favorable periods for spiders in this region: Varet et al., 2013Varet et al., ) in 2003Varet et al., and in 2004.The numbers of individuals caught in traps were divided by the number of days traps were active (Luff, 1975;Curtis, 1980).Hand collections were carried out to sample all visible ground-living spiders, including some less mobile species that would have been few or not sampled by traps (Churchill, 1993).Hand collections were time-standardized (two collectors for 10 min in each unit represent one sample, for a total of eight samples).Vegetation-dwelling spiders were collected using sweep-netting (40 cm diameter) along two parallel transects (20 m + 10 m and 45 sweeps, which represents one sample altogether), two times in each unit (June and September 2003 and 2004, for a total of eight samples).

Identification and conservation of specimens
Spiders were sorted and stored in tubes with 70 % ethanol (University of Rennes 1, France).Adults were identified to species using Roberts (1987), Heimer and Nentwig (1991) and Roberts (1995).

Data analysis
Differences in species composition between years and management were visualized by a Venn diagram of shared vs. exclusive species of grazed and un-grazed units before and after the management took place, and tested using an analysis of similarity (ANOSIM) completed by IndVal calculations (Dufrêne and Legendre, 1997) on species abundances using individuals sampled using pitfall traps.A detrended correspondence analysis (DCA) was done on individual counts as response variables and year/management as predictors.A redundancy (RDA) or a constrained correspondence analysis (CCA) was then chosen according to the axis length of the DCA, < 3 or > 4, respectively (Legendre and Gallagher, 2001).Here, a CCA (first axis of the DCA = 1.40) was performed using vegan R package (Oksanen et al., 2013).Monte Carlo tests with 999 permutations were carried out to test the significance of the two factors and constrained analyses axes.Hunting guilds were defined according to Uetz et al. (1999), and assigned to species according to the families they belong to.Guild composition differences between years and management were studied using x²-tests for each sampling method.
As the same pitfall traps were operative in 2003 and 2004, abundances and species richness were pair-matched over time and consequently com-pared using repeated analysis of variance (Pétillon et al., 2010;Lafage and Pétillon, 2014).Tests were performed using spider activity-density and species richness as dependent variables, management (grazed or non-grazed) as a fixed factor, and date (pre-vs.post-grazing) as a within subject effect.If the interaction between fixed factors was not significant (in model 1), a second GLM with logit link and negative binomial distribution (model 2, see O'Hara and Kotze, 2010) was used to test significant effects of separated fixed factors, without their interaction.If the interaction was significant, t-tests were used to detect significant differences between sampling periods.In the case of grazing effects, a significant interaction between management and date was expected (i.e. the within subject factor being expressed differentially for the two units due to grazing effects in one of them).For each analysis, the level of statistical significance used was α = 0.05.
All statistical analyses were performed using R 3.2.3(R Core Team, 2015).

Results
A total of 1990 spiders belonging to 55 species were collected in 2003 and 1,040 spiders belonging to 66 species were collected in 2004, respectively (see table 1s in supplementary material).Most spiders were sampled using pitfall traps (1,801 individuals), followed by sweep-net and hand-collection (702 and 537 individuals, respectively).
There was a slight increase in the number of species shared between the two units from 2003 to 2004 (22 to 24 species: fig.1).Overall, no difference in species composition was found between experimental units in 2003 (R = -0.02,P = 0.537) and 2004 (R = 0.31, P = 0.056).Only Pachygnatha degeeri had a significant IndVal, with a preference for the grazed vs. non-grazed unit (IndVal = 4.24 vs. 2.49 respectively, P = 0.028).Furthermore, no significant difference was found in species composition of the non-grazed unit between 2003 and 2004 (R = 0.47, P = 0.057).Conversely, a significant difference in species composition of the grazed unit was found between 2003 and 2004 (R = 0.84, P = 0.026); this was confirmed by the CCA (significant effect of year along the axis 1, F 1,13 = 2.17, P < 0.001; fig.2).CCA was significant (F 2,13 = 1.76,P < 0.001), explaining 34.5 % of the observed variance with the first axis being significant (F 1,13 = 2.29, P = 0.003).
A significant interaction between 'management' and 'date' effects was found for total activity-den-    Total activity-density decreased in the two units (non-grazed: t = 5.56, df = 3, P = 0.011; grazed: t = 8.85, df = 3, P = 0.003) but the decrease was more important in the grazed unit (fig.6).No significant interaction between 'management' and 'date' effects was found for total species richness (F 1,6 < 0.001, P > 0.999).No significant effect of year (F 1,13 = 2.049, P = 0.387) or management was found on species richness (F 1,13 = 2.049, P = 0.176).β-diversity was nearly constant over time in each unit, yet the nestedness increased in the non-grazed unit (table 2), whereas its species turnover decreased.Differences between the units were similar in the two years of sampling, i.e. before and after management by grazing took place.

Species composition
Considering the species that were not shared between the two units in 2004 but that were shared in 2003, the most numerous species was Linyphiids (e.g.Bathyphantes gracilis or Lepthyphantes ericaeus), a family well known for its long-distance dispersal abilities, at both young and adult stages (e.g.Blandenier, 2009;Simonneau et al., 2016).This change is therefore probably due to inter-annual variations in ballooning and/or under-sampling due to their small size.Almost all new shared species in 2004 were also Linyphiids (e.g.Erigone atra, Meioneta mollis, M. beata and Pelecopsis radicicola), and therefore unlikely attributable to any management effect.Among the species that disappeared after the management took place, some are strongly dependent on vegetation structure (a key factor in shaping spider assemblages: e.g.Hatley and Macmahon, 1980), and were thus likely disfavored by the grazing.This could be the case of several ambush hunters (Xysticus cristatus, X. erraticus and X. tortuosus) and web-builders (e.g.Enoplognatha ovata).It is harder to link the disappearance of the other species from the grazed unit to the grazing, because they were not initially in high numbers and/or they are not directly linked to the vegetation.However, the appearance of several thermophilous species (either at ground level or at vegetation level: Myrmarachne formicaria and Tegenaria agrestis, Cyclosa oculata, Dictyna latens and Hypsosinga albovittata respectively; Harvey et al., 2002) can be explained by a more open micro-habitat under grazing (e.g.Gibson et al., 1992).This tendency should be verified by longer-term monitoring.
Overall, few, if any (see e.g. the preference of Pachynatha degeeri, an ubiquist species, for the grazed meadow), significant change in species composition were observed after horses were introduced on the dry meadow, a finding in agreement with a few other studies on the impact of grazing on spider composition (Pozzi et al., 1998;Dennis et al., 2001;Pétillon et al. 2007).No significant changes were found after the grazing took place in spider hunting guilds, although web-builders tended to be reduced in Hemm and Höffer, 2012;Ford et al., 2013), mainly explained by a reduced vegetation complexity (Churchill and Ludwig, 2004;Kovac and Mackay, 2009).

Structural diversity
A decrease in total activity-density was observed in both units over time, but the rate was higher in the grazed unit.This is likely the result of the deleterious effects of grazing on ground-dwelling spiders (see above) and inter-specific competition-mediated changes (Wise, 2006;Van Klink et al., 2015).Competition among spider species is indeed higher in simple habitats (Hurd and Fagan, 1992;Marshall and Rypstra, 1999).Closer attention to change in abiotic and biotic conditions would be necessary to disentangle direct and indirect effects of habitat change on spider abundance (local factors are likely more important here, in an extensively managed land: Horváth et al., 2015).
Contrary to several previous studies on other large herbivores (e.g., Pozzi and Borcard, 2001;Pétillon et al., 2007), no significant effect of grazing was found on spider species richness.Indeed, as for many other taxa, and as with many other disturbances, grazing by large herbivores is usually reported to be negative on spiders (Bell et al., 2001; see also the meta-analysis of Prieto-Benítez and Méndez, 2011).Mechanisms usually involved in such a rich-Table 2. Partition of beta-diversity into species turnover and nestedness in both grazed and non-grazed units, before (2003) and after (2004) management took place (data from pitfall traps: βt, β turnover; βn, β nestedness).
Hunting guilds are usually affected by grazing, with several studies showing a decrease in web-builders (Gibson et al., 1992;Kirby, 1992;Bell et al., 2001;

Ground runners
Web-builders ness decrease encompass local habitat simplification (litter and vegetation strata: Dennis et al., 2001) and related food reduction (Kruess and Tscharntke, 2002; for soil functioning: Koppel et al., 1997).However, Lafage et al. (2014) recently demonstrated that spiders were not influenced by the plant biomass (either in terms of abundance or species richness).Finally, it is possible that the study period was too short to show effects.
Our hypothesis about increased β-diversity was not verified.This negative result has two main explanations: the study period on the effects of grazing was too short to be visible (although arthropods, and especially spiders, are known to quickly react to changes in habitat structure: Pétillon et al., 2014), and/or the effects of grazing were counterbalanced by inter-annual variations.We consider our second explanation is the most likely because it would explain the increased nestedness in the ungrazed unit.Plant heterogeneity is usually higher under grazing treatment (e.g.Gallet and Rozé, 2001;Van Klink et al., 2015), which reinforces the idea that plants and arthropods, here spiders, react in a different way to management practices (probably due to differences in mobility: Lafage et al., 2015;Lafage and Pétillon, 2016).We should stress that our sampling design did not necessarily encompass the spatial heterogeneity resulting from grazing in general (e.g. for spiders: Bonte et al., 2000), and especially by horses (Loucougaray      et al., 2004), and would deserve a higher replication for all the sampling methods to properly assess spider diversity (including beta-diversity, see e.g.Klimek et al., 2008;Báldi et al., 2013) Concluding remarks In this study, despite the existence of true replicates within each unit, units were confounded with the management treatment, which can be considered as a case of pseudoreplication in the sense of Hurlbert (1984).
Comparing stations between different sites is likely to increase inter-class variance through the existence of other co-varying factors (Oksanen, 2001).Here, even at a small scale, we showed a high variance between stations, with differences between units before the grazing took place.This underlines the importance of carrying out diachronic approaches, also because the effects of management were sometimes only visible over time, and not when comparing grazed vs. nongrazed units.Although spiders present high dispersal ability and high mobility (Lafage and Pétillon, 2014) that may hide or decrease the effects of management (Pech et al., 2015), they can be considered a relevant group for monitoring biodiversity consequences of management, bringing complementary information to changes in vegetation structure.
Fig. 1.Venn diagram of exclusive vs. shared species of grazed and non-grazed units, before (2003) and after (2004) management took place.
Fig. 3. Percentage activity density of each hunting guild sampled by pitfall trapping in 2003 and 2004.

Fig. 4 .
Fig. 4. Percentage abundance of each hunting guild sampled by sweep netting in 2003 and 2004.

Fig. 5 .
Fig. 5. Percentage abundance of each hunting guild sampled by hand collecting in 2003 and 2004.

Fig. 6 .
Fig. 6.Activity-density of spiders in the grazed and non-grazed units in 2003 and 2004 (data from pitfall traps).
Fig. 2. Projection of significant variables from the CCA on spider species.Pitfall traps are represented by circles and species by crosses.