Fish Invasions in the World’s River
Systems: When Natural Processes Are
Blurred by Human Activities
Fabien Leprieur1*, Olivier Beauchard2, Simon Blanchet3, Thierry Oberdorff4, Se ´bastien Brosse1
1 Laboratoire Evolution and Diversite ´ Biologique, UMR 5174, CNRS-Universite ´ Paul Sabatier, Toulouse, France, 2 Faculty of Sciences, Department of Biology, Ecosystem
Management Research Group University of Antwerp, Antwerpen (Wilrijk), Belgium, 3 De ´partement de Biologie, Centre Interuniversitaire de Recherche sur le Saumon
Atlantique (CIRSA) and Que ´bec-Oce ´an, Universite ´ Laval, Sainte-Foy, Quebec City, Quebec, Canada, 4 Institut de Recherche pour le De ´veloppement (UR131), Antenne au
Muse ´um National d’Histoire Naturelle, Paris, France
Because species invasions are a principal driver of the human-induced biodiversity crisis, the identification of the major
determinants of global invasions is a prerequisite for adopting sound conservation policies. Three major hypotheses,
which are not necessarily mutually exclusive, have been proposed to explain the establishment of non-native species:
the ‘‘human activity’’ hypothesis, which argues that human activities facilitate the establishment of non-native species
by disturbing natural landscapes and by increasing propagule pressure; the ‘‘biotic resistance’’ hypothesis, predicting
that species-rich communities will readily impede the establishment of non-native species; and the ‘‘biotic acceptance’’
hypothesis, predicting that environmentally suitable habitats for native species are also suitable for non-native
species. We tested these hypotheses and report here a global map of fish invasions (i.e., the number of non-native fish
species established per river basin) using an original worldwide dataset of freshwater fish occurrences, environmental
variables, and human activity indicators for 1,055 river basins covering more than 80% of Earth’s surface. First, we
identified six major invasion hotspots where non-native species represent more than a quarter of the total number of
species. According to the World Conservation Union, these areas are also characterised by the highest proportion of
threatened fish species. Second, we show that the human activity indicators account for most of the global variation in
non-native species richness, which is highly consistent with the ‘‘human activity’’ hypothesis. In contrast, our results do
not provide support for either the ‘‘biotic acceptance’’ or the ‘‘biotic resistance’’ hypothesis. We show that the
biogeography of fish invasions matches the geography of human impact at the global scale, which means that natural
processes are blurred by human activities in driving fish invasions in the world’s river systems. In view of our findings,
we fear massive invasions in developing countries with a growing economy as already experienced in developed
countries. Anticipating such potential biodiversity threats should therefore be a priority.
Citation: Leprieur F, Beauchard O, Blanchet S, Oberdorff T, Brosse S (2008) Fish invasions in the world’s river systems: When natural processes are blurred by human activities.
PLoS Biol 6(2): e28. doi:10.1371/journal.pbio.0060028
The deliberate or accidental introduction of species out-
side their native range is a key component of the human-
induced biodiversity crisis, harming native species and
disturbing ecosystems processes [1–3]. The greater the
introduction of non-natives in a region, the higher the
probability that some of them become invasive and will hence
cause ecological or economic damage [4,5]. Patterns of non-
native species richness are therefore relevant in forecasting
the overall impact of invasions on a global scale  and
should help management authorities to adopt sound,
effective conservation policies [5–7].
The process of species invasion consists of three successive
stages: initial dispersal, establishment of self-sustaining
populations, and spread into the recipient habitat. The last
two stages are contingent upon the first one, i.e., if initial
dispersal is interrupted, establishment and spread do not
occur . Three major hypotheses, which are not necessarily
mutually exclusive, have been proposed to explain invasion
patterns: the ‘‘human activity’’ , ‘‘biotic acceptance’’ ,
and ‘‘biotic resistance’’  hypotheses. The ‘‘human activity’’
hypothesis refers to the three stages of the invasion process
(initial dispersal, establishment, and spread), whereas the
‘‘biotic resistance’’ and ‘‘biotic acceptance’’ hypotheses
address only the establishment and spread stages . With
regards to the establishment stage, the ‘‘human activity’’
hypothesis predicts that, by disturbing natural landscapes and
increasing propagule pressure (i.e., the number of individuals
released and the frequency of introductions in a given
habitat), human activities facilitate the establishment of non-
native species [9,13,14]. Everything else being equal, a positive
relationship is therefore expected between non-native
species richness and quantitative surrogates of propagule
pressure and habitat disturbance (e.g., gross domestic
product [GDP], percentage of urban area, and human
Academic Editor: Daniel Simberloff, University of Tennessee, United States of
Received September 10, 2007; Accepted December 20, 2007; Published February
Copyright: ? 2008 Leprieur et al. This is an open-access article distributed under
the terms of the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original author
and source are credited.
Abbreviations: GDP, gross domestic product; GLM, generalised linear model;
IUCN, The World Conservation Union; NPP, net primary productivity
* To whom correspondence should be addressed. E-mail: email@example.com
PLoS Biology | www.plosbiology.org February 2008 | Volume 6 | Issue 2 | e280001
P PL Lo oS S BIOLOGY
population density ). Then, the ‘‘biotic acceptance’’
hypothesis predicts that the establishment of non-native
species would be greatest in areas that are rich in native
species and with optimal environmental conditions for
growth (i.e., ‘‘what is good for natives is good for non-natives
too’’ ). Everything else being equal, native and non-native
species richness should co-vary positively with environmental
factors such as energy availability and habitat heterogeneity,
which are already recognised as the primary global determi-
nants of native species richness [15,16]. In contrast, the
‘‘biotic resistance’’ hypothesis predicts that species-poor
communities will host more non-native species than spe-
cies-rich communities, the latter being highly competitive
and hence readily impede the establishment of non-native
species [11,17]. Therefore, a negative relationship is expected
between native and non-native species richness. To date, the
relative importance of these hypotheses in explaining the
variation in non-native species richness had never been
tested at the global scale.
We tested these hypotheses and report a global map of fish
invasions (i.e., the number of non-native fish established per
river basin) by using an extensive worldwide dataset of
freshwater fish occurrences (i.e., more than 40,000 occurrences
of 9,968 fish species) on the river basin scale (1,055 basins
covering more than 80% of Earth’s surface). Freshwater fish
offer a unique opportunity to identify factors that are
responsible for large-scale gradients in non-native species
richness for at least two main reasons. First, among vertebrate
groups, freshwater fish have been widely introduced over the
world , which often had subsequent negative consequences
on native species and ecosystems integrity [19–23]. Second, as
rivers are separated from one another by barriers insur-
mountable for freshwater fish (land or ocean), they form kind
of ‘‘biogeographical islands’’, whose space is delimited .
This implies that the natural and human factors shaping global
patterns of non-native species richness can be easily separated.
Our results revealed six global invasion hotspots where
non-native species represent more than a quarter of the total
number of species per basin: the Pacific coast of North and
Central America, southern South America, western and
southern Europe, Central Eurasia, South Africa and Mada-
gascar, and southern Australia and New Zealand (Figure 1A).
According to The World Conservation Union (IUCN) Red
List , these areas were also characterised by the highest
proportion of fish species having a high risk of extinction in
the wild (Figure 2).
Analysing the absolute number of species, we found that
river basins of the Northern Hemisphere host the highest
number of non-native fish species (Figure 1B). The human
factors considered here to test the ‘‘human activity’’
hypothesis (GDP, population density, percentage of urban
area) were found to be positively related to non-native
species richness (Table 1), after controlling for the effects of
environmental conditions and native species richness. In
contrast, the positive correlation between native and non-
native richness that was expected by the ‘‘biotic acceptance’’
hypothesis was not significant after controlling for the effects
of propagule pressure and habitat disturbance (Table 1).
Indeed, the environmental factors displayed either no (net
primary productivity) or a weak positive correlation (altitu-
dinal range and basin area) with non-native species richness,
after controlling for the effects of propagule pressure and
habitat disturbance (Table 1). The negative correlation
between native and non-native richness, expected by the
‘‘biotic resistance’’ hypothesis, was not significant after
controlling for the effects of environmental conditions,
propagule pressure, and habitat disturbance (Table 1).
Then, we applied hierarchical partitioning [26–28] that
aims to quantify the independent explanatory power of each
variable by considering all possible submodels. The deviance
explained by the 128 submodels computed in hierarchical
partitioning accounted in average for 52% of the total
deviance (67% standard deviation [SD], min ¼ 37%, max ¼
67%). The human factors had together the greatest inde-
pendent effect on non-native species richness (70%, Table 2).
Among the human factors, the GDP (an economical index of
human activities ) had the greatest independent explan-
atory power (43%; Table 2). To a lesser extent, the habitat
heterogeneity (i.e., basin area and altitudinal range) and the
number of native species also contribute to the variation in
non-native species between river basins (Table 2).
To test for potential bias in our results due to differences
in sampling effort between continents, bootstrap analysis was
performed by applying hierarchical partitioning to 1,000
random subsets of 100 basins. For each variable, the
independent effect observed did not differ from the 95%
bootstrap percentile confidence interval (Table 2), testifying
that potential differences in sampling effort between con-
tinents hardly affected the results.
By using an explanatory modelling approach, we showed
that the human activity indicators of the world’s river basins
were positively related to the number of established non-
native fish species. In addition, they account for most of the
global variation in non-native species richness, giving support
for the ‘‘human activity’’ hypothesis. More particularly, we
highlight that the level of economic activity of a given river
basin (expressed by the GDP) strongly determines its
invasibility. Three non-exclusive mechanisms may account
PLoS Biology | www.plosbiology.org February 2008 | Volume 6 | Issue 2 | e280002
Fish Invasions in the World’s River Systems
As one of the major threats to biodiversity, the detrimental
consequences of biological invasions are widely recognised. Despite
this, a global view of invasion patterns and their determinants is still
lacking in aquatic ecosystems, reducing our ability to initiate
practical actions. Here we report the global patterns of freshwater
fish invasion in 1,055 river basins covering more than 80% of Earth’s
continental surface. This allows us to identify six major invasion
hotspots where non-native species represent more than a quarter of
the total number of species. According to the World Conservation
Union, these areas are also characterised by the highest proportion
of threatened fish species. We also show that the natural factors
controlling global biodiversity do not influence the number of non-
native species in a given river basin. Instead, human activity–related
factors, and particularly economic activity, explain why some river
basins host more non-native species. In view of our findings, we fear
massive invasions in developing countries with a growing economy
as already experienced in developed countries. This constitutes a
serious threat to global biodiversity.
for this pattern. First, economically rich areas are more prone
to habitat disturbances (e.g., dams and reservoirs modifying
river flows) that are known to facilitate the establishment of
non-native species [7,23,29]. Second, high rates of economic
exchanges increase the propagule fluxes of non-native species
[6,9] via ornamental trade, sport fishing, and aquaculture .
Third, the increased demand for imported products associ-
ated with economic development increase the likelihood of
unintentional introductions through the import process .
The ‘‘biotic resistance’’ hypothesis cannot explain the
pattern of fish invasions observed, because no negative
relationship between native and non-native species richness
was found after controlling for the effects of environmental
conditions, propagule pressure, and habitat disturbance. This
means that regional species-rich communities are not
necessarily a barrier against the establishment of non-native
species . Our results are consistent with several studies
showing that species-rich fish communities can support
higher species richness if the pool of potential colonisers is
increased by species introductions [24,30,31]. More generally,
our results agree with studies on various taxa that do not
report biotic resistance at broad spatial scales [10,11]. Then,
we provide no real support for the alternative ‘‘biotic
acceptance’’ hypothesis  even if native and non-native
species richness do respond similarly to some of the
environmental gradients tested (i.e., altitudinal range and
basin area). Actually, the absence of a significant positive
relationship between native and non-native species richness
implies that species-rich river basins do not support more
non-native species than basins with a low native species
richness (i.e., ‘‘the rich do not get richer’’). This contrasts with
Figure 1. Worldwide Distribution of Non-Native Freshwater Fish
(A) The percentage of non-native species per basin (i.e., the ratio of non-native species richness/total species richness) and (B) the non-native species
richness per basin. Each basin was delimited by a GIS using 0.5830.58 unit grid. The maps were drawn using species occurrence data for 9,968 species
in 1,055 river basins covering more than 80% of continental areas worldwide. Invasion hotspots are defined as areas where more than a quarter of the
species are non-native (red areas on map (A)), leading to define six invasion hotspots: the Pacific coast of North and Central America, southern South
America, western and southern Europe, central Eurasia, South Africa and Madagascar, southern Australia, and New Zealand.
Figure 2. Percentage of Threatened Species for the Three Invasion Levels
Threatened species were identified from the IUCN Red List (vulnerable,
endangered, critically endangered). We calculated the percentage of
threatened species, listed in the IUCN Red List, for the three invasion
levels considered in Figure 1A. Each invasion level expessed as the
percentage of non-native species. ([ 0%–5% ], ]5%–25%], ]25%–95%])
account for 8,363, 2,257, 1,241 native species and 544, 240, 271 river
PLoS Biology | www.plosbiology.org February 2008 | Volume 6 | Issue 2 | e280003
Fish Invasions in the World’s River Systems
numerous continental and regional-scale studies on plants
and animals that report a strong matching between native
and non-native species richness [10,32–35]. More generally,
our results do not agree with the expectation that native and
non-native species richness covary positively at macroeco-
logical scales .
The interpretation of the exact role of human activities
(i.e., propagule pressure and habitat disturbance) in driving
broad-scale patterns of non-native species richness faced
major difficulties in previous continental and regional-scale
studies due to covariations between human and natural
factors [9,13,34,35]. Indeed, because humans may have
preferred to settle in areas providing diverse natural
resources, human population was found to be largest in
regions with high levels of habitat heterogeneity and energy
availability that favour species-rich native fauna and flora
[34,37]. This therefore makes it difficult to determine whether
the often-reported positive relationship between native and
non-native species richness is driven by (i) common responses
to habitat heterogeneity and energy availability or (ii)
increased propagule pressure and habitat disturbance. Such
difficulties were probably related to the spatial extent
considered (i.e., a continental or regional extent). Indeed,
we found a weak covariation between environmental and
human descriptors of the world’s river basins at the global
scale (Pearson’s correlation coefficients: r , 0.35, Table S1).
This allowed us to clearly disentangle the relative roles of
human activities and environmental conditions in shaping
the global pattern of fish invasions. We show that the
biogeography of fish invasions at the global scale matches
the geography of human impact but not the biogeography of
Because increasing the number of non-native species
increases the risk of biodiversity loss [4,5], our results have
two major implications for future conservation strategies.
First, the six global invasion hotspots identified here account
for the highest proportion of threatened fish species listed on
the IUCN Red List . These areas are also recognised as
being biodiversity hotspots (particularly southern Europe,
South Africa and Madagascar, southern Australia, and New
Zealand [38,39]). Although species classified on the IUCN Red
List are threatened by various sources of disturbance (e.g.,
habitat loss, pollution, species invasion, and overexploitation
), non-native species are recognised as a major threat to
biodiversity after habitat loss [25,40]. For example, 20% of the
680 species extinctions listed by the IUCN were directly
caused by species invasions . Freshwater fish follow the
same tendency, as 20% of the species listed by the IUCN are
Table 1. Spearman Rank Correlation (rs) between the Number of Non-Native Fish Species (Residuals) and Each Explanatory Variable
Related to the ‘‘Human Activity,’’ ‘‘Biotic Acceptance,’’ and ‘‘Biotic Resistance’’ Hypotheses (n ¼ 597)
Human activity hypothesis
Gross domestic product
Percentage of urban area
Number of native species
Net primary productivity
Number of native species
Biotic acceptance hypothesis
Biotic resistance hypothesis
For each hypothesis, the relationship between the number of non-native fish species and the explanatory variables considered was quantified by controlling for the effects of the
explanatory variables relevant to the other hypotheses (see Materials and Methods for more details).
* p , 0.006 (Bonferroni correction, a ¼ 0.006).
Table 2. Independent Effect of Each Environmental and Human Activity–Related Variable on the Number of Non-Native Species per
Independent Effect (%) (n ¼ 597) 95% Boostrap Confidence Interval (n ¼ 100)
Gross domestic product
Percentage of urban area
Number of native species
Net primary productivity
Hierarchical partitioning was applied to the 597 basins for which the seven variables selected to test the ‘‘human activity,’’ ‘‘biotic acceptance,’’ and ‘‘biotic resistance’’ hypotheses were
available. The independent effect of a variable was expressed as a percentage of the total independent contribution associated with the seven variables. To test potential bias due to
sample size, hierarchical partitioning was run on 1,000 random subsets of 100 basins among the total of 597 basins. For each variable, the independent effect based on 597 basins did not
differ from the 95% bootstrap percentile confidence interval, testifying that sample size hardly affected the results. Both analyses underline the predominant role of the three human
variables that together represent more than 70% of the independent effect.
PLoS Biology | www.plosbiology.org February 2008 | Volume 6 | Issue 2 | e280004
Fish Invasions in the World’s River Systems
threatened by non-native species . In that context, we
recommend that non-native species importations in the six
invasion hotspots be prohibited without detailed risk and
long term cost-benefits assessments . Special attention
should also be given to these areas to design efficient control
programs of already-established non-native species.
Second, as we provide strong evidence for the ‘‘human
activity’’ hypothesis (with a special emphasis on economic
activity), we expect that river basins of developing countries
will host an increasing number of non-native fish species as a
direct result of economic development. This constitutes a
serious threat to global biodiversity, because rivers of most
developing areas (e.g., southern Asia, western and central
Africa) are characterised by high levels of endemism .
Anticipating potential biodiversity threats should therefore
be a priority, because once they are established, the
eradication of a non-native species is extremely difficult
and result in high economic costs .
Despite the increasing literature on non-native species, this
study is, to our knowledge, the first to provide a global map of
species invasions for a given taxonomic group and should
stimulate others to test the generality of these findings for
other taxa at this spatial scale. Such broad-scale analyses
would help local researches to focus on non-native species
control in the most sensitive areas (e.g., the six invasion
hotspots we identified here for freshwater fish). This study
should also stimulate researches on freshwater ecosystems by
combining the existing global scale databases of physical
disturbances [44,45] and the global pattern of fish invasions
given here. This would permit to quantify river basins threats
by considering simultaneously different sources of disturb-
ance. Such an approach is urgently needed as rivers are
among the most threatened ecosystems of the world  and
as freshwater fish constitute a major source of protein for a
large part of the world population .
Materials and Methods
Databases. We conducted an extensive literature survey of native
and non-native freshwater fish species check lists. Only complete
species lists at the river basin scale were considered, and we discarded
incomplete check lists such as local inventories of a stream reach or
based only on a given family. The resulting database was gathered
from more than 400 bibliographic sources including published
papers, books, and grey literature databases (references available
upon request). Our species database contains species occurrence data
for the world’s freshwater fish fauna at the river basin scale (i.e., 80%
of all freshwater species described  and 1,055 river basins covering
more than 80% of Earth’s surface). It constitutes the most
comprehensive global database for freshwater fish occurrences at
the river basin scale and, to our knowledge, the largest database for a
group of invaders. We considered as non-native a species (i) that did
not historically occur in a given basin and (ii) that was successfully
established, i.e., self-reproducing populations. Estuarine species with
no freshwater life stage were not considered in our analyses.
The environmental and human databases contain seven variables
selected to test (i) the ‘‘human activity’’ hypothesis: human
population density (number of people km?2), percentage of urban
area and purchase power parity GDP (in US$); (ii) the ‘‘biotic
acceptance’’ hypothesis: number of native fish species, basin area
(km2), altitudinal range (m), net primary productivity (NPP in kg-
carbon m?2year?1), and (iii) the ‘‘biotic resistance hypothesis’’:
number of native fish species. The area of each river basin was taken
from published and unpublished data. The altitudinal range for each
river basin was determined from a geographical atlas. We calculated
the mean value of NPP, human population density, GDP, and
percentage of urban area over the surface area of each basin from
0.58 3 0.58 grid data available in the Center for International Earth
Science Information Network (CIESIN) and the Atlas of Biosphere
[48,49]. The surface area and altitudinal range at the river basin scale
are used as quantitative surrogates for habitat heterogeneity ,
which is known to influence native freshwater fish species richness
[15,16]. Net primary productivity is used as a quantitative surrogate
to river basin energy availability  and strongly correlates to native
freshwater fish species richness [15,16]. This is verified in our data, as
we found that both basin area and NPP are positively correlated to
native species richness (partial Pearson’s correlation coefficient: r ¼
0.592 and p , 0.0001 for basin area while controlling for the effect of
NPP; r¼0.514 and p , 0.0001 for NPP while controlling for the effect
of the basin area). Then, the human population density, percentage of
urban area, and GDP were used as quantitative surrogates for
propagule pressure and habitat disturbance [5,9,33]. The GDP
measures the size of the economy and is defined as the market value
of all final goods and services produced within a region in a given
period of time.
Fish invasions mapping. We first mapped the worldwide distribu-
tion of (i) the non-native species richness per basin and (ii) the
percentage of non-native species per basin (i.e., the ratio of non-
native species richness/total species richness). To do that, each basin
was delimited by a geographic information system (GIS) using a grid
reference of 0.58 latitude and 0.58 longitude and then reported on a
world map. We used three classes of percentage (Figure 1A) and
richness (Figure 1B) of non-native species to draw colour maps. Other
maps with more classes were tried and provided similar results. We
selected the one that minimised differences in sample size (i.e.,
number of river basins) between classes. The percentage of non-
native species per basin was used to define invasion hotspots where
more than a quarter of the species are non-native (i.e., the third class
of percentage of non-native species; red areas in Figure 1A). It was
preferred to the richness in non-natives due to its independence
from native richness and basin area. For each of the three levels of
fish invasion ([ 0%–5% ], ]5%–25%], ]25%–95% ]), we determined the
percentage of species facing a high to extremely high risk of
extinction in the wild, i.e., the vulnerable, endangered, and critically
endangered fish species according to the IUCN Red List . The
percentage of threatened species should be regarded with caution,
because the IUCN Red List for freshwater fish is still incomplete. The
percentages of threatened species for the three levels of fish invasion
are therefore probably underestimated. Although we recognise the
potential biases and limitations of the IUCN listing procedure, the
IUCN Red List of threatened species remains the most objective and
authoritative system for classifying species in terms of the risk of
extinction at the global scale [41,50]. The list of basins for the three
levels of invasion is provided in Dataset S1.
Modelling method. In this study, to test the three hypotheses (i.e.,
‘‘human activity’’, ‘‘biotic acceptance’’, and ‘‘biotic resistance’’), we
did not build the best single and parsimonious model by using
stepwise selection of a subset of independent variables having a
significant effect on the number of non-native species per basin (i.e.,
predictive approach). Indeed, a single best model is not necessarily
the best explanatory model, because minimizing the overall differ-
ence between the observed and predicted values does not necessarily
equate to determining probable influence in a multivariate setting
[26–28,51,52]. In addition, a simple regression model cannot identify
situations in which potentially important independent variables are
suppressed by other variables due to their high colinearity. When
there is colinearity between independent variables, the direct
response of the dependent variable to a independent variable may
in fact only be an indirect effect owing to high dependence of the
considered variable with one or many others .
In our dataset, the seven environmental and human variables are
not independent (Pearson’s correlation coefficient values ranging
from ?0.25 to 0.79, Table S1). We therefore evaluated the
independent explanatory power of each environmental and human
variable by using hierarchical partitioning [26–28,51,52], a method
based on the theorem of hierarchies in which all possible models in a
multiple regression setting are considered jointly to attempt to
identify the most likely causal factors (explanatory approach).
If we consider k, the number of explanatory variables (X1, Xi,...,
Xk), there are 2kpossible models (i.e., 128 submodels by considering
the seven explanatory variables), including the null model (M0). The
Riis a measure of fit between one independent variable Xiand the
dependent variable Y. The fit between each of the seven explanatory
variables and the dependent variable Y (number of non-native fish
species per basin) was measured by the reduction of deviance
generated by introducing a given variable into all of the possible
models built with the six other variables within the considered
hierarchies. We used a generalised linear model (GLM) with a Poisson
error to treat our count data (i.e., the number of non-native fish
PLoS Biology | www.plosbiology.org February 2008 | Volume 6 | Issue 2 | e280005
Fish Invasions in the World’s River Systems
species per basin). Each explanatory variable was log-transformed to
meet the assumptions of normality and homoscedasticity.
We consider k! hierarchical orderings of models that always begin
with M0and end with Mx1;2;3:::k. For any given initial variable Xi, there
are (k – 1)! possible hierarchies containing k(k – 1)! models in which Xi
appears. For each hierarchy, we evaluate the influence of Xion each
of the k models including Xi (increase in model fit generated by
including the variable Xi within each model). The independent
influence (Ii) of Xion Y was obtained by averaging all of the k(k – 1)!
increases of fit. This averaging alleviates multicolinearity problems
that are ignored by using a simple regression model [26–28]. The joint
component Ji(effect caused jointly with the k – 1 other variables) is
obtained by subtracting Iifrom Ri, with Ri¼ Iiþ Ji. If all explanatory
variables were completely independent of one another, there would
be no joint contributions . For each variable, the independent
and joint contributions are expressed as the percentage of the total
explained deviance (R)
R ¼ I þ J ¼
In our models, the total independent contribution accounts for
75% of the total explained deviance, which means that the joint
contribution of each explanatory variable was weak in explaining the
global variation in non-native species richness (Figure S1). We
therefore quantified the independent effect (IEi) of each variable on
the dependent variable Y as the percentage of the total independent
contribution, i.e. IEi¼ Ii=Rk
effect (IEi) of each variable was determined by a randomization
approach (n¼100) which yielded Z-scores . Statistical significance
was based on an upper confidence limit of 0.95. Each variable display
a significant independent effect.
We applied hierarchical partitioning to a subsample of 597 basins
(Afrotropical: 72; Australian: 94; Nearctic: 127; Neotropical: 68;
Oriental: 29; Palearctic: 207) for which all seven environmental and
human variables used were available. To test potential bias due to
differences in sampling effort between continents, hierarchical
partitioning was run on 1,000 random subsets of 100 basins among
the total of 597 basins. For each variable, we calculated the 95%
bootstrap percentile confidence interval of the independent effect
(IEi). Hierarchical partitioning was conducted using the ‘hier.part’
package  version 1.0–1 implemented on the open source R
software . Hierarchical partitioning implemented for linear
relationships was relevant to our data, because preliminary analyses
did not detected any significant effect of polynomial terms. The
hierarchical partitioning results were compared with those obtained
with another method (i.e., variation partitioning, ). Overall, the
results of the two methods were similar, and the variables highlighted
as significant by the two approaches were the same.
Hierarchical partitioning does not provide information on the
form of the relationship (positive or negative) between the number of
non-native species and each explanatory variable. To test the ‘‘human
activity’’ hypothesis, we analysed the form and the significance of the
relationship between each variable related to the ‘‘human activity’’
hypothesis (GDP, percentage of urban area, and population density)
and the residuals from a GLM with a Poisson error. This model
explains the number of non-native species by using independent
variables related to the ‘‘biotic resistance’’ and ‘‘biotic acceptance’’
hypotheses (number of native species, altitudinal range, basin area,
and net primary productivity). This allowed us to control for the
effects of environmental conditions and native species richness.
Then, to test the ‘‘biotic acceptance’’ hypothesis, we analysed the
form and the significance of the relationship between each variable
i¼1Ii:The significance of the independent
related to the ‘‘biotic acceptance’’ hypothesis (i.e., number of native
species, altitudinal range, basin area, and net primary productivity)
and the residuals from a GLM explaining the number of non-native
species by using the human activity–related variables (i.e., GDP,
percentage of urban area, and population density). This allowed us to
control for the effects of propagule pressure and habitat disturbance.
Lastly, to test the ‘‘biotic resistance’’ hypothesis, we analysed the form
and the significance of the relationship between the number of native
species and the residuals from a GLM explaining the number of non-
native species by using independent variables related to the ‘‘biotic
acceptance’’ and ‘‘human activity’’ hypotheses (i.e., altitudinal range,
basin area, net primary productivity, GDP, percentage of urban area,
and population density). This allowed us to control for the effects of
environmental conditions, propagule pressure and habitat disturb-
ance. To test the relationship between the model residuals and each
explanatory variable, we performed a Spearman rank correlation
test, because the model residuals were not normally distributed.
Dataset S1. Names and Invasion Levels of the 1,055 River Basins
The three invasion levels are those used in Figure 1A (i.e., the
percentage of non-native species per basin). (i) [ 0%–5% ]; (ii) ]5%–
25%]; (iii) ]25%–95% ]. Longitude and latitude at the river mouth was
also provided for the 1,055 river basins.
Found at doi:10.1371/journal.pbio.0060028.sd001 (92 KB XLS).
Figure S1. Results from Hierarchical Partitioning Analysis Illustrating
the Independent and Joint Contributions of the Explanatory
Variables in Accounting for the Variation in Non-Native Species
Richness between River Basins (n ¼ 597)
Values are presented as the percentage of the total explained
deviance extracted from a GLM with a Poisson error. The total
independent contribution of the explanatory variables accounts for
75% of the total explained deviance.
Found at doi:10.1371/journal.pbio.0060028.sg001 (46 KB PDF).
Table S1. Pearson’s Correlation Coefficient (r) between Each
NSR: native species richness; AR: altitudinal range; BA: basin area;
NPP: net primary productivity; GDP: gross domestic product; PUA:
percentage of urban area; PD: population density. Bold values
indicate a significant correlation p , 0.002 (Bonferroni correction,
a ¼ 0.002).
Found at doi:10.1371/journal.pbio.0060028.st001 (52 KB PDF).
We thank J. Chave, E. Danchin, C.R. Townsend, P. Winterton, and
three anonymous referees for their insightful comments, which have
improved the manuscript.
Author contributions. F. Leprieur, O. Beauchard, S. Blanchet, and
S. Brosse conceived and designed the experiments, performed the
experiments, and analyzed the data. F. Leprieur, O. Beauchard, S.
Blanchet, T. Oberdorff, and S. Brosse contributed reagents/materials/
analysis tools and wrote the paper.
Funding. This work was supported by the National Agency for
Research (ANR) Freshwater Fish Diversity (ANR-06-BDIV-010).
Competing interests. The authors have declared that no competing
1.Vitousek PM, Mooney HA, Lubchenco J, Melillo JM (1997) Human
domination of Earth’s ecosystems. Science 278: 494–499.
2.Clavero M, Garcı ´a-Berthou E (2005) Invasive species are a leading cause of
animal extinctions. Trends Ecol Evol 20: 110.
3.Byrnes JE, Reynolds PL, Stachowicz JJ (2007) Invasions and extinctions
reshape coastal marine food webs. PLoS ONE 2: e295. doi:10.1371/journal.
4. Jeschke JM, Strayer DL (2005) Invasion success of vertebrates in Europe and
North America. Proc Natl Acad Sci U S A 102: 7198–7202.
5. Pys ˇek P, Richardson DM (2006) The biogeography of naturalization in alien
plants. J Biogeogr 33: 2040–2050.
6. Levine JM, D’Antonio CM (2003) Forecasting biological invasions with
increasing international trade. Conserv Biol 17: 322–326.
7. Williamson M (1996) Biological invasions. London: Chapman & Hall. 244 p.
8. Puth M, Post DM (2005) Studying invasion: have we missed the boat? Ecol
Lett 8: 715–721.
Taylor BW, Irwin RE (2004) Linking economic activities to the distribution
of exotic plants. Proc Natl Acad Sci U S A 101: 17725–17730.
10. Fridley JD, Stachowicz JJ, Naeem S, Sax DF, Seabloom EW, et al. (2007) The
invasion paradox: reconciling pattern and process in species invasions.
Ecology 88: 3–17.
11. Levine JM (2000) Species diversity and biological invasions: relating local
process to community pattern. Science 288: 852–854.
12. Lockwood JL, Hoopes MF, Marchetti MP (2007) Invasion Ecology. Oxford:
Blackwell Publishing. 304 pp.
13. Chown SL, Gremmen NJM, Gaston KJ (1998) Ecological biogeography of
southern ocean islands: species–area relationships, human impacts, and
conservation. Am Nat 152: 562–575.
PLoS Biology | www.plosbiology.orgFebruary 2008 | Volume 6 | Issue 2 | e28 0006
Fish Invasions in the World’s River Systems
14. Meyerson LA, Mooney HA (2007) Invasive alien species in an era of Download full-text
globalization. Front Ecol Environ 5: 199–208.
15. Oberdorff T, Gue ´gan JF, Hugueny B (1995) Global scale patterns of fish
species richness in rivers. Ecography 18: 345–352.
16. Gue ´guan JF, Lek S, Oberdorff T (1998) Energy availability and habitat
heterogeneity predict global riverine fish diversity. Nature 391: 382–384.
17. Kennedy TA, Naeem S, Howe KM, Knops JMH, Tilman D, et al. (2002)
Biodiversity as a barrier to ecological invasion. Nature 417: 636–638.
18. Lever C (1996) Naturalized Fishes of the World. London: Academic Press.
19. Vander Zanden MJ, Casselman JM, Rasmussen JB (1999) Stable isotope
evidence for the food web consequences of species invasions in lakes.
Nature 401: 464–467.
20. Kolar CS, Lodge DM (2002) Ecological predictions and risk assessment for
alien fishes in North America. Science 298: 1233.
21. McDowall RM (2006) Crying wolf, crying foul, or crying shame: alien
salmonids and a biodiversity crisis in the southern cool-temperate
galaxioid fishes? Rev Fish Biol Fish 16: 233–422.
22. Leprieur F, Hickey M, Arbuckle CJ, Closs G, Brosse S, et al. (2006)
Hydrological disturbance benefits a native fish at the expense of an exotic
fish. J Appl Ecol 43: 930–939.
23. Light T, Marchetti M (2007) Distinguishing between invasions and habitat
changes as drivers of diversity loss among California’s freshwater fishes.
Conserv Biol 21: 434–446.
24. Gido KB, Brown JH (1999) Invasion of North American drainages by alien
fish species. Freshw Biol 42: 387–399.
25. IUCN (2006) 2006 IUCN Red List of Threatened Species. Available: http://
www.iucnredlist.org. Accessed 3 January 2008.
26. Chevan A, Sutherland M (1991) Hierarchical partitioning. Am Stat 45: 90–
27. Mac Nally R (2002) Multiple regression and inference in conservation
biology and ecology: further comments on identifying important predictor
variables. Biodivers Conserv 11: 1397–1401.
28. Heikkinen RK, Luoto M, Kuussaari M, Po ¨yry J (2005) New insights to
butterfly–environment relationships with partitioning methods. Proc R Soc
Lond B 272: 2203–2210.
29. Havel JE, Lee CE, Vander Zanden MJ (2005) Do reservoirs facilitate
invasions into landscapes? BioScience 55: 518–252.
30. Smith SA, Bell G, Bermimgham E (2004) Cross-Cordillera exchange
mediated by the Panama Canal increased the species richness of local
freshwater fish assemblages. Proc R Soc Lond B 271: 1889–1896.
31. Leprieur F, Beauchard O, Hugueny B, Grenouillet G, Brosse S (2008) Null
model of biotic homogenization: a test with the European freshwater fish
fauna. Divers Distrib: In press. doi:10.1111/j.1472–4642.2007.00409.x
32. Stohlgren TJ, Barnett DT, Kartesz JT (2003) The rich get richer: patterns of
plant invasions in the United States. Front Ecol Environ 1: 11–14.
33. Marchetti MP, Light TS, Moyle PB, Viers J (2004) Invasion and extinction in
California fish assemblages: testing hypotheses using landscape patterns.
Ecol Appl 14: 1507–1525.
34. Evans KL, Warren PH, Gaston KJ (2005) Does energy availability influence
classical patterns of spatial variation in exotic species richness? Glob Ecol
Biogeogr 14: 57–65.
35. Chown SL, Hull B, Gaston KJ (2005) Human impacts, energy availability
and invasion across Southern Ocean Islands. Glob Ecol Biogeogr 14: 521–
36. Melbourne BA, Cornell HV, Davies KF, Dugaw CJ, Elmendorf S, et al. (2007)
Invasion in a heterogeneous world: resistance, coexistence or hostile
takeover? Ecol Lett 10: 77–94.
37. Rejma ´nek M (2003) The rich get richer – responses. Front Ecol Environ 1:
38. Moyle PB, Cech JJ (2004) Fishes: An introduction to ichthyology. New
Jersey: Prentice-Hall. 726 p.
39. Wilcove DS, Rothstein D, Bubow J, Phillips A, Losos E. 1998. Quantifying
threats to imperiled species in the United States. BioScience 48: 607–615.
40. Myers N, Mittermeier RA, Mittermeier CG, Da Fonsega GAB, Kent J (2000).
Biodiversity hotspots for conservation priorities. Nature 403: 853–858.
41. Olden JD, Hogan ZS, Vander Zanden JV (2007) Small fish, big fish, red fish,
blue fish: size-biased extinction risk of the world’s freshwater and marine
fishes. Global Ecol Biogeo 16: 694–701.
42. Lodge DM, Williams S, MacIaac HJ, Hayes KR, Leung B, et al. (2006)
Biological invasions: recommendations for U.S. policy and management.
Ecol Appl 16: 2035–2054.
43. Pimentel D, Zuniga R, Morrison D (2005) Update on the environmental and
economic costs associated with alien–invasive species in the United States.
Ecol Econ 52: 273–88.
44. Vo ¨ro ¨smarty CJ, Green P, Salisbury J, Lammers RB (2000) Global water
resources: vulnerability from climate change and population growth.
Science 289: 284–288.
45. Nilsson C, Reidy CA, Dynesius M, Revenga C (2005) Fragmentation and flow
regulation of the World’s large river systems. Science 308: 405–408.
46. Millennium Ecosystem Assessment (2005) Ecosystems and human well
being: current state and trends. Freshwater (Vol 1, Chapter 7). In:
Rijsberman F, Costanza R, Jacobi P, editors. Washington (D.C.): World
Resources Institute. pp. 165–207.
47. Le ˆve `que C, Oberdorff T, Paugy D, Stiassny M, Tedsco PA (2008) Global
diversity of fish (Pisces) in freshwater. Hydrobiologia: In press. doi:10.1007/
48. Center for International Earth Science Information Network (CIESIN)
(2005) Gridded Gross Domestic Product (GDP). Available: http://islscp2.
sesda.com/ISLSCP2_1/html_pages/groups/soc/gdp_xdeg.html. Accessed 3
49. Centre for Sustainability and the Global Environment (SAGE) (2002) Atlas
of the Biosphere. Available: http://www.sage.wisc.edu/. Accessed 3 January
50. Rodrigues ASL, Pilgrim JD, Lamoreux JF, Hoffmann M, Brooks TM (2006)
The value of the IUCN Red List for conservation. Trends Ecol Evol 21: 71–
51. Pont D, Hugueny B, Oberdorff T (2005) Modelling habitat requirement of
European fishes: do species have similar responses to local and regional
environmental constraints? Can J Fish Aquat Sci 62: 163–173.
52. MacNally R, Walsh CJ (2004) Hierarchical partitioning public-domain
software. Biodivers Conserv 13: 659–660.
53. Ihaka R, Gentleman RJ (1996) R: a language for data analysis and graphics. J
Comput Graph Stat 5: 299–314.
54. Borcard D, Legendre P, Drapeau P (1992) Partialling out the spatial
component of ecological variation. Ecology 73: 1045–1055.
PLoS Biology | www.plosbiology.org February 2008 | Volume 6 | Issue 2 | e280007
Fish Invasions in the World’s River Systems