Endemic plant species are more palatable to introduced herbivores than non-endemics

Abstract

Islands harbour a spectacular diversity and unique species composition. This uniqueness is mainly a result of endemic species that have evolved in situ in the absence of mammal herbivores. However, island endemism is under severe threat by introduced herbivores. We test the assumption that endemic species are particularly vulnerable to generalist introduced herbivores (European rabbit) using an unprecedented dataset covering an entire island with enormous topographic, climatic and biological diversity (Tenerife, Canary Islands). With increasing endemism, plant species are more heavily browsed by rabbits than non-endemic species with up to 67% of endemics being negatively impacted by browsing, indicating a dramatic lack of adaptation to mammal herbivory in endemics. Ecosystems with high per cent endemism are most heavily browsed, suggesting ecosystem-specific vulnerability to introduced herbivores, even within islands. Protection of global biodiversity caused by disproportionally high endemism on oceanic islands via ecosystem-specific herbivore control and eradication measures is of utmost importance.

1. Introduction

The European rabbit (Oryctolagus cuniculus) is considered one of the introduced species that has been most detrimental to island floras worldwide [1,2]. It has been successfully introduced to at least 800 islands, making it one of the most widely distributed animal species on Earth [3]. On islands, endemic plant species have often evolved in the absence of mammalian herbivores [1]. Thus, endemics are considered to be particularly vulnerable to browsing by generalist herbivores such as rabbits as a result of lacking herbivore defence mechanisms [4]. However, an island and ecosystem-scale quantification of the browsing preference of introduced rabbits and how it affects the distribution of diversity and endemism on oceanic islands is still missing.

Islands, especially in the tropics and subtropics, are hot spots of global biodiversity [5]. Although islands only cover approximately 5% of the global terrestrial surface, they contain around 17% of all plant species [6]. This remarkable contribution to global biodiversity is mainly a result of the many range-restricted endemics found on islands [7] that are a result of evolutionary processes [8]. However, this exceptional contribution of islands to biodiversity is threatened by a multitude of different and often interacting threats such as land use change [9], invasive species [10] and climate change [11]. Indeed, globally most species extinctions have been reported from islands with extinction rates that are currently accelerating [12].

Besides these threats, introduced herbivores, exemplified by the detrimental effects of the European rabbit, are among the most serious threats to island biota and insular systems [1]. Thereby, rabbits are considered ‘ecosystem engineers’ because their activities produce remarkable ecosystem-level effects including changes in soil structure and composition [13] and modifying richness and diversity of plant species [14]. Indeed, rabbits exhibit high grazing selectivity in areas where they have been introduced [15]; hence, palatable or preferred plants are less abundant where rabbits reach high numbers [16]. Rabbit damage to native vegetation is primarily due to selective browsing that suppresses preferred species and favours less preferred competitors [17]. Rabbits thereby show extremely high search efficiency for preferred seedlings that constitute a very small proportion of available plant biomass [18]. However, they contribute largely to the total biomass reduction through their effects on regeneration [19]. This has led to the fact that certain endemic plant species have been pushed to the brink of extinction [19,20].

We use the entire island of Tenerife, the largest and most diverse island of the Canary Islands and part of the Mediterranean Basin biodiversity hotspot sensu [5], as a model system for the following reasons: (a) Tenerife can be considered a climatic mini-continent with continental-scale climatic gradients on the scale of a medium-sized island [21], thus allowing more general conclusions than from a less heterogeneous island. (b) Tenerife contains a spectacular array of endemic plant species of different endemism categories (single-island endemic (SIE), archipelago endemic, Macaronesian endemic [22,23]) making it the most diverse oceanic island in the Atlantic Ocean and enabling us to accurately assess browsing damage per endemism category. (c) The island harbours a large set of very different ecosystems (ranging from desert-like coastal areas through evergreen subtropical cloud forest to arid alpine scrub [24]), thus, allowing us to assess rabbit damage in very distinct ecosystem types. (d) An abundant rabbit population exists throughout the whole island, even reaching the highest elevations [19,20].

Based on this background, we investigate the following hypotheses: (i) endemic plant species are more palatable to introduced rabbits than non-endemic plant species. (ii) Higher rabbit browsing damage is expected in ecosystems dominated by endemics. (iii) A combination of environmental and biotic variables determines rabbit browsing damage throughout the island of Tenerife. To achieve this, we use a state-of-the-art indirect sampling method enabling us to assess island-wide rabbit density [17], rabbit browsing damage [25] and plant diversity and endemism, resulting in an unprecedented spatial coverage of rabbit browsing effects on an entire oceanic island. We apply structural equation models (SEMs) to disentangle the direct and indirect effects of environmental and biotic variables on rabbit browsing damage.

2. Material and methods

(a). Study area

This study was conducted on the mountainous and topographically variable island of Tenerife (Canary Islands, Spain). Tenerife reaches a maximum elevation of 3718 m.a.s.l., and covers an area of 2034 km². Tenerife is 11.9 Ma years old [26]. Our study was conducted within an elevation range from 17 to 3538 m.a.s.l., thus covering virtually all relevant environmental gradients and all major ecosystems of the island (figure 2). As a result of the influence of the NE trade winds, the windward northern side of the island receives more precipitation than the dry to the arid southern leeward side. On the windward side, trade wind stratocumulus clouds form a cloud layer between 800 and 1500 m.a.s.l. [27]. Inversion prevents the clouds from reaching higher elevations leading to arid and cool conditions that can regularly receive snow and ice during winter above the inversion layer.

Figure 2.

Spatial pattern and per-ecosystem rabbit browsing damage and rabbit density. (a) Spatial distribution of rabbit browsing and (b) rabbit density on Tenerife (n = 215 sites). Sample sites cover the entire island and all major ecosystems. Circle sizes are proportional to the values, which also increase from light to dark. Boxplots show (c) rabbit browsing damage per ecosystem and (d) rabbit density per ecosystem. Interestingly, rabbit browsing damage per ecosystem and rabbit density per ecosystem do not necessarily overlap. Colours as in (a) and (b), respectively. Boxplots in (c) and (d) are ordered from highest mean value to lowest. For more information on the ecosystems see electronic supplementary material, table S2.

Tenerife possesses wide climatic gradients ranging from areas with a mean temperature of 22°C and less than 200 mm of annual precipitation in the driest coastal areas to around 5°C and 400 mm in summit regions, as well as mid-elevation laurel forest areas where precipitation can rise to more than 800 mm (electronic supplementary material, figure S1). As a result, Tenerife harbours a wide range of quite distinct ecosystems strongly influenced by its Mediterranean to subtropical position in the Atlantic Ocean. We classify 11 major ecosystems based on [24]. See a complete list of all 11 ecosystems and their description in electronic supplementary material, table S2.

As a result of these diverse environmental conditions, Tenerife is the most diverse oceanic island of the archipelago (with 297 Canary endemics, of which 135 are SIE and 31 Macaronesian endemics, among a total of 1468 vascular plants) and of all oceanic islands in the Atlantic Ocean, making it a major contributor to the Mediterranean biodiversity hotspot [5]. Tenerife is an oceanic island, i.e. it has never had contact with the mainland. Thus evolutionary processes such as adaptive radiation have created a unique and spectacularly diverse assembly of endemic plant species [7], including several endemic genera (e.g. Bystropogon, Gonospermum, Kunkeliella, Spartocytisus or Todaroa).

(b). Sampling method

We randomly preselected 215 sample sites covering the entire island in unpopulated and non-urban areas with an inclination of less than 45° in ArcGIS (figure 2). At each site, we used circular transects avoiding areas with more than 30° slope inclination. Inclination of less than 30° was chosen as a threshold because rabbits prefer deep soils for burrowing [19] which are only found in flat to moderately steep areas. Each circular transect measured 400–500 m in length (depending on site conditions) and was surveyed on both sides (5 m at each side) by walking along it until completing the rabbit density method (see below). Within each site, we recorded and classified all species according to their origin, i.e. as endemics (including single-island endemic, Canary endemic, and Macaronesian endemic), non-endemic natives and non-native species [22,23]. In total, we recorded 316 vascular plant species. Of these 164 were non-endemic native and non-native species, 27 Macaronesian endemics, 98 Canary endemics and 27 single-island endemics. Also, per site we recorded abiotic variables: elevation, slope, area of rock cover and soil cover. Mean annual precipitation and mean annual temperature were taken from climate maps (period 1981–2010) for each site [28]. The sampling was undertaken in June and July 2017.

(c). Measures of species richness and endemism

We calculated single-island endemic richness, archipelago endemic richness and Macaronesian endemic richness (most of them restricted to Canaries and Madeira) as the sum of all species per site falling into the respective categories. Thereby, we defined levels of endemism in a strict sense, i.e. Macaronesian endemics are only those species that are not single-island or Canary endemics (in contrast to wider definitions, e.g. [8]). Similarly, we calculated per cent single-island endemics, per cent archipelago endemics and per cent Macaronesian endemics by dividing the respective richness value through total species richness. Per cent endemism is an indicator of the importance of endemic species in a plant community and can be used to identify hotspots of endemism important for conservation purposes [21].

(d). Rabbit browsing damage and rabbit density

The European rabbit is a generalist herbivore and one of the most widespread and most invasive species on Earth [3]. Under favourable conditions, rabbits have high reproduction rates and show high adaptability to a wide range of climates and diets [29]. The European rabbit was introduced to the Canary Islands in the 15th century AD by Castilian settlers [29] and is still used for recreational hunting purposes.

We identified rabbit browsing damage per species per site mainly by the oblique chisel-like cuts through stems and the presence of rabbit pellets at the basis of the plants [25]. The clean 45° cut distinguishes their bite marks from others left by rats (small, irregular bite marks) and ruminants such as goats and mouflon (ruminant browsing destroys large parts of the plant). By standing on their rear legs, adult rabbits can damage up to heights of 40–50 cm above the ground [25]. Because we could clearly distinguish rabbit browsing from other types of browsing (very rare), we exclusively focused on rabbit browsing. Based on these observations, we categorized rabbit browsing damage on a scale from 0 (no damage) to 5 (damaged twigs and bark for all individuals) per species following recommendations by [25]. Rabbit browsing damage characterizes browsing damage on the entire plant, including leaves, branches, stems and twigs. Per-species rabbit browsing damage was calculated as the mean of all recorded damages per species, while per-site rabbit browsing damage was calculated as the mean of all browsing damages recorded for all species per site.

We estimated rabbit abundance during June and July 2017, when rabbit populations on the Canary Islands reach their maximum densities [30] which is typical for Mediterranean-type ecosystems [30]. To estimate rabbit density, we counted rabbit pellets (droppings) in 150 × 0.1 m² circular plots randomly distributed along the above-mentioned line transect at each site [17,31]. Each plot was thoroughly surveyed, and the total number of pellets counted. This method provides an estimate of rabbit density derived predominantly from pellets deposited mostly in the previous 12–24 months, and with a lesser contribution of pellets deposited previously. As a result, estimates have shown to reflect rabbit density across a 2-year period [17], thus accounting for short-term population fluctuations. It is important to note that under humid conditions rabbit pellets disintegrate faster than under dry conditions [32]. This might lead to a certain degree of bias towards dry areas in our study. However, we consistently sampled rabbit densities during the dry summer months, where precipitation is generally much lower than in winter. Indeed, summer months have been reported to be the optimal period for counting rabbit pellets [32]. Previous studies have shown that comparable sampling methods to ours are very effective in estimating rabbit density, also in comparison of different climatic conditions [31].

In addition, to ensure that the European rabbit was the only invasive herbivore affecting our sites, we also counted dung pellets of feral goats (Capra hircus L.) and European mouflons (Ovis orientalis Gmelin)—two other important introduced herbivores—in the same plots. A low number of goats and mouflon pellets in only a few of the sites allowed us to relate the browsing damage effects exclusively to rabbits.

(e). Data analysis

We tested significant differences in species-specific rabbit browsing damage according to endemism using the non-parametric Kruskal–Wallis test (R package pgirmess) because the data were not normally distributed, even after transformation. Based on the Kruskal–Wallis test, we assigned significant groups if differences in p-values were non-significant using a p > 0.05. To test for alternative influences, we used generalized linear mixed effects models (GLMMs) with a Poisson-family distribution within the framework of the R package lme4. We explained individual browsing damage per species on the site level with endemism (our hypothesis) but included several other, possibly confounding fixed effects: (i) site elevation representing changes of endemism and habitat along the elevation gradient, (ii) nitrogen-fixer because N-fixing species have higher nutritional value for herbivores and might be preferentially browsed [14], (iii) plant life form (see electronic supplementary material, table S3 for details), because certain plant life forms might be preferentially browsed by herbivores (e.g. herbs versus shrubs/trees), (iv) succulence because succulent species might be preferentially browsed in water-limited systems, and (v) mechanical defences (e.g. presence/absence of thorns and spines) because these defences would deter herbivores from browsing on such species [4]. As random effects, we included species name to account for uneven sampling across species and plant family to account for possible phylogenetic effects because endemic, and therefore particularly palatable species, might aggregate within certain plant families as a result of evolutionary processes such as adaptive radiation that are common for islands [8]. In a stepwise procedure all non-significant fixed effects were removed from the model, resulting in a most parsimonious model.

To assess the complex interdependencies between environmental and biotic variables affecting rabbit browsing damage, we built a structural equation model (SEM) based on a conceptual path dependency model (electronic supplementary material, figure S4). SEMs have the advantage that they are able to account for direct and indirect effects in complex systems [33]. Our conceptual path dependency model assumes that environmental variables can either directly affect rabbit browsing damage or indirectly via biotic variables. The SEM was implemented in the R package lavaan, which lets the user create individual links between variables and then calculates (i) a standardized model estimate (that can be seen as a correlation coefficient) per linkage as well as a p-value, and (ii) an R² value for all dependent variables. In the first step, a full SEM was calculated for each rabbit browsing damage category (all species, single-island endemics, Canary endemics, Macaronesian endemics) including all predefined links from the conceptual path dependency model. Note that for each rabbit browsing damage the respective richness values and, where applicable, also per cent endemism values were adjusted accordingly (e.g. for rabbit browsing damage on single-island endemics we used single-island endemic richness and per cent single-island endemics in the SEM). Then, we removed all non-significant links and ran the best-fit SEM again in order to obtain the most parsimonious SEM. In the second step, we tested if there were significant differences between the full SEM and the best-fit SEM for each rabbit browsing damage category using a χ²-test [34]. The best-fit SEMs for all species, single-island endemics, Canary endemics and Macaronesian endemics had a good fit with their respective data (in all cases: p(χ²) > 0.05), i.e. the result of the χ²-test was non-significant. This indicates that these were the most parsimonious SEMs.

We also calculated the direct and indirect effects of each environmental and biotic variable on each rabbit browsing damage category to identify the pathways of influence for each environmental and biotic variable using the best-fit SEM for each rabbit browsing damage category. The direct influence is defined as the direct link between an independent variable (environmental or biotic) and rabbit browsing damage. In contrast, we calculated the indirect influence by summing up the product of each combined environmental and biotic link influencing rabbit browsing damage [35]. For example, the indirect effect of slope on rabbit browsing damage for Canary endemics (electronic supplementary material, table S5) was calculated as: (0.18 × (−0.30)) + ((−0.20) × (−0.15)) = −0.02. Links undefined in the conceptual path dependency model were assigned ‘n.a.’.

All analyses were conducted in R Statistics (Version 3.4.0) [36].

3. Results

(a). Browsing selectivity of rabbits and level of endemism

Endemic species were significantly more browsed than non-endemic species and non-native species (p < 0.001), and were thus considered to be more palatable to introduced rabbits than non-endemic species (figure 1a). Indeed, among the endemics, 56% showed rabbit browsing damage (mean rabbit browsing damage for endemics = 0.65); while only 30% of the non-endemic species and non-native species were damaged (mean rabbit browsing damage for non-endemics = 0.25). When looking at endemism in more detail, rabbit browsing damage tended to increase with the degree of endemism (i.e. Macaronesian endemics < Canary endemics < single-island endemics; figure 1b). Rabbit browsing damage was present in 67% of the single-island endemics (mean rabbit browsing damage = 0.87), 54% of the Canary endemics (mean rabbit browsing damage = 0.65), and 52% of the Macaronesian endemics (mean rabbit browsing damage = 0.49). We found no significant difference in rabbit browsing damage between non-endemic natives and non-natives. All in all, this suggests that palatability increases with an increasing level of endemism.

Figure 1.

Rabbit browsing damage per level of endemism. (a) Endemic species (n = 150) are significantly more damaged by browsing than non-endemic native and non-native species (n = 169). Here, endemism is defined as all species endemic to Macaronesia. This includes single-island endemics as well as Canary endemics and Macaronesian endemics. (b) A tendency exists that with increasing endemism status browsing damage increases, indicating that palatability increases towards increasing range restriction of endemics. Here, endemism is explicit, i.e. Macaronesian endemics consist only of species endemic to Macaronesia but no Canary endemics nor single-island endemics. Endemism are single-island endemics only (n = 27), Canary endemics only (n = 98), Macaronesian endemics only (n = 27), non-endemic natives (n = 123) and non-natives and ruderals (n = 44). Asterisks indicate a high level of significance (*** = p < 0.001) in (a). Lowercase letters indicate significant groups in (b) (significance level: p < 0.05).

When accounting for plant family and species name as random effects, the results from the most parsimonious GLMM showed that endemism contributed significantly to individual browsing damage per species on the site level, together with site elevation and N-fixers (electronic supplementary material, table S6). In contrast, plant life forms, mechanical defences and succulence did not significantly contribute to the GLMM and were therefore discarded from the model. The estimates from the GLMM show that endemism and N-fixers had a positive effect, while elevation had a negative effect on browsing damage.

(b). Rabbit browsing damage and rabbit density

Interestingly, rabbit browsing damage was not randomly distributed throughout the major island ecosystems (figure 2a,c). Open and semi-open ecosystems found at high and low elevations (e.g. the summit scrub or coastal succulent scrub) generally had higher rabbit density than forest ecosystems (pine forest, Erica–Morella woodland and laurel forest; figure 2c,b). A strong positive relationship existed between browsing damage and per cent endemism (R² = 0.46***, n = 215). A spatial concentration of rabbit browsing damage at high elevations was clearly visible (figure 2a), whereas the humid evergreen laurel forest with its high share of Macaronesian endemics showed both low damage and density. Between rabbit browsing damage and rabbit density there was only a very weak positive correlation (R² = 0.04**, n = 215). In fact, rabbit density can be quite high in ecosystems that showed little or no browsing damage (figure 2a,b). Thus, rabbit density was only a weak overall predictor of rabbit browsing damage.

(c). Direct and indirect effects of environmental and biotic variables

In looking at rabbit browsing damage for all species, we found a positive direct effect of elevation on rabbit browsing damage and a negative indirect effect through species richness (figure 3a; electronic supplementary material, table S5). Mean annual precipitation directly affected rabbit browsing damage negatively and had an adverse effect on rabbit density. Slope and soil cover had positive direct effects on species richness but did not have direct effects on rabbit browsing damage. Slope also had a negative direct effect on rabbit density. As mentioned above, rabbit density had a slightly positive effect on rabbit browsing, while species richness had a negative effect, which is related to the richness of non-endemic and non-native species in the areas with higher species richness. By combining environmental and biotic variables the SEM can successfully explain rabbit browsing damage as well as species richness for all species, as the high R² values show, while it can only explain a small proportion of variance for rabbit density.

Figure 3.

Results of the structural equation model (SEM) for different levels of endemism explaining rabbit browsing damage. SEMs were constructed for (a) all species, (b) single-island endemics, (c) Canary endemics and (d) Macaronesian endemics. Blue arrows indicate positive, red arrows negative relationships. Values next to the arrows are standardized model estimates indicating the strength of the relationship. This strength is reflected in the proportional width of arrows. Only values greater than 0.2 or less than −0.2 are shown to increase readability of SEMs (for all values see electronic supplementary material, table S5). R² values indicate the combined explanatory power of all dependent variables pointing at the respective variable. Note that some variables were log-transformed or quadratic-transformed.

When looking at the different levels of endemism, we see a largely consistent picture with regard to the environmental variables, although slight differences exist (figure 3b–d; electronic supplementary material, table S5). Elevation showed a positive influence on rabbit browsing damage for single-island endemics and Canary endemics but a negative influence for Macaronesian endemics. Elevation also had a strong positive influence on per cent endemism (SIE and Canary endemics). However, elevation had a negative effect on per cent endemism for Macaronesian endemics, because this species group mainly occurs in the laurel forest at low to mid elevations—a habitat where only few non-endemic and non-natives occur. Elevation had a positive effect on single-island endemic richness and a negative one on Canary endemic richness and Macaronesian endemic richness. This is likely to be caused by the fact that high elevation areas harbour many single-island endemics, while many laurel forest endemics at mid- to low elevations are shared with the island of Madeira [37]. Mean annual precipitation had a direct negative effect on rabbit browsing damage for Canary and Macaronesian endemics as well as on rabbit density (figure 3b–d; electronic supplementary material, table S5). Again, this has to do with the fact that rabbits tend to avoid humid areas such as the laurel forest. Although we only sampled slopes up to 30° inclinations, rabbits preferred flat areas, as the negative effect of slope on rabbit density indicates, because flat areas offer deeper soils better suited for burrowing [38]. Soil cover had only a slight direct negative effect on endemic richness.

Per cent endemism displayed a positive relationship with rabbit browsing damage (except for Macaronesian endemics located at highest precipitation areas), while the direct influence of rabbit density only was weak or non-existent (figure 3b–d; electronic supplementary material, table S5). The direct influence of rabbit density on browsing damage had only a weak positive effect on Canary endemics and did not exist for single-island endemics and Macaronesian endemics (figure 3b–d). Only Canary endemic richness had a negative effect on rabbit browsing damage. In general, we found high explanatory power to explain rabbit browsing damage on endemic plant species based on environmental and biotic variables, whereas rabbit density is only poorly explained in the SEM.

4. Discussion

We comprehensively demonstrate that endemic plant species are more vulnerable to browsing damage from introduced herbivores than non-endemic and non-native plant species on the scale of an entire, very heterogeneous island harbouring multiple distinct ecosystems and a large variety of endemic plant species. Our study identifies a surprising mismatch between browsing damage and introduced herbivore density with major implications for conservation efforts and endangered species management on Tenerife, and likely also other islands. Based on these findings and owing to its widespread distribution on islands, the dramatic impact of the European rabbit on endemic plant species is well documented and, in many cases, has completely reshaped island systems and their species assemblages worldwide (e.g. [1,13,14]).

Browsing damage increases with increasing endemism (figure 1). Indeed, the endemism-browsing relationship seems stable as other, potentially additional factors such as plant life forms, mechanical defences and succulence were not significant factors. On Tenerife, like on most oceanic islands, endemic plant species have evolved in the absence of mammalian herbivory [4,39]. Thus, it is likely that through evolutionary processes, e.g. by reallocating resources formerly used for herbivore defence to other physiological processes, endemic plant species have lost (or never developed) specific herbivore defence mechanisms [4]. This makes them particularly palatable to introduced generalist herbivores such as the European rabbit, mouflon or feral goats, and explains why endemic plant species are preferentially targeted for browsing by introduced herbivores [13,14,40]. In contrast, non-endemic plant species are significantly damaged less by browsing on Tenerife. As a result, continuing browsing pressure of introduced herbivores on endemic plant species reduces or even completely eliminates regeneration in endemics [13,19,20]. In fact, the combined effect of introduced herbivore pressure and invading alien plant species [10] strongly increases the extinction risk of endemics, which are often already subject to increased vulnerability by land-use change [9] and climate change [11]. In order to conserve the spectacular diversity presented by islands around the world combined control measures targeting introduced herbivores as well as alien plant species are of utmost importance in reducing the already exceedingly high threat of extinction of many island species [10].

We find a clear spatial pattern of browsing damage on Tenerife that is likely a result of the positive relationship between endemism and rabbit browsing damage. With increasing endemism rabbit browsing damage tends to increase (figure 3), which reflects the higher palatability of endemic plants for rabbits compared to non-endemics natives and non-natives. An increase of per cent endemism, especially SIE, with elevation is a global phenomenon on islands due to topography-driven isolation [41]. Additionally, harsh climate conditions prevent the establishment of most non-endemic native and non-native plants related to disturbance at high elevations [42], thus increasing per cent endemism at high elevations. Our results show that, under current conditions, high elevation areas are those with highest overall browsing damage. Interestingly, high elevation areas are often perceived as untouched or conserved, especially because of large distances to human settlements and overall relatively low human influence. However, they are among the systems most disturbed by introduced herbivores, even in protected areas (Haleakala NP on Maui, Hawaii [43], Tongariro NP, North Island, New Zealand [44] or Fogo NP on Fogo, Cape Verde [45]).

Beside the presence of endemic species, island environmental features are also responsible for the distribution of browsing damage on Tenerife. Precipitation has a negative effect on rabbit density for all considered groups as well as on browsing damage for the Canary and Macaronesian endemics (figure 3). Interestingly, in this ecosystem we find both the lowest rabbit densities as well as the lowest rabbit browsing damage of all ecosystems of the island, indicating that this type of habitat is unsuitable for rabbits, at least in the densest and best preserved forest areas. Also, among the Macaronesian endemics, a common group in the laurel forests [41], nearly half did not show rabbit browsing damage. As a result of its Mediterranean origin, the European rabbit prefers mesic to dry habitats and avoids humid forests [46] such as the laurel forest that is located on the windward side of the island facing the incoming trade wind clouds [24]. This might result in long-term protection of the laurel forest from the detrimental effects of rabbit browsing. However, climate change will likely alter island-scale precipitation regimes [11], potentially threatening plant species and systems previously considered safe from introduced herbivore browsing.

Interestingly, rabbit density is a remarkably poor indicator of browsing damage on island plant communities (figures 2 and 3). Our results clearly illustrate that ecosystems with high rabbit density are not necessarily the most heavily browsed ones and vice versa. Most strikingly visible is this mismatch for the economically and ecologically important pine forest ecosystem (figure 2). Of all 11 ecosystems considered on the island of Tenerife, the pine forest has the third lowest rabbit density but the second highest browsing damage. Thus, depending on the ecosystem in focus, rabbits can already have a detrimental effect on endemic species at low population densities. Indeed, even in their native range, huge damage to crops by rabbits can occur in areas with low rabbit density [47]. At the same time, we find areas with high rabbit densities but low browsing damage. This might be the result of a saturation effect where the damage reaches a maximum and cannot increase further [48]. The saturation effect might lead to a high ratio of adapted versus non-adapted plant species to rabbit browsing, helping rabbits to maintain high densities but masking their overall browsing effects, as apparently occurs when planting alternative food to avoid damage to crops [49].

Traditional conservation efforts are directed towards reducing introduced herbivore densities on entire islands or within individual systems based on the assumption that introduced herbivore densities are an indirect measure of browsing damage [40,50]. Yet, we show that island systems react in a rather selective manner with regard to introduced herbivore densities, i.e. showing system-specific density–damage relationships [40], depending on the share of endemic species per system and the environmental conditions. Thus, we argue that instead of using introduced herbivore densities, it is more effective to directly assess the browsing damage per system and then develop control targets and management strategies based on the damage assessment for individual systems.

5. Conclusion

Our study adds to the mounting evidence suggesting that it is paramount to control or, if possible, even eradicate introduced herbivores on oceanic islands to preserve native and endemic plant species on islands. Islands are disproportionally rich in range-restricted endemic species [7], which are—as we show—particularly threatened by generalist-introduced herbivores such as the European rabbit. Indeed, studies report that the European rabbit is regarded as a direct threat to more than 30 endemic plant species on the Canary Islands, which are considered to be endangered or critically endangered by the IUCN [50]—a number that is currently increasing [19]. In addition, rabbits can disrupt native seed dispersal networks while at the same time promote the dispersal of invasive plant species [39] and change plant species composition, making rabbits effective ‘ecosystem engineers’ within their introduced range [13,19,20]. Our results show that on Tenerife more than 40% of all plant species, and up to two thirds of the endemics, are negatively affected by rabbit browsing, emphasizing the profound impacts of introduced herbivores shaping native island floras. However, eradication measures targeting introduced herbivores such as the European rabbit on other islands have shown promising results for the recuperation of native and endemic plant species [51]. We argue that conservation managers and practitioners should pursue such eradication efforts because rabbit eradications are feasible and successful in 84% of all documented cases (106 successful eradications out of 126 documented eradications according to the Database of Island Invasive Species Eradications [52]). However, on such a big and complex island as Tenerife, eradication is not always feasible but, for effective rabbit control, managers and practitioners should still consider the system-specific vulnerabilities to browsing as well as the spatial distribution of endemism on a target island to ensure the long-term survival of unique island species and thus help to sustainably preserve substantial contributors to global biodiversity.

Data accessibility

All biotic and abiotic data used in this study can be accessed via the Dryad Digital Repository: https://doi.org/10.5061/dryad.55hk639 [53].

Authors' contribution

J.M.G.-M. designed the study; all authors collected the data; S.D.H.I. and J.C. analysed the data; S.D.H.I., J.C. and J.M.G.-M. lead the writing with all co-authors contributing. All authors gave final approval for publication.

Competing interests

We have no competing interests.

Funding

Funding for the project was provided by the Ministerio de Agricultura, Alimentación y Medio Ambiente (REF 1621/2015). J.C. received a PhD scholarship from La Laguna University.