royalsocietypublishing.org/journal/rspbResearch
Cite this article: Luskin MS, Johnson DJ, Ickes
K, Yao TL, Davies SJ. 2021 Wildlife disturbances
as a source of conspecific negative
density-dependent mortality in tropical trees.
Proc. R. Soc. B 288: 20210001.
https://doi.org/10.1098/rspb.2021.0001Received: 1 January 2021
Accepted: 5 February 2021Subject Category:
Ecology
Subject Areas:
ecology
Keywords:
plant–animal interactions, wild boar
(Sus scrofa) pigs, Malaysia, Southeast Asia,
plant recruitment, tree diversityAuthor for correspondence:
Matthew Scott Luskin
e-mail: m.luskin@uq.edu.auElectronic supplementary material is available
online at https://doi.org/10.6084/m9.figshare.
c.5309956.© 2021 The Author(s) Published by the Royal Society. All rights reserved.Wildlife disturbances as a source of
conspecific negative density-dependent
mortality in tropical trees
Matthew Scott Luskin1,2,3, Daniel J. Johnson4, Kalan Ickes5, Tze Leong Yao6
and Stuart J. Davies2
1School of Biological Sciences, The University of Queensland, Brisbane, Queensland 4072, Australia
2Forest Global Earth Observatory, Smithsonian Tropical Research Institute, Washington, DC, USA
3Asian School of the Environment, Nanyang Technological University, Singapore, Singapore
4School of Forest Resources and Conservation, University of Florida, Gainesville, FL, USA
5Department of Biological Sciences, Clemson University, Clemson, SC, USA
6Forest Research Institute Malaysia (FRIM), Kepong, Selangor Darul Ehsan, Malaysia
MSL, 0000-0002-5236-7096; DJJ, 0000-0002-8585-2143
Large vertebrates are rarely considered important drivers of conspecific
negative density-dependent mortality (CNDD) in plants because they are
generalist consumers. However, disturbances like trampling and nesting
also cause plant mortality, and their impact on plant diversity depends on
the spatial overlap between wildlife habitat preferences and plant species
composition. We studied the impact of native wildlife on a hyperdiverse
tree community in Malaysia. Pigs (Sus scrofa) are abnormally abundant at
the site due to food subsidies in nearby farmland and they construct birthing
nests using hundreds of tree saplings. We tagged 34 950 tree saplings in a
25 ha plot during an initial census and assessed the source mortality by
recovering tree tags from pig nests (n = 1672 pig-induced deaths). At the
stand scale, pigs nested in flat dry habitats, and at the local neighbourhood
scale, they nested within clumps of saplings, both of which are intuitive for
safe and efficient nest building. At the stand scale, flat dry habitats contained
higher sapling densities and higher proportions of common species, so pig
nesting increased the weighted average species evenness across habitats.
At the neighbourhood scale, pig-induced sapling mortality was associated
with higher heterospecific and especially conspecific sapling densities.
Tree species have clumped distributions due to dispersal limitation and
habitat filtering, so pig disturbances in sapling clumps indirectly caused
CNDD. As a result, Pielou species evenness in 400 m2 quadrats increased
105% more in areas with pig-induced deaths than areas without disturb-
ances. Wildlife induced CNDD and this supported tree species evenness,
but they also drove a 62% decline in sapling densities from 1996 to 2010,
which is unsustainable. We suspect pig nesting is an important feature shaping
tree composition throughout the region.1. Introduction
Plant community dynamics are shaped by plant–animal interactions such as
seed dispersal, seed predation, herbivory and trampling [1–3]. The potential
for large (greater than 10 kg) wildlife to increase tropical tree diversity has
been demonstrated by the positive effects of seed dispersal [1,4]. However,
the effect of large wildlife consuming or trampling plants on tree diversity as
rarely been considered because large animals are rarely species-selective
[5–9]. Non-selective wildlife disturbances can still support tree diversity if the
spatial patterns of habitat use disproportionately affect common species and
help limit local or community dominance [10].
Local plant diversity is maintained through a variety of mechanisms, such as
niche differentiation, and conspecific negative density dependence (CNDD).
2
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 20210001CNDD occurs when a species’s mortality rate increases with
its density, and this supports diversity and coexistence by lim-
iting local dominance [11]. CNDD can either arise from
stronger intraspecific competition than interspecific compe-
tition or higher attack rates by plant enemies for common
and/or clumped species [8,11]. For example, the Janzen–Con-
nell hypothesis posits that there are higher attack rates by
specialist natural enemies (mostly host-specific arthropods
and fungi) in offspring clumps nearby parents, thereby indu-
cing CNDD [8]. Small mammals that have non-selective
diets have only occasionally been found to induce CNDD,
and only at the seed stage [12,13]. Larger herbivorous and
omnivorous vertebrates greater than 1 kg (hereafter ‘wildlife’)
with broader generalist diets and higher mobility have not
been considered important sources of CNDD for seeds, seed-
lings or saplings [6,8].
In addition to consumptive effects from herbivory,
wildlife also physically disturb plants, but how this influ-
ences plant diversity has rarely been assessed. In tropical
forests, a study monitoring artificial seedlings found that
59–77% were severely damaged in unhunted forests over a
12-month period, which was 350% higher than in hunted
forests, and suggests that trampling is an important source
of damage for tree seedlings and saplings [14]. Non-
random patterns in wildlife habitat use at the local scale
and across a landscape can result from behaviours such as
the avoidance of steep slopes or areas with high predation
risk (landscapes of fear) [9,15]. Wildlife may also avoid
areas with denser tree saplings for the ease of navigation,
or alternatively, wildlife may prefer dense patches of saplings
for cover from predators or for nesting opportunities. This
effects of wildlife selective habitat use on plant species
depend on the underlying plant community composition.
To better understand the role wildlife may play in shaping
tree composition, we examined (i) if the spatial patterns of
wildlife disturbances overlapped with common tree species
distributions, (ii) if wildlife targeted heterospecific or conspe-
cific sapling aggregations, and (iii) if areas with wildlife
disturbances experienced different trends in species richness
and species evenness.
Tropical forest tree species saplings tend to be spatially
aggregated at the stand scale (i.e. 25 ha [16]) due to abiotic
habitat filtering [17–19] and at the local neighbourhood scale
due to dispersal limitation (i.e. 0.04 ha). Thus, wildlife disturb-
ances in specific habitats could affect a subset of species and
disturbances in sapling clumps may disproportionately kill
locally abundant species (figure 2). If common species are
more likely to die from wildlife than rare species due to their
habitat preferences, this would increase species evenness
of the stand scale, but may not affect coexistence. Within a
habitat, if a species likelihood of dying from wildlife dis-
turbances increases with their local density, then wildlife
disturbances would induce CNDD at the local neighbourhood
scale, and this may contribute to species evenness and
coexistence [20].
Our study focuses on disturbances from native wild boars
(Sus scrofa, hereafter ‘pigs’) in a primary tropical rainforest
called the Pasoh Research Forest in central Peninsular
Malaysia. Pasoh is a large forest reserve with continuous con-
nectivity to other forests and possesses a thriving wildlife
community [21,22]. In Southeast Asia, native pig species
often subsidize their omnivorous diet by crop raiding on
fruits in oil palm plantations, leading to much higher densitiesthan in interior forest areas [22–25]. Pigs can suppress tree
regeneration through seed predation, rooting and seedling
trampling [22,26] but have minor consumptive effects on
larger saplings because they are not browsers (nor are they
seed dispersers [27]). In Peninsular Malaysia, pigs’ most
important impacts on tree saplings come from their nests,
which pregnant pigs build to provide cover during birth and
the first days of nursing (figure 1). A single pig nest can include
hundreds of tree saplings collected from a small area (approx.
0.025 ha) [28]. At Pasoh, nest building by crop-raidingpigswas
identified as the main driver of a 62% decline in tree sapling
densities from 1986 to 2010 [22], but it remains unclear how
pig nesting affects tree diversity.
We examined the patterns of tree sapling mortality due to
pig nesting and linked this to changes in tree diversity over a
5-year period. We were able to identify pig-induced mortality
by directly retrieving previously tagged saplings from pigs
nests. We tested if the spatial patterns of pig nest-induced
mortality differed from other sources of mortality using
individual-based mortality models. We tested if pig nesting
was associated with specific habitats and tested for CNDD
by including sapling neighbourhood heterospecific and con-
specific densities as covariates in the mortality models. We
used three models (different response variables) to compare
patterns of mortality from pigs, mortality from other sources
and if pig-induced mortality patterns differed from other
sources. In separate models, we tested if pig nesting induced
higher species richness and evenness by comparing diversity
changes in areas with and without pig nesting damage. This
approach allowed us to test three hypotheses about how the
spatial distribution of pig disturbances could alter sapling
diversity (detailed in electronic supplementary material and
illustrated in figure 2):
(1) Pigs nest in abiotically defined habitats (dry flat areas). If
these preferred nesting areas overlap with the habitat
associations of more common sapling species, then the
per capita pig-induced death rate may be higher for
common species, supporting species evenness of the
stand (greater than 10 ha).
(2) At the local neighbourhood scale (less than 1 ha), pigs
choose nesting sites where there are many saplings
(clumps) to efficiently construct their nests. Saplings
with higher neighbourhood densities (stems from all het-
erospecifics) will face higher mortality from pigs, thereby
reducing the spatial variation in stem densities, but not
causing CNDD or increasing species evenness.
(3) Since species tend to be spatially aggregated due to abiotic
habitat filtering and dispersal limitation, if pigs target
any sapling clumps for nesting materials, then saplings
with higher conspecific neighbourhood densities may
face higher risk of mortality from pigs. This will cause
‘pig-induced CNDD’ and increase neighbourhood scale
species evenness.
2. Methods
(a) Study site
Our study was conducted at the 1813 ha Pasoh Research Forest
located within the 12 507 ha Pasoh Forest Reserve in Negri
Sembilan, Malaysia (hereafter Pasoh; 2°580470 N, 102°180290 E).
The primary lowland forest has a 40–60 m tall canopy dominated
by trees in the Dipterocarpaceae [29]. Our study made use of the
3
Figure 1. The birthing nest of a native wild boar (S. scrofa) at the Pasoh Research Forest in Peninsular Malaysia. This study documented how nesting disturbances
were associated with the homogenization of tree sapling densities and increases in species evenness. (Online version in colour.)
stand scale saplings that were not saplings that were
(a) (b) (c)
H1—pigs nest in specific habitats used in pig nests later used in pig nests
500
100 100
400
95 95
300
90 90
200
85 85
100
80 80
0.0 2.5 5.0 7.5 0.0 2.5 5.0 7.5
0
slope (°) slope (°)
0 100 200 300 400 500
(d) 2 (e)R  = 0.021, P < 0.001 R2 = 0.023, P < 0.001
neighbourhood scale 30 1000 b
H2—pigs select H3—pigs select a a c
mixed species conspecific 20 a bclumps clumps d 100
c
10
10
0
H4—pigs nest randomly
= pig nest
Figure 2. Locations of pig nests, hypotheses about pig nesting habitat preferences and habitat characteristics of saplings that were found in nests. (a) Locations of
pig nests (black points) in the 25 ha study area, coloured by soil habitat type, with legend matching colours in d and e). In the inset, local sapling neighbourhoods
are shown with tree species represented by different colours. Hypotheses are described in the text and electronic supplementary material, table S1. (b) The 33 278
sapling stems that were not used in pig nests, plotted in elevation-slope habitat niche-space and coloured by soil types (matching colours in d and e). Wet and dry
alluvial soils are flat low-lying areas, laterite-derived and shale-derived soils are found on hills. (c) The 1672 stems that were later recovered in pig nests, showing a
clustering in low elevation, low slope sites with dry alluvial soils (orange colour). (d ) Mean sapling density in different in soil habitats. Letters indicate significant
differences at the p < 0.05 level assessed using LMMs. (e) Average species stand-wide abundance for stems found in different soil habitats, showing dry alluvial soils
contain more common species. (Online version in colour.)
saplings density
(all species, 5 m radius) elevation
swamp alluvial
dry alluvial
laterite-derived
shale-derived
average
stand−wide abundance
of a stem found on this soil elevation
swamp alluvial
dry alluvial
laterite-derived
shale-derived
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 20210001
Table 1. Characteristics of sapling communities in each of the four main soil habitats. Stand-wide species abundance is the total number of saplings 1–1.5 cm 4
dbh in the 25 ha plot for each species, and the table shows the median value among stems located in each habitat. Habitats defined by dry alluvial soils were
flat and unflooded, potentially suitable making them most suitable for pig nest construction.
stems
total found in % stems stand-wide
saplings in pig nests sapling found in species Pielou
habitat (1995– density in pig elevation slope abundance species species
soil habitat (1995) 2000) 1995 (ha−1) nests (mean) (mean) (median) richness evenness
wet alluvial 8034 237 1551 2.950 81.65 1.15 124 533 0.849
dry alluvial 20 207 1261 1687 6.240 82.88 0.95 234 575 0.791
laterite-derived 4288 120 1319 2.799 89.80 5.14 139 439 0.866
shale-derived 2421 54 1441 2.230 90.84 4.25 118 362 0.877
totals 34 950 1672
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 2021000150 ha forest dynamics plot (FDP) established in 1985 [30]. The
plot is in the centre of the reserve, greater than 1.35 km from
any edge. Elevation within the 50 ha plot ranges from 80 to
110 m.a.s.l. The plot includes four distinct edaphic habitats:
hills had well-drained shale-derived soils and laterite-derived
soils, and low-lying and flatter areas ‘dry alluvial soils’ and
‘wet alluvial soils’ [30].
Pasoh has a diverse wildlife community, including several
species of primates, deer, mousedeer, hornbills, civets, as well
tapirs (Tapirus indicus), sun bear (Helarctos malayanus) and a
number of predatory cat species including leopards (Panthera
pardus) [22]. The seed-dispersing community has not appreciably
changed from 1986 to present [22]. However, there is an over-
abundance of native pigs (S. scrofa), whose densities were
estimated from distance sampling at 27–47 individuals km−2
during the study period [23].
(b) Tree sapling censuses
All tree stems greater than or equal to 1 cm dbh were identified,
mapped and measured in a 25 ha area in 1995 and 2000 [29].
In calculating the neighbourhood statistics for all saplings, we
had to exclude the borders of the plot and this left a 22.08 ha
study area where all stems could be associated with all habitat
and neighbourhood variables. We focused on 34 950 stems in
the 1.0–1.5 cm dbh size class, hereafter called saplings, because
pigs primarily use this size class and smaller for their nests
[22,28]. We categorized 1995 saplings into three groups: (i) ‘sur-
vivors’ (saplings with dbh≥ 1 cm in the 1995 and 2000 censuses);
(ii) ‘pig-induced mortality’ (saplings with dbh≥ 1 cm in the 1995
whose stem tags were found in pig nests); and (iii) ‘other deaths’
(saplings in the 1995 that were marked ‘dead’ or ‘missing’ in the
2000 census but not found in pig nests).
(c) Pig nest surveys
We identified pig-induced mortality by excavating the tree tags
from the FDP found within pig nests in a 25 ha study area in
1996 and 1998 (figure 1) [28]. At Pasoh, each pig nest contains
an average of 267 (s.d.: 86) woody saplings with most smaller sap-
lings being uprooted and larger saplings being snapped [28]. The
tags linked saplings to the FDP dataset provided information on
the identities, sizes and original locations, enabling us to assess
how a sapling’s abiotic and biotic characteristics in the initial
census affected their likelihood of being later used for nest build-
ing. Stems whose tags were found in nests were assumed to suffer
‘pig-induced mortality’. As nests were only excavated in two of
the 5 years between censuses, the other deaths include some
stems used for pig nesting in unsampled years (electronicsupplementary material, table S4). Stems that were found in
nests but did not have tags were not included in the study since
their species identity or original location could not be verified.
We were interested in the spatial patterns of pig-induced
sapling mortality more than the actual nesting sites themselves.
These locations are similar since saplings are only taken from
an area of 244 m2, with most stems coming from within a 5m
radius around the nest site [28]. We considered that stems
found in nests were killed, and although many stumps created
by pigs resprout these face much higher mortality than unbroken
stems [28,31].
(d) Tree sapling mortality analyses
We tested our three hypotheses about the impact of pig disturb-
ances on sapling mortality using individual-based generalized
linearmixedmodels (GLMMs). Specifically, we compared the like-
lihood of saplings being found in pig nests (pig-induced deaths)
versus likelihood of dying due to other causes. In the GLMMs,
there was a binary response variable (0, 1) where 0 corresponded
to survival and 1 for death, and we used binomial-family error
distributions. We included explanatory variables (covariates)
to evaluate which factors influenced the odds of mortality
where all continuous variables were centred on the mean and
standardized by 1 s.d. Each parameter estimated by the
GLMM (on the logit scale) represents the change in log odds that
a sapling died in the study period with a 1 s.d. increase in the
corresponding variable.
We constructed three GLMMs to test the influence of the
covariates on the odds of mortality: (GLMM1) pig-induced
deaths versus survivors, wherein we set the binary response to
one for pig-induced deaths and zero for survivors, (GLMM2)
other deaths versus survivors, wherein we set the binary response
one for other deaths and zero for survivors and (GLMM3) pig-
induced deaths versus other deaths of saplings, wherein we set
the binary response to one for pig-induced deaths and zero other
deaths. This lastmodel assessed if the effects frompigs significantly
differed from backgroundmortality. All GLMMmodels contained
the same set of predictor variables so they could be comparedusing
model selection.
Our three hypotheses are tested in each model by assessing
the significance of three key covariates. Firstly, to test the hypoth-
esis that pig mortality varied between habitats, we included the
abiotic habitat type (elevation, slope and soil habitat type) where
each tagged sapling found in a nest was located in all models.
We assessed if local abiotic factors were associated with the
odds of mortality (table 1 and figure 2). For the soil type variable
(descriptions in the Study site section), dry alluvial soil was
included as a categorical variable and compared to the other
5
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 20210001three soil types combined (grouped), because the other three soil
types showed no difference in their effects on mortality and we
used AIC model selection to confirm reducing the levels of this
categorical variable improved model fit. Secondly, to assess if
pigs caused higher mortality in clumps, we included the hetero-
specific sapling abundance within 5 m radius neighbourhoods
(calculated for each sapling) around each sapling in all models.
Thirdly, to assess if pigs caused conspecific negative density
dependence (CNDD), we included the conspecific density
around each sapling in all models. We tested if common species
faced higher mortality by including the species stand-level
abundance (total saplings of each species in the 25 ha study area).
All models included the sapling dbh as a predictor of mor-
tality because previous work has shown pigs prefer smaller
saplings [22]. To address whether certain life-history strategies
were more prone to pig-induced mortality, we included whether
the species was pioneer versus a seral or late-successional
species, as identified by Davies et al. [29], as since these species
are known to have higher recruitment and mortality rates. We
did not consider how species birth, survival or intrinsic growth
rates (r) affected mortality from pigs versus other sources, as
this was not directly related to our hypotheses.
We assessed if there was support for interactions using
model selection by comparing Akaike information criterion
(AIC) values. There was no support for interactions among vari-
ables in the models presented in this study. To assess the
significance of predictors included in the GLMMs, we used Sat-
terthwaite approximations for the degrees of freedom and
report z-values and p-values (lmerTest package in [32]). Degrees
of freedom are not provided in the lme4 package for GLMMs
for conceptual reasons, although it should be noted that our
sample sizes were very large (34 950 stems) and we limited our
predictors to less than 10 variables that included less than 20
total factor levels.
In all GLMMS, we accounted for spatial autocorrelation
among nearby stems by including a random intercept for each
10 × 10 m subplot (n = 2196) to allow for shared responses within
localized areas [19,33]. We choose not to include species-level
random effects in models presented in the main text because we
had no predictions about pigs being selective. Instead, we tested
if pig-driven mortality could be understood relative to their
environment, so we considered how the likelihood of mortality
of stem to be predicted by its abiotic and biotic neighbourhood
and not species (electronic supplementary material, table S1
andboxS1).Wenote there are also technical issueswhenmodelling
species-level mortality with habitat or neighbourhood density,
since all three variables are highly conflated (figure 2; electronic
supplementary material, figures S2 and S5). As a result, the
species-level random effects absorb the variation in mortality
among conspecifics located in different habitats or densities, caus-
ing the models to become uninformative for our main hypotheses.
Therefore, we present results including species-level random
effects in the electronic supplementary material.
(e) Tree diversity in quadrats with and without
pig nesting
We assessed changes in sapling density and diversity between
1995 and 2000within 400 m2 quadrats (20 × 20 m; hereafter ‘quad-
rats’) using linear regressions (LR). The LR response variables
were differences in sapling density, species richness or species
evenness between the censuses, measured for each 400 m2 quadrat
(n = 554). We included a dummy variable (0 or 1) to assess if
changes in density and diversity differed between quadrats with
greater than or equal to 1 pig-induced sapling deaths versus
those with zero pig-induced sapling deaths. The evenness
values were calculated as the Pielou index (J0), which is derived
from the Shannon index and ranges from 0.0 to 1.0, with highervalues representing more even species distributions [34]. A key
aspect of Pielou evenness is that it represents the relative evenness,
which is the proportion of maximum possible evenness given a
particular number of species within the sample, and is unbiased
to differences in sample sizes and species richness [35].
( f ) Determining if rare species were spared from
pig nests
We assessed if the composition of saplings in pig nests was repre-
sentative of the community using the Wilcoxon rank-sum test
comparing the species abundance distributions among species
found within nests differed those for species that were never
found in nests. We evaluated if the actual species richness found
within nest saplings differed from a random sample of the entire
1995 baseline community using the Kolmogorov–Smirnov test.
For the latter, we created a null distribution of expected species
richness among nest saplings—assuming pigs harvested ran-
domly—by simulating 10 000 random subsets of 1672 stems
(number of stems recovered from pig nests) taken from the full
community of 34 950 saplings. All statistical analyses were
conducted using the statistical program R v. 3.5.1.3. Results
(a) Tree sapling mortality
Therewere 34 950 saplings (1.0–1.5 cm dbh) alive in 1995 from
661 species in the 22.08 ha study area. In the 2000 census, 8911
(25.5%) had died or were broken below measurement height
(assumed to have died), of which 1672 (18.76%) were found
in pig nests (pig-induced deaths) in the 1996 and 1998 surveys
and 7239 (81.2%) were not associated with pig nests in these
two years (other deaths).
Pigs nesting removed 6.24% of stems from dry alluvial soil
habitats and 2–3% of stems in each other habitat. There were
higher sapling densities in lower elevations, flatter areas and
in dry alluvial soils (table 1 and figure 2; electronic supplemen-
tarymaterial, figure S2). Dry alluvial soils also containedmore
common species than the other habitat types, as measured by
the average stand-wide abundance of stems found in each soil
type. Pielou species evenness was also 6.9–8.7% lower in allu-
vial habitats than the other three soil types (LM: d.f. = 34 948,
t =−109.4, p < 0.001). Species stand-wide abundancewas posi-
tively correlated with both conspecific and heterospecific
densities, meaning that common species are more likely to
be found in denser areas of saplings (table 1; electronic
supplementary material, figure S2).
(b) Pig-induced mortality patterns differed from other
sources of mortality
Within the 1.0–1.5 dbh size class of saplings, those with smal-
ler dbh were significantly more likely to suffer pig-induced
deaths (GLMM coefficient: z-value =−20.234, p < 0.001 in
GLMM 1) and other deaths (z =−17.744, p < 0.001 in GLMM
2; figure 3; electronic supplementary material, table S5). Pig-
induced sapling deaths showed clear abiotic associations
revealed by positive relationships with topographic position
(slope), elevation and dry alluvial soil habitats (z <−5 and
p < 0.001 for all three), and all three abiotic variables were
stronger predictors (larger parameter magnitude) of pig-
induced deaths than other deaths (z > 4 and p < 0.001 for all
three). The effect of being a pioneer species was not a
nesting mortality differed effect on likelihood of sapling being found in pig nest 6
from other deaths (p < 0.05)* effect on likelihood of sapling dying for other reasons
individual-level stem diameter *
effects higher likelihood of a stem dying -->
and being found in a pig nest
sapling density within 5 m 
biotic *
effects
conspecific density within 5 m
dry alluvial soils *
abiotic 
effects higher elevation *
flatter (less slope) *
species-level pioneer species
effects
species abundance
0.5 1 2
mortality odds ratios
Figure 3. Predictors of sapling mortality due to pigs (orange, GLMM1) versus other sources (GLMM2), assessed with an individual-based mortality models with
random effects to account for spatial pseudoreplication. Odds ratios above 1 indicate increased mortality among the 34 950 saplings. There were 1672 stems found in
pig nests in 1996 or 1998 and 7285 stems that died for other reasons. Values are the GLMM coefficient estimates and error bars show 95% CI. Closed circles show
results with p < 0.05 and open circles show results with p > 0.05 for the GLMM1 and GLMM2. All continuous variables were scaled (mean of 0 and s.d. of 1).
Heterospecific and conspecific densities are measured within 5 m radius neighbourhoods of every stem. Significant differences between the variables were assessed
using GLMM3 (pig-induced deaths versus other deaths) and is noted the asterisk next to the variable name. As an example of how to interpret the terms, the
negative stem diameter coefficients indicate that larger stems had lower chance of dying, and this relationship was significantly stronger for pig-induced deaths than
other deaths (denoted by the asterisk). Full model output is presented in electronic supplementary material, table S4. (Online version in colour.)
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 20210001significant predictor of pig-induced deaths (z = 1.840, p =
0.066), but pioneers did have greater odds of other deaths
(z = 3.148, p = 0.002).
The GLMM1 results (pig-induced deaths versus survivors,
including abiotic habitat variables) revealed that higher
neighbourhood conspecific densities increased the odds
of pig-induced mortality (z = 4.760, p < 0.001). There was
no significant effect of heterospecific density on the odds
of pig-induced mortality (z = 1.165, p = 0.244; electronic
supplementary material, table S5) because heterospecific
neighbourhood densities were correlated with abiotic habitat
variables (electronic supplementary material, figure S2).
A reduced version of GLMM1 that omitted abiotic variables
revealed that neighbourhood heterospecific density increased
the odds of pig-induced mortality (z = 2.899, p = 0.034;
electronic supplementary material, figure S3 and table S6).
A reduced version of GLMM1 including only stem diameter
and species abundance also suggested that species with
higher stand-level abundance faced higher odds of pig-
induced mortality (z = 3.911, p = 0.019). The significance of
these parameters in reduced models but not full models
is explained by conflation of habitat variables (elevation,
slope and soils) with neighbourhood heterospecific densityand species stand-level abundance (figure 2; electronic
supplementary material, figure S2).
Other work as suggested that CNDD may vary with
species abundance [20]. We tested this by including an inter-
action term between neighbourhood conspecific density and
species stand-level abundance GLMM1. We found no sup-
port that CNDD from pigs varied with species stand-level
abundance when controlling habitat covariates (z = 0.098,
p = 0.922) or excluding habitat covariates (z = 0.398, p = 0.690).
The GLMM 2 results (other deaths versus survivors) also
revealed that neighbourhood conspecific density increased
the odds of other deaths (GLMM2: z = 3.051, p = 0.002), albeit
this effect size was significantly smaller than for pig-induced
deaths (GLMM3: z = 2.886, p < 0.001). However, increasing
heterospecific neighbourhood densities reduced odds of mor-
tality (GLMM2: z =−7.967, p < 0.001), indicating dense
neighbourhoods had higher odds of survival, and which
was the opposite direction as its effect on pig-induced mor-
tality (figure 3; electronic supplementary material, table S5).
Stand-level species abundance was negatively associated
with odds of other deaths (GLMM2: z = 3.068, p = 0.002),
suggesting common species suffer lower odds of mortality,
again the opposite effect as for pig-induced mortality. Most
(a) a (b)
45 (c) c 7
a
70 b 0.96 b b quadrats with
40 pig-induced deaths
b
60 0.95 quadrats without
d a
35 pig-induced deaths
c
0.94
50 c c
30
0.93
40 25
1995 2000 1995 2000 1995 2000
(d) 0.8 (e) (f) 0.08
species never found in nests 2000
species found in nests actual
0.6 species
found 0.04
1500 in nests
0.4
1000
0
0.2 500
−0.04
0 0
0.1 0.2 0.5 1 2 320 340 360 380 100 400 2500 10 000
stand-wide species abundance species in random draw of 1626 stems subplot size (m2)
(% of stems in 1995) (10 000 replications)
Figure 4. Impacts on pig nesting on tree sapling compositions and diversity. (a) Sapling density, (b) species richness and (c) Pielou species evenness in 400 m2
quadrats in 1995 and 2000, separated by those quadrats with at least one pig-induced death (n = 293; orange) versus quadrats with no known pig-induced deaths
(n = 261; green). Letters indicate significant differences at the p < 0.05 level, assessed using linear mixed models. Note, this quadrat-level summary of community
composition changes is different than the previous individual-level mortality models (GLMMs 1, 2 and 3). (d ) Species abundance distributions of the 339 species
found in nests (green histogram) versus the 332 that were never found in nests (orange histogram; Kolmogorov–Smirnov test, d = 0.114, p < 0.001). The histogram
area for each group is equal to one. (e) The predicted number of species that would be found in pig nests in our study (grey histogram) versus the actual number
(322 species, orange; Wilcoxon rank-sum test, p = 0.002). The grey histogram is the null expectation produced by random drawing 1672 stems (actual tags recovered
from nests) from the 34 950 possible stems available in the 22.08 ha study area. ( f ) Changes in Pielou’s species evenness, measured in square subplots. We used
LMM to test if the change in diversity differed from zero and all four subplot sizes showed highly significant differences ( p < 0.001 for all). (Online version in
colour.)
relative frequency sapling density (400 m−2)
species richness (400 m−2)
frequency
Pielou species evenness (400 m−2)
change in pielou evenness
(per subplot)
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 20210001covariate estimates significantly differed between pig-induced
deaths and other deaths (GLMM3: p < 0.02 for all comparisons
in electronic supplementary material, table S5 and S6),
showing pigs produced distinct patterns of sapling mortality.
The results were qualitatively similar when we included a
species-level random effect into GLMMs 1, 2 and 3, except
conspecific density was no longer significant (electronic sup-
plementary material, figure S4). However, this change was
only due to strong correlations between the species-level
random effects (differences in mortality rates) and conspecific
densities (electronic supplementary material, figure S5). The
interpretation is that species aggregated in clumps had higher
mortality from pigs, especially in dry alluvial habitat. We also
re-ran the models only including data from dry alluvial habitat
and found similar results showing significantly higher pigmor-
tality for stems in both heterospecific and conspecific clumps
(electronic supplementary material, figure S6).(c) Diversity in quadrats with and without pig nesting
Of the 554 quadrats in the sample area, Pielou evenness
increased in most quadrats, but increased 105% more in the291 quadrats with pig-induced deaths than without pig-
induced deaths (LR: d.f. = 1101, t = 4.02, t =−13.977, p <
0.001; electronic supplementary material, table S7). In the
initial census, quadrats that latter suffered pig-induced
deaths were 19.8% more dense (LR: d.f. = 1104, t =−12.06,
p < 0.001), had 9.1% higher species richness (LR: d.f. = 1104,
t =−11.27, p = 0.001) and had 1.5% lower Pielou evenness
(LR: d.f. = 1104, t = 4.25, p < 0.001) than quadrats without
pig-induced deaths (figure 4; electronic supplementary
material, table S7). Between the two censuses, the sapling den-
sities declined 38.1% more in quadrats with pig-induced
deaths (LR: d.f. = 1101, t =−5.435, p < 0.001). Among the quad-
rats with pig-induced deaths, sapling species richness only
declined 2.72% despite a 27.8% decline in the mean sapling
abundance during the study period.(d) Rare species were spared from pig nests
Of the 661 species of saplings in 1995, only 322 species were
found in nests, statistically fewer than expected by chance
(figure 4e). The species stand-level abundance distributions
differed between nest-species and those species never used
8
royalsocietypublishingin nests (Wilcoxon rank-sum test, p = 0.002; figure 4f ), with
more abundant species suffering higher pig-induced death
rates (Kolmogorov–Smirnov test, d = 0.114, p < 0.001;
figure 4c). The average stand-level sapling abundance for
species suffering pig-induced deaths was 452.3 stems, 17.8%
higher than the average abundance of a surviving stem
(384 stems) and 16.4% higher than other-death saplings
(388.5 stems; LR: d.f. = 34 948, t = 10.197, p < 0.001)..org/journal/rspb Proc. R. Soc. B 288: 202100014. Discussion
Our study illustrates that wildlife can alter tropical tree diver-
sity through stand-scale and neighbourhood-scale habitat
preferences. At our primary rainforest forest in Malaysia,
native wild pigs removed woody and herbaceous stems in
small areas to build nests. Pigs nested in flat but unflooded
areas of the landscape and disproportionately utilized aggre-
gated saplings. At the stand scale, pigs’ habitat selection
overlapped with the habitat associations of more common
tree species, inducing higher tree mortality that increased the
weighted average species evenness across habitats. Rarer
species from smaller habitats may not replace those common
stems killed by pigs due to habitat filtering, so this mechanism
may not support coexistence. At the neighbourhood scale (e.g.
within a single habitat), pig-induced mortality was associated
with heterospecific and conspecific density, the latter inducing
CNDD, which increased species evenness. This source of
CNDD may support coexistence, but the most appropriate
analogy would be similar to a high-intensity but patchy fire,
as opposed to species-specific mortality from host-specific
enemies within the Janzen-Connell framework.
Pigs’ habitat preferences for nest building were logical for
safe and efficient construction, and thus supported our
hypotheses 1–3. First, pigs preferred flat dry habitats, which
would intuitively reduce the likelihood of nests flooding or
toppling. Second, pig-induced sapling mortality was also
associated with local sapling densities, which would be
suitable for optimal forging of nest materials. Tree sapling
of the same species are often aggregated due to dispersal
limitation and fine scale habitat filtering [16–19,36], and we
predicted and found that stems with nearby conspecifics
were more likely to suffer pig disturbances. This indirectly
caused CNDD.
These results are ecologically significant because pig nest-
ing was responsible for at least 18% of sapling mortality at
our site, and quadrats with nest mortality showed 105%
greater increases in species evenness than quadrats without
nest mortality. Pielou’s evenness increased by 0.49% in
20 m neighbourhoods around pig-induced death saplings
over just 5 years, which is equivalent to 9.8% per century.
Pigs also had a small negative effect on species richness,
which slightly declined by 2.72% due to sampling effects
(fewer total stems). While our study focused on the processes
explaining shifts in diversity from 1995 to 2000, another study
from site found sapling species evenness increased consist-
ently between 1986 and 2010 [22], suggesting pigs’ impacts
on diversity are a persistent feature of this forest.
(a) Pigs as tree enemies
Since pigs kill saplings, they can be considered as tree ene-
mies, but their impacts differ from the impacts of generalist
herbivores that can cause CNDD even through relativelynon-selective grazing or browsing [37,38]. Most realizations
of Janzen–Connell hypotheses ascribe CNDD in saplings
clumps nearby parents to growing populations of dispersal-
limited and host-specific arthropods and fungi enemies [8].
Our focal enemies (pigs) are a generalist mobile omnivore
and their populations do not increase in conjunction with a
particular host tree species. Therefore, our case represents a
new variety of enemy-mediated CNDD that is more akin to
localized disturbances—such as a patchy fire, lightning
strikes or landslides—and provides a robust and intuitive
example of how wildlife can induce CNDD that is missing
in the literature.(b) Positive density dependence among non-pig
afflicted saplings
Pig-induced sapling mortality showed opposing patterns
compared to mortality from other sources. This illustrates
wildlife can inflict fundamentally different processes, and
not just reinforce other mortality processes. For example,
both pig-induced mortality and other sources of mortality
showed significant CNDD, so in this case, pigs would
reinforce CNDD processes that might have arisen due to
enemy-mediated Janzen–Connell effects or intraspecific com-
petition [33,39]. However, while pig-induced mortality was
positively associated with heterospecifics densities, other
sources of mortality showed lower mortality in clumps,
indicating heterospecific positive density dependence [40].(c) Results presented are conservative
Our results likely underestimate the effect of pig-induced
increases on tree sapling evenness for three reasons. Firstly,
most stems used in nests are less than 1 cm dbh so, by focus-
ing on larger saplings, we only captured a fraction of the total
pig-induced mortality [22]. Secondly, a considerable number
of tags from pig-induced death saplings were not found in
the pig nests, as some would have fallen off when the stem
was snapped and carried to nests. This was apparent because
the number of saplings greater than 1 cm found in nests was
much higher than the actual number of tags recovered, mean-
ing we underestimated all pig-induced deaths. Thirdly, nest
surveys were not available for 1997, 1999 and 2000 and con-
sequently, saplings labelled ‘other deaths’ probably include
many pig-induced deaths from these years yielding another
source of underestimation.(d) Longer-term implications for tree recruitment
However, we do not suggest that pig nesting disturbances
observed at Pasoh are necessarily representative of other
sites, or that the observed impacts at Pasoh will have lasting
effects on diversity, because pig densities are elevated at our
site and wildly fluctuate between decades [22]. We narrowly
focused on trees in this study, but pig disturbances at Pasoh
have also been shown to increase the relative abundance of
lianas [41], which suppress tree survival and growth [42],
and contributes to spreading invasive species [43]. Finally,
we suggest tree species with more aggregated distributions
may be a higher risk of pig nesting damage, such as ballistic
and wind-dispersed species that are more aggregated than
vertebrate-dispersed species [17].
9
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 20210001(e) Conservation implications
The findings presented here show that elevated pig popu-
lations can affect forest diversity over a relatively short
time, which may have regional and global conservation
repercussions. Ongoing pig-inflicted mortality would even-
tually reduce richness and cause negative cascading
impacts on forest. On the contrary, defaunated forests may
be missing these wildlife disturbances that contribute to the
maintenance of hyperdiverse plant communities.
Our previous work from the site has shown that oil palm
provides food subsidies to crop-raiding pigs and this triggers
ecological cascades affecting tree densities that reach greater
than 1 km into the interior of protected areas [22]. This
study extends this work in showing that subsidy cascades
can also influence CNDD and tree diversity. Currently,
there are approximately 7 Mha of oil palm and approximately
15 Mha of forest in Malaysia, with most forests being frag-
mented and close to oil palm plantations [44,45]. Pigs are
now over-abundant in many forests in Southeast Asia [23–
26] and pigs are invasive in other oil palm-growing regions
[27]. Mitigating the impacts from pigs on forests could be
accomplished through fencing farmland or managing pigs,such as through culling or controlled hunting. The lethal
onslaught of African swine fever across Asia may severely
reduce pig densities in the coming years and provides oppor-
tunities for natural experiments about pigs impacts on
ecosystems [46].Data accessibility. Nesting data are available at https://doi.org/10.6084/
m9.figshare.14068229.v1 and tree census data are available at
https://forestgeo.si.edu/sites/asia/pasoh.
Authors’ contributions. M.S.L. designed the study and analysed the data.
K.I. conducted the pig nest surveys, T.L.Y. and S.J.D. manage the tree
census database, and D.J.J. assisted with the data analysis. M.S.L.,
S.J.D. and D.J.J. wrote the paper. All authors contributed to the
final version.
Competing interests. We declare we have no competing interests.
Funding. Funding support for M.S.L. provided by the Asian School of
the Environment, Nanyang Technological University.
Acknowledgements. We thank FRIM, the Negeri Sembilan Forestry
Department and the Pasoh Research Committee for research per-
mission. We acknowledge ForestGEO and FRIM for maintaining
and supporting the Pasoh 50 ha forest dynamics plot. We thank
M. Detto, M. Visser, L. Hulsman, R. Chisholm, T. Fung, S. J.
DeWalt, S. Hart, A. Letten, M. Yamamichi and three anonymous
reviewers provided input on statistics or reviewed drafts.References1. Wright SJ. 2003 The myriad consequences of
hunting for vertebrates and plants in tropical
forests. Perspect. Plant Ecol. Evol. Syst. 6, 73–86.
(doi:10.1078/1433-8319-00043)
2. Kurten EL. 2013 Cascading effects of
contemporaneous defaunation on tropical forest
communities. Biol. Conserv. 163, 22–32. (doi:10.
1016/j.biocon.2013.04.025)
3. Dirzo R, Young HS, Galetti M, Ceballos G, Isaac NJ,
Collen B. 2014 Defaunation in the Anthropocene.
Science 345, 401–406. (doi:10.1126/science.
1251817)
4. Villar N, Siqueira T, Zipparro V, Farah F,
Schmaedecke G, Hortenci L, Brocardo CR, Jordano P,
Galetti M. 2020 The cryptic regulation of diversity
by functionally complementary large tropical forest
herbivores. J. Ecol. 108, 279–290. (doi:10.1111/
1365-2745.13257)
5. Westoby M. 1974 An analysis of diet selection by
large generalist herbivores. Am. Nat. 108, 290–304.
(doi:10.1086/282908)
6. Owen-Smith RN. 1988 Megaherbivores: the influence
of very large body size on ecology. Cambridge, UK:
Cambridge University Press.
7. Clark CJ, Poulsen JR, Levey DJ. 2012 Vertebrate
herbivory impacts seedling recruitment more than
niche partitioning or density-dependent mortality.
Ecology 93, 554–564. (doi:10.1890/11-0894.1)
8. Terborgh J. 2012 Enemies maintain hyperdiverse
tropical forests. Am. Nat. 179, 303–314. (doi:10.
1086/664183)
9. Terborgh J, Davenport LC, Niangadouma R, Dimoto
E, Mouandza JC, Scholtz O, Jaen MR. 2016
Megafaunal influences on tree recruitment in
African equatorial forests. Ecography 39, 180–186.
(doi:10.1111/ecog.01641)10. Żywiec M, Fedriani JM, Kurek P, Holeksa J. 2019
Non-trophic plant–animal interactions mediate
positive density dependence among conspecific
saplings. Oikos 128, 1041–1050. (doi:10.1111/oik.
06071)
11. Song X, Lim JY, Yang J, Luskin MS. 2021 When do
Janzen–Connell effects matter? A phylogenetic
meta-analysis of conspecific negative distance and
density dependence experiments. Ecol. Lett. 24,
608–620. (doi:10.1111/ele.13665)
12. Paine CE, Beck H. 2007 Seed predation by
neotropical rain forest mammals increases diversity
in seedling recruitment. Ecology 88, 3076–3087.
(doi:10.1890/06-1835.1)
13. Fricke EC, Tewksbury JJ, Rogers HS. 2014 Multiple
natural enemies cause distance-dependent mortality
at the seed-to-seedling transition. Ecol. Lett. 17,
593–598. (doi:10.1111/ele.12261)
14. Rosin C, Poulsen JR, Swamy V, Granados A. 2017
A pantropical assessment of vertebrate physical
damage to forest seedlings and the effects of
defaunation. Glob. Ecol. Conserv. 11, 188–195.
(doi:10.1016/j.gecco.2017.06.001)
15. Gaynor KM, Brown JS, Middleton AD, Power ME,
Brashares JS. 2019 Landscapes of fear: spatial
patterns of risk perception and response. Trends
Ecol. Evol.
16. Condit R et al. 2000 Spatial patterns in the
distribution of tropical tree species. Science 288,
1414–1418. (doi:10.1126/science.288.5470.1414)
17. Seidler TG, Plotkin JB. 2006 Seed dispersal and
spatial pattern in tropical trees. PLoS Biol. 4, e344.
(doi:10.1371/journal.pbio.0040344)
18. Russo SE, Brown P, Tan S, Davies SJ. 2008
Interspecific demographic trade-offs and soil-related
habitat associations of tree species along resourcegradients. J. Ecol. 96, 192–203. (doi:10.1111/j.
1365-2745.2007.01330.x)
19. McFadden IR, Bartlett MK, Wiegand T, Turner BL,
Sack L, Valencia R, Kraft NJ. 2019 Disentangling the
functional trait correlates of spatial aggregation in
tropical forest trees. Ecology 100, e02591. (doi:10.
1002/ecy.2591)
20. Comita LS, Muller-Landau HC, Aguilar S, Hubbell SP.
2010 Asymmetric density dependence shapes
species abundances in a tropical tree community.
Science 329, 330–332. (doi:10.1126/science.
1190772)
21. Ickes K, Thomas SC. 2003 Native, wild pigs (Sus
scrofa) at Pasoh and their impacts on the plant
community. In Pasoh (eds T Okuda et al.),
pp. 507–520. Berlin, Germany: Springer.
22. Luskin MS, Brashares JS, Ickes K, Sun I-F, Fletcher C,
Wright SJ, Potts MD. 2017 Cross-boundary subsidy
cascades from oil palm degrade distant tropical
forests. Nat. Commun. 8, 2231. (doi:10.1038/
s41467-017-01920-7)
23. Ickes K. 2001 Hyper-abundance of native
wild pigs (Sus scrofa) in a lowland dipterocarp rain
forest of Peninsular Malaysia. Biotropica 33,
682–690. (doi:10.1111/j.1744-7429.2001.
tb00225.x)
24. Luskin MS, Christina ED, Kelley LC, Potts MD. 2014
Modern hunting practices and wild meat trade in
the oil palm plantation-dominated landscapes of
Sumatra, Indonesia. Hum. Ecol. 42, 35–45. (doi:10.
1007/s10745-013-9606-8)
25. Love K, Kurz DJ, Vaughan IP, Ke A, Evans LJ,
Goossens B. 2018 Bearded pig (Sus barbatus)
utilisation of a fragmented forest–oil palm
landscape in Sabah, Malaysian Borneo. Wildl. Res.
44, 603–612. (doi:10.1071/WR16189)
10
royalsocietypublishing.org/journal/rspb Proc. R. Soc. B 288: 2021000126. Curran LM, Leighton M. 2000 Vertebrate responses
to spatiotemporal variation in seed production of
mast-fruiting Dipterocarpaceae. Ecol. Monogr. 70,
101–128. (doi:10.1890/0012-9615(2000)070[0101:
VRTSVI]2.0.CO;2)
27. Barrios-Garcia MN, Ballari SA. 2012 Impact of wild
boar (Sus scrofa) in its introduced and native range:
a review. Biol. Invasions 14, 2283–2300. (doi:10.
1007/s10530-012-0229-6)
28. Ickes K, Paciorek CJ, Thomas SC. 2005 Impacts of
nest construction by native pigs (Sus scrofa) on
lowland Malaysian rain forest saplings. Ecology 86,
1540–1547. (doi:10.1890/04-0867)
29. Davies SJ, Noor NSM, LaFrankie JV, Ashton PS. 2003
The trees of Pasoh Forest: stand structure and floristic
composition of the 50-ha forest research plot. In
Pasoh (eds T Okuda et al.), pp. 35–50. Berlin,
Germany: Springer.
30. Okuda T, Manokaran N, Matsumoto Y, Niiyama K,
Thomas SC, Ashton PS. 2013 Pasoh: ecology of a
lowland rain forest in Southeast Asia. Berlin,
Germany: Springer.
31. Ickes K, Dewalt SJ, Thomas SC. 2003 Resprouting
of woody saplings following stem snap by
wild pigs in a Malaysian rain forest. J. Ecol.
91, 222–233. (doi:10.1046/j.1365-2745.2003.
00767.x)
32. Kuznetsova A, Brockhoff PB, Christensen RHB. 2015
Package ‘lmerTest’. R package version 2.33. Johnson DJ, Condit R, Hubbell SP, Comita LS. 2017
Abiotic niche partitioning and negative density
dependence drive tree seedling survival in a tropical
forest. Proc. R. Soc. B 284, 20172210. (doi:10.1098/
rspb.2017.2210)
34. Pielou EC. 1966 Species-diversity and pattern-
diversity in the study of ecological succession.
J. Theor. Biol. 10, 370–383. (doi:10.1016/0022-
5193(66)90133-0)
35. Jost L. 2010 The relation between evenness and
diversity. Diversity 2, 207–232. (doi:10.3390/
d2020207)
36. Harrison RD, Tan S, Plotkin JB, Slik F, Detto M,
Brenes T, Itoh A, Davies SJ. 2013 Consequences of
defaunation for a tropical tree community. Ecol.
Lett. 16, 687–694. (doi:10.1111/ele.12102)
37. Borer ET et al. 2014 Herbivores and nutrients
control grassland plant diversity via light
limitation. Nature 508, 517–520. (doi:10.1038/
nature13144)
38. Murphy S, Comita LS. In press. Large mammalian
herbivores contribute to conspecific negative density
dependence in a temperate forest. J. Ecol.
39. Adler PB, Smull D, Beard KH, Choi RT, Furniss T,
Kulmatiski A, Meiners JM, Tredennick AT, Veblen KE.
2018 Competition and coexistence in plant
communities: intraspecific competition is stronger
than interspecific competition. Ecol. Lett. 21,
1319–1329. (doi:10.1111/ele.13098)40. Liang M, Johnson D, Burslem D, Yu S, Fang M,
Taylor J, Taylor A, Helgason T, Liu X. In press. Soil
fungal networks maintain local dominance of
ectomycorrhizal trees. Nat. Commun.
41. Luskin MS, Ickes K, Yao TL, Davies SJ. 2019 Wildlife
differentially affect tree and liana regeneration in a
tropical forest: an 18-year study of experimental
terrestrial defaunation versus artificially abundant
herbivores. J. Appl. Ecol. 56, 1379–1388. (doi:10.
1111/1365-2664.13378)
42. Schnitzer SA, Bongers F. 2002 The ecology of lianas
and their role in forests. Trends Ecol. Evol. 17,
223–230. (doi:10.1016/S0169-5347(02)02491-6)
43. Fujinuma J, Harrison RD. 2012 Wild pigs (Sus scrofa)
mediate large-scale edge effects in a lowland
tropical rainforest in Peninsular Malaysia. PLoS ONE
7. (doi:10.1371/journal.pone.0037321)
44. Haddad NM et al. 2015 Habitat fragmentation and
its lasting impact on Earth’s ecosystems. Sci. Adv. 1,
e1500052. (doi:10.1126/sciadv.1500052)
45. Cheng Y, Yu L, Xu Y, Lu H, Cracknell AP, Kanniah K,
Gong P. 2019 Mapping oil palm plantation
expansion in Malaysia over the past decade
(2007–2016) using ALOS-1/2 PALSAR-1/2 data.
Int. J. Remote Sens. 40, 7389–7408. (doi:10.1080/
01431161.2019.1580824)
46. Luskin MS, Meijaard E, Surya S, Walzer C, Linkie M.
2020 African Swine fever threatens Southeast Asia’s
11 endemic wild pig species. Conserv. Lett. e12784.