Seed predation by rodents results in directed dispersal of viable seed fragments of an endangered desert shrub

. Seed predation and seed dispersal are important ecological processes with antagonistic effects on plant recruitment. In the southern edge of the Atacama Desert in Chile, Myrcianthes coquimbensis is an endangered, large-seeded, vertebrate-dispersed shrub that in the present-day has no known dispersers. Native rodents hoard and eat the seeds of M. coquimbensis but leave viable seed fragments at the hoarding sites; soil interspaces within rock outcrops where seedlings recruit. Here we examined whether rodents act as effective dispersers of M. coquimbensis by discarding viable seed fragments in sites suitable for recruitment. We simulated different levels of endosperm loss to determine if seedlings could develop from seed fragments. We assessed how frequently rodents discarded fragments, and the probability that these fragments produced seedlings. Finally, we compared emergence and seedling survival at the hoarding sites and in two other habitats where seeds arrive to evaluate the suitability of the hoarding sites. Seeds of M. coquimbensis developed seedlings even after 87 % of their storage tissue was removed. Rodents left seed fragments in more than 50 % of the trials; almost 60 % of the discarded fragments produced seedlings. Seedlings did not emerge from open ground habitats, and emergence was higher under M. coquimbensis shrubs than in rock habitats. Survival of two-year-old seedlings was higher in rock habitats than under conspecific adult shrubs. Our results suggest that rodents may play a dual role in the recruitment dynamics of M. coquimbensis , acting simultaneously as seed predators and effective dispersers. Therefore, though seed predators impose costs, their net effect on plant fitness in this system—where dispersers of large-seeded species have been lost—is likely positive.


INTRODUCTION
Seed predation and seed dispersal are important ecological processes with antagonistic effects on plant population dynamics. Seed predation leads to a reduction of seed densities, and can limit plant recruitment and abundance (e.g., Orrock et al. 2006, Ferreira et al. 2011, Bricker and Maron 2012. Conversely, seed dispersal is a process that can increase seed survival and, ultimately promote plant establishment (e.g., Fragoso et al. 2003, Blendinger et al. 2011. Therefore, the loss of dispersers can hinder the regeneration of plants that depend on them (Loayza and Knight 2010, Calviño-Cancela et al. 2012) and may eventually increase their chance of becoming extinct (McConkey andDrake 2002, Babweteera et al. 2007).
Nonetheless, there are some examples of plant species with large-seeded fruits that have presumably lost their dispersers, but have been able to persist without them (Janzen andMartin 1982, Guimarães et al. 2008). The mechanism proposed to explain the persistence of these species is that seed dispersal by scatter-hoarding rodents may have substituted the services provided by the original dispersers (Guimarães et al. 2008). Scatter-hoarding rodents bury intact seeds in shallow caches, and those seeds that are not retrieved by the animals, are protected from invertebrate predation, and can germinate and establish (Briggs et al. 2009. Thus, seed dispersal by scatter-hoarding rodents results from forgotten cached or re-cached intact seeds. Generally, however, rodents will not disperse intact or undamaged seeds; instead, they remove the seeds and either (1) eat them completely or (2) partially consume them, discarding uneaten seed fragments (e.g., Steele et al. 1993, Perea et al. 2011, Shiels and Drake 2011. The fates of these seed fragments has been rarely addressed in the literature, but the empirical data available shows that although partial consumption can result in seed death or decreased seed survival (Vallejo-Marín et al. 2006, Shiels andDrake 2011), discarded seed fragments can also germinate and establish (e.g., Perea et al. 2011). Therefore, effective seed dispersal by rodents may also occur, even if the seed has been partially consumed.
In the southern limit of the Atacama Desert, Myrcianthes coquimbensis is a fleshy-fruited, largeseeded species that has no present-day dispersers, but whose fruit traits strongly suggest adaptations for vertebrate dispersal. Along its area of distribution, natural seedling recruitment of M. coquimbensis is extremely low (García-Guzmán et al. 2012), and limited to soil spaces found within large rock outcrops (rock habitats hereon). Seeds, however, cannot reach these habitats unless dispersed there by an animal vector. Rodents usually hoard seeds to these habitats, and either gnaw and destroy them completely, or eat part of the seed leaving fragments of different sizes at the hoarding sites; they do not cache the seeds and few, if any, are left intact at the sites (A. P. Loayza, D. E. Carvajal, P. García-Guzmán, personal observations). Nonetheless, seedlings occasionally emerge from seed fragments (Fig. 1). Here we tested the hypothesis that by hoarding seeds of M. coquimbensis to rock habitats, rodents act as effective dispersers of this species even after partially destroying the seed. To test this hypothesis we collected four sets of independent empirical data. First, we evaluated whether seeds with large fractions of the endosperm removed could produce seedlings. Second, we quantified how frequently rodents leave seed fragments when feeding. Third, we planted seed fragments discarded by rodents after feeding to assess how often these fragments produced seedlings. Finally, to determine if the rock habitats where rodents hoard the seeds and leave the fragments are suitable sites for plant recruitment, we (1) compared emergence in these habitats with those of two other habitats where seeds can arrive and, (2) followed and compared the fate of naturally occurring seedlings of M. coquimbensis for two years.

Study area and species
We conducted this study in the southern limit of the Atacama Desert, 26 km south of La Serena, Chile (30804 0 S, 71814 0 W; 0-33 m.). The area is dominated by Eulychnia acida, Echinopsis coquimbensis, Copiapoa coquimbana, Oxalis gigantea, Lithraea caustica, Heliotropium stenophyllum, Fuchsia licyodes and Myrcianthes coquimbensis. Shrub canopies cover between 25-30% of the ground; the rest of which is devoid of plants, except after rains when annuals cover the desert floor. The landscape is also characterized by the presence of large granite rock outcrops (i.e., rock habitats); these outcrops and the crevices within them generate shady habitats where soil moisture (x ¼ 14.02%, SD ¼ 7.64) is higher than in open habitats (x ¼ 8.54%, SD ¼ 6.50, t ¼À3.8, df ¼ 93, P , 0.001) (A. P. Loayza, unpublished data). Water in this system comes from occasional rains, as well as from coastal fogs. Mean annual temperature is 14.28C; annual rainfall ranges from 21 to 122 mm (75 6 33, mean 6 SD), and falls predominantly between June and August (CEAZAMET 2004(CEAZAMET -2012. Myrcianthes coquimbensis (Barnéoud) Landrum and Grifo (Myrtaceae) is an endangered, evergreen shrub, endemic to the Elqui Province in Chile. It has an extremely narrow distribution, occupying a coastal strip of 83 km long by approximately 1 km wide (García-Guzmán et al. 2012). M. coquimbensis flowers from October through February, ripe fruits are available from late August through January, with the peak of the fruiting season in October. Fruit production per individual plant is highly variable, ranging from ,10 to .1500 fruits. Ripe fruits are large (2.5 3 3.5 cm), fleshy, red berries that generally contain only one seed (0.8-2.5 cm diameter, weighing on average 2.6 g, but reaching up to 6 g), although a single fruit may occasionally have up to three seeds. Fruits possess a sugary water-rich pulp, and have a low pericarp:seed mass ratio (0.38). Seeds are monoembryonic and recalcitrant (i.e., they are large and photosynthetically active); they can germinate epigeally from fallen fruits and do not persist in the soil seed bank for more than a few weeks before drying out. Seedling emergence in the field depends on the amount of annual rainfall, but it is generally low and usually restricted to rock habitats. After emergence, there is extensive mortality due to summer droughts.

Emergence from seed fragments
We simulated the effect of rodent predation on emergence by experimentally cutting portions of the seed's storage tissue and comparing seedling emergence among different cutting treatments. Specifically, we selected 80 seeds of approximately the same size (x ¼ 2.64, SD ¼ 0.53) and randomly assigned each seed to one of four cutting treatments: (1) intact seeds (no cutting); (2) seeds cut in half; (3) seeds cut into four pieces or; (4) seeds cut into eight pieces. Before and after being cut, each seed was weighed individually in an analytical balance (Scaltec, SBA31) to calculate the percentage of storage tissue removed. Cutting treatments resulted in a reduction of the storage tissue by 31-46% for seeds cut in half, by 64-79% for seeds cut in four, and by 85-87% for seeds cut into eight pieces. In each case we were careful to leave one of the seed fragments with the embryo intact; these fragments were sown in plastic pots, watered twice a week and checked for seedling emergence once a week for five months (August-December).

Rodent trials
To determine if seed fragments discarded by native rodents could germinate and develop seedlings, we captured rodents present at the study site and fed them M. coquimbensis seeds. In November 2012 we conducted a trapping session during two consecutive nights and days using a 5 3 6 trapping grid (25-m station intervals), with two Sherman-style aluminum folding live traps (30 cm 3 11 cm 3 9 cm; Sherman Traps, Tallahassee, Florida, USA) per station (n ¼ 60). Each trap was baited with rolled oats and checked before sunrise and before sunset. We recorded sex, reproductive condition, and body mass of each captured individual, and brought them into the laboratory. Each individual was then caged singly and provided with water, rolled oats, and four whole M. coquimbensis seeds. At the end of the day or night (depending when the individual was fed), we checked the cages to see if seeds had been eaten. When seeds were eaten, all the remaining fragments (with or without an embryo) were collected and sown individually in plastic pots. Pots were watered twice a week and emergence was checked once a week for four months (November-February). All individuals were returned to their capture site after the feeding trial ended.

Seedling emergence and survival
During the peak of the fruiting seasons of 2011 and 2012, we sowed a group of 10 seeds in each of three habitats: (1) rock habitats (where rodent hoards are found); (2) under adult M. coquimbensis shrubs (where the majority of fruits fall); and (3) open ground (where fruits can occasionally arrive by rolling down slopes). Seeds were extracted manually from ripe fruits, and since they are not regularly buried, they were only partially covered with a thin layer of soil (,0.5 cm). Each habitat was replicated 10 times, and in each replicate seeds were protected from predators with 25 3 15 3 10 cm wire cages with a mesh size of ca. 1 cm 2 . To mimic the fate of seeds processed by rodents, emergence was examined once a month for six months and the presence of emerged seedlings recorded in each survey.
To test if survival of the early stages of M. coquimbensis differed between habitats, we monitored the fate of naturally occurring seedlings in permanent plots established in 2011. We established four permanent 25 m 3 25 m plots in each of seven localities across the distribution range of M. coquimbensis (García-Guzmán et al. 2012). Within each plot, all M. coquimbensis plants were individually tagged, classified into stages, and the habitat where they were present was recorded. Seedling fate was followed for all marked seedlings once a year during two years.

Statistical analyses
We examined whether experimental removal of the storage tissue affected seedling emergence using generalized linear models with Poisson error distributions. We considered cutting treatment as the main factor, and number of emerged seedlings as the response variable; significance was tested using a chi-square likelihood ratio test. Additionally, we evaluated if the removal of seed storage tissue affected temporal patterns of seedling emergence with Kaplan-Meir survival analysis, where the fraction of emerged seedlings in each cutting treatment is analyzed as a function of time. We assessed whether the proportion of emerged seedlings from discarded fragments differed among rodent species with a chi-square. We used generalized linear models, with quasi-Poisson (for the 2011 data) or Poisson (for the 2012 data) error distributions to determine if the probability of seedling emergence differed among habitats. For both years, habitat was the main factor, and number of emerged seedlings the response variable. We examined habitat-specific differences in survival of naturally occurring seedlings in the field with a proportions test. Finally, we explored early recruitment dynamics of M. coquimbensis by estimating habitat-specific transition probabilities (TPs), calculated as the mean number of individuals completing a stage divided by the number of individuals entering that stage (Rey and Alcántara 2000). We then calculated the cumulative probability of early recruitment for each habitat as the product of the individual TPs. All statistical analyses were conducted using the v www.esajournals.org R statistical environment, version 3.1.1 (R Development Core Team 2013).

RESULTS
Cutting treatment had no effect on either the number of seedlings emerged (v 2 ¼ 0.49, df ¼ 78, P ¼ 0.48) or the temporal pattern of seedling emergence (v 2 ¼ 2.00, df ¼ 3, P ¼ 0.57). Consequently, M. coquimbensis seedlings are able to emerge from seed fragments that are up to 87% smaller than the original seed, as long as the embryo is unharmed.
We captured a total of 24 individuals of the three rodent species: Octogon degus (n ¼ 1), Abrothrix olivaceus (n ¼ 13), and Phyllotis darwini (n ¼ 10). On average, A. olivaceus and P. darwini left seed fragments on ca. 70% of the trials, which suggests that these species commonly discard some portions of a seed. Of 54 seed fragments collected from A. olivaceus and P. darwini during the experiment, 59% (n ¼ 32) developed a seedling; there were no differences in the proportion of seedlings produced by seed fragments discarded by either species (v 2 ¼ À0.02, df ¼ 1, P ¼ 0.89). In the feeding trial with the only individual captured of O. degus, we collected three seed remnants; of those two produced a seedling.
Only 4% (Fig. 2). Natural recruitment of M. coquimbensis in the field is extremely low. In 2011, only 20 of ca. 1200 plants marked within the permanent plots were seedlings (,0.02% of the population). Of these seedlings, nine were under shrubs and 11 in rock habitats; no seedlings were recorded in open ground sites. The proportion of seedlings that survived to one-year saplings did not differ between rock habitats and under M. coquimbensis shrubs (Z ¼ À1.82, P ¼ 0.07). However, the proportion of seedlings that survived to two-year saplings was higher in rock habitats than under adult M. coquimbensis shrubs (Z ¼ À3.30, P , 0.001); in fact, no seedlings survived after one year under conspecific plants. Thus, our results suggest that although emergence is higher under M. coquimbensis, recruitment is ultimately limited to rock habitats (Table 1).

DISCUSSION
We show that rodents are effective dispersers v www.esajournals.org of M. coquimbensis seeds, even after partially destroying them. By discarding viable seed fragments at the hoarding sites, rodents are effectively transporting embryos to suitable sites for recruitment. Because M. coquimbensis has no present-day dispersers, rodents are acting as substitute dispersers of this endangered plant. However, unlike other studies, in which effective seed dispersal by rodents is ultimately the result of forgotten cached or re-cached intact seeds, our results provide empirical evidence of a system where the almost complete destruction of a seed via seed consumption can, nonetheless, result in effective dispersal.
Seeds of M. coquimbensis have the ability to develop roots and seedlings from seed fragments up to 87% smaller than the original seed. The ability to germinate from partially destroyed seeds has been reported for several large-seeded species (e.g., Dalling et al.1997, Joshi et al. 2006, Vallejo-Marín et al. 2006, Pérez et al. 2008, Teixeira and Barbedo 2012. This ability is likely retained because larger seeds have greater quantities of nutritional tissue, which allows them to withstand different levels of endosperm loss (Dalling et al. 1997, Mack 1998). If so, then it is possible that the majority of M. coquimbensis seeds that germinate after being partly consumed, are those in the larger size range. If seed size is heritable, this could eventually result in a shift towards shrubs producing larger seeds.
In our study, we did not quantify the percentage of discarded fragments in which the embryo was unharmed; therefore, we assume that fragments that produced seedlings were those where the damage was located in non-embryonic tissue. Avoidance of the seed parts that contain the embryo has been found in other studies (e.g., Perea et al. 2011), and may be related to higher tannin contents in these parts (Steele et al. 1998). The presence of the embryo, however, is not always a requirement for a seed fragment to produce a seedling. For example, among the Myrtaceae, Teixeira and Barbedo (2012) showed that seed fragments of several species of Eugenia (E. brasiliensis, E. cerasiflora, E. involucrata, E. pyriformis, and E. uniflora) could produce a seedling whether or not they had the embryo, which indicates the presence of meristematic tissues in the seeds capable of differentiating and forming new embryos. Similarly, Joshi et al. (2006) found that fragments of Garcinia gummigutta seeds, irrespective of their size, could produce both roots and shoots. Although, in our study we did not test explicitly whether fragments of a single seed could produce different seedlings, we have anecdotal evidence that suggests that M. coquimbensis may also have this capability. If so, this would increase the potential of seed fragments leftover by rodents at the hoarding sites to develop into seedlings.
The fruit traits of M. coquimbensis are strongly indicative of vertebrate dispersal; this is consistent with other Myrcyanthes species in the Neotropics, which are dispersed by birds and/ or monkeys (Barberis et al. 2002, Pizo 2002, Wilms and Kappelle 2006. However, as mentioned previously there are no present-day dispersers of M. coquimbensis. Along its range, some species of birds (e.g., Mimus tenca, Turdus sp.) peck the pulp of ripe M. coquimbensis fruits that are still attached to the plant, but none are large enough to either carry a fruit or swallow a seed. It is possible then, that the animals that originally dispersed its seeds are no longer present. Whether M. coquimbensis represents an anachronic dispersal system is unknown, but there is ample archeological evidence of fauna that went extinct in the area towards the end of the Pleistocene (Moreno et al. 1994, Borrero 2009, Jackson et al. 2011) that may have consumed the fruits of M. coquimbensis. For example, there are fossil records of several herbivores (e.g., Cuvieronius, Megatherium, Macrauchenia, among others), which have been recognized as potential dispersers of other extant large-seeded species (Janzen and Martin 1982). Emergence and seedling survival of M. coquimbensis in the field is very low and contingent not only on the amount of rainfall, but also on the habitat. In our study, emergence was higher under adult conspecific shrubs than in rock habitats. This could be because the soil underneath M. coquimbensis shrubs accumulates large quantities of leaf litter, which can influence germination cues, such as moisture and temperature (Carson and Peterson 1990), as well as release leachates that can be a source of mineral nutrients (Facelli and Pickett 1991). Moreover, the presence of leaf litter can be important to maintain soil moisture when rainfall is low, particularly for species with recalcitrant seeds that cannot form a seed bank (Becerra et al. 2004). However, the spatial pattern of initial recruitment was not maintained through later plant stages, as seedlings did not survive underneath conspecific shrubs. Although leaf litter may also negatively impact seedling survival (Facelli and Pickett 1991), all of the seedlings that were marked under adult M. coquimbensis shrubs eventually dried out, which suggests that competition for water with the adult conspecific was the most likely cause of mortality. In contrast, seedlings survived more in rock habitats, probably because soils in these habitats retain moisture for several months after the winter rains (A. P. Loayza, unpublished data), allowing seedlings the time to develop deeper root systems. Given that our data on seedling survival come from only 20 naturally occurring seedlings in the field, we cannot conclude with certainty that rock habitats are always better recruitment sites for M. coquimbensis as survival may be context-dependent. Nonetheless, because almost 70% of all of the plants marked across the distribution range of the species are associated to rock habitats, it is feasible that recruitment is strongly limited to these habitats.
The role of rodents as both seed predators and dispersers remains largely unexplored. Partial consumption of seeds may have important consequences on plant community structure and population dynamics in several systems (e.g., Pearson and Theimer 2004, Perea et al. 2010, Shiels and Drake 2011. Our results show that rodents play a dual role in the recruitment dynamics of M. coquimbensis, acting simultaneously as seed predators and effective dispersers of predated seeds. The relative importance of each role, however, will probably depend on the environmental context. For example, although rodents discarded fragments in ca. 70% of the feeding trials, in the field in years of low fruit production (and consequently lower seed availability), rodents are unlikely to satiate and may discard fragments less frequently, if at all. Conversely, seed fragments may be discarded more frequently when fruit and seed abundances are high. To fully discern the role of rodents on the recruitment of M. coquimbensis, this hypothesis needs to be tested.
Finally, though seed predators certainly impose costs, their net effect on plant fitness in this system, where the dispersers are absent, is not necessarily negative. Our results highlight that the costs and benefits of an animal-plant interaction are not always evident Drake 2011, Arnan et al. 2012), and in our case a predominantly antagonistic interaction between rodents and M. coquimbensis seeds, may be considered a mutualism in the sense that rodents are currently the only vertebrates that effectively promote recruitment of M. coquimbensis.