This study investigated the impacts of mammal community composition on seed predation, germination, and seedling herbivory in Panamanian forests. Experiments were conducted at sites with differing mammal communities: mainland forests with a full mammal community including top predators; Barro Colorado Island lacking the two largest cats; medium islands with rats, agouti, rabbit and paca present; and small islands with only rats. Seed removal, post-removal seed fate, and seedling herbivory were examined for three large-seeded tree species. Results showed seed removal did not differ between forest types, but post-removal seed fate and seedling herbivory varied significantly, with higher predation and herbivory on smaller islands with

1. 941 REPORTS Ecology, Vol. 78, No. 3
Ecology, 78(3), 1997, pp. 941–946
1997 by the Ecological Society of America
DOES MAMMAL COMMUNITY COMPOSITION CONTROL RECRUITMENT
IN NEOTROPICAL FORESTS? EVIDENCE FROM PANAMA
NIGEL M. ASQUITH,1,4 S. JOSEPH WRIGHT,2 AND MARIA J. CLAUSS3
1 Department of Biological Sciences, University of Illinois, 845 W. Taylor Street, Chicago, Illinois 60607 USA
2Smithsonian Tropical Research Institute, Apartado 2072, Balboa, Ancon, Republic of Panama
3Department of Ecology and Evolutionary Biology, University of Arizona, Tucson, Arizona 85721 USA
Abstract. Patterns of seed predation, germination, and seedling herbivory were in-
vestigated in Panamanian forests. We hypothesized that seed and seedling survival would
vary with differences in mammal community composition. We tested this hypothesis at five
sites in mainland forests adjacent to Gatun Lake, full terrestrial mammalian granivore/
herbivore communities with top predators; at five sites on Barro Colorado Island (BCI),
also a full mammalian granivore/herbivore community but without the two largest cats; at
one site each on five medium-sized islands, with rats, agouti, rabbit, and paca present; and
on five small islands that support rats only. Experiments were replicated for Dipteryx
panamensis, Gustavia superba, and Virola nobilis, all of which have large seeds. To assess
seed removal, seeds were placed in wire exclosure cages and nearby outside the cages.
There was no difference in removal rates between forest types, with almost all unprotected
seeds removed at all sites. To assess post-removal seed fate, seeds of Gustavia and Virola
were attached to threads and placed on the forest floor. All threaded seeds were victims of
predation on small islands, whereas 34, 43, and 77% of threaded seeds were dispersed and
buried on BCI, medium islands, and the mainland, respectively. To assess seedling her-
bivory, half of the wire exclosure cages were removed after germination, and seedling
survival was assessed after 13–14 mo. Protection from mammals increased seedling sur-
vivorship by more than sixfold on the smallest islands, by threefold on the medium islands,
by twofold on the mainland, and by less than twofold on BCI.
The absence of the two largest cats and the exclusion of poachers from BCI was
associated with lower seedling herbivory and higher seed predation than observed on the
mainland. In contrast, extreme mammal defaunation on the small and medium islands had
large and consistent effects on seedling recruitment, including increased seed predation
and increased seedling herbivory relative to sites with more intact mammal communities.
Key words: Dipteryx panamensis; exclosure experiments; forest fragmentation; Gustavia superba;
herbivory; islands; mammals; neotropical forests; Proechimys semispinosus; seedling recruitment;
seed predation; Virola nobilis.
Ecologists have long debated the role of top-down portant impacts on vegetation structure and species di-
and bottom-up forces in structuring biological com- versity (Brown and Heske 1990).
munities (Hairston et al. 1960, Paine 1966). Although Terborgh (1992) hypothesized that a top-down in-
top-down effects have more often been observed in teraction may structure vegetation dynamics in neo-
aquatic systems (e.g., Carpenter et al. 1985), recent tropical forests. Predators such as jaguar and puma ap-
studies have found equally strong top-down forces in pear to regulate the population densities of medium-
terrestrial systems (e.g., McLaren and Peterson 1994). sized mammals, which in turn may limit seedling
In some circumstances, top-down forces can have im- recruitment. Changes in mammal community compo-
sition may therefore have profound effects on tree spe-
cies diversity (Terborgh 1992). Much of the evidence
Manuscript received 12 February 1996; revised 7 July relating to this hypothesis comes from islands in Gatun
1996; accepted 13 July 1996; final version received 8 August Lake (Panama), isolated in 1913, which now lack com-
1996. ponents of a full mammal community.
4 Present address: Department of Zoology, Duke Univer-
sity, Box 90325, Durham, North Carolina 27708-0325 USA,
Barro Colorado Island (BCI), an island with neither
and Smithsonian Tropical Research Institute, Apartado 2072, jaguar nor puma resident, appears to support very high
Balboa, Ancon, Republic of Panama. densities of some medium-sized mammals (Glanz

2. 942 REPORTS Ecology, Vol. 78, No. 3
TABLE 1. The terrestrial mammalian granivore/herbivore community and predators.
Medium Small
Mainland BCI islands islands
Spiny rat (Proechimys semispinosus) x x x x
Agouti (Dasyprocta punctata) x x x
Rabbit (Sylvilagus brasiliensis) x x x
Paca (Agouti paca) x x x
Collared peccary (Pecari tajacu) x x
Red-tailed squirrel (Sciurus granatensis) x x
White-tailed deer (Odocoileus virginianus) x x
Red brocket deer (Mazama americana) x x
Tapir (Tapirus bairdii) x x
Jaguarundi (Herpailurus yaguarundi) x x
Ocelot (Leopardus pardalis) x x
Margay (Leopardus wiedii) ? x
Jaguar (Panthera onca)† x
Puma (Puma concolor) ?
Sources: Glanz 1982, 1990, 1991, Adler and Seamon 1991.
† Jaguars are known to visit BCI frequently (S. J. Wright, personal observations).
1982, 1991; but see Wright et al. 1994 for an alternative tographs suggest that many Gatun Lake islands have
view). DeSteven and Putz (1984) and Sork (1987) been forested since isolation (Leigh et al. 1993). An-
found that seed predation was much higher and seed- nual rainfall on BCI (9 10 N, 79 51 W) is 2600 mm,
ling survival much lower on BCI than on the adjacent with a 4-mo dry season beginning in December.
mainland (but cf. Terborgh and Wright 1994). Other We conducted experiments in four forest types
islands in Gatun Lake are too small ( 1.2 ha) to sup- around Gatun Lake: five sites in mainland forests ad-
port any terrestrial mammals and are now dominated jacent to the lake, five sites on BCI (1600 ha), and one
by a restricted subset of tree species (Leigh et al. 1993). site each on five medium-sized islands (15–80 ha) and
Leigh et al. (1993) postulate that two factors have five small islands ( 2 ha). All experimental sites were
been of critical importance in the reduction of tree spe- within 1 km of BCI, except for four of the medium
cies diversity from small islands in Gatun Lake. Seeds islands, which were 4–5 km distant. Further details of
that require burial by agouti (Dasyprocta punctata) to the forest types are given in Leigh et al. (1982, 1993),
escape insect predation are unable to germinate on Wright (1985), and Adler and Seamon (1991). Our
small islands (see also Forget 1990). The wind-influ- small islands were islands 43, 44, 46, 48, and 49 in
enced island microclimate then favors the seedlings of Adler and Seamon (1991). All granivorous and her-
other species, such as the large-seeded Protium pan- bivorous mammals 1 kg in body mass found on the
amense (Leigh et al. 1993). As an alternative mecha- mainland are also present on BCI. The terrestrial mam-
nism, Terborgh (1992) suggests that as medium-sized malian granivore/herbivore community in each of the
mammals are absent from small islands, large-seeded different forest types is shown in Table 1. Mainland
tree species can escape predation and are subsequently forests and four medium islands are subject to limited
able to outcompete other species. poaching, while BCI, one medium island, and the small
In this paper we compare rates of seed predation, islands are not (Glanz 1991, S. J. Wright, personal
germination, and seedling herbivory in Panamanian observation).
forests with different mammal communities. We ex-
perimentally examine factors that influence early mor- Tree species
tality of three large-seeded tree species and test two We investigated three native tree species. Dipteryx
specific predictions: (1) predation of large seeds is low
panamensis (Pitt.) Rec. & Mell (Leguminosae) is a
on small islands, allowing escape of large-seeded tree
common, large (40–50 m) canopy tree. Its drupes have
species; (2) germination rates are low on small islands.
a thin exocarp, and a hard thick-walled endocarp that
We then discuss the mechanisms that appear to have
encloses a single, large (4 cm) seed (DeSteven and Putz
caused the loss of tree species diversity from small
1984). Gustavia superba (H.B.K.) Berg. (Lecythida-
Gatun Lake islands, and evaluate the hypothesis of
ceae) is a small (typical dbh 10–15 cm) pioneer tree,
mammalian control of neotropical forest dynamics
common in the understory of secondary forest on BCI.
(Terborgh 1992).
Its large fruits (150–600 g) contain 5–50 seeds, each
Study sites weighing 3–15 g (Sork 1987). Virola nobilis A. C.
The islands and peninsulas surrounding Gatun Lake Smith (Myristicaceae) is a dioecious canopy tree. A
are covered by tropical moist forest. Early aerial pho- fibrous capsule dehisces to expose a gray seed (2

3. April 1997 REPORTS 943
1.5 cm) enclosed by a red 1 mm thick aril (Howe 1993). after the first experiment was initiated, and within 10
Primary dispersal of V. nobilis seeds is by birds and m of the original sites. After 7 d, threads were relocated
mammals (Howe 1993), D. panamensis seeds by bats by searching within a 5 m radius of the site. If the
(Bonnacorso et al. 1980), and G. superba fruits by thread was relocated with the seed still attached and it
gravity (Sork 1987). Seeds of all three species are con- had been moved and/or buried, secondary dispersal was
sumed by mammalian granivores and may be second- indicated. If the thread alone was recovered, we as-
arily dispersed and buried by scatterhoarding agouti sumed that the seed had been predated. We tested this
(Dasyprocta punctata) (Forget and Milleron 1991, For- assumption by attaching similar threads to Astrocaryum
get 1992, 1993). standleyanum seeds, which unlike G. superba and V.
nobilis have an inedible endocarp. Almost 80% (23) of
Methods
the threads from which the A. standleyanum seed had
Five experimental sites were randomly located in been removed (n 29) were within a few centimeters
each of the four forest types. Each site was within 50 of an opened, predated endocarp. The fates of the G.
m of the lake because more distant sites were unavail- superba and V. nobilis seeds were pooled by forest type
able on small islands. During the 1st wk of March 1993, and subjected to three-way log-linear contingency anal-
we placed arrays of 10 intact and apparently viable ripe ysis, with the main effects being tree species, forest
D. panamensis seeds in four wire exclosure cages at type, and seed fate. Seeds that had not been removed
each site. The cages were constructed of galvanized and still remained in situ, and those for which neither
wirecloth and were 1 m tall by 0.7 m diameter. They seed nor thread were recovered, were excluded from
were open at the top and their bases were staked firmly the analysis.
to the soil to prevent passage of small mammals un- To assess germination success in the absence of
derneath. Although climbing mammals and large mam- mammals, after 90% of seeds in the growing-house had
mals such as deer had access through the open tops, germinated, we recorded the proportion of seeds that
Terborgh and Wright (1994) showed the loss of D. pan- had germinated in each exclosure, pooled values for
amensis seeds from similar exclosure cages to be each site and species, and analyzed the arcsine-trans-
10% (see also the results of this study). We thus formed data by ANOVA.
consider seed loss to mammals from within the cages After the seed removal and germination experiments
to be minimal. For each wire exclosure, a correspond- had been completed (June 1993 D. panamensis; Oc-
ing array of 10 seeds was placed unprotected on the tober 1993 G. superba and V. nobilis) we removed half
forest floor. The location of the unprotected array was of the wire exclosure cages at each site. This resulted
marked with a stake, 2–3 m from the caged array. Each in two groups of seedlings protected from mammalian
seed was within 0.3 m of the stake. herbivores, and two unprotected groups at each site.
We repeated the procedure with arrays of eight ripe The difference between seedling mortality inside and
G. superba seeds during the 1st wk of July 1993 and outside the cages provided an indicator of rates of
arrays of eight ripe V. nobilis seeds during the 1st wk mammalian herbivory. The proportions of surviving
of August 1993. The number of seeds in each array seedlings after 13–14 mo were pooled by treatment,
mimicked the densities that might result from primary site, and species and arcsine-transformed before anal-
dispersal away from the parent tree. At the time that ysis by ANOVA. Sites at which no seeds survived to
seeds of each species were set out in the field, 100 the seedling stage were excluded from analysis, as was
seeds were also placed in a screen growing-house on one site where the wire cages had disappeared.
Barro Colorado Island. When 90% of the growing-
house seeds had germinated and seedlings were no lon- Results
ger dependent on endosperm, removal from the exper-
imental arrays was assessed. Seedlings and intact seeds The proportion of intact seeds and seedlings after
remaining were counted in June 1993 (D. panamensis) 90% of growing-house seeds had germinated are pre-
and October 1993 (G. superba and V. nobilis). The sented in Table 2a. While many protected seeds re-
replicate exclosures and controls were pooled for each mained in situ, almost all unprotected seeds were re-
site and species. As many groups showed no variance, moved at all sites (F 865.0, df 1, 96, P 0.001).
the proportions of seeds remaining were ranked before There was no difference in removal rates between forest
analysis by ANOVA (Conover and Iman 1981). types (F 2.41, df 3, 96, P 0.071). However,
To assess the fate of removed seeds, individual G. there were significant differences in the fates of re-
superba and V. nobilis seeds were threaded with 60 cm moved seeds among forest types (G 66.1, df 6, P
long lengths of cotton thread and placed on the forest 0.01) (Table 2b). On small islands there was no
floor (Forget and Milleron 1991). Two arrays of eight evidence that any removed seeds had been dispersed.
seeds of each species were set out at each site, 2 wk On BCI and medium islands, 34% and 43% of the

4. 944 REPORTS Ecology, Vol. 78, No. 3
TABLE 2. Fate of seeds/seedlings.
a) Proportion of intact seeds/seedlings after 90% of growing house seeds germinated (V. nobilis after 3 mo, G. superba and
D. panamensis after 4 mo).
Mainland BCI Medium islands Small islands
Protect- Protect- Protect- Protect-
ed Open ed Open ed Open ed Open
D. panamensis 0.89 0.01 0.74 0.00 0.78 0.01 0.46 0.00
G. superba 0.88 0.02 0.99 0.00 0.98 0.00 0.96 0.16
V. nobilis 0.72 0.00 0.32 0.00 0.69 0.00 0.45 0.00
b) Fate of removed seeds after 1 wk (proportions).
Mainland BCI Medium islands Small islands
Dis- Dis- Dis- Dis-
Eaten persed n Eaten persed n Eaten persed n Eaten persed n
G. superba 0.18 0.82 33 0.70 0.30 46 0.53 0.47 38 1.00 0.00 2
V. nobilis 0.29 0.71 35 0.63 0.37 32 0.62 0.38 45 1.00 0.00 28
c) Fate of seeds protected from mammals (proportion surviving to the seedling stage).
Mainland BCI Medium islands Small islands
D. panamensis 0.64 0.55 0.62 0.23
G. superba 0.79 0.98 0.98 0.90
V. nobilis 0.64 0.23 0.54 0.11
d) Proportion of seedlings surviving to 13–14 mo.
Mainland BCI Medium islands Small islands
Protect- Protect- Protect- Protect-
ed Open ed Open ed Open ed Open
D. panamensis 0.37 0.19 0.30 0.23 0.41 0.13 0.25 0.00
G. superba 0.81 0.41 0.96 0.57 0.98 0.28 1.00 0.15
V. nobilis 0.35 0.00 0.09 0.00 0.76 0.00 0.28 0.00
removed seeds were dispersed before being buried. On P 0.001) (Table 2d). The species effect was signif-
the mainland 77% were dispersed. icant (F 59.39, df 2, 80, P 0.001) but total
There were significant differences in germination survivorship did not differ significantly between forest
success between species (F 57.52, df 2, 48, P types (F 2.46, df 3, 80, P 0.069). The species
0.001) and between forest types (F 9.25, df 3, 48, treatment interaction was significant ( F 8.21, df
P 0.001) (Table 2c), but the species forest type 2, 80, P 0.001), because although moderate num-
interaction was also significant (F 3.73, df 6, 48, bers of unprotected D. panamensis and G. superba
P 0.004). We tested the following post-hoc com- seedlings survived, no unprotected V. nobilis did. The
parisons within each species using a Bonferroni-ad- forest type treatment interaction was also significant
justed significance level of P 0.006: small islands (F 5.02, df 3, 80, P 0.003). Protection from
vs. other sites; medium islands vs. BCI and mainland mammalian herbivores increased D. panamensis and
sites; and BCI vs. mainland sites. Most G. superba G. superba seedling survivorship by less than twofold
germinated in all forest types (no post-hoc comparisons on BCI, by twofold on the mainland, by threefold on
significantly different). D. panamensis had moderate the medium islands and by more than sixfold on the
germination at mainland sites and on BCI and medium smallest islands.
islands, and very low germination on small islands
(small islands vs. other sites, F 14.0, df 1, 48, P Discussion
0.001; other comparisons not significant). V. nobilis Germination of protected seeds and 1-yr survivor-
had very low germination on both small islands and ship of protected seedlings were similar for medium
BCI (small islands vs. other sites, F 15.2, df 1, islands, BCI, and the mainland (Table 2c and d). Phys-
48, P 0.001; medium islands vs. BCI and mainland ical differences among these three forest types were
sites, not significant; and BCI vs. mainland sites, F unimportant in the early stages of tree regeneration. In
12.14, df 1, 48, P 0.001). contrast, the germination of protected seeds of both D.
Protection from mammalian herbivores increased panamensis and V. nobilis was reduced on the small
survivorship in all forest types (F 95.6, df 1, 80, islands (Table 2c). This effect was limited to the es-

5. April 1997 REPORTS 945
tablishment phase when seedlings were still dependent that both the magnitude and the direction of such ef-
on seed reserves. Protected seedlings of all three spe- fects are contingent on the biota involved.
cies survived equally well in all forest types (Table The results of more limited changes in mammal com-
2d). Low rates of germination on the small islands is munity composition are even less predictable. Al-
probably a consequence of their desiccating physical though the composition of the terrestrial mammalian
environment. Microclimate thus appears to play a crit- herbivore/granivore community is the same on BCI and
ical role in regulating small-island tree recruitment the adjacent mainland, predation pressure is different.
(Leigh et al. 1993). Jaguar and puma are resident only on the mainland,
Mammals influenced the regeneration of the three and in addition, poachers are wholly excluded from
tree species in all four forest types. Unlike Sork (1987), BCI. The consequences for tree regeneration varied
who found rates of seed removal (of G. superba) to with life history stage, survivorship of unprotected
differ between BCI and the nearby Gigante peninsula, seeds being greater on the mainland (Table 2b), and
we found no such differences between BCI and main- 1-yr survivorship of unprotected seedlings being great-
land sites. Rather, seed removal was virtually complete er on BCI (Table 2d). In an earlier comparison of BCI
in all forests, despite the radical differences in mammal and Manu, Peru, mammal exclosures had dramatic and
community composition (Table 2a). quantitatively similar effects on seedling recruitment
On small islands, all of the removed seeds were pre- and survivorship, despite apparent differences in the
dated (Table 2a and b), and mortality due to mammalian population densities of most mammal species (Ter-
seedling herbivory was also extraordinarily high (Table borgh and Wright 1994). In contrast to the effects of
2d). The one resident terrestrial mammal species ( Proe- the complete extirpation of mammals, the relatively
chimys semispinosus) maintained extremely high den- subtle differences in mammal population densities at-
sities on these islands (up to 58 individuals/ha on a tributable to cats and limited poaching have no clear
nearby 1.8-ha island; Adler, in press). The highest lev- effects on plant regeneration.
els of seed and seedling predation occurred on the However, the loss of certain critical mammal species
smallest islands in the absence of mammals 1 kg body can cause important top-down changes in plant re-
mass. cruitment (Brown and Heske 1990, McLaren and Pe-
The second largest impact of mammals occurred on terson 1994). Almost complete loss of terrestrial mam-
the medium islands. Here, 57% of V. nobilis and G. mals from Panamanian forests released the one re-
superba seeds removed by terrestrial mammals were maining mammal species (Proechimys semispinosus)
predated (Table 2b), and 1-yr survivorship of unpro- from competition and predation, allowing it to prevent
tected seedlings of D. panamensis and G. superba was regeneration of three large-seeded tree species. An in-
lower than for BCI and the mainland (Table 2d). As termediate reduction in mammal community diversity
the mammal communities in these forests were sim- also reduced recruitment of these tree species. In a
plified, both seed predation and the loss of seedlings Mexican forest, similar reductions in the mammal com-
to browsers increased. munity had equally significant but opposite effects on
Predatory mammals and browsing mammals 1 kg seedling recruitment. Poachers and habitat fragmen-
body mass have also been largely extirpated at Los tation are currently affecting mammal communities
Tuxtlas, Mexico (Dirzo and Miranda 1991). The effects throughout the Neotropics (Redford 1992). These hu-
on tree regeneration have, however, been nearly the man-induced changes in top-down processes may well
opposite of those observed on small and medium Gatun have profound effects on patterns of forest regenera-
Lake islands. Early regeneration has been enhanced at tion.
Los Tuxtlas, and dense monospecific carpets of seed- ACKNOWLEDGMENTS
lings cover the forest floor (Dirzo and Miranda 1991).
We thank Henry Howe, Joel Brown, Bruce Patterson, Fran-
Pit vipers (Bothrops spp.) occur at extremely high den- ces White, and David Mertz for assistance and advice, and
sities at Los Tuxtlas and consume small rodents (Perez- John Terborgh, Greg Adler, Pierre-Michel Forget and two
Higarda 1978; H. Greene, personal communication). reviewers for improving the manuscript. N. Asquith was sup-
This may check small-rodent populations and permit ported by a Smithsonian Graduate Fellowship.
the development of dense seedling carpets. Interest- LITERATURE CITED
ingly, similar monospecific seedling stands occasion- Adler, G. H. In press. The island syndrome in isolated pop-
ally cover the forest floor of small Gatun Lake islands. ulations of a tropical forest rodent. Oecologia.
Such occurrences appear to coincide with periods of Adler, G. H., and J. O. Seamon. 1991. Distribution and abun-
low spiny rat densities (G. H. Adler, personal com- dance of a tropical rodent, the spiny rat, on islands in Pan-
ama. Journal of Tropical Ecology 7:349–360.
munication). It is clear that extreme mammal defaun- Bonaccorso, F. J., W. E. Glanz, and C. M. Sandford. 1980.
ation alters seedling regeneration. The contrast between Feeding assemblages of mammals at fruiting Dipteryx pan-
Los Tuxtlas and the small Gatun Lake islands indicates amensis (Papilionaceae) trees in Panama: seed predation,

6. 946 REPORTS Ecology, Vol. 78, No. 3
dispersal and parasitism. Revista de Biologia Tropical rests. Yale University Press, New Haven, Connecticut,
28(1):61–72. USA.
Brown, J. H., and E. J. Heske 1990. Control of a desert- . 1991. Mammalian densities at protected versus hunt-
grassland transition by a keystone rodent guild. Science ed sites in Central Panama. Pages 163–173 in J. G. Rob-
250:1705–1708. inson and K. H. Redford, editors. Neotropical wildlife use
Carpenter, S. R., J. F. Kitchell, and J. R. Hodgson. 1985. and conservation. University of Chicago Press, Chicago,
Cascading trophic interactions and lake productivity. Illinois, USA.
BioScience 35:634–639. Hairston, N. G., F. E. Smith, and L. B. Slobodkin. 1960.
Conover, W. J., and R. L. Iman. 1981. Rank transformations Community structure, population control and competition.
as a bridge between parametric and nonparametric statis- American Naturalist 94:421–425.
tics. American Statistician 35:124–129. Howe, H. F. 1993. Aspects of variation in a neotropical seed
DeSteven, D., and F. E. Putz. 1984. Impact of mammals on dispersal system. Vegetatio 107:149–162.
early recruitment of a tropical canopy tree, Dipteryx pan- Leigh, E. G. Jr., A. S. Rand, and D. M. Windsor. 1982. The
amensis, in Panama. Oikos 43:207–216. ecology of a tropical forest. seasonal rhythms and long-
Dirzo, R., and A. Miranda. 1991. Altered patterns of her- term changes. Smithsonian Institution Press, Washington,
bivory and diversity in the forest understory: a case study D.C., USA.
of the possible consequences of contemporary defaunation. Leigh, E. G. Jr., S. J. Wright, E. A. Herre, and F. E. Putz.
Pages 273–287 in P. W. Price, T. M. Lewinsohn, G. W. 1993. The decline of tree diversity on newly isolated trop-
Fernandes, and W. W. Benson, editors. Plant-animal inter- ical islands: a test of a null hypothesis and some impli-
actions: evolutionary ecology in tropical and temperate cations. Evolutionary Ecology 7:76–102.
regions. John Wiley & Sons, New York, New York, USA. McLaren B. E., and R. O. Peterson. 1994. Wolves, moose
and tree rings on Isle Royale. Science 266:1555–1558.
Forget, P.-M. 1990. Seed-dispersal of Vouacapoua ameri-
Paine, R. T. 1966. Food web complexity and species diver-
cana (Caesalpiniaceae) by caviomorph rodents in French
sity. American Naturalist 100:65–75.
Guiana. Journal of Tropical Ecology 6:459–468.
Perez-Higarda, G. 1978. Reptiles and amphibians from the
. 1992. Seed removal and seed fate in Gustavia su-
Estacion Biologia Tropical ‘‘Los Tuxtlas’’ (UNAM), Ve-
perba (Lecythidiaceae). Biotropica 24:408–414. racruz, Mexico. Bulletin of the Maryland Herpetological
. 1993. Post-dispersal predation and scatterhoarding Society 14:67–74.
of Dipteryx panamensis (Papilionaceae) seeds by rodents Redford, K. H. 1992. The empty forest. BioScience 42:412–
in Panama. Oecologia 94:255–261. 422.
Forget, P-M., and T. Milleron. 1991. Evidence for secondary Sork, V. L. 1987. Effects of predation and light on seedling
seed dispersal by rodents in Panama. Oecologia 87:596– establishment in Gustavia superba. Ecology 68:1341–1350.
599. Terborgh, J. 1992. Maintenance of diversity in tropical for-
Glanz, W. E. 1982. The terrestrial mammal fauna of Barro ests. Biotropica 24(2b):283–292.
Colorado Island: censuses and long-term changes. Pages Terborgh, J., and S. J. Wright. 1994. Effects of mammalian
455–468 in E. G. Leigh Jr., A. S. Rand, and D. M. Windsor, herbivores on seedling recruitment and survivorship in two
editors. The ecology of a tropical forest: seasonal rhythms neotropical forests. Ecology 75:1829–1833.
and long-term changes. Smithsonian Institution Press, Wright, S. J. 1985. How isolation affects rates of turnover
Washington, D.C., USA. of species on islands. Oikos 44:331–340.
. 1990. Neotropical mammal densities: how unusual Wright, S. J., M. E. Gompper, and B. DeLeon. 1994. Are
is the community on Barro Colorado Island, Panama? Pages large predators keystone species in neotropical forests? The
287–313 in A. H. Gentry, editor. Four neotropical rainfo- evidence from Barro Colorado Island. Oikos 71:279–294.