316 Jo hn Terborg hand Kennet hFeeley
Consumer populations regulated from the bot-
tom u p(i.e., by food resources) should show
evidence of nutritional stress, a prediction we
were able to test with two species: tortoises and
howler monkeys. Tortoises belonging to a hyper-
dense island population grew at roughly half the
rate of uncrowded mainland tortoises (Aponte
et al. 2003); likewise, hyperdense howler mon-
keys weighed a third less as adults and reproduced
at lower rates than their large landmass counter-
parts (K. Glander unpublished data,Terborghet al.
2001). Howlers confined to a small island lacked
many food resources available to howlers living on
a large island and fed heavily on tree species that
were ignored or consumed only sparingly by the
latter (Orihuela-Lopezet al. 2005).
Possible effects of edge and increased
exposure
“Edge effects,” particularly exposure to desiccat-
ing convectional drafts, are commonly invoked
to explain vegetation changes in other tropical
forest fragment systems (Lauranceet al. 2002,
Laurance 2004). Forest margins around Lago
Guri are exposed to persistent trade winds (Feeley
2004), suggesting the possibility that die-back
and/or recruitment failure along exposed edges
was contributing to the demographic decline of
tree stands. We conducted a series of tests to inves-
tigate this possibility. In comparisons of sapling
plots situated on windward versus leeward slopes
of islands, we found no discernible effect of expo-
sure on the number, mortality, or recruitment of
either small or large saplings. We also failed to
find significant effects on the density, species rich-
ness, or composition of seedling cohorts growing
on the windward versus leeward margins of small
islands. Similarly, we found no effects of proximity
to exposed edges on either the growth or mortality
of trees≥10 cm dbh (Terborghet al. 2006).
DISCUSSION
We tested the top-down hypothesis of Hairston
et al.(1960) by taking advantage of a mega-
experiment initiated in 1986 with the creation
of Lago Guri in the lower Caroní Valley of Bolívar
State, Venezuela. Lago Guri islands smaller than
ca. 15 ha provided predator-free habitats support-
ing hyperabundant densities of species belong-
ing to a non-random selection of functional
groups. Previous predator-exclusion experiments
have produced less dramatic results because the
spatial scale of the experiment was small (usually
onlyafewsquaremeters;reviewedinSchmitzet al.
2000), the exclusion of predators was only partial
(Sinclairet al. 2000), or the duration of exclusion
was brief (usually<1 year; Schmitzet al. 2000).
The Lago Guri experiment largely overcame these
difficulties, providing replicate isolates of natural
habitat over spatial scales ranging from less than
1 ha to more than 100 ha and isolated for more
than a decade.
The presence of water barriers effectively
trapped populations of non-volant animals in
emergent patches of a previously continuous
tropical dry forest. A majority of vertebrate species
apparently disappeared from islands of less than
15 ha within the first few years of isolation, but
some species survived. Among these, predators of
vertebrates were conspicuously absent. Released
from predation and unable to disperse or emi-
grate, trapped populations of several vertebrates
and some invertebrates increased in density by as
much as one to two orders of magnitude.
Our research compared small (≥0.5 ha,<2 ha),
medium (≥4 ha,<15 ha), and large (≥80 ha)
landmasses, the faunas of which conformed to
nearly perfect nested subsets, presumably a con-
sequence of the differing area requirements of
persistent species (Terborgh 1992, Terborghet al.
1997a, Feeley 2003). The loss of some ecologi-
cal functions (predation, seed dispersal), reduction
of others (pollination), and exaggeration of still
others (folivory) resulted in pronounced func-
tional imbalances in the residual animal commu-
nities of small and medium islands, unleashing a
complex trophic cascade with both top-down and
bottom-u pcom ponents.
Hyperabundant consumers marginally in-
creased tree and sapling mortality on S islands
compared with M or L landmasses. However,
recruitment of small saplings on S and M islands
was only about 20% of that on the L landmasses
and insufficient to offset mortality (Terborgh