Ecological filtering or random extinction? Beta‐diversity patterns and

Oikos 124: 206–215, 2015
doi: 10.1111/oik.01018
© 2014 The Authors. Oikos © 2014 Nordic Society Oikos
Subject Editor: Paulo Guimaraes Jr. Editor-in-Chief: Dries Bonte. Accepted 8 June 2014
Ecological filtering or random extinction? Beta-diversity patterns
and the importance of niche-based and neutral processes following
habitat loss
Thomas Püttker, Adriana de Arruda Bueno, Paulo I. Prado and Renata Pardini
T. Püttker ([email protected]) and R. Pardini, Depto de Zoologia, Inst. de Biociências, Univ. de São Paulo, Rua do Matão, 101,
trav. 14, CEP 05508-090, São Paulo, SP, Brazil. – A. de Arruda Bueno, Fundação Florestal – Planos de Manejo, Rua do Horto 931,
CEP 02377-000, São Paulo, SP, Brazil. – P. I. Prado, LAGE, Depto de Ecologia, Inst. de Biociências, Univ. de São Paulo, Rua do Matão,
101, trav. 14, CEP 05508-090, São Paulo, SP, Brazil.
Although both niche-based and neutral processes are involved in community assembly, most models on the effects of
habitat loss are stochastic, assuming neutral communities mainly affected by ecological drift and random extinction. Given
that habitat loss is considered the most important driver of the current biodiversity crisis, unraveling the processes underlying the effects of habitat loss is critical from both a theoretical and an applied perspective. Here we unveil the importance
of niche-based and neutral processes to species extinction and community assembly across a gradient of habitat loss,
challenging the predictions of neutral models. We draw on a large dataset containing the distribution of 3653 individuals
of 42 species, representing 35% of the small mammal species of the Atlantic Forest hotspot, obtained in 68 sites
across three continuously-forested landscapes and three adjacent 10 000-ha fragmented landscapes differing in the amount
of remaining forest (50%, 30% and 10%). By applying a null-model approach, we investigated β-diversity patterns
by detecting deviations of observed community similarity from the similarity between randomly assembled communities.
Species extinction following habitat loss was decidedly non-random, in contrast to the notion that fragmented communities are mainly driven by ecological drift. Instead, habitat loss led to a strong biotic homogenization. Moreover, species
composition changed abruptly at the same level of landscape-scale habitat loss that has already been associated with a
drastic decline in species richness. Habitat loss, as other anthropogenic disturbances, can thus be seen as a strong ecological filter that increases (rather than decreases) the importance of deterministic processes in community assembly. As such,
critical advances for the development of conservation science lie on the incorporation of the relevant niche traits associated
with extinction proneness into models of habitat loss. The results also underscore the fundamental importance of pro-active
measures to prevent human-modified landscapes surpassing critical ecological thresholds.
Understanding the processes underlying community assembly
has been a long-lasting challenge in community ecology
(Vellend 2010). While niche-based, deterministic processes,
such as environmental filtering and species interactions, have
traditionally been seen as of foremost importance, consensus
among ecologists now exists on the non-negligible role of
neutral and stochastic processes, such as random extinctions
and ecological drift, on community assembly (Adler et al.
2007, Vellend 2010, Chase and Myers 2011). However, the
strength of the influence of these two classes of processes is
likely to vary in space and time depending on abiotic conditions as well as on taxonomic group (Vergnon et al. 2009,
Vellend 2010).
Among others (e.g. predation, Chase et al. 2009; size of
the species pool, Fukami 2004; interaction between species,
Orrock and Watling 2010), one of the drivers of the importance of stochastic (neutral) and deterministic (niche-based)
processes to community assembly is disturbance (Chase
2003, 2007, Trexler et al. 2005). Natural disturbance regimes
206
have been shown to increase the importance of deterministic
processes by acting as environmental filters that lead to nonrandom, reduced-membership communities (Wright et al.
1998, Chase 2003, 2007). Likewise, anthropogenic disturbances have been shown to increase similarity in community
composition among sites leading to biotic homogenization
(Baeten et al. 2012, Karp et al. 2012, Tabarelli et al. 2012).
However, empirical investigations of the effects of anthropogenic disturbance on the importance of deterministic and
stochastic processes to community assembly are rare, particularly at relevant spatial scales (but see Karp et al. 2012).
At the landscape scale, disturbance is typically represented by habitat loss and fragmentation, which alters
not only the quantity and quality of available habitat but
also the rate of dispersal among habitat patches (Fischer
and Lindenmayer 2007). Notably, most frameworks used
to model the effects of habitat loss are stochastic theories
that assume ecological equivalence among species, and neutral communities that are mainly affected by ecological drift
and random extinction. The theory of island biogeography
(MacArthur and Wilson 1967), for instance, is a neutral
model assuming random species extinction with equal probability across species. Likewise, species–area relationships
(SAR) and species–abundance-distributions (SAD) deduced
from models of community dynamics have been criticized
not to take into consideration species identity (Mac Nally
2007, Alonso et al. 2008, Banks-Leite et al. 2012), and
to assume that species extinction follows the sequence of
before-fragmentation abundance ranks. These theories
predict that the reduction and isolation of populations in
remaining habitat patches would increase the probability of
random extinctions and the importance of ecological drift
(i.e. independent, random changes in species relative abundance), leading to an increased dissimilarity (β-diversity)
between communities in different patches. This dissimilarity is expected to increase with time since isolation, the
velocity depending on the size of the remaining populations (the smaller the populations the faster the increase in
dissimilarity, Hubbell 2001). It is well known though that
species occurrence and/or abundance after habitat loss are
not necessarily positively related to abundance before habitat
loss (Mac Nally 2007, Banks-Leite et al. 2012), indicating
that habitat loss can filter species according to their traits,
and thus extinctions may have a non-random, deterministic
component. Given that habitat loss is considered the most
important driver of the current increased rates of extinction
(Pereira et al. 2010), unraveling the processes underlying
species extinction and community assembly following
habitat loss has profound implications from theoretical
and practical standpoints (e.g. conservation and restoration
practices, Legendre et al. 2005).
While the detailed study of processes underlying community assembly requires datasets with temporal replication
(Adler et al. 2007), which are difficult to obtain at large or
relevant spatial scales, the examination of β-diversity patterns across space offers a useful alternative to gain insight
on the importance of stochastic (neutral) and deterministic
(niche-based) processes. A common practice is the partitioning of the variance in community composition to distinguish
between underlying spatial autocorrelation (due to dispersal limitation) and environmental control (Legendre et al.
2005, Smith and Lundholm 2010). More recently, however,
Chase and colleagues (Chase et al. 2011, Chase and Myers
2011) called attention to the utility of null-models for
disentangling the contribution of these two classes of processes by comparing the observed similarity in community
composition to that expected assuming random assembly. The approach, originally described by Raup and Crick
(1979; hereafter called βRC) also corrects for differences in
species richness between localities, which is known to influence β-diversity measures (Jost 2007, Kraft et al. 2011).
Therefore, the βRC-approach is especially indicated when
changes in β-diversity are concomitant with differences in
α- and/or γ-diversity, as in the case of habitat loss. Further,
due to the independence of βRC from species richness, it measures β-diversity associated exclusively with species turnover,
while controlling for differences in nestedness (Baeten et al.
2012). Therefore, βRC is also considered a powerful metric
for detecting biotic homogenization (Vellend et al. 2007,
Lôbo et al. 2011, Baeten et al. 2012).
Here we focus on disentangling the contribution of
deterministic (niche-based) and stochastic (neutral) processes to community assembly following habitat loss. We
draw on a dataset from a species-rich, tropical community,
including up to 35% of all small mammal species of the
Atlantic Forest hotspot. Data collection was designed to
address habitat loss at the landscape scale by sampling 68
forest sites across three continuously-forested landscapes and
three adjacent 10 000-ha fragmented landscapes varying in
the amount of remaining forest (50%, 30% and 10% forest
cover). After exploring the effects of habitat loss on species
richness (α- and γ-diversity) with this dataset (Pardini et al.
2010), we here used a null-model approach to investigate
patterns of within- and between-landscape β-diversity, aiming at addressing the predictions of neutral models about the
effects of habitat loss. We focus on three novel inter-related
questions: 1) does the importance of stochastic processes
increase with habitat loss as predicted by neutral models,
leading to an increase in β-diversity within landscapes?
Or alternatively 2) does habitat loss lead to a decrease in
β-diversity within landscapes and thus to a biotic homogenization? Finally 3) do changes in species composition (βdiversity between landscapes) increase abruptly at high levels
of habitat loss, accompanying the drastic decline in species
richness observed below 30% of habitat (Pardini et al. 2010,
Hanski 2011)?
Material and methods
Study area and sites
The study comprise three regions (municipalities of Tapiraí/
Piedade, Cotia/ Ibiúna and Ribeirão Grande/ Capão Bonito)
in the Atlantic Plateau of São Paulo, Brazil, which encompasses one of the largest tracts of Atlantic Forest (Ribeiro
et al. 2009), and represents one of the most diverse areas
of the biome (Costa et al. 2000). Within each of the three
regions, we sampled one 10 000-ha fragmented landscape
and one adjacent continuously-forested landscape, adding
up to six landscapes (Fig. 1). In total, we sampled 68 sites: 50
in forest fragments across the three fragmented landscapes,
and 18 in continuously-forested landscapes (six per region;
Fig. 1). Sampled sites were all in similar stages of regeneration, and presented similar variation in forest structure,
fragment area and distance from the edge across the three
fragmented landscapes (for detailed description of study sites
and study site selection see Pardini et al. 2010).
All three continuously-forested landscapes are part of
one of the largest continuous remnants of Atlantic forest,
while past anthropogenic deforestation led to variation in
the proportion of remaining forest across the three fragmented landscapes (11% in Ribeirão Grande/Capão Bonito,
31% in Cotia/Ibiúna, and 49% in Tapiraí/Piedade; hereafter called 10, 30 and 50% forest cover landscapes; Fig. 1).
The decrease in forest cover across fragmented landscapes
resulted in expected differences in landscape variables closely
tied to habitat loss, such as the decrease in the size of the
largest patch, in the number and mean size of patches, and in
the proximity among patches, and the increase in mean
distance to nearest patch (Supplementary material Appendix 1
207
Figure 1. Distribution of forest remnants in three continuously-forested and three adjacent fragmented landscapes. All landscapes
are located at the Atlantic Plateau of São Paulo, Brazil, with continuously-forested landscapes being part of one of the largest continuous
remnants of Atlantic forest. (a) Landscapes in Ribeirão Grande/Capão Bonito (fragmented landscape with 10% forest cover). (b)
Landscapes in Cotia/Ibiúna (fragmented landscape with 30% forest cover). (c) Landscapes in Tapiraí/Piedade (fragmented landscape
with 50% forest cover). Numbers above landscapes are mean monthly temperatures (left) and mean annual precipitation (right) between
municipalities of each region. Forest is shown in gray, and the 68 sampling sites as dots.
Table A1). In contrast landscape metrics associated to
fragmentation per se (sensu Fahrig 2003) did not vary
between the three fragmented landscapes (clumpiness index;
Supplementary material Appendix 1 Table A1).
Despite the differences in forest cover, human use
resulted in similar land-use patterns in the three fragmented
landscapes, with converted areas dominated by pastures
and annual crops across all of them (Pardini et al. 2010).
Moreover, although the trajectory of deforestation and regeneration in most tropical landscapes is highly dynamic, and total
forest cover varied over time in the three studied fragmented
landscapes, in all of them total forest cover was already similar
to the observed today in 1962, the date of the oldest available
forest cover map (Lira et al. 2012). While forest cover was
equal (11%) or lower (43%) to the current value in the 10
and 50% forest cover landscape, only in the 30% forest cover
landscape it was slightly higher (44%). This indicates that
populations of small mammals in our fragmented landscapes
have experienced similar level of forest loss as observed today
for several decades (at least 50 years, in most cases). Given
the considerably short lifespan (∼1 year) of small mammals,
this should represent sufficient time for dissimilarity between
communities caused by ecological drift to be detected, if this
process is indeed important to community assembly. Indeed,
Lira et al. (2012) found no evidence for extinction debt in
small mammals in any of the three landscapes, with current
species richness being best explained by current landscape
structure (instead of past landscape structure) in all of them,
suggesting that small mammal population dynamics has
already responded to forest cover changes.
208
Although the three regions are located relatively close
together in the Atlantic Plateau (parallel to the coastline
within a distance of 150 km; Fig. 1), there is some variation
in climate. Mean annual rainfall varies from 1257.6 mm in
Ribeirão Grande/Capão Bonito to 1581.2 mm in Tapiraí/
Piedade, and mean monthly temperature between 19.2°C in
in Tapiraí/ Piedade and 21.1°C in Ribeirão Grande/Capão
Bonito (Fig. 1). However, altitude is similar, varying between
800 and 1000 m in all regions, and forest type is the same
(‘Lower Montane Atlantic Rain Forest’, Oliveira-Filho and
Fontes 2000). Despite these abiotic differences, the three
regions belong to the same biogeographic southeast component of the Atlantic Forest biome regarding the small mammal fauna (Costa 2003). Indeed, although the geographical
range of most Atlantic forest small mammal species is not
yet precisely determined, all captured species are expected to
occur along the entire Atlantic Plateau of São Paulo based on
the records available in the literature (Rossi 2011), thereby
including the three study regions.
Data collection and small mammal communities
At each of the study sites, small mammals were captured
following a standardized capture protocol consisting of a
100-m long transect with 11 60-l pitfall traps every 10 m
connected by a 50-cm high drift fence. Four 8-day trapping
sessions, two per summer in two consecutive summers, were
carried out per site, totaling in 23 936 trap-nights across
the six landscapes. Traps were checked each morning and
captured animals were marked with numbered ear tags
at first capture and released. A detailed description of the
capture protocol can be found in Pardini et al. (2010).
Sampling coverage (percentage of observed richness relative
to the richness estimated by ACE) varied between 0.42 and
1 among the 68 sites, and mean sampling coverage did not
differ significantly between landscapes (ANOVA F ⫽ 0.72,
p ⫽ 0.61; Supplementary material Appendix 1 Fig. A1;
results were similar considering sampling coverage based on
richness estimated by Chao1).
Species identification was based on a reference collection
of specimens obtained at the same sites in a pilot study, which
were determined by experienced taxonomists. However,
species distinction in the field by external morphological traits
was not possible for three pairs of congeners (Juliomys pictipes
and J. ossitenuis, Monodelphis scalops and M. americana, and
Phyllomys nigrispinus and P. sulinus), which were therefore
considered in conjunction in the analyses. Total sampling
effort resulted in 42 species of rodents and marsupials (39
morphotypes identifiable in the field), representing 35%
of the known small mammal species of the entire Atlantic
Forest biome (Paglia et al. 2012). Previous analysis of α- and
γ-diversity patterns have shown a non-linear, drastic decrease
in species richness across landscapes with decreasing proportion of remaining forest (Pardini et al. 2010).
Estimation of b-diversity
Occurrence-data
We calculated β-diversity using a null-model approach and
a modified Raup–Crick metric following Raup and Crick
(1979) and Chase et al. (2011). The Raup–Crick metric (βRC)
between two sites is calculated by repeated random sampling
(using species frequencies as weights) of the observed number of species at each site from the regional species pool,
followed by the comparison of the pairwise similarity (i.e.
number of shared species) between these random samples
to the observed similarity between sites. The βRC metric is
the proportion of random samples that results in equal or
higher number of shared species compared to the observed.
Thus, βRC measures the deviation from the null expectation
that community assembly is stochastic (i.e. mainly influenced by random extinction and ecological drift; Raup and
Crick 1979, Chase et al. 2011), allowing to evaluate the
role of deterministic (niche-based) and stochastic (neutral)
processes in community assembly. A scaling proposed by
Chase et al. (2011) leads to a metric ranging from –1 to 1,
indicating whether two communities are more similar than
(approaching –1), less similar than (approaching 1) or as
similar as (close to 0) expected by chance. Hence, a mean
value of βRC among a given group of sites that is different
from 0 (either positive or negative) indicates a deterministic
process in community assembly, while a mean value close
to zero indicate a community assembly that is mainly stochastic in relation to the regional species pool. For a detailed
description of the protocol see Chase et al. (2011).
We considered each landscape as one sampling unit. To
evaluate if the importance of stochastic processes increases
with habitat loss as predicted by neutral models, leading to
an increase in β-diversity within landscapes (question 1), we
compared the mean βRC between sites ‘within’ landscapes
across the gradient of habitat loss. To investigate if changes
in species composition (between-landscape β-diversity)
increase abruptly at high levels of habitat loss (question
3), we compared the mean βRC between sites of ‘different’
landscapes. More specifically, we compared the mean βRC
between sites in each landscape and sites in the continuouslyforested landscapes.
Following Chase et al. (2011), the regional species pool
was defined as all species that are potentially able to colonize the study sites. As a recent compilation of the ecological
and taxonomic literature on the small mammal fauna of the
Atlantic Plateau of São Paulo (Rossi 2011) indicated that
the geographical distribution of all captured species includes
the entire study system (i.e. encompassing all three study
regions), we defined the species pool as the set of all captured species. However, to test the sensitivity of our results
to changes in the definition of the regional species pool, we
repeated the analyses separately for each of the three regions,
thereby relying on a restricted and more conservative species pool, which included only the species captured at each of
the three study regions (Supplementary material Appendix 2).
The original method uses the total number of occupied
sites as a proxy of the frequency of species in the theoretical species pool, from which the random samples are drawn
(i.e. use species frequencies as sampling weights; Chase
et al. 2011). In our study, however, the number of surveyed
sites was unequal among landscapes. Consequently, a species recorded in many sites in a landscape where fewer sites
were surveyed could be considered as frequent in the species
pool as a species recorded at the same number of sites but in
a landscape where more sites were surveyed. We therefore
corrected the frequencies of species in the species pool following Vellend et al. (2007), first standardizing the observed
number of occupied sites within landscapes, dividing it
by the total number of sampled sites, and then using the
mean proportion of occupied sites among landscapes as the
frequency of each species in the species pool.
Abundance data
Biotic homogenization (question 2) might be underestimated when exclusively relying on presence/absence data
(Cassey et al. 2008). We thus evaluated biotic homogenization (i.e. a decrease in mean within-landscape β-diversity
across landscapes with decreasing forest cover) by calculating
βRC between sites ‘within’ landscapes including abundance
information (hereafter βRC-abund). We followed the procedure
described in Chase et al. (2011), with the difference that
we used a pool of individuals (instead of species) to define
the regional pool and drew randomly individuals instead of
species from the pool. In analogy to the procedure described
earlier, the weights for sampling from the theoretical pool
were defined as the number of individuals captured per
species corrected for differences in the number of sampled
sites among landscapes: we divided the number of captured
individuals by the total number of sampled sites in each
landscape, and then calculated the mean among landscapes
to define the sampling weights. Analyses taking into account
species abundance were also run considering both the
less and the more restricted species pool described above
(Supplementary material Appendix 2).
In both analyses (i.e. βRC and βRC-abund) we ran 10 000
random samples without reposition from the theoretical
209
pool for each pair of sites. For graphical representations of
βRC and βRC-abund, we used two- and three-dimensional nonparametric multidimensional scaling (NMDS). Additionally, after calculating for each site the βRC and βRC-abund to the
n-th most similar site (n-th nearest neighbor), we computed
the mean of each of these two metrics among sites for each
rank. The resulting plot of mean β-diversity as a function of
the neighbor rank is an effective way to depict the structure
of similarities among objects (Inger and Colwell 1977). All
analyses were conducted in R environment, ver. 2.15.0. We
calculated βRC using the R-script provided by Chase et al.
(2011). For calculation of βRC-abund we used a modified
R-script, which can be obtained from the first author.
Results
The total number of species per landscape decreased with
decreasing forest cover. Of the 39 species or morphotypes,
we recorded 19 species in the continuously-forested landscape adjacent to the 50% forest cover landscape, 21 in the
continuously-forested landscape adjacent to the 30% forest cover landscape, and 26 in the continuously-forested
landscape adjacent to the 10% forest cover landscape. In
the three fragmented landscapes 24, 20 and 13 species were
recorded within the 50, 30 and 10% forest cover landscape,
respectively.
All the results described below are based on the first
analysis considering the entire set of 39 species captured in
at least one of the three study regions as the regional species
pool. Results from the second analysis using the conservative species pool restricted to the species captured in each of
the three regions separately were qualitatively the same
(Supplementary material Appendix 2).
Does the importance of stochastic processes
increase with habitat loss as predicted by neutral
models, leading to an increase in b-diversity ‘within’
landscapes?
Despite differences in species richness, mean βRC between
sites ‘within’ all six landscapes were similar (–0.82 to –0.95;
Fig. 2a, Supplementary material Appendix 3 Fig. A3a). Mean
βRC between sites was not correlated with mean distance
between sites across the six landscapes (Supplementary material Appendix 4 Fig. A5a). In contrast, despite the variation in
the mean distances between sites within landscapes, mean βRC
values were close to –1 within all landscapes, indicating that
community composition was more similar between sites within
all landscapes than expected by chance. Hence, there was no
sign of increased ‘within’-landscape β-diversity (all mean βRC
values similar), nor of community composition being more
influenced by stochastic processes (all mean βRC values close to
–1), in fragmented than in continuously-forested landscapes or
in fragmented landscapes with lower forest cover.
Alternatively, does habitat loss lead to a decrease in
b-diversity within landscapes and thus to a biotic
homogenization?
Mean βRC ‘within’ all six landscapes were similar, not indicating any decrease or increase in ‘within’-landscape βdiversity with decreasing forest cover (size of polygons similar
in Supplementary material Appendix 3 Fig. A3a). However,
inclusion of abundance information resulted in a different pattern (Fig. 2b, Supplementary material Appendix 3
Fig. A3b). As expected, all ‘within’-landscape mean βRC-abund
were larger than ‘within’-landscape mean βRC, since inclusion of the number of individuals increases the chance
Figure 2. Non-metric multi-dimensional scaling (NMDS) ordinations of sites within six landscapes with different amount of forest based
on βRC (a) and βRC-abund (b) between sites. Polygons enclose sites within the same landscape. Numbers close to or within polygons are mean
βRC (a) or mean βRC-abund (b) among all pairs of sites within each landscape. Full symbols: continuously-forested landscapes; open symbols:
fragmented landscapes. Triangles: landscapes in Tapiraí/Piedade (fragmented landscape with 50% forest cover); circles: landscapes in Cotia/
Ibiúna (fragmented landscape with 30% forest cover); squares: landscapes in Ribeirão Grande/Capão Bonito (fragmented landscape with
10% forest cover). Note that sizes of polygons are not directly proportional to mean βRC (a) or mean βRC-abund (b) due to the projection of
the NMDS in two dimensions (see Supplementary material Appendix 3 Fig. A3 for 3D-version of Fig. 2).
210
of dissimilarity between sites. Although a species might
occur at different sites, and therefore account for similarity
between these sites when relying on occurrence-data only, it
is unlikely that this species occurs at equal abundances at all
sites, thereby accounting for dissimilarity between sites when
relying on abundance information. This was particularly the
case in the most forested landscapes (50% forest cover landscape and all continuously-forested landscapes), where mean
βRC-abund was positive (0.69–0.75; Fig. 2b, Supplementary
material Appendix 3 Fig. A3b), indicating that community
structure was more dissimilar between sites than expected
by chance within these forested landscapes. At the 30%
forest cover landscape, however, mean βRC-abund was much
lower. An even stronger decrease in mean βRC-abund, and the
only negative value, was found within the most deforested
landscape, indicating – in contrast to all other landscapes
– a community structure more similar between sites than
expected by chance (smaller polygon Supplementary material Appendix 3 Fig. A3b).
Likewise, means of ranked βRC were generally low at all
landscapes, indicating high within-landscape similarity in
community composition (Fig. 3a). In contrast, means of
ranked βRC-abund spanned over a wider range of values, and
clearly ordered landscapes from the most to the least forested
(Fig. 3b), with the more forested landscapes showing lower,
and the most deforested landscape the highest similarity in
community structure. In the 10% forest cover landscape,
few sites accounted for an increase in βRC-abund, while most
sites presented very similar βRC-abund-values (plateau at lower
ranks within the 10% forest cover landscape; Fig. 3b).
These results are not related to differences in distance
between sites within landscapes, since mean βRC-abund between
sites was not correlated with mean distance between sites
across the six landscapes (Supplementary material Appendix
4 Fig. A5b). Instead, they indicate a decrease in ‘within’-
landscape β-diversity across the gradient of habitat loss when
abundance data is taken into account, with a strong biotic
homogenization within the most deforested landscape.
This homogenization was caused by the proliferation of few
species in most sites of the most deforested landscape: three
species showed a consistent increase in abundance compared
to all other landscapes (Oligoryzomys nigripes, Gracilinanus
microtarsus and Monodelphis kunsi; Supplementary material
Appendix 3 Fig. A4b). The terrestrial rodent O. nigripes in
particular dominated all sites, representing 71% of captured
individuals at this landscape.
Do changes in species composition (b-diversity
‘between’ landscapes) increase abruptly at high
levels of habitat loss, accompanying the drastic
decline in species richness observed below 30% of
habitat?
Mean βRC between sites of the two more forested fragmented
landscapes (i.e. the 50% and the 30% forest cover landscapes)
and sites of all three continuously-forested landscapes were negative and comparable to the mean βRC between sites of the three
different continuously-forested landscapes (Fig. 4). In contrast,
mean βRC between sites in the 10% forest cover landscape and
sites in all three continuously-forested landscapes were close
to zero or even positive (Fig. 4). This indicates that changes
in species composition relative to the continuously-forested
landscapes are not proportional to the amount of forest
lost, being larger in the 10% forest cover landscape (Fig. 4).
Again, this result is not related to differences in the distance
between landscapes (Fig. 4). Instead, change in species composition increased abruptly at the most deforested landscape
(Fig. 4), accompanying the drastic decline in species richness
at this landscape (Pardini et al. 2010). Accordingly, seven
species (including four very common species – Marmosops
Figure 3. Rank-dissimilarity curves between sites within six landscapes with different amount of forest. (a) Rank-dissimilarity curves
using species presence and (b) species abundance. Each curve shows, for each landscape, the mean among the pair-wise dissimilarity (βRC
or βRC-abund) between each site and its n-th most similar neighbor as a function of the rank of dissimilarity. Full symbols: continuouslyforested landscapes; open symbols: fragmented landscapes. Triangles: landscapes in Tapiraí/Piedade (fragmented landscape with 50% forest
cover); circles: landscapes in Cotia/Ibiúna (fragmented landscape with 30% forest cover); squares: landscapes in Ribeirão Grande/Capão
Bonito (fragmented landscape with 10% forest cover).
211
Figure 4. Mean βRC and mean distance ⫾ SD (open circles) of sites in each landscape relative to sites in the three continuously-forested
landscapes as a function of forest cover at the landscape scale.
incanus, Delomys sublineatus, Monodelphis spp. and Brucepattersonius soricinus) occurred at all landscapes except the most
deforested landscape, while no species was exclusively absent
from other landscapes (Supplementary material Appendix 3
Fig. A4a). At the same time, the most deforested landscape
harbored the largest number of exclusive species (three
species not captured at other landscapes – Monodelphis kunsi,
Mus musculus and Oxymycterus delator; Supplementary
material Appendix 3 Fig. A4b).
Discussion
Our results indicate the prevalence of niche-based, deterministic processes relative to neutral, stochastic processes for
community assembly at all landscapes. Although ecological
drift may occur in these communities, we found no sign of
increased within-landscape β-diversity or increased influence
of stochastic processes on community assembly across the
gradient of habitat loss, as it would be expected if communities were neutral and species extinction completely random.
Instead, habitat loss can be seen as an ecological filter, leading
to a strong decrease in within-landscape β-diversity, and thus
to a homogenization of communities, at the most deforested
landscape. Further, changes in species composition between
landscapes were not proportional to forest loss, with relatively low turnover of species across forested landscapes (i.e.
continuously-forested, 50 and 30% forest cover landscapes),
but an abrupt change in species composition at the most
deforested landscape.
It is important to highlight though that as any observational study our work provides weak inference on the causal
role of forest cover on the observed patterns. Especially
the relatively large number of sites sampled per landscape
– an essential aspect of our study design for allowing the
evaluation of both within and between-landscape β-diversity
– set limits to our ability to increase the number of sampled regions and landscapes. However, our sampling design
presents three important strengths: 1) each of the three
fragmented landscapes are paired (i.e. located adjacent) to a
212
continuously-forested landscape, increasing the probability
that the small mammal species were originally present at the
fragmented landscapes before habitat loss; 2) the three study
regions are close together (within a 150-km strip parallel to
the coastline along the Atlantic Plateau) and sampled sites
were chosen in order to maintain similar environmental heterogeneity within different landscapes and regions; and 3)
the best available compilation of the natural history of the
small mammals of the study system indicates that all captured species do indeed originally occupy the entire study
area. Hence, considering the evidence value that observational surveys at large spatial scale can provide, our results
support the prevalence of deterministic processes in community assembly following habitat loss, as well as the biotic
homogenization and abrupt change in species composition
in highly deforested landscapes.
In the following paragraphs we discuss the implications
of each of these three inter-related findings.
Non-random extinctions due to habitat loss
Small mammal communities at all landscapes were not a
random sample from the overall species pool. Although only
about half of the number of species occurred in the most
deforested landscape compared to continuously-forested
landscapes, once controlling for species richness, similarity
in species composition between sites within landscapes did
not decrease across the gradient of habitat loss, as it would
be expected under random community drift. Indeed, a
considerable number of species that became extinct at the
most deforested landscape were highly abundant in more
forested landscapes, suggesting that the chance of extinction is neither equal among species nor negatively correlated
with species abundance before habitat loss, as predicted by
neutral models. Although in the present work we did not
test for niche-based correlates of extinction proneness, a
previous analysis of α- and γ-diversity of this dataset suggested that niche breadth in terms of habitat requirements
is associated with species response to habitat loss: species
richness decreased across the gradient of habitat loss only
among forest specialists but not among habitat generalists
(Pardini et al. 2010). Extinction due to habitat loss is thus
likely influenced by deterministic, niche-related processes, in
this case apparently associated with species niche breadth,
which has been suggested as one of the most important
ecological correlates of extinction proneness in general
(Stork et al. 2009). Previous empirical investigations of the
effects of habitat loss on β-diversity that did not control for
species richness commonly encountered a decrease in similarity (increase in β-diversity) within communities in smaller
patches or landscapes with lower habitat cover (Harrison
1997, Didham et al. 1998, Pardini et al. 2005, Kattan et al.
2006), thereby suggesting an increased importance of stochastic processes to community assembly with habitat loss.
The importance of controlling species richness during the
analysis of β-diversity has received attention only recently
(Jost 2007), and currently most studies on β-diversity that
control for species richness do not focus on habitat loss
(Smith et al. 2009, Kraft et al. 2011, Márquez and Kolosa
2013). To our knowledge the only two studies that focused
on the effects of habitat loss using this approach also did not
find an increase in β-diversity in tree communities (Lôbo
et al. 2011) and plants (trees, shrubs, lianas, palms and
herbs; Arroyo-Rodriguez et al. 2013). These results show the
fragility of the assumptions of random extinction included
in several important models of species-area relationships.
Albeit their importance to general theoretical frameworks
(e.g. Island Biogeography, MacArthur and Wilson 1967),
these models do not fully describe the processes underlying
species extinction following habitat loss.
Biotic homogenization at highly deforested
landscapes
Not only did we not observe an increase in β-diversity within
landscapes across a gradient of habitat loss when considering
only species occurrence, but also and more importantly we
observed a strong decrease in within-landscape β-diversity
when abundance information was included. Considering
species abundance, the similarity between sites within landscapes increased across fragmented landscapes, especially at
the most deforested landscape, where communities were
more similar than expected by chance. Increased similarity
in community structure is the opposite of what is expected
assuming neutral communities, ecological drift and random
extinction and is thus not predicted by the commonly used
theoretical frameworks on the effects of habitat loss. In contrast, our results suggest an increased importance of deterministic, niche-based processes with habitat loss, leading to
biotic homogenization (Lôbo et al. 2011, Banks-Leite et al.
2012, Tabarelli et al. 2012), which has also been shown to
be a consequence of other anthropogenic disturbances, such
as land use intensification (Karp et al. 2012), forest regeneration (Vellend et al. 2007), or human-assisted invasions
(Olden et al. 2011).
Biotic homogenization is thought to be the result of
ecological filters leading to the dominance of a similar subset of species able to resist harsh conditions (Chase 2007),
either due to high competitive ability or competitive release
(Azevedo et al. 2012), and is expected to influence various
ecosystem functions and services (Olden et al. 2011). In our
study, although community composition is equally similar
within all landscapes, few species became universally abundant in the more deforested landscape. This is the case of
Oligoryzomys nigripes, a widespread generalist species, which
is one of the main reservoirs of the hantavirus associated
with the fatal hantavirus pulmonary syndrome in the Atlantic Forest (Suzuki et al. 2004). Thus, the decrease in forest
cover in our study region acts as an environmental filter,
leading to a biotic homogenization that may affect disease
risk in humans.
Besides pointing at a biotic homogenization within the
most deforested landscape, the inclusion of abundance data
also indicates that sites within more forested landscapes,
although very similar in community composition, varied
more than expected by chance in species abundance. In other
words, species dominating each community differed between
sites within more forested landscapes. A similar result was
also observed for plants (Arroyo-Rodriguez et al. 2013).
These results suggest non-random intra-specific aggregations
within these forested landscapes, which as such should not
be caused by stochastic, neutral processes like priority effects
alone (Chase 2003, Márquez and Kolosa 2013). Instead, it
might be caused by different non-mutually exclusive nichebased factors, such as small scale heterogeneity in habitat
quality (Allouche et al. 2012), inter-specific competition
(Götzenberger et al. 2012), or differential dispersal limitation among species (Gilbert and Lechowicz 2004).
An abrupt change in species composition
When considering only species occurrence, withinlandscape community dissimilarity did not change, but
between-landscape community dissimilarity increased
abruptly across the gradient of habitat loss. Species composition in the most deforested landscape was disproportionately
dissimilar relative to species composition in the continuously-forested landscapes. A drastic decrease in species persistence and species richness below 30% of remaining habitat
is expected due to the exponential increase in the distance
between patches at around ∼20% habitat cover (Fahrig
2003), and has been observed in empirical studies (Radford
et al. 2005, Hanski 2011, Estavillo et al. 2013), including
in our study landscapes (Pardini et al. 2010). However,
as far as we are aware, no previous study evaluated if this
drastic increase in species extinction is accompanied by a
disproportional change in species composition. Although
we were not able to test this directly given the limited number of sampled landscapes, our results are in agreement
with an abrupt change in species composition and a drastic decline in species richness occurring simultaneously at
the 10% forest cover landscape. Because we controlled for
differences in richness, our results suggest that communities
at the most deforested landscape are not rarefied samples of
species-rich communities in forested landscapes; rather these
communities experience a high turnover of species, being
disproportionately dissimilar to species-rich communities in
more forested landscapes.
Beside the loss of seven relatively common species that
occurred at all other landscapes, the most deforested landscape, despite presenting the lowest γ-diversity, harbored the
213
highest number of exclusive species (not captured at other
landscapes). These species are invading exotic species (Mus
musculus) or species typical from open, savanna-like biomes
(M. kunsi and O. delator). Likely, adequate habitat condition (i.e. high proportion of open anthropogenic habitats)
as well as the expansion of competition-free space inside forest patches (Azevedo et al. 2012), caused by the extinction
of the most common species in more forested landscapes,
enable exotic and open-area specialist species to occupy forests
at highly deforested landscapes. Together with the exposed
above, our results suggest that biological communities in
forest patches at highly deforested landscapes are not only
different in terms of species composition from those at forested landscapes, but also much more homogeneous in terms
of species abundance, being highly dominated by few species.
Conclusions and implications
Methodologically, our work calls attention to the usefulness
of null-model approaches to identify patterns of change in
community composition and structure independent from
differences in richness (Kraft et al. 2011), thus allowing
inference on the underlying processes (Chase et al. 2011).
It also highlights the importance of including abundance
information in measures of β-diversity in order to detect
otherwise unseen consequences of disturbances on community assembly (Cassey et al. 2008).
From a theoretical perspective, our results indicate that
species extinction due to habitat loss is mainly non-random,
and influenced by deterministic, niche-based processes.
As such, extinction should be associated with ecological
correlates that define extinction proneness (Smith et al.
2009) instead of abundance ranks in continuous habitat
(Mac Nally 2007). Habitat loss, as other anthropogenic
disturbances (Vellend et al. 2007, Karp et al. 2012), can
thus be seen as a strong ecological filter that increases (rather
than decreases) the importance of deterministic processes in
community assembly (Chase 2007, Smith et al. 2009).
Most models on the effects of habitat loss are stochastic, neutral theories that assume communities are mainly
affected by ecological drift and random extinction. Given
our results on the prevalence of deterministic, niche-based
processes for community assembly following habitat loss,
the incorporation of the relevant niche traits associated with
extinction proneness into these models represents critical
advances for the application of ecological knowledge and
the development of conservation science. Equally important
from the applied point of view is our finding that habitat
loss simultaneously led to a strong within-landscape biotic
homogenization, an abrupt change in community composition and a drastic decline in species richness. This result corroborates the importance of pro-active measures to prevent
human-modified landscapes to go beyond these ecological
thresholds (Pardini et al. 2010, Hanski 2011), and on the
limitations of ecological restoration in highly deforested
landscapes (Tscharntke et al. 2005, Pardini et al. 2010).
Acknowledgements – We are grateful to Cristina Banks-Leite,
Elizabeth Nichols and Leandro Reverberi Tambosi for critical
comments on previous versions of the manuscript, as well as to the
Subject Editor Paulo Guimarães for the comments that greatly
214
improved the paper. This work was funded by FAPESP - Fundação
de Amaparo à Pesquisa do Estado de São Paulo (05/56555-4), and
CNPq - Conselho Nacional de Desenvolvimento Científico e Tecnológico/ BMBF - Bundesministerium für Bildung und Forschung
(690144/01–6). TP had a postdoctoral fellowship from FAPESP
(09/54052-6), AAB a doctoral fellowship from CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior, and RP and
PIP a research fellowship from CNPq (Bolsa de Produtividade
306715/2011-2 and 305326/2011-2), during the development of
this work.
References
Adler, P. B. et al. 2007. A niche for neutrality. – Ecol. Lett. 10:
95–104.
Allouche, O. et al. 2012. Area–heterogeneity tradeoff and the
diversity of ecological communities. – Proc. Natl Acad. Sci.
USA 104: 17495–17500.
Alonso, D. et al. 2008. The implicit assumption of symmetry and
the species abundance distribution. – Ecol. Lett. 11: 93–105.
Arroyo-Rodriguez, V. et al. 2013. Plant beta-diversity in
fragmented rain forests: testing floristic homogenization and
differentiation hypotheses. – J. Ecol. 101: 1449–1458.
Azevedo, F. et al. 2012. Competitive release and area effects. – Ecol.
Complex. 11: 154–159.
Baeten, L. et al. 2012. Distinguishing between turnover and
nestedness in the quantification of biotic homogenization.
– Biodivers. Conserv. 21: 1399–1409.
Banks-Leite, C. et al. 2012. Unraveling the drivers of community
dissimilarity and species extinction in fragmented landscapes.
– Ecology 93: 2560–2569.
Cassey, P. et al. 2008. The varying role of population abundance
in structuring indices of biotic homogenization. – J. Biogeogr.
35: 884–892.
Chase, J. M. 2003. Community assembly: when should history
matter? – Oecologia 136: 489–498.
Chase, J. M. 2007. Drought mediates the importance of stochastic
community assembly. – Proc. Natl Acad. Sci. USA 104:
17430–17434.
Chase, J. M. and Myers, J. A. 2011. Disentangling the importance
of ecological niches from stochastic processes across scales.
– Phil. Trans. R. Soc. B 366: 2351–2362.
Chase, J. M. et al. 2009. Predators temper the relative importance
of stochastic processes in the assembly of prey metacommunities. – Ecol. Lett. 12: 1210–1218.
Chase, J. M. et al. 2011. Using null models to disentangle variation
in community dissimilarity from variation in a-diversity.
– Ecosphere 2: 1–11.
Costa, L. P. 2003. The historical bridge between the Amazon and
the Atlantic Forest of Brazil: a study of molecular phylogeography with small mammals. – J. Biogeogr. 30: 71–86.
Costa, L. P. et al. 2000. Biogeography of South American forest
mammals: endemism and diversity in the Atlantic Forest.
– Biotropica 32: 872–881.
Didham, R. K. et al. 1998. Beetle species responses to tropical
forest fragmentation. – Ecol. Monogr. 68: 295–323.
Estavillo, C. et al. 2013. Forest loss and the biodiversity threshold:
an evaluation considering species habitat requirements and the
use of matrix habitats. – PLoS ONE 8: e82369.
Fahrig, L. 2003. Effects of habitat fragmentation on biodiversity.
– Annu. Rev. Ecol. Evol. Syst. 34: 487–515.
Fischer J, and Lindenmayer, D.B. 2007. Landscape modification
and habitat fragmentation: a synthesis. – Global. Ecol. Biogeogr. 16: 265–280.
Fukami, T. 2004. Community assembly along a species pool
gradient: implications for multiple-scale patterns of species
diversity. – Popul. Ecol. 46: 137–147.
Gilbert, B. and Lechowicz, M. J. 2004. Neutrality, niches, and
dispersal in a temperate forest understory. – Proc. Natl Acad.
Sci. USA 104: 7651–7656.
Götzenberger, L. et al. 2012. Ecological assembly rules in plant
communities – approaches, patterns and prospects. – Biol.
Rev. 87: 111–127.
Hanski, I. 2011. Habitat loss, the dynamics of biodiversity, and a
perspective on conservation. – Ambio 40: 248–255.
Harrison, S. 1997. How natural habitat patchiness affects the
distribution of diversity in Californian serpentine chaparral.
– Ecology 78: 1898–1906.
Hubbell, S. P. 2001. The unified neutral theory of biodiversity and
biogeography. – Princeton Univ. Press.
Inger, R. and Colwell, R. K. 1977. Organization of contiguous
communities of amphibians and reptiles in Thailand. – Ecol.
Monogr. 47: 229–253.
Jost, L. 2007. Partitioning diversity into independent alpha and
beta components. – Ecology 88: 2427–2439.
Karp, D. S. et al. 2012. Intensive agriculture erodes beta-diversity
at large scales. – Ecol. Lett. 15: 963–970.
Kattan, G. H. et al. 2006. Spatial components of bird diversity in
the Andes of Colombia: implications for designing a regional
reserve system. – Conserv. Biol. 20: 1203–1211.
Kraft, N. B. et al. 2011. Disentangling the drivers of β diversity
along latitudinal and elevational gradients. – Science 333:
1755–1758.
Legendre, P. et al. 2005. Analyzing beta diversity: partitioning the
spatial variation of community composition data. – Ecol.
Monogr. 75: 435–450.
Lira, P. K. et al. 2012. Evaluating the legacy of landscape history:
extinction debt and species credit in bird and small mammal
assemblages in the Brazilian Atlantic Forest. – J. Appl. Ecol.
49: 1325–1333.
Lôbo, D. et al. 2011. Forest fragmentation drives Atlantic forest
of northeastern Brazil to biotic homogenization. – Divers.
Distrib. 17: 287–296.
MacArthur, R. H. and Wilson, E. O. 1967. The theory of island
biogeography. – Princeton Univ. Press.
Mac Nally, R. 2007. Use of the abundance spectrum and relativeabundance distributions to analyze assemblage change in massively altered landscapes. – Am. Nat. 170: 319–330.
Márquez, J. C. and Kolosa, J. 2013. Local and regional processes
in community assembly. – PLoS ONE 8: e54580.
Olden, J. D. et al. 2011. Biological invasions and the homogenization of faunas and floras. – In: Ladle, R. J. and Whittaker, R.
J. (eds), Conservation biogeography. Wiley, pp. 224–243.
Oliveira-Filho, A. T. and Fontes, M. A. L. 2000. Patterns of floristic differentiation among Atlantic forests in southeastern
Brazil and the influence of climate. – Biotropica 32: 793–810.
Orrock, J. L. and Watling, J. I. 2010. Local community size mediates ecological drift and competition in metacommunities.
– Proc. R. Soc. B 277: 2185–2191.
Paglia, A. P. et al. 2012. Annotated checklist of Brazilian mammals,
2nd edn. – Conservation International.
Pardini, R. et al. 2005. The role of forest structure, fragment size
and corridors in maintaining small mammal abundance and
diversity in an Atlantic forest landscape. – Biol. Conserv. 124:
253–266.
Pardini, R. et al. 2010. Beyond the fragmentation threshold
hypothesis: regime shifts in biodiversity across fragmented
landscapes. – PLoS ONE 5: e13666.
Pereira, H. M. et al. 2010. Scenarios for global biodiversity in the
21st century. – Science 330: 1496–1501.
Radford, J. Q. et al. 2005. Landscape-level thresholds of habitat
cover for woodland-dependent birds. – Biol. Conserv. 124:
317–337.
Raup, D. M. and Crick, R. E. 1979. Measurement of faunal similarity in paleontology. – J. Paleontol. 53: 1213–1227.
Ribeiro, M. C. et al. 2009. The Brazilian Atlantic Forest: how
much is left, and how is the remaining forest distributed?
Implications for conservation. – Biol. Conserv. 142:
1141–1153.
Rossi, N. F. 2011. Pequenos mamíferos não-voadores do Planalto
Atlântico de São Paulo: identificação, história natural e
ameaças. – Univ. de São Paulo.
Smith, K. G. et al. 2009. Selecting for extinction: nonrandom
disease-associated extinction homogenizes amphibian biotas.
– Ecol. Lett. 12: 1069–1078.
Smith, T. W. and Lundholm, J. T. 2010. Variation partitioning
as a tool to distinguish between niche and neutral processes.
– Ecography 33: 648–655.
Stork, N. et al. 2009. Vulnerability and resilience of tropical forest
species to land-use change. – Conserv. Biol. 23: 1438–1447.
Suzuki, A. et al. 2004. Identifying rodent hantavirus reservoirs,
Brazil. – Emerg. Infect. Dis. 10: 2127–2134.
Tabarelli, M. et al. 2012. The ‘few winners and many losers’
paradigm revisited: emerging prospects for tropical forest
biodiversity. – Biol. Conserv. 155: 136–140.
Trexler, J. C. et al. 2005. Disturbance frequency and community
structure in a twenty-five year intervention study. – Oecologia
145: 140–152.
Tscharntke, T. et al. 2005. Landscape perspectives on agricultural
intensification and biodiversity – ecosystem service management. – Ecol. Lett. 8: 857–874.
Vellend, M. 2010. Conceptual synthesis in community ecology.
– Q. Rev. Biol. 85: 183–206.
Vellend, M. et al. 2007. Homogenization of forest plant communities and weakening of species–environment relationships via
agricultural land use. – J. Ecol. 95: 565–573.
Vergnon, R. et al. 2009. Niches versus neutrality: uncovering the
drivers of diversity in a species-rich community. – Ecol. Lett.
12: 1079–1090.
Wright, D. H. et al. 1998. A comparative analysis of nested subset
patterns of species composition. – Oecologia 113: 1–20.
Supplementary material (available online as Appendix
oik.01018 at ⬍www.oikosjournal.org/readers/appendix⬎).
Appendix 1–4.
215