Herbivores and granivores represent one of the most influential drivers of plant abundance and population dynamics. Their effect can be, in turn, modulated by biotic or abiotic factors such as community composition, habitat characteristics or space heterogeneity.
Recent approaches to the study of herbivore and granivore impacts on plants have considered the combined action of multiple herbivore species, the effect of herbivores on several plant life stages or the effect of environmental gradients on these interactions. However, studies addressing the effect of multiple herbivore species on different plant life stages are still lacking.
We estimated the combined effect of multiple exotic herbivores (European rabbits, Oryctolagus cuniculus; black rats, Rattus rattus; and house mouse, Mus musculus) on four different life stages of an endangered plant species (Medicago citrina, Fabaceae). Mortality for seed, seedling and sapling was estimated at three types of plots (open, rat exclusion and rat + rabbit exclusion) replicated at four sites (N = 3 per site and treatment) within Cabrera Island (Balearic Islands, western Mediterranean). Browsing of reproductive adults was simulated under common-garden conditions (Sóller Botanic Garden, Mallorca Island) and its effect on reproductive effort and success measured.
European rabbits and black rats had complementary impacts on the different life stages of M. citrina. These included independent effects on different life stages (seed predation by rats, seedling predation by rabbits), which resulted in multiplicative increases in plant mortality, and concurrent effects on the same life stage (sapling predation). In addition, the simulated-herbivory experiment showed that a low rate of canopy removal (25% of initial biomass) already causes a strong decrease in fruit set (from 54% to 30%), but increasing rates of canopy removal do not increase this effect.
Synthesis. Our results stress the importance of considering the combined effects of different herbivores on several life stages of the plant's life cycle and their consecutive effects on population dynamics. From an applied point of view, future reintroduction attempts of M. citrina in Cabrera Island should consider measures to either control the populations of both exotic herbivores or mitigate their impacts on the earlier recruitment stages of the plant (seeds, seedlings and saplings).
Plant recruitment is modulated by biotic and abiotic factors, which determine the survival of individuals between vital phases and ultimately condition population dynamics (Streng, Glitzenstein & Harcombe 1989; Bossard 1991; Clark et al. 1999; Barberá, Navarro-Cano & Castillo 2006). The analysis of such factors has therefore been a recurrent topic in population ecology from the early years of the discipline (e.g. Duncan 1954; Sacchi & Price 1992). Among biotic factors, the interaction between plants and herbivores has repeatedly been reported to be an intense modulator of plant population dynamics (e.g. Belsky 1986; Marquis & Braker 1994). Although the impact of herbivores is particularly significant during the initial stages of a plant's ontogeny (Hendrix 1988; Kauffman & Maron 2006; Bricker, Pearson & Maron 2010), their effects can be considerable over ulterior stages – saplings, juveniles and adults (Hester et al. 2000; Tanentzap et al. 2009).
Traditionally, plant–herbivore studies have focused on assessing the impact of a specific herbivore on a particular life stage of a focal plant species. However, recent work has started to address more complex cases, such as the impact of several herbivores on single plant life stages (e.g. Hulme 1993; Louda & Potvin 1995) or that of single herbivores acting on multiple life stages (Russell, Zippin & Fowler 2001). Different herbivores often show contrasting effects on the performance of consumed plants (Owen 1980; Rhoades 1985; Belovsky 1997), which may be mediated by contrasting impacts on different plant life stages (Maron & Kauffman 2006). Their combined action may result in either positive (Bakker et al. 2006), negative (Inouye 1982) or neutral (i.e. purely additive effects; Anderson & Paige 2003) accumulated impacts, or any combination thereof (e.g. Strauss 1991). In addition, the strong impact of a given herbivore type on particularly sensitive plant stages may determine ‘recruitment bottlenecks’ that can limit plant population dynamics – masking the putative effects of other herbivores in ulterior life stages of the plant (Louda 1983; Louda & Potvin 1995).
These complex effects can be reduced or amplified by the concurrent effect of the (abiotic and biotic) environment. Aside from the more generic effects of habitat dependency (e.g. Kauffman & Maron 2006; Maron & Kauffman 2006; but see Russell, Rose & Louda 2010), plant responses to herbivory have been shown to depend on resource availability (e.g. Hawkes & Sullivan 2001; Wise & Abrahamson 2007), physiognomic gradients (e.g. Louda 1982, 1983; Reader 1992), safe-site availability (e.g. Maron & Gardner 2000; García, Obeso & Martínez 2005) and community composition (Callaway et al. 2005; Baraza, Zamora & Hódar 2006).
Hence, to understand the conditions determining the effects of herbivores on plant performance and their ultimate bearing on plant population dynamics (Halpern & Underwood 2006), a deeper understanding of the effects of multiple herbivores on the various stages of plant development is required. Studies addressing such effects should ideally address the spatial context (e.g. habitat structure and/or community composition) in which they take place. Recent studies have extended their scope to deal with the differential effect of multiple herbivores (Strauss 1991; Hulme 1994; Gómez & Zamora 2000) and/or distinguish their effect on different plant life stages (Fagan & Bishop 2000; García & Ehrlén 2002; Horvitz & Schemske 2002), and others have analysed the effect of environmental gradients (Maron, Combs & Louda 2002; Maron & Kauffman 2006; Lau et al. 2008; Rose, Russell & Louda 2011) or density dependence (Parmesan 2000; Sullivan 2003) on herbivore impact. However, we are only aware of a handful of studies assessing the effects of multiple herbivores when acting on different life stages of their food plant (e.g. Lotze & Worm 2000; Vázquez 2002; Warner & Cushman 2002), as well as their interaction with spatial heterogeneity in habitat and/or community composition (Louda 1982; Gómez, García & Zamora 2003; Traveset et al. 2003).
In this study, we used a series of field and common-garden experiments to estimate the relative role of different herbivores in different stages of the life cycle of a perennial plant. We focused on the impact of three exotic herbivores with different functional roles (a herbivore, a plant- and seed-eating omnivore, and a granivore) acting on four different life stages (seed, seedling, sapling and reproductive adult) of the endangered, endemic shrub Medicago citrina (Font-Quer) Greuter (Fabaceae). Specifically, we aimed to (i) estimate the impact of the different herbivores on the different life stages of M. citrina and their cumulative effect on total plant recruitment (from seed to adult); (ii) identify ‘plant recruitment bottlenecks’, that is, plant life cycle stages particularly prone to suffer herbivory impacts and therefore limit recruitment; and (iii) evaluate whether these effects vary between two habitat types.
Our study served an applied purpose, related to the impact of exotic herbivores on insular plant endemisms. M. citrina's status as endangered, protected species results from the combination of reduced geographical range and the putative impact of several exotic herbivores (goat, rabbit, black rat) introduced to its natural habitats (Juan 2002). By quantifying the impact of the main exotic herbivores present at our study site (the Cabrera Archipelago National Park) and comparing it to that of native herbivores, we present sound guidelines for the conservation of the existing species populations and the design of future reintroduction attempts.
Materials and methods
Fieldwork took place at the Cabrera Island, situated just south of the island of Mallorca (Balearic Islands, eastern Spain) and included in the Cabrera Archipelago National Park (Figure S1, Supporting Information). Cabrera, the main island of the archipelago (11.5 out of 13.2 km2), is surrounded by 18 islets, three of which host small populations of the western Mediterranean micro-endemic M. citrina. These islets have a calcareous lithology and semi-arid warm Mediterranean weather (Emberger 1955) and are covered by Mediterranean garrigue under a discontinuous cover of Aleppo pine, Pinus halepensis L. (Rita & Bibiloni 1993). Cabrera's central valley (39°08′38.04″ N, 2°56′10.92″ E) hosted several agricultural plots which were abandoned in the mid-1960s and are now covered by tall grasses, thistles and spurges (Euphorbia spp.) interspersed with scattered, small bushes (hereafter referred to as ‘grassland’). The study area (Figure S1) comprised four independent sites located within these abandoned agricultural plots, at either valley-bottom areas of grassland or sloping areas colonized by open garrigue (two sites each).
Medicago citrina (Leguminosae) is a perennial medium-sized woody shrub (≤3 m height) with a total distribution of < 10 km2 comprised of only 11 populations in small islets. It is distributed in the Balearic and Columbretes Archipelagos, and a few small populations on the Eastern coast of Spain (Juan 2002). Most of these populations are composed of a few hundred individuals and show a continuous decrease in the number of reproductive adults (Juan 2002). In Cabrera Archipelago, it is present in three small islets (< 0.53 ha) situated around Cabrera: Ses Bledes, S'Estell de Coll and S'Estell de Fora (Palmer & Pons 2001; Juan et al. 2004). M. citrina differs from its two closest congeners (M. arborea and M. strasseri) in several morphological features (e.g. larger seeds and fruits, larger and tenderer leaves; González-Andrés et al. 1999; Sobrino et al. 2000), which tend to limit its dispersal capacity and make it more susceptible to natural enemies. It tolerates high levels of aridity, enabling it to survive in coastal rocky slopes with little to no soil (Pérez-Bañón et al. 2003; Juan et al. 2004; Crespo et al. 2007). It also shows a high regrowth potential, no vegetative reproduction and abundant seed production, although the seed set of small-islet populations may be lowered by a deficit of pollinators (Pérez-Bañón et al. 2003).
Similar to the rest of Balearic Islands, Cabrera has a long history of anthropogenically facilitated biological invasions (first human settlements date from 2500 to 2300 BC; Alcover et al. 2001; Calvo, Guerrero & Salvà 2002). The invasion complex currently established on Cabrera Island includes three potential consumers of the study plant: a herbivore (the European rabbit, Oryctolagus cuniculus L.), a granivore (the house mouse, Mus musculus L.) and a plant- and seed-eating omnivore (the black rat, Rattus rattus L., Amengual 2000). Feral goats and sheep also roamed freely on the island until their elimination in the 1930s–1940s (Alcover 1993) and 1991 (Frontera et al. 2000), respectively. The exotic scale insect Icerya purchasi, reported to attack M. citrina at Columbretes Islands (Juan 2002), has not been documented on Cabrera Island. Although we are not aware of any published information on M. citrina's native granivores and herbivores, these probably include molluscs (mainly snails, including both native and exotic species; Altaba 1993), insects (mainly granivorous ants, which may also function as dispersal vectors, although the seeds lack elaiosomes and their large size, 20 mg on average, largely hampers transportation; Reader & Beisner 1991; Reader 1993) and granivorous birds (e.g. chaffinch, brambling, greenfinch and goldfinch; Traveset 1993).
Our study initially focused on the potential impact of the two most widespread herbivores, the European rabbit and the black rat (excluding the house mouse, assuming its distribution to be restricted to the immediate vicinity of human settlements). Black rats have permanent populations on the two largest islands/islets of the Cabrera Archipelago (e.g. Cabrera Island and Conillera Islet), while smaller islets experience sporadic colonizations followed by local extinctions (e.g. Illa de Ses Bledes, one of the islets hosting a M. citrina population; Palmer & Pons 2001). The National Park authority has applied successful rat eradication programmes in several small islets, while in the larger ones, eradication has not succeeded (Conillera Islet; Moreno 2009) or been attempted to date (Cabrera; J. Amengual, pers. comm.). European rabbits, native to the Iberian Peninsula, were introduced into the Balearic Islands in the Talaiotic age (Alcover 1993). They show stable populations on Cabrera Island and Conillera Islet. To date, there has been no attempt to evaluate its impact on the native biota and/or control or eradicate its populations.
All experiments used four sets of exclosures installed in the study area to evaluate the impact of rodents (mainly rats, but also mice) and rabbits on native vegetation. Each of the four sets was installed at a different site, two of them in open garrigue and two in grassland vegetation (See Fig. S1). Habitat types (open garrigue and grassland) were selected to span the range of colonization sites available to M. citrina and variability in abundance of exotic herbivores.
Each set of exclosures included nine plots of 10 × 10 m, arranged to form a 3 × 3 square grid, with one of three possible treatments randomly assigned to each plot (thus making a total of three replicates per treatment per site): (i) Rabbit exclosure: plot fenced with wire mesh (1 m height, 5 × 10 cm mesh size buried at 30–50 cm depth), to prevent the entrance of rabbits while allowing easy access to rats. Although the youngest rabbits (c. 1 month of age) could probably cross this fence, systematic mark counts and baited live-trapping (inside and around the plots) carried out during the 3 years previous to this experiment indicated that entrances were extremely rare (a single mark recorded in a single exclosure plot and no trapping, while marks and captures around the plots were abundant throughout the entire period). (ii) Rat + rabbit exclosure: plot fenced with wire mesh (1 m height, 2 × 2 cm mesh size buried at 30–50 cm depth) to prevent the entrance of rabbits and rats, combined with bromadiolone-based toxic baits (Notrac Blox®) applied inside the plot (within baiting devices tailored to prevent the entrance of native vertebrates). Although occasional entrances were revealed by evidence of bait consumption, 2-year monitoring of baited traps following the installation of the exclosures indicated no rat presence (no trapping after the first campaign, in which three individuals were removed). (iii) Control: open plot.
At each site, abundance of both exotic herbivores (rabbits and rats) had been assessed by means of capture–mark–recapture during the 2 years that preceded the experiments (four sessions per year, Santamaría et al. 2007). Despite the strong seasonal and moderate interannual variation, differences between plots remained comparable across the 2-year period; hence, the data provide a good indication of spatial differences in herbivore abundances. As previously mentioned, we did not estimate mice abundance in our experimental plots because, according to local managers and practitioners (M. McMinn and J.A. Amengual, pers. comm.), they were only present nearby human settlements. Two isolated captures at the rat traps suggested, however, that they were able to enter the grassland plots (hence, we set an extra treatment to account for their effect on seed predation).
To evaluate the impact of the different (native and exotic) seed predators on seed survival, we offered groups of 20 seeds in plastic-mesh trays (10 × 10 cm) placed directly on the ground. At each site, we placed three trays (one per each of three treatments) in each of the rat + rabbit exclosures, as follows: (i) Control (open) trays were placed outside the exclosure, but in its immediate vicinity (< 5 m distance), and were accessible to all native (birds + ants) and exotic (rats + mice) seed predators. (ii) Rat exclosure trays were placed inside the exclosure; hence, seed were only accessible to native predators (birds + ants) and, potentially, mice. (iii) Rat + natives exclosures were placed to evaluate the impact of mice (as we suspected, they could enter the rat exclosure but did not reach the bromadiolone bite due to their small size); trays were fixed on two wooden sticks (2 cm diameter by 20 cm long) impregnated with ant-repelling, contact insecticide (to exclude ants) and placed inside a wire-mesh cage (24 mm mesh size) that excluded birds but allowed mice to enter. Sticks held the trays at c. 2 cm from the soil, preventing the access of ants (unless walking on them, from which they were repelled by the insecticide) but allowing easy access to the mice (without having to contact the sticks or encountering them before accessing the tray's content). All trays (4 sites × 3 treatments × 3 replicates = 36 trays and 720 seeds) were set in October 2005 and monitored monthly until September 2006 (except for January, March and April, when weather prevented access to the island). At each visit, we recorded the number of intact, partially consumed, damaged (attacked by fungi) and germinated seeds and considered as predated the sum of those absent and partially consumed. Damaged seeds represented only a small portion (< 4%) of the seeds offered and were excluded from the analyses.
Seed germination and seedling survival
To evaluate the effect of environmental variation (across the different habitats) on seed germination and the subsequent effects of native and exotic herbivores on seedling survival (i.e. of seedling predation), we sowed 20 M. citrina seeds within each experimental plot (i.e. 20 seeds × 3 treatments × 3 replicates × 4 sites, making a total of 720 seeds; see above for a description of the different treatments in the plots). Seeds were sown in October 2005 at 1 cm depth in the soil and spaced regularly over a 20 × 30 cm grid (with specific positions assigned at random). A wire mesh nailed to the floor (2 cm mesh size) provided protection against seed predators and was used to localize each individual seed (by means of a coordinate system). Germination and seedling survival were monitored at monthly intervals from November 2005 until September 2006 and revised again in subsequent years (April and May 2007 and March 2009) to account for long-term survival and delayed germination. However, the low germination rates observed prevented the analysis of seedling survival at the different treatments (see 'Results').
To evaluate the effect of consumption by the herbivores on seedling survival, we germinated M. citrina seeds within plastic trays filled with soil in a greenhouse and offered them over a 4-week period in the experimental plots (5–10 seedlings per tray, one tray per plot, placed at a randomly chosen place within the plot's surface, making a total of 282 seedlings). Trays were buried to level their surfaces with the surrounding ground, and their substrate was maintained wet using a diffusive irrigation system (a 100-mL bottle filled with water, buried next to the tray and connected to its substrate through a cotton wick that transferred water by capillarity) to prevent seedlings from withering during the 4-week period. Trays were installed in April 2007 and monitored once after 28 days. Predated seedlings showed clear signs of herbivory, such as complete defoliation or consumption of stems at ground level.
Herbivory on saplings
To evaluate the effect of consumption by two of the herbivores (rats and rabbits) on sapling survival, we offered M. citrina saplings (2-year juveniles with woody stems and 60–170 cm height) in the experimental plots (two saplings per plot, placed at randomly chosen places within the plot's surface, making a total of 72 saplings). Saplings had been cultivated at a nearby site (the greenery of the National Park's Botanical Garden) and were planted by burying their pots at ground level. They were installed in April 2007 and monitored 35 days later to score sapling survival and herbivory damage (proportion of biomass removed, using a semi-quantitative scale).
Effect of simulated herbivory on reproductive success
Simulated herbivory was applied to 4-year-old adult individuals grown in a common-garden setting at Sóller Botanical Garden (Mallorca). In 2004, 68 2-year-old individuals grown in individual pots from seeds collected at the Cabrera Archipelago were randomly interspersed and planted in the common-garden plot. Two years later, we measured the basal diameter of each individual (as a surrogate of plant size), grouped the individuals by size (17 groups of 4 individuals) and randomly assigned one individual from each group to one of the four simulated-herbivory treatments: 25%, 50% and 75% of canopy (leaves + branches) biomass removed, and a control (no clipping). Biomass removal treatments were chosen to span the complete range of herbivore pressure reported for the species (i.e. from little to no herbivory on small islets, to high to complete consumption by rabbits in the Columbretes Islands; Pérez-Bañón et al. 2003). To ensure a homogeneous removal of canopy biomass across the plant, we removed complete branches from the base to the tip of the plant, starting from a randomly chosen branch among the four basal ones, and alternating removed and nonremoved branches as required by the corresponding treatment (i.e. 25% removal = one branch of each consecutive four, 50% removal = two branches of each consecutive four, etc.). Removed biomass was weighed and used to estimate total plant biomass (based on the fraction of biomass removed), which showed a high correlation with the plant's basal diameter (r = 0.77, P < 0.005; control plants not included).
Simulated-herbivory treatments were applied at the beginning of the growth season (December 2006), and plants were monitored throughout the flowering and fruiting period (February-April 2007) to estimate reproductive effort and success (fruit and seed set). For this purpose, we recorded weekly the number of flower buds, closed flowers, unvisited open flowers, visited flowers and fruits in three branches per individual plant (marked before the onset of flowering, to avoid biases towards more productive branches). Unvisited and visited flowers were distinguished visually, since M. citrina flowers have an explosive tripping mechanism (i.e. the sexual column is released from the two-petal keel when a visiting insect presses the corolla tube for the first time, striking the stigma against it) that prevents pollination of unvisited flowers (McGregor 1976). On the last visit, we collected 30 (randomly selected) fruits per plant and counted and weighed their seeds to estimate seed set.
All analyses were carried out using Generalized Linear Mixed Models (GLMM; procedure GLIMMIX) in SAS v.9 (SAS Institute 2000). Instead of pre-selecting a given error distribution and link function, we fitted all available error distributions and link functions and selected the one that minimized the residuals' dispersion and provided a better fit (based on the AIC score). All models included treatment and habitat (garrigue vs. grassland) as fixed factors and site and plot as random factors. Post hoc comparisons were subjected to sequential Bonferroni correction (Holm 1979).
Data were fitted to GLMM for seed survival (by the end of the experiment) and for life expectancy of predated seeds (number of days from the beginning of the experiment to the predation of each seed, therefore describing predation rate) with binomial and normal error distributions and logit and identity link, respectively. Both models included an additional random factor to account for the covariance of the seeds belonging to the same tray.
Seed germination and seedling survival: Differences in seed germination between habitat types were analysed using a Fisher's exact test using SPSS v. 16.0 (SPSS Inc. 2001) (after pooling all seeds across replicate plots due to the small number of germinated seeds). Seedling survival could not be analysed as a result of small sample size (low germination frequency).
The survival of seedlings (offered in trays) was also fitted with a binomial error distribution and a probit link. We included tray as a random factor to account for the dependency among seedlings in the same tray.
Herbivory on saplings
All saplings attacked by herbivores suffered great biomass losses and died; hence, the only measure of herbivore damage analysed was sapling mortality. One of the treatments had no variance (all juveniles in the rat + rabbit exclosure survived), which prevented the adjustment of parametric models to assess the effect of the three treatments on sapling survival. Instead, we used a 2 × 3 Fisher's exact test (Joosse 2011). Following the detection of significant differences between treatments, we performed 3 2 × 2 Fisher's exact tests to obtain pairwise contrasts between treatments and adjusted the results using sequential Bonferroni correction. In addition, we evaluated the effect of sapling size on survival by fitting, only to data from the control and rabbit exclosure, a GLMM with plant size (number of leaves per plant) as a continuous covariate, treatment and habitat (garrigue vs. grassland) as fixed factors, site and plot as random factors, a binomial error distribution and a logit link.
The effect of simulated herbivory on reproductive effort (flower and fruit production per branch) and success (fruit and seed set, seed weight) was analysed by means of GLMMs with treatment as a fixed factor, basal diameter as a continuous covariate and plant as random factor. A normal error distribution and log link function was used for seed weight, a binomial error distribution and probit link function for fruit set, and a negative-binomial error distribution and log link function for flower production, fruit production and seed set.
Based on naive estimates (catch per unit effort, i.e. per trap and trapping nigh; CPUE hereafter) obtained by a previous capture–mark–recapture study (Santamaría et al. 2007; average across all sampling sessions), we estimated the relationship between rat and rabbit abundances at each of the four sites and the mortality of M. citrina seeds, seedlings and saplings. Mortality estimates caused by rats and rabbits assumed additivity of effects (i.e. they were based on the difference between the mortalities recorded at the corresponding pair of exclosure treatments: rabbit = control – rabbit exclosure, rat = rabbit exclosure – rat + rabbit exclosure) and were regressed on herbivore abundance. Given the low sample size, these relationships are primarily shown to illustrate the effect of herbivore abundance upon the different life stages of our focal plant species and should not be taken to provide robust statistical relationships.
Only 16% of the 680 seeds were still present in the trays at the end of the experiment. 27 individual seeds were infected by mould, and therefore survival to seed predation was estimated on a total of 653 seeds.
Seed survival (by the end of the experiment) varied significantly between habitats (F1,2 = 18.69, P < 0.05), but not between treatments (treatment effect: F2,4 = 0.98, P > 0.10, habitat*treatment effect: F2,4 = 0.55, P > 0.10). Survival probability was much lower in grassland (0.89% on average) than in garrigue (24.97%).
In contrast, life expectancy (i.e. time to predation) of predated seeds varied significantly among treatments (F2,4 = 14.03, P > 0.1), but not between habitats (habitat effect: F1,2 = 0.91, P > 0.1; habitat*treatment effect: F2,4 = 2.81, P > 0.1). In rat and rat + natives exclosures, life expectancy was comparable (140 ± 16 and 162 ± 15 days, respectively) and doubled that in open trays (59 ± 16 days; Fig 1).
Seed Germination and Seedling Survival
One year after planting seeds in the field sites, only 5% of the seeds had germinated. Seeds sowed in garrigue had more successful germination than those sowed in grassland (34 vs. 2 germinated seeds; Fisher's exact test: P < 0.0001). Seedling survival after 1 year was almost zero, mainly due to the effect of summer drought (but also to predation: see below); however, three of the seedlings found in 2006–07 within rat + rabbit exclusions grew to middle-sized saplings by 2011.
Seedling survival varied significantly among treatments (F2,4 = 8.46, P < 0.05; Fig. 2). Survival increased around threefold from the open control to the rabbit and rat + rabbit exclosures (30%, 83% and 98% survival, respectively). Assuming additive effects, rabbits, rats and native herbivores (insects and molluscs) were responsible for 53%, 15% and 2% of seedling mortality, respectively.
All saplings placed in rat + rabbit exclosures survived after 1 month, and we found no evidence of predation by native (invertebrate) herbivores. 2 × 3 Fisher's exact test indicated a significant effect of treatment on sapling survival (P < 0.005). Subsequent 2 × 2 tests between pairs of treatments indicated that the only significant difference was the 33% increase in survival from the controls to the rat + rabbit exclosures (P < 0.005). Survival in the rabbit exclosures was intermediate and did not differ significantly from that in the other two treatments (P > 0.1 in both cases; Fig. 3, upper panel). Sapling survival increased marginally with the number of leaves (a surrogate of sapling size; F1,41 = 3.69, P = 0.0668), but this relationship saturated for moderately large saplings (>300 leaves; Fig. 3 lower panel).
Relationship with Herbivore Abundance
Estimates of the separate effect of each herbivore on M. citrina (assuming additivity of effects) only scaled to herbivore abundance for the sapling stage (Fig. 4). While both rat and rabbit increased their respective impacts on saplings proportionally to their abundance (R2 = 0.99, P < 0.05 and R2 = 0.95, P < 0.05, respectively), rat abundance was not significantly correlated with either seed (R2 = 0.02, P > 0.1) or seedling (R2 = 0.02, P > 0.1) mortality caused by rats. Similarly, rabbit abundance was not correlated with seedling mortality caused by rabbits (R2 = 0.25, P > 0.1). Correlations between the combined effect of both herbivores (control – rat + rabbit exclosure) and their cumulative abundance (number of rats + number of rabbits) showed comparable results: significant for sapling mortality (R2 = 0.98, P < 0.05) but not for seed or seedling mortality (R2 = 0.17, P > 0.1 and R2 = 0.01, P > 0.1; respectively).
Effect of Simulated Herbivory on Reproductive Success
We estimated the effect of (simulated) herbivory on the reproductive success of adults with the assumption that, for adults, herbivory by rats and rabbits will not cause plant death (i.e. it will not represent a predation event, as it did for saplings and seedlings). The smallest basal diameter of the individuals at the onset of the experiment (1.13 cm) would result in 88% survival in the sapling-predation experiment (based on the results shown in Fig. 3, and the relationship between the basal diameter and the number of leaves of the plants used in the experiment) and, as most plants used in the experiment had a diameter larger than 1.98 cm (77%), they would have shown a 98% survival in the sapling-predation experiment.
Flowering peaked in the third week of May, with a display of 9.52 ± 0.82 flowers per branch (average ± SE). Almost every flower was pollinated during anthesis, indicating a lack of pollinator limitation in our common-garden set-up. While simulated herbivory did not significantly affect flower production (F3,63 = 0.37, P > 0.5), seed set (number of seeds per fruit: F3,29 = 0.20, P > 0.5) or seed weight (F3,28 = 0.56, P > 0.5), it resulted in a significant decrease in fruit set (54% less ripe fruits per flower; F3,52 = 4.79, P < 0.01), which was comparable across all three herbivory treatments (i.e. 25%, 50% and 75% removal; Fig. 5). Plant size (basal diameter) was strongly correlated with flower production (F1,129 = 22.28, P < 0.001), but not with fruit set (F1,52 = 3.45, P > 0.5), seed set (F1,698 = 1.38, P > 0.1) or seed weight (F1,28 = 0.09, P > 0.5).
The different herbivores had complementary impacts on the different life stages of M. citrina, which may result in important cumulative impacts at the population level. While native herbivores appeared to pose little to no risk for M.citrina, rodents (black rats and mice) and rabbits severely reduced plant performance through their complementary impacts on its different life stages. These included independent effects on different life stages (seed predation by rodents, seedling predation by rabbits) that resulted in multiplicative increases in plant mortality, and concurrent effects on the same life stage (sapling predation). Seed predation varied between habitats, while seedling and sapling herbivory effects did not. Differences in germination and seedling establishment could modulate herbivore impacts on plant recruitment (e.g. delayed germination could increase the impact of seed predation but also reduce the impact of seedling herbivory, through reduced appearance effects). In addition, the simulated-herbivory experiment showed that the reduction in flower production per plant (proportional to biomass removal, as flower production per branch did not differ between clipped and unclipped plants) is compounded by a strong decrease in fruit set, already observed at the lowest (25%) biomass removal rate.
Predation by rats reduced the life expectancy of Medicago citrina seeds (Fig. 1), although after 1 year seed survival was comparable in open and excluded plots. This suggests the existence of different predation rates between rats and other granivores (largely mice, but also native granivores), whereby rats consume the available seeds more quickly. However, sustained predation by other granivores may suffice to deplete them within a single growth season (limiting the build-up of a dormant seedbank). At any rate, early predation by rats will severely reduce seed germination in the subsequent autumn, as it decreased seed life expectancy (59 days) well below the minimum period between seed production and germination (120–150 days, from April-May to October-November). All together, rodents had a large impact on seed survival, with overall survival rates (16%) much lower than the average, particularly for a species with large seeds (20 mg on average; Reader 1993) than the figures for the rest of life stages studied. These results support Maron & Crone's (2006) contention that, contrary to previous suggestions, granivore impact on plant populations is greater than that of other herbivores. In fact, rodents have been reported to be key determinants of seed fate and seedbank dynamics across most bioclimatic regions (Heithaus 1981; Brown & Heske 1990; Hulme 1994, 1997) and a consumption of only 40% of the seeds produced may already lead to population declines (Bricker, Pearson & Maron 2010).
By contrast, seedling survival seemed to be unaffected by the presence of rats, while rabbits were responsible for a high mortality rates at this early stage. This pattern was different still for sapling survival, where rabbits had a greater impact than rats, but the combined effect of both herbivores was required to detect significant effects (which involved a 40% reduction in sapling survival). In summary, we detected two important bottlenecks at early stages of plant ontogeny; one caused by rodents (through seed predation) and one caused by rabbits (seedling survival), indicating that their combined action can impose a severe constraint on plant reproduction. This early risk is later moderated for small saplings and disappears for large ones (> 300 leaves). Admittedly, demonstrating the impact of herbivores on plant individuals does not lead to a complete understanding of their effect on plant population dynamics (Maron & Crone 2006), yet having focused on lethal effects, we can expect them to translate to reduced population growth rates.
Once the plant is large enough to overcome this high mortality risk, herbivory effects are likely to be dominated by sublethal impacts, which might still lead to changes at population level (Maron & Crone 2006). We focused on the effect on plant reproduction and found that flower production per branch was unaffected by biomass loss during the early part of the growing season – that is, plants neither expressed a trade-off between reproduction and vegetative regrowth (e.g. Edwards 1985) nor over-compensated for vegetative-biomass losses with an increased reproductive effort (Obeso & Grubb 1993; Agrawal 1998). Hence, flower production per plant decreased in direct proportion to the amount of biomass removed. Additionally, fruit set decreased after defoliation and, because seed set (per fruit) did not change; the reduction in seed production per plant can be expected to double the proportion of biomass consumed by herbivores. Similar results have commonly been reported which show a negative effect of herbivory either on fruit (McCarthy & Quinn 1992; Tong, Lee & Morton 2003; Hladun & Adler 2009) or seed production (Lee & Bazzaz 1980; Islam & Crawley 1983; Marquis 1984; Lehtilä & Syrjanen 1995). This effect was not linked to a decrease in pollination efficiency (flower visitation did not differ among treatments); hence, the reduction in fruit output could be the consequence of a change in resource availability or distribution (largely mediated by fruit abortion; data not shown) as a response to herbivory. It is important to note, however, that the response of our experimental plants could be conditioned by the favourable environment in which they are grown (in terms of soil quality and water supply), and it may change considerably under field conditions. In addition, because we only monitored the plant's response during the same growth season in which we defoliated them, long-term responses could compensate or exacerbate (particularly, if herbivory takes place every year) those reported here.
The effects of multiple herbivores on plant performance and population dynamics have received increasing attention in the past two decades, and the existence of common interactions between multiple herbivore species and complementary and synergistic effects of these different species is now widely acknowledged (Hulme 1996; Juenger & Bergelson 1998; Gómez & Zamora 2000; Hufbauer & Root 2002). Our study supports this evidence by providing an example of a synergistic effect of two different herbivores acting on different life stages of the plant cycle. In our system, the combination of rat and rabbit effects on early stages and the effect of herbivory on adult reproduction lead to an important decrease in plant recruitment in natural settings. Herbivore abundances that would cause moderate damage to the plant foliage (only 25% of foliage removal) would already translate into a reduction of 60% of fruit and seed production. This effect would be exacerbated by subsequent seed, seedling and sapling predation. Hence, for the moderate rat and rabbit abundances of Cabrera, herbivory would severely reduce the number of individuals surviving until reproduction.
Surprisingly, few of the herbivory effects studied differed between habitats or replicate sites (with the exception of seed predation, which decreased by 25% from grassland to garrigue, and sapling predation, which increased proportionally to rat and rabbit abundances; Fig. 4). These differences could be further distinctive as a result of the differing rates of seed germination and seedling survival between habitats, which are likely to modulate the effect of the detected herbivory bottlenecks on plant demography (Bonsall, van der Meijden & Crawley 2003). For example, low germination in grassland will exacerbate the consequences of stronger seed predation by increasing the period at which seeds are vulnerable to granivores. Both habitats do also differ sharply on resource availability (soil quality and water availability), a factor that has been reported to determine herbivory compensation (Hawkes & Sullivan 2001). Woody species, in particular, tend to show stronger regrowth responses to herbivory in resource-poor habitats (Hawkes & Sullivan 2001), which could help garrigue plants to better compensate for herbivory damage. Despite the lack of relationship found for seed and seedling predation, density dependence of herbivore impacts on saplings agree with previous literature (e.g. Pearson & Callaway 2008) and indicates that herbivore abundance surveys may provide useful information for managers interested in mitigating their impacts on vegetation.
These results stress the importance of considering the combined effects of different herbivores and assessing their consecutive effects on the various phases of the plant's life cycle (flower and seed production, as well as seed, seedling and sapling survival). In established populations with large numbers of adults and recruits, such a study would be best approached through the assessment of cumulative survival probabilities, from seeds to adults (Jordano & Herrera 1995; Shea & Kelly 1998; Lázaro, Traveset & Castillo 2006). This possibility is precluded by the characteristics of our study system: an endangered plant putatively driven to extinction by the impact of the invasion complex to which the studied herbivores belong. Instead, we resorted to a series of independent, short-term experiments that served the triple purpose of illustrating the complementary impacts of the different herbivores on different life stages, identifying the most sensitive ones and evaluating the potential of exclosure plots as reintroduction sites.
Here, we give a final note to the relative effects of M.citrina's exotic and native herbivores and the potential reintroduction of the species on Cabrera Island. We found negligible effects of native herbivores and granivores, as compared to those of small exotic mammals. Although measured impacts on plant performance do not necessarily translate to changes in population dynamics, our results suggest that the combined impact of herbivores may have contributed to the extinction of M. citrina on the largest islands of Cabrera's Archipelago (Rita & Bibiloni 1993; Juan et al. 2004). Furthermore, the bottlenecks imposed by both exotic herbivores would likely hinder future reintroduction attempts (see also Pérez-Bañón et al. 2003; Mestre, González & Del Señor 2010). It is possible that, in M. citrina populations reaching a reasonable proportion of large individuals, the levels of seed/seedling/sapling predation and adult herbivory imposed by (moderate abundances of) rats and rabbits could be tolerated. The evidence presented here suggests, however, that reintroduction efforts are unlikely to succeed in the presence of moderate, and even low, abundances of these herbivores, owing to the small population size and the strong dependence of new populations on the early recruitment phases (seed production, seedling and sapling survival). Reintroduction efforts should address the control of both exotic herbivores or the mitigation of their impacts and also make use of the habitat preferences and transition probabilities identified here. According to these, the open garrigue shrubland is the most adequate habitat for seed germination and provides the highest survival to seed predation. However, it is also the habitat where both herbivores are most abundant and where desiccation risk is probably higher. Measures to prevent or mitigate seedling and sapling predation will therefore be required, be it in the form of rat/mice/rabbit eradication or control or by means of mechanical structures (such as fences or individual exclusions) that provide protection to the earliest life stages.
The Cabrera National Park, provided permits, accommodation, transportation and technical support during fieldwork. The Sóller Botanical Garden provided saplings, common-garden facilities and assistance during the simulated-herbivory experiment. The greenery of the National Park's Botanical Garden kindly provided and cultivated the individuals used in the saplings experiment. We are grateful to J. Amengual, C. Cejudo, C. Celedón, B. Cid, B. Gozalo, P.F. Méndez, Á. Moraña, J.J. Pericás, M. Piazzon, L. Royo and M. Sobral for their assistance in the field. Amanda M. Dorsett revised the English text. During the preparation of the manuscript, ARL was financially supported by the JAEDoc program (Spanish Research Council), cofunded by the European Science Fund. Funding by the National Parks Office of the Spanish Ministry of Environment (HERBIMPACT project, ref. 050/2002) is also gratefully acknowledged.