ArticlePDF Available

Abstract and Figures

Communities of vertebrates tend to appear together under similar ranges of environmental features. This study explores whether an explicit combination of vertebrates and their contact rates with a tick vector might constitute an indicator of the prevalence of a pathogen in the quest for ticks at the western Palearctic scale. We asked how ‘indicator’ communities could be ‘markers’ of the actual infection rates of the tick in the field of two species of Borrelia (a bacterium transmitted by the tick Ixodes ricinus). We approached an unsupervised classification of the territory to obtain clusters on the grounds of abundance of each vertebrate and contact rates with the tick. Statistical models based on Neural Networks, Random Forest, Gradient Boosting, and AdaBoost were detect the best correlation between communities’ composition and the prevalence of Borrelia afzelii and Borrelia gariniii in questing ticks. Both Gradient Boosting and AdaBoost produced the best results, predicting tick infection rates from the indicator communities. A ranking algorithm demonstrated that the prevalence of these bacteria in the tick is correlated with indicator communities of vertebrates on sites selected as a proof-of-concept. We acknowledge that our findings are supported by statistical outcomes, but they provide consistency for a framework that should be deeper explored at the large scale.
Content may be subject to copyright.
Is composition of vertebrates an indicator of the prevalence of tick-borne
Agustín Estrada-Peña
and Natalia Fernández-Ruiz
Department of Animal Health. Faculty of Veterinary Medicine, University of Zaragoza, Zaragoza, Spain;
Instituto Agroalimentario de
Aragón (Ia2), Zaragoza, Spain
Communities of vertebrates tend to appear together under similar ranges of environmental
features. This study explores whether an explicit combination of vertebrates and their contact
rates with a tick vector might constitute an indicator of the prevalence of a pathogen in the
quest for ticks at the western Palearctic scale. We asked how ‘indicator’ communities could be
‘markers’ of the actual infection rates of the tick in the field of two species of Borrelia (a
bacterium transmitted by the tick Ixodes ricinus). We approached an unsupervised classifica-
tion of the territory to obtain clusters on the grounds of abundance of each vertebrate and
contact rates with the tick. Statistical models based on Neural Networks, Random Forest,
Gradient Boosting, and AdaBoost were detect the best correlation between communities’
composition and the prevalence of Borrelia afzelii and Borrelia gariniii in questing ticks. Both
Gradient Boosting and AdaBoost produced the best results, predicting tick infection rates
from the indicator communities. A ranking algorithm demonstrated that the prevalence of
these bacteria in the tick is correlated with indicator communities of vertebrates on sites
selected as a proof-of-concept. We acknowledge that our findings are supported by statistical
outcomes, but they provide consistency for a framework that should be deeper explored at
the large scale.
Received 4 November 2021
Accepted 30 December 2021
Keystone communities;
vertebrates; Europe; tick-
borne pathogens; indicator
Communities are sets of organisms that tend to
appear together in ecosystems, from microbes [i.e.
1] to higher vertebrates or plants. Members of
a community interact and can colonize a range of
environmental conditions, with species exhibiting dif-
ferent relative abundances. Interactions among spe-
cies provide a view of the community in equilibrium;
this is the ‘optimal composition and size’ of
a community under a given set of (a)biotic con-
straints. Other than the use of communities of organ-
isms as bioindicators of the health of ecosystems [2],
they can be used to assess the quality of human health
and the predicted future response to changes in cli-
mate [3]. Hawkins et al. [4] stated that the determi-
nants of local biodiversity and variation of organisms
are a central aim of modern ecology, and that a basic
set of temperature, rainfall, or evapotranspiration
‘can account for much of the variation in plant and
animal species diversity across space.’ [4].
It has been recognized that parasites influence host
communities in different ways [5,6] but the opposite
view, i.e. how the vertebrates’ community affects
parasites, has been less addressed [i.e., 7, 8]. There
are excellent examples, generally performed on the
regional scale, about the impact of the community
structure of vertebrates on the faunal composition of
parasites [9,10]. Regarding the effects of vertebrates’
communities on the circulation of vector-borne
pathogens, it has been demonstrated that the relative
composition of communities may influence the pat-
terns of tick-borne pathogen infection rates [sum-
marized by 11]. As an example, it has been pointed
out that some carnivores may play only a small role
in the circulation of the tick-borne pathogen Borrelia
spp, but they are able to influence the density of small
mammals and birds by a predator–prey cascade of
effects [13] on the rodents and birds that are reser-
voirs of the pathogen. The composition of the verte-
brates affects the entire complex of vertebrates-ticks-
pathogens [14,15].
The issue surrounding tick-borne pathogens
revolves around the questions of: (i) whether there
is a combination of vertebrates behind the range of
existing field observations of prevalence of a tick-
borne pathogen, (ii) whether it is a reproducible find-
ing for each ecosystem, and (iii) how the alterations
of these natural communities may affect the circula-
tion of the tick(s) or the pathogen(s). There are
obvious logistical issues in conducting large-scale
CONTACT Agustín Estrada-Peña Miguel Servet 177, Zaragoza 50013, Spain Department of Animal Health, Faculty of
Veterinary Medicine, University of Zaragoza,
Supplemental data for this article can be accessed here.
2022, VOL. 12, 2025647
© 2022 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
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 work is properly cited.
tick surveys [14]. By way of example, a large meta-
study was carried out on the reservoir capacities of
some well-studied reservoirs of tick-borne bacteria of
the genus Borrelia, correlating the abundance of
I. ricinus and the prevalence of B. burgdorferi s.l.
with morphological and physiological traits of the
animals, concluding that ‘few vertebrate species dom-
inate the B. burgdorferi s.l. life cycle’ [16]. Other
studies have conducted research on the significance
of different vertebrates in the transmission rates of
tick-borne pathogens [i.e. 17, 18].
A pending issue in this field of research is the
translation of an explicit composition of vertebrates
into an epidemiologically coherent indicator for the
circulation of a tick-borne pathogen on a large scale.
It must be noted that some vertebrates just feed the
tick, while others contribute to the circulation of
pathogens (acting as reservoirs). The rate of ticks
feeding on reservoirs is a spatially variable feature
generated by the community of available vertebrates
that could dramatically change the prevalence of
a pathogen in ticks. It depends upon the relative
abundance of each vertebrate and its ‘attractiveness’
to the tick, plus the status of the vertebrate as
a reservoir or not. The idea of the importance of
vertebrates’ communities on the complex patterns of
dilution or amplification of prevalence of
B. burgdorferi s.l. in ticks has been discussed and
summarized by 19. These authors enumerated the
issues regarding the impact of vertebrates’ composi-
tion on the infection rates of tick-borne pathogens
and wrote ‘combined with data on host feeding utili-
zation, infection prevalence and duration and magni-
tude of infectiousness [the data on hosts and
reservoirs] could be used to make predictions of
nymphal infection prevalence across space or time.’
The impact of the joint contribution of
a community of vertebrates on the prevalence of
a tick-borne pathogen has been systematically
neglected for large regions. The relative composition
of a community of vertebrates changes along spatial
scales as a response to gradients of environmental
conditions that also impact contact rates with ticks.
Knowledge of the effects of the faunal composition of
vertebrates across a geographical gradient on tick-
borne pathogens would provide an unparalleled fra-
mework, helping to evaluate the relative importance
of vertebrates’ composition on the circulation of tick-
borne pathogens. Such a description of an ‘indicator
community’ could then be expected to describe the
prevalence of a pathogen in questing ticks.
This study aims first to classify the territory of the
western Palearctic into clusters, using the known dis-
tribution of 165 species of vertebrates reported as
hosts for the tick I. ricinus, in a method known as
bioregionalization. Each cluster results from similar
presence and abundance patterns of a set of
vertebrate’s species, statistically different from those
found in other clusters. We calculated both the spe-
cies richness and the phylogenetic diversity of the
communities per cluster. We further correlated the
composition of the communities against the preva-
lence of either Borrelia afzelii or Borrelia garinii, two
major pathogens transmitted by that tick, as a proof-
of-concept, using data about questing nymphs of
I. ricinus at the European scale. The purpose is not
to predict infection rates in ticks in the territory, but
to demonstrate that an indicator community exists,
correlating with the prevalence of a tick-borne patho-
gen even on a large scale. The main novelty of our
approach is the building of spatial regions according
to the vertebrates’ composition, thus proposing
a method to de-correlate the animals’ assemblages
with the abiotic factors.
Material and methods
Following the concepts provided by [20], we refer to
the exercise of obtaining spatial units with a similar
composition of vertebrates as ‘bioregionalization’. It
is a classification technique. This study focuses on the
characterization of clusters that have a similar com-
position of vertebrates in the western Palearctic,
weighted by contact rates with the tick I. ricinus
[21]. As a proof-of-concept, we outlined the indicator
community in selected clusters of the target region
against the infection rate of two species of
B. burgdorferi s.l. in questing I. ricinus nymphs, as
previously compiled and reported [22].
Compiling the reported distribution of
vertebrates and I. ricinus in the target territory
We collected information about the recorded distri-
bution of the vertebrate hosts for I. ricinus in Europe,
with coordinates, originally published by Estrada-
Peña and de la Fuente (2016) (data available at
dryad.2h3f2). These data produced the maps of the
predicted distribution of vertebrates, also available in
the same link. We updated the outcome of that pre-
vious analysis only for the species of vertebrates for
which the number of records reported had increased
by more than 10% since the date of the previous
publication (2016) using new records from GBIF
(, last accessed March 2020).
Since the proof-of-concept of this study uses the
distribution of the tick I. ricinus, vertebrates were
selected to include only those species hosting the
tick (166 species reported). We acknowledge that
this does not reflect the complete distribution of all
the vertebrates in the western Palearctic, but rather,
those that have an impact on the circulation of the
selected pathogens. In total, we handled more than
3 million geo-referenced records of vertebrates and
more than 14,000 records of I. ricinus.
Mapping the distribution of vertebrates
For calculations of the predicted distribution of both
the tick and vertebrates recorded as hosts or reser-
voirs of Borrelia, we used a series of monthly values
of temperature, soil humidity, and water vapor deficit
between the years 1980 and 2018, from the
TerraClimate repository (http://www.climatologylab.
org/terraclimate.html, last accessed March 2020). The
complete time series was summarized as the monthly
average of each variable. Each set of average monthly
values was subjected to harmonic regression. The use
of harmonic regression coefficients has been pre-
viously validated [23] since they are free of the fre-
quent issues of spatial correlation and
multicollinearity between layers. Harmonic regres-
sion produces the best fit for seasonal variability of
each variable, and each regression curve has several
coefficients. We used the first three coefficients of the
harmonic regression for each climate variable as
explanatory variables for predictive mapping (total:
9 explanatory variables).
We independently modeled the presence of each
species using the niche modeling algorithm MaxEnt
integrated in the ‘dismo’ package [24] for R [25].
Models were developed with linear and quadratic
features, using a variable number of background
points (10,000–100,000), 10 replicates per species
were modeled, and 70% of points for training pur-
poses. We used cross-validation to compare the
resulting models. The variable number of background
points was proportional to the number of actual
records. This strategy was implemented because
some species have many records in the target region,
while others are poorly represented. Each model was
replicated 100 times, partitioning the data into repli-
cate folds, with each fold used in turn to test the
model. The regularization multiplier was set to 1.
We evaluated the performance of the models using
the Boyce index implemented in the ‘ecospat’ package
26 for R [25]. We did not use the usual index of the
Area Under the Curve (AUC) since it has received
criticism as being affected by background area.
Considering that we are dealing with vertebrates
that may have a relatively restricted distribution, the
ratio between the size of the background and the
actual vertebrate’s distribution may have an impact
on modeling [27]. Rotllan-Puig and Traveset 28 com-
mented on the rationale behind the Boyce index that
varies between – 1 and +1. The only species of verte-
brate that had poor modeling values according to the
Boyce index was Luscinia luscinia, which was
dropped from the final dataset, resulting in the total
of 165 distribution maps. However, this new model-
ing exercise did not change the original conclusions
on the predicted distribution (maximum change <1%,
recorded for Capreolus capreolus; values of change for
the remaining species well below the 0.5%); therefore,
we continue considering the maps of Estrada-Peña
and de la Fuente (2016) as a valid picture of the
predicted distribution of vertebrates and I. ricinus
over the target territory. The final maps intended
for bioregionalization (see point 2.4) display the
expected distribution of each vertebrate and its envir-
onmental suitability (translated here as ‘abundance’).
We then explicitly addressed the contact rates
between each vertebrate and I. ricinus as reported
[21], calculating the overlap of habitat throughout
the complete target territory on a pixel-by-pixel
basis. This was done in two steps, first using the
function ‘pno’ (predicted niche occupancy) in the
package ‘phyloclim’ [29] for R, following the concepts
by [30], on which the function ‘niche.overlap’ calcu-
lates the percent of the environmental niche that is
shared by any pair of species. The result represents
a measure of the amount of habitat shared by any
tick-vertebrate combination [21].
To note, we did not consider the actual contribu-
tion of each species of vertebrates to support the
feeding of the tick because ‘tick preferences’ to feed
on different vertebrates, obtained from field or
laboratory data, as summarized by [16]. We think
that reliable field data are available for only the
most surveyed vertebrates; they are thus unavailable
for more than 90% of the species included in this
study. Therefore, the inclusion of the empirical data
on host’s preferences by the tick for the few available
species would introduce a distorting variable because
it could not be applied to every pair of associations
tick-vertebrate. The tick preferences towards each
vertebrate were obtained from two previous reports,
derived from the centrality index of a network ana-
lysis on published records of I. ricinus on hosts [21];
the raw files to build the networks are available at
Clustering the distribution of vertebrates into
spatial units
A stack consisting of 165 layers of vertebrates’ raster
maps was used for an unsupervised classification of
the territory. The purpose is to classify the territory
into clusters based on the abundance of each verte-
brate and the contact rates with the ticks by pixel. We
applied an unsupervised classification that used the
aforementioned stack of maps but did not supply any
response data (that is, we did not identify any pixel as
belonging to a particular class). This technique is
useful when we have no prior knowledge of the study
area. We used the k-means clustering algorithm to
process a set of maps that resulted in the bioregiona-
lization. To implement a k-means classification algo-
rithm, the target number of regions (k) was
determined by maximizing the cluster validity index.
The Calinski–Harabasz Variance Ratio Criterion
(VRC) 31 was used to measure within-group and
between-group dispersions. The classification pro-
duced a set of areas representing a unique combina-
tion of vertebrates’ species, their abundance (from
modeling), and contact rates with the tick. Clusters
of the same category are statistically inseparable, and
clusters belonging to different categories are statisti-
cally different. The optimal number of categories for
the target territory was 36, although two of them were
returned as ‘empty clusters’ due either to the absence
of vertebrates or of I. ricinus.
Measuring the phylogenetic diversity of the
We asked if the phylogenetic diversity of each
vertebrates’ community is correlated with certain
traits of the tick presence or abundance (modeled).
The Open Tree of Life is an online phylogenetic
tree of life that is updated by adding published and
curated phylogenetic trees of any organism. The
project integrates these new trees as they are pub-
lished into the mega-tree hosted by the website. It
is thus possible to query the complete mega-tree
and a subset of a number of species or other taxa
ranks to prepare an ad hoc synthetic tree. These
subset trees are not ultrametric, and it is necessary
to calculate the branch length between any pair of
species to have the phylogenetic measures of inter-
est. We used data from the Open Tree of Life
(, last accessed
March 2021) for phylogenetic calculations, acces-
sing its API using the ‘rotl’ package [32] for R.
We calculated the phylogenetic diversity (PD) and
species richness (SR) existing in each cluster of the
target territory using the package ‘picante’ [33] for
R. While SR is a pure count of species, PD estimates
the amount of phylogenetic variability in a cluster
using the sum of the lengths of the branches of the
phylogenetic tree of vertebrates present in that clus-
ter. For the main calculations, we used only the
generic name of each vertebrate to improve the solid-
ness of the outcome. Supplemental material 1
includes the phylogenetic trees of the vertebrates’
species together with the use of these hosts by
I. ricinus and includes a ‘readme’ file (in PDF format)
with information about all the files. Supplemental
material 2 includes these same phylogenetic trees of
vertebrates’ species together with the spatial context
of the target area.
Proof-of-concept: discriminating the prevalence
of Borrelia spp. in questing ticks using the
composition of communities
We aimed to demonstrate that, at a rough scale of
landscape, there is an indicator community of verte-
brates that could describe the infection rates of B. afzelii
and B. garinii in questing ticks. We chose these two
bacteria because they are widely distributed in Europe,
they have different vertebrate reservoirs (in general
terms, birds for B. garinii, rodents for B. afzelii), and
reports point to a role of community composition on
the infection rates 34. It is not possible to develop these
calculations for the complete target territory (due to
lack of data on prevalence in ticks), or to apply the
hypothesis to the points (coordinates) in which
Borrelia spp. have been reported. Therefore, we covered
the target territory with a hexagonal tessellation with
a radius of 0.25º, selecting only those cells in which data
of prevalence in ticks of either B. afzelii or B. garinii (or
both) have been published. The choice of the diameter
is not unintentional: higher cell sizes blurred the results
(many species of vertebrates in the same cell) and smal-
ler cell sizes commonly overfitted the models (too few
species of vertebrates present in the cell). Data on the
distribution of Borrelia spp. in Europe were obtained
from [22], accounting for the prevalence of the patho-
gens in questing ticks. This produced 549 records for
B. garinii, 555 records for B. afzelii, and 319 cells of the
different clusters further used for modeling exercises
(Figure 1)]. No limitations on the sample size of ticks
were introduced as constraints in the selection, aiming
to increase the number of sites to test.
The dataset on prevalence rates of Borrelia in
ticks has been used as published [22]. In short,
the dataset was compiled from scientific literature,
between the years 1990 and 2017. All the details
about the bibliographical query are described in the
original study [22]. To summarize, only data from
molecular detection of the pathogen(s) on questing
nymphs of I. ricinus were used. We excluded every
report of ticks collected while feeding, since it is
not possible to ascertain if the DNA of the patho-
gen was already in the tick or was acquired with
the blood meal. We also excluded field data on
larvae or adults, the former because there is no
transovarial transmission of these pathogens and
the larvae hatch free of them; the latest because
adults are more difficult to collect, and surveys
tend to produce consistently a fewer number of
specimens. As mentioned, only molecular tests
(qPCR, RT-PCR) were used; other tests, like dark
field microscopy or xenodiagnoses (i.e., infection of
naïve hosts allowing ticks collected in the field to
feed), while useful in its own context, are not
comparable among them, thus biasing the statistical
The purpose of modeling is two-fold, namely (i)
to explain the prevalence of Borrelia spp. in quest-
ing for nymphal ticks and (ii) to delineate the
indicator community that shapes the recorded
infection rates in ticks. Modeling was done in
the Orange Programming Environment (which is
freely available from https://orangedatamin
Figure 1. General background of distribution of Borrelia spp. in western Palearctic and the individual sets of clusters used for
further analyses. a: The coordinates of records of B. afzelii in questing nymphal I. ricinus as reported in the published literature.
b: The coordinates of records of B. garinii in questing nymphal ticks as reported in the published literature. For both a and b,
compilation finished in the year 2018. Color and size of the dots mean for the reported prevalence. c: The sites used for
statistical evaluations between the communities of vertebrates and infection rates of Borrelia spp. were calculated. The color of
each point (which is actually an hexagon whose diameter is 0.25º) corresponds to the correlative numbering of the clusters
obtained from the bioregionalization of vertebrates.
Modeling the vertebrates’ communities that might
drive the prevalence of Borrelia in the vector
We purposely chose sites of the same cluster category,
displaying differences in infection rates with other
clusters and with a minimum of 15 independent
surveys of questing ticks. We ultimately selected clus-
ters 32 and 20 for B. afzelii (average reported pre-
valence in questing nymphs 10.5% and 22.4%,
respectively) and clusters 35 and 20 for B. garinii
(average reported prevalence in questing nymphs
7% and 15%, respectively). It was difficult to find
clusters with lower prevalence of Borrelia spp. (i.e.,
lower than 7%) because the scarcity of these records
in published literature (less than seven different sur-
veys) compromising the quality of the statistical out-
comes. It was impossible to select a significant
number of clusters with a reported prevalence of ‘0.’
The mere lack of reporting at a site (thus preva-
lence = 0) could mean that surveys have never been
conducted at that site.
All algorithms for model development were avail-
able in the Orange programming environment.
Supplemental material 3 includes the scripts using
a graphical interface for repeating the modeling exer-
cises or issuing new ones under different conditions.
The ecological meaning of the calculations is also
shown as separate charts in the body of the text. We
used four different modeling approaches to correlate
a given combination of vertebrates with the preva-
lence of Borrelia spp. in questing ticks: (i) Neural
Networks, (ii) Random Forest, (iii) Gradient
Boosting, and (iv) AdaBoost. All are algorithms of
‘regression and classification’ that operate on numer-
ical data to obtain a response. Neural networks are
comprised of node layers, containing an input layer,
one or more hidden layers, and an output layer. Each
node connects to another with a weight and
a threshold. If the output of the node is above the
specified threshold value that node is activated, send-
ing data to the next layer of the network. The specific
combination of ‘on-off’ nodes provides the solution
[35]. For Neural Networks, we used 100 neurons per
hidden layer, the ReLu algorithm, the Adam solver
and 200 iterations. Random Forest, an ensemble
learning method developed by 36, builds a set of
decision trees. Each tree is developed from
a bootstrap sample from training data. For each indi-
vidual tree, an arbitrary subset of attributes is drawn
from which the best attribute for the split is selected.
The final model is based on the majority of votes
from individually developed trees in the forest [37].
For Random Forests, we included 10 trees (i.e., the
number of decision trees will be included in the
forest), five trees to be split, which specifies the num-
ber of attributes that are arbitrarily drawn at each
node at every step of the tree’s development, without
balance of classes, and three replicates of each model.
Gradient boosting [38] is a method for creating an
ensemble that starts by fitting an initial model (e.g.,
a tree or linear regression) to the data. A second
model is then built, focusing on improving predic-
tions where the first model performs poorly. The
combination of these two models is expected to be
better than either model alone. The process is then
repeated many times, each successive model attempt-
ing to correct for the shortcomings of the combined
boosted ensemble of all previous models. For
Gradient Boosting, we used 100 trees with
a learning rate of 0.1. AdaBoost (short for ‘adaptive
boosting’) is a machine-learning algorithm, formu-
lated by 39,that uses learning algorithms and itera-
tively tries to improve the solution in an adaptive way
(tweaking weak learners in favor of those instances
misclassified by previous classifiers.]. For AdaBoost,
we used 50 estimators, a learning rate of 1, the
SAMME.R classification algorithm and a linear
regression lost function.
Detection of indicator communities and infection
rates by Borrelia.
The modeling algorithms mentioned in the previous
point are addressed to predict the prevalence of the
pathogen in the questing tick using the abundance
and contact rates of the tick vector with the verte-
brates. Here, we explicitly asked for the ‘indicator
community’: the subset of vertebrates that better
explains these infection rates, removing species that
have little or no significance in the outcome. Our
purpose is not to state the individual roles of each
vertebrate, but rather, the effects of the whole com-
munity on the observed prevalence of the pathogen(s)
in ticks. We used a ‘rank filter’ based on RReliefF [a
method originally developed by 40]. The filter
employs a stand-alone modeling algorithm to extract
the set of candidate subsets of vertebrates that con-
tribute most to the modeling results. In other words,
the rank filter extracts variables with the highest
impact on the results and promotes them as the
‘best set of vertebrates’ that are behind the observed
prevalence of Borrelia spp. in questing ticks.
As mentioned, Supplemental data 3 includes the
scripts for Orange and the input files necessary for
reproducing the complete set of calculations
explained in section 2.6 of Methods. Interested read-
ers should have a basic knowledge of Orange pro-
gramming environment to reproduce the results.
Clusters of vertebrates and I. ricinus in western
Palearctic follow a gradient of climate
The unsupervised classification of the territory using
the k-means algorithm produced 36 clusters, shown
in Figure 2. Note that clusters are defined by the
composition of 165 species of vertebrates and their
relative abundance weighted by contact rates with
I. ricinus. Clusters 1 and 2 were restricted to the
coldest mountain regions of Scandinavia and either
the vertebrate species targeted in the study or the tick
are predicted to be absent for these areas. According
to the clustering methods, areas with the same color
in Figure 3 have a more similar vertebrate composi-
tion/contact rates within them than among other
areas. Clusters have similar communities of verte-
brates to the level of significance p = 0.05. All clusters
depicted in Figure 2 are statistically different from the
faunal composition of other clusters in terms of ver-
tebrate communities. Clusters of the same color may
be spatially separated by other clusters.
High contact rates of tick and vertebrates are
concentrated in clusters with highest
phylogenetic diversity
Figure 3 represents the calculated environmental suit-
ability/contact rates for vertebrates in clusters of the
target region. The heatmap also includes two dendro-
grams: one for vertebrates (linking those that tend to
appear together) and one for clusters (linking those
that tend to have similar vertebrate composition).
Note that there is a clear gradient of spatial variabil-
ity. Of interest (comparing Figures 2 and 3) is the
poor variability of some territories, including mainly
desert areas of northern Africa and contiguous Asia,
and the higher species diversity in others. These pat-
terns are not only affected by the ‘abundance’ of each
vertebrate but also by the contact rates with I. ricinus,
showing low values in clusters where niche overlap
between tick and vertebrates is low. Note that many
clusters in the territory (grouped mainly in north,
central, and Western Europe) are predicted to carry
large fractions of the complete set of vertebrates in
this study, suggesting both a substantial contact with
I. ricinus in these areas.
We calculated the PD and SR of each cluster
(Figures 4 and 5). The values of both indexes do
not perfectly overlap because species richness is not
the same as phylogenetic diversity: areas of high SR
may have a low PD because existing vertebrates are
phylogenetically close. In general terms, most of the
western Palearctic has a PD higher than 10 (a value
considered high), meaning there is a wide range of
potential, phylogenetically distant hosts for I. ricinus.
Most of Central Europe, the Baltic countries, and
southern Scandinavia, as well as parts of northern
Spain and other mountain ranges (i.e., in Italy or
Romania) displayed a high PD, suggesting large
communities of phylogenetically separated verte-
brates that could interact with the tick. Further on
this, the highest environmental suitability for
I. ricinus overlaps the territories with the highest
PD of vertebrates (R
: 0.897, F: 1287.14, p < 0.05).
Values of SR and PD tend to attenuate in the eastern
range of the map and in northern Africa. Since the
contact rate of the tick with the vertebrates is part of
Figure 2. Clustering and the regions resulting from bioregionalization of the expected distribution of vertebrates in the target
region, the expected distribution of Ixodes ricinus and its niche overlap. The map was obtained using an unsupervised
classification using k-means on the raster maps of the predicted distribution of 165 species of vertebrates and I. ricinus and
calculating the predicted niche occupancy of every pair of combinations vertebrate – tick. The unsupervised classification
returned 36 categories, of which the 1 and 2 are in northern Scandinavia, western Russia, and eastern Turkey, where I. ricinus is
absent. We kept the remaining 34 categories. Colors of the figure are random.
our strict definition of communities, such attenua-
tion of values should be expected because I. ricinus is
predicted to be mostly absent from the mentioned
Correlations between the prevalence of B. afzelii or
B. garinii with the PD or SR of each cluster were far
from significant (PD: R
0.0009 for B. afzelii and
0.0012 for B. garinii; SR: R
0.0014 for B. afzelii and
0.0125 for B. gariniiI; p > 0.8 in both cases) indicating
that the prevalence of the pathogens is not only
correlated with the contact rates of I. ricinus with
any vertebrate. Therefore, the mere co-existence of
large populations of the tick and a high number of
available vertebrates, is not a hallmark for the circu-
lation of the chosen pathogens. This suggests that the
pathogens could be linked to peculiar combinations
of vertebrates feeding the tick.
Proof-of-concept: detecting the communities
driving the prevalence of B. afzelii and B. garinii
in selected clusters
We asked whether indicator communities exist as
the best index of infection rates by either B. afzelii
or B. garinii in I. ricinus. This is not to evaluate
whether each cluster resulting from bioregionali-
zation carries a unique indicator species of verte-
brate shaping high or low values of prevalence of
Borrelia spp. in questing ticks. The goal is to find
communities displaying statistically solid relation-
ships with the patterns of prevalence in ticks. This
analysis cannot be done on single sites belonging
to a given cluster but rather by using all sites
belonging to the same category of clusters. With
these cautionary words, we first concluded that
the infection rates in questing ticks are statistically
different among the clusters, as detected by an
ANOVA test (B. afzelii: F: 77.08; p < 0.0001;
B. garinii: F: 1042.3; p < 0.0001).
The ecological meaning of the proof-of-concept
is schematized in Figure 6. It represents the com-
munities of vertebrates in the three different clus-
ters selected for testing by the modeling
algorithms (whose spatial distribution is shown
in the accompanying maps) expressing the contact
rates with I. ricinus corrected by the hosts prefer-
ences, and the prevalence of Borrelia spp. reported
in questing nymphal ticks. At a first view, it is
difficult to observe a pattern. The task of the
algorithms is twice: (i) evaluate the combinations
of vertebrates’ species to obtain a better correla-
tion with tick prevalence, and (ii) deduce a better
combination of vertebrates’ species that produces
that outcome and display it in a reduced space of
principal components. To note, vertebrates ‘com-
pete’ for the tick, and it may have high contact
rates with a non-reservoir vertebrate (therefore
‘diluting’ the circulation of the pathogen) or with
a prominent reservoir. The modeling must
‘remove the noise’ leaving only the most impor-
tant vertebrates whose joint contribution describes
the field findings.
The performance of the models: explaining the
prevalence of Borrelia spp. through the vertebrates’
The results of the modeling algorithms for each clus-
ter and both B afzelii and B. garinii are included in
Table 1. Several algorithms displayed high reliability,
Figure 3. A heatmap representing the abundance of verte-
brates, weighted by the contact rates with Ixodes ricinus in
the western Palearctic at each cluster of the territory. The
values in the heatmap show two dendrograms, one for the
vertebrates that tend to appear together (left of the figure)
and the other for sites that tend to support similar fauna of
vertebrates (top of the figure). Specific names for every
vertebrate are included, even if the phylogenetic tree of
vertebrates has been made using only generic
accounting for the effect of the combination of some
species on the prevalence of Borrelia spp. in questing
ticks. Some of the algorithms systematically demon-
strated a poorer ability to discern the faunal compo-
sition linked to the prevalence.
Modeling results for the clusters selected as of
high or low prevalence (bold typeface in Table 1)
clearly supports the fact that there is an indicator
community of vertebrates that results in good
correlations with infection rates by Borrelia spp. in
I. ricinus. Gradient Booster provided the best algo-
rithm, with an R
value of more than 0.9 for each
tested condition. Figures 7 and 8 expand the ecolo-
gical explanation of the proof-of-concept and sum-
marize the resulting communities of vertebrates for
each condition (cluster+pathogen), separated in the
reduced space of principal components after the
ranking algorithm. Although these results came
Figure 4. Species richness of vertebrates in the target territory. The value must be interpreted as the number of vertebrates
reported as hosts of I. ricinus that are expected to be present in the territory and available for the tick because they share
portions of the environmental niche.
Figure 5. The phylogenetic diversity of the vertebrates in the target territory. The value must be interpreted as the phylogenetic
diversity of the vertebrates reported as hosts of I. ricinus that are expected to be present in the territory, measured by the
method of Faith. Supplementary Research Data contains the details of the spatial distribution of each cluster, and the use of
portions of the phylogenetic tree of the vertebrates by the tick in each cluster.
from pure modeling, there is agreement between the
species included as most/less prominent in each
indicator community, and the observed field rates
of prevalence of Borrelia spp.
Regarding B. afzelii (Figure 7(a)), sites belonging
to cluster 20 (high prevalence of Borrelia reported)
have as most contributing vertebrates the Yellow-
necked mouse (Apodemus flavicollis) and the Bank
vole (Myodes glareolus). Some Insectivora, ungulates
like the Red deer (Cervus elaphus), the Chamois
(Rupicapra rupicapra) and the Mountain sheep
(Ovis ammon), carnivores, and a few birds are
Figure 6. The modelled suitability of each species of vertebrate included in this study, weighted by the contact rates with the
tick I. ricinus and the preferences of the tick for each host (histograms). Only data for areas included in the categories 20, 32 and
35 are shown, since they represent the most contrasting sites regarding prevalence of Borrelia spp. in the target territory. These
ar ethe sites that were subjected to modeling. A small map at right shows the spatial extension of these territories. All these
data were obtained from Supplemental material 3.
Table 1. Outcome of the modeling algorithms between the vertebrates’ community and the reported prevalence of Borrelia
afzelii and Borrelia garinii in questing nymphs of Ixodes ricinus ticks, including clusters 12 to 36, for which there are available
data. Cluster (left column) is a consecutive numbering of the unsupervised classification carried out on the target territory as
shown in Figure 1(c). The column ‘prevalence’ indicates the averaged reported prevalence of either B. afzelii or B. garinii
together with the number of reported surveys in that cluster in parentheses. Each other column indicates the percent of correct
classification of the prevalence of Borrelia in questing ticks by the regression and classification algorithms, separately for the two
pathogens tested.
B. afzelii
boosting AdaBoost
B. garinii
boosting AdaBoost
12 10.55 (3) Only 3
5.00 (3) Only 3
13 11.18 (17) 0.625 0.481 1 0.997 5.38 (13) 0.607 0.464 1 1
14 16.58 (14) 0.666 0.338 1 0.947 6.02 (14) 0.816 0.359 1 1
15 14.1 (26) 0.522 0.564 1 0.999 6.83 (26) 0.837 0.639 1 0.995
17 9.21 (11) 0.613 0.761 1 0.999 10.5 (11) 0.985 0.927 1 1
18 8.93 (32) 0.778 0.589 1 0.993 8.46 (32) 0.801 0.737 1 0.998
19 3.62 (10) 1 0.601 1 1 14.03 (10) 0.899 0.855 1 0.998
20 10.56 (44) 0.811 0.791 1 0.988 18.61 (44) 0.562 0.691 1 0.998
22 13.45 (9) 1 0.377 1 1 8.83 (9) 1 0.597 1 1
23 7.04 (8) 0.994 0.778 1 0.999 2.67 (8) 1 0.743 1 1
27 3.81 (8) 1 0.426 1 1 8.94 (8) 0.999 0.422 1 1
28 1.09 (5) 0.999 0.133 1 0.555 5.58 (5) 0.994 0.186 1 1
29 21.45 (8) 1 0.487 1 1 20.19 (8) 1 0.543 1 0.989
31 6.18 (8) 1 0.887 1 1 1.06 (8) 0.999 0.777 1 1
32 22.39 (64) 0.443 0.655 0.999 0.999 15.5 (64) 0.598 0.716 1 0.994
33 0 (1) Only 1
1.4 (1) Only 1
34 12.53 (7) 1 0.25 1 1 8.4 (7) 1 0.455 1 1
35 11.13 (33) 0.866 0.685 1 0.982 7.21 (33) 0.941 0.713 1 0.998
36 23.08 (11) 0.890 0.448 1 0.998 12.17 (11) 0.998 0.535 1 1
shown as secondary components of the indicator
community in that cluster. However, the community
defining sites of cluster 32 (low prevalence) is poorer
and dominated by the European pine vole (Microtus
subterraneus), and other species of rodents (i.e., Mus
spp., Apodemus spp.), and Insectivora (i.e., Crocidura
Figure 7. The indicator community of vertebrates projected on the reduced space in two areas of the target territory reporting
different rates of infection by B. afzelii in questing nymphs of I. ricinus.
Figure 8. The indicator community of vertebrates projected on the reduced space in two areas of the target territory reporting
different rates of infection by B. garinii in questing nymphs of I. ricinus.
spp.) that appear distant from the main set of domi-
nant species and closely grouped among them
(Figure 7(b)). There is a strong joint occurrence of
these secondary species with a low contribution to the
reported prevalence of the pathogen in the tick.
A different indicator community was detected for
B. garinii (Figure 8). Sites reporting low-to-medium
prevalence (cluster 35, Figure 8(a)) have several spe-
cies of mammals, which rank prominently as the
main species driving the results of modeling. The
only bird among these dominant species is the short-
toed treecreeper (Certhia brachydactyla). Other spe-
cies of birds, which are reservoirs of the bacterium,
form a community that ranks second in the cluster
definition (top of the chart), and that group together,
out of the main group of dominant vertebrates. Sites
belonging to cluster 20 (highly reported infection
rates by B. garinii in questing ticks, Figure 8B) are
dominated by birds (which are the reservoirs of the
pathogen). Some ungulates and Insectivora are part
of the community. It is of interest to note that the
Chamois, Capreolus capreolus, is detected by the
modeling algorithms as a prominent part of the indi-
cator community (bottom of the chart) but the Red
deer, Cervus elaphus, (top of the chart) is not.
Summarizing, the indicator community of sites
with high prevalence of B. afzelii has a large compo-
nent of Rodentia and Insectivora (its reservoirs), with
birds located in secondary positions. An inverse
situation has been detected at sites with low-
medium prevalence by the pathogen. Clusters with
high prevalence of B. garinii are dominated by their
reservoir birds, with a significant increase in rodents
in clusters in which prevalence is smaller. Carnivora
and Ungulata are always secondary members of the
indicator community because of their role as tick
feeders, contributing to the population of ticks, but
not reservoirs.
We demonstrated that a bioregionalization of the
western Palearctic can be built with an epidemiologi-
cal focus on tick-borne pathogens, based on the dis-
tribution and abundance of hundreds of species of
vertebrates and the contact rates with the tick
I. ricinus. The classification of modeled distribution
maps resulted in clusters reflecting specific combina-
tions of vertebrates and different contact rates with
the tick vector. This could provide a strict determina-
tion of the impact of changing climate conditions on
the predicted distribution of both the tick and the
vertebrates, and thus the contact rates and the result-
ing epidemiological consequences. Previous
approaches have aimed at evaluating the background
behind the prevalence of Borrelia spp. in ticks, con-
sidering only abiotic features (Estrada-Peña et al.,
2011). The effects of the variability of vertebrates on
the infection rates of a tick-transmitted pathogen
have never been addressed at anything beyond the
regional scale, based on the gold standard based on
field or laboratory protocols calculating the preva-
lence of questing nymphs. However, this study pro-
posed a statistical procedure that establishes the
impact of the communities of vertebrates on the
infection rates of Borrelia spp. in questing I. ricinus
nymphs, and that is well correlated with the recorded
situation in the target territory [22]. This study is
a proposal, open to major improvements, that pin-
points an area to be addressed also for other tick-
borne pathogens.
The relative importance of several vertebrates in
the epidemiology of Lyme borreliosis in Europe has
been addressed in literally dozens of studies: not only
the key role of rodents or birds as reservoirs of some
species [i.e., 16, 41, 42] but also the dual role played
by some taxa like large ungulates on the amplifica-
tion/dilution of the pathogen [i.e., 43, 44]. In the
USA, research has been partly focused on the life
history traits of different vertebrates, aiming to find
a correlate of their contributions to the infection rates
of ticks by B. burgdorferi s.l. 45 The list of references
above is far from complete but provides an appraisal
of open debates on the topic. To note, the evaluation
of the effect of the vertebrates’ communities on infec-
tion rates by Borrelia spp. in questing ticks has
already been proposed by Mysterud et al. (2019a)
but using field experiments.
Normally, only a few species of hosts in foci of
tick-transmitted pathogens are studied in the field,
most likely due to the impressive logistical issues
involved in such surveys or because of the difficulty
in collecting scarce vertebrates or protected species
[46]. One of the major issues in conducting separate
modeling of individual species of organisms is that
they may interact in different ways [47,48]. The strat-
egy ‘predict first, then cluster’ as adhered to here,
seems to be a good method when modeling groups
of co-occurring organisms [20]. Interactions among
species are already included in datasets from which
the predicted maps are derived: if two or more spe-
cies compete for a resource and one ends up dis-
placed by the competition phenomena, a lack of
records of the affected species will be noticed when
the competing species is present.
We observed that higher contact rates of I. ricinus
with the vertebrates are well correlated for sites in
which the phylogenetic diversity of vertebrates is
high. This is an important finding since the phyloge-
netic diversity of an area could be important when
species differ in their contribution to the support of
the populations of ticks and pathogens. Thus, the
phylogenetic composition rather than the list of spe-
cies in an ecosystem could be of particular
importance for understanding the always complex
epidemiology of tick-borne pathogens. According to
Webb et al. [49], the interpretation is that ‘the more
distantly related two species are, the greater the like-
lihood that they differ ecologically’ (summarized by
Cadotte et al. [50]). At sites in which phylogenetic
diversity is high, the tick has literally dozens of verte-
brates to feed upon since every site in the environ-
mental niche of vertebrates is suitable for the tick. We
state that the system, under these conditions, may be
highly redundant: the absence of a few key species is
replaced by the presence of others that could not be
significant under a different community composition.
The concept of indicator species has deep ecologi-
cal roots in multivariate statistics (Legendre et al.
[51]). According to Legendre and De Cáceres [52],
a species is ‘an indicator of a group of sites if the
indicator value of the species is the highest for that
group of sites and is statistically significant at
a preselected significance level.’ Our study was
based on that concept (revisited by De Cáceres et al.
[53]) but aimed to pinpoint indicator communities of
vertebrates instead of single species. These indicator
communities are expected to change in both space
and time because of their intrinsic requirements, the
availability of resources, and the occasional replace-
ment of species due to trends of climate [54,55]. We
previously demonstrated that there are highly signifi-
cant statistical differences in the infection rates of
ticks among sites, for both B. afzelii and B. garinii
56. However, climate shapes the occurrence of verte-
brates and delineates the gradients of contact of the
co-occurring species with the vector. This is the intri-
cate niche epidemiology of Borrelia spp. that has been
already elaborated, using a network analysis [21].
We acknowledge the gaps in this study. Some
issues may affect the calculations, such as the sample
size of surveyed questing ticks and reported preva-
lence, the season of the year when the survey was
done, local vegetation conditions or landscape frag-
mentation, or even the method for detecting the
bacterium (e.g., either pools of ticks or specimens
processed individually). Our study also ignored the
specific contributions of individual vertebrates’ spe-
cies to the feeding of the tick or transmission of the
pathogen. This is an important point since these data
are commonly obtained in field or laboratory proto-
cols [11,57]. A method to evaluate the individual
importance of each vertebrate in the circulation of
Borrelia, aimed to replace the field surveys, and that
is based on the entire network of relationships among
vertebrates and the tick has been already proposed
[21]. This procedure corresponds well with the iden-
tification of keystone taxa in ecological studies [58].
Since we are looking for indicator communities, the
centrality of the network derived from the matrix of
interactions among partners looks like the obvious
value to reflect the tick preferences for a host [follow-
ing 58]. However, no studies have yet linked both
field-derived and network-derived data given the
paucity of data for many vertebrates. We consider
that the validation of the network approach is
a necessary step before taking a deeper dive into the
reservoir capacity of tick-borne pathogens by verte-
brates. The key concept is, ‘how does the combina-
tion of several vertebrates affect transmission rates to
a single vector?’ At least for Borrelia spp., factors
allowing speciation are commonly linked to the phy-
logeny of the reservoir 59,60–63) which fully supports
our selection of species and comments on reservoirs.
Results are consistent with the epidemiology of the
pathogen, circulating only among some of the verte-
brates that feed the tick vector. It is nevertheless of
interest to note that the ranking of the vertebrates of
each indicator community matches the current
knowledge of the most common reservoirs of tested
species of Borrelia, and points to the dual role played
by Carnivora or Ungulata [12]. Most importantly,
variability of the communities, in either species com-
position or abundance, is detected as key factors
shaping infection rates in ticks, even if recognized
reservoirs are present, but ranked as less important
in the community. We think that results are compa-
tible with the knowledge of the ecology of Borrelia
spp. and their reservoirs. There is still considerable
room for improvement of the methods, linking net-
work approaches with spatial modeling and ranking
algorithms as well as the basic assumptions, but
results are encouraging.
It did not escape our attention that the highest
infection rates of both species of Borrelia in the
three selected groups of clusters have always been
found in a more fragmented area (compare, i.e. the
spatial fragmentation of clusters in Figure 6). The
effect has not been conceptualized in our models
since we focused on a purely biological approach
regarding the community composition of vertebrates,
and because the scale of work would not allow to
capture these fine differences. However, the effect has
been mentioned in the literature on the topic and
even pinpointed as one of the most important effects
of preliminary assessments on the distribution of
Borrelia spp. in Europe 56. On a small scale, the
importance of landscape structure has been pointed
out as affecting prevalence of some species of
Borrelia, most probably because of the impact on
the diversity of vertebrates’ communities [64]. The
scale of our study cannot outline or reject these
field studies, but this is an open field that deserves
interesting findings when compared with the prob-
able relative rarefaction of some key vertebrates.
We aimed for an ecologically sound and radically
different approaches to explain the infection rates of
a tick-borne pathogen in the vectors: could
a bioregionalization including contact rates among
ticks and vertebrates be correlated with the prevalence
of Borrelia spp. in ticks? Our analysis added an extra
dimension that may be of interest to the study of the
dynamics of tick-borne pathogens. We anticipate that
a wide-open field of research remains ahead of this
view: just to cite the example of Tick-borne encephalitis
(TBE) that is also transmitted by the same tick and
observes a puzzling pattern of distribution, seasonality
and (re)emergence of foci [i.e., 65]. We hope this
approach can provide innovative ways to approximate
the complex epidemiology of many tick-borne patho-
gens using a synthetic background [42,44,56,66–68].
This study did not receive any specific funds. NFR was
supported by a grant from the Regional Government of
Aragón (Spain). This study is part of the research by the
network of excellence “Zoonoses and Emerging Diseases in
Public Health” (A16_20R) of the Instituto Agroalimentario
de Aragón (IA2, Zaragoza, Spain). Our special acknowl-
edgment goes to Ana Ramo, who downloaded large por-
tions of the data from GBIF, before the year 2016.
Disclosure statement
The authors declare that the research was conducted in the
absence of any commercial or financial relationships that
could be construed as a potential conflict of interest.
Author contributions
AEP conceptualized the work, calculated the networks,
wrote the first versions of the draft, outlined the figures,
obtained, and processed climate data, programmed the
scripts in Orange, and wrote some of the R scripts for
calculation of habitat suitability. NFR did part of the lit-
erature search and updated existing data on climate; she
prepared several figures from scripts in R and wrote parts
of the draft document. Both authors wrote and agreed on
the final version of the manuscript.
The author(s) reported that there is no funding associated
with the work featured in this article.
[1] Konopka A. What is microbial community ecology?
ISME J. 2009;3(11):1223–1230.
[2] Cairns J, McCormick PV, Niederlehner BR.
A proposed framework for developing indicators of
ecosystem health. Hydrobiologia. 1993;263(1):1–44.
[3] Bertollo P. Assessing ecosystem health in governed
landscapes: a framework for developing core
indicators. Ecosystem Health. 1998;4(1):33–51.
[4] Hawkins BA, Field R, Cornell HV, et al. Energy, water,
and broad-scale geographic patterns of species
richness. Ecology. 2003;84(12):3105–3117.
[5] Bordes F, Morand S. The impact of multiple infections
on wild animal hosts: a review. Infect Ecol Epidemiol.
[6] Mouritsen KN, Poulin R. Parasites boosts biodiversity
and changes animal community structure by trait-
mediated indirect effects. Oikos. 2005;108(2):344–350.
[7] Mouritsen KN, Poulin R. Parasitism, community
structure and biodiversity in intertidal ecosystems.
Parasitology. 2002;124(7):101–117.
[8] Watve MG, Sukumar R. Parasite abundance and
diversity in mammals: correlates with host ecology.
Proc Nat Acad Sci. 1995;92(19):8945–8949.
[9] Poulin R. The functional importance of parasites in
animal communities: many roles at many levels?
Int J Parasitol. 1999;29(6):903–914.
[10] Poulin R, Krasnov BR, Mouillot D, et al. The com-
parative ecology and biogeography of parasites. Philos
Trans R Soc B. 2011;366(1576):2379–2390.
[11] Mannelli A, Bertolotti L, Gern L, et al. Ecology of
Borrelia burgdorferi sensu lato in Europe: transmission
dynamics in multi-host systems, influence of molecu-
lar processes and effects of climate change. FEMS
Microbiol Rev. 2012;36(4):837–861.
[12] Hofmeester TR, Krawczyk AI, Van Leeuwen AD, et al.
Role of mustelids in the life-cycle of ixodid ticks and
transmission cycles of four tick-borne pathogens.
Parasit Vectors. 2018;11(1):1–13.
[13] Terraube J. Can protected areas mitigate Lyme disease
risk in Fennoscandia? EcoHealth. 2019;16(2):184–190.
[14] Krasnov BR, Stanko M, Morand S. Host community
structure and infestation by ixodid ticks: repeatability,
dilution effect and ecological specialization. Oecologia.
[15] Linske MA, Williams SC, Stafford III KC, et al. Ixodes
scapularis (Acari: ixodidae) reservoir host diversity
and abundance impacts on dilution of Borrelia burg-
dorferi (Spirochaetales: spirochaetaceae) in residential
and woodland habitats in Connecticut, USA. J Med
Entomol. 2018;55(3):681–690.
[16] Hofmeester TR, Coipan EC, Van Wieren SE, et al.
Few vertebrate species dominate the Borrelia burgdor-
feri s.l. life cycle. Environ Res Lett. 2016;11(4):43001.
[17] Hofmeester TR, Jansen PA, Wijnen HJ, et al.
Cascading effects of predator activity on tick-borne
disease risk. Philos Trans R Soc B. 2017;284:20170453.
[18] Takumi K, Hofmeester TR, Sprong H. Red and fallow
deer determine the density of Ixodes ricinus nymphs
containing Anaplasma phagocytophilum. Parasit
Vectors. 2021;14(1):1–9.
[19] Kilpatrick AM, Dobson AD, Levi T, et al. Lyme dis-
ease ecology in a changing world: consensus, uncer-
tainty and critical gaps for improving control. Philos
Trans R Soc B. 2017;372(1722):20160117.
[20] Hill N, Woolley SN, Foster S, et al. Determining marine
bioregions: a comparison of quantitative approaches.
Meth Ecol Evolut. 2020;11(10):1258–1272.
[21] Estrada-Peña A, Sprong H, Cabezas-Cruz A, et al.
Nested coevolutionary networks shape the ecological
relationships of ticks, hosts, and the Lyme disease
bacteria of the Borrelia burgdorferi (s.l.) complex.
Parasit Vectors. 2016;9(1):1–15.
[22] Estrada-Peña A, Cutler S, Potkonjak A, et al. An
updated meta-analysis of the distribution and preva-
lence of Borrelia burgdorferi s.l. in ticks in Europe.
Int J Health Geogr. 2018;17(1):1–16.
[23] Estrada-Peña A, Estrada-Sánchez A, de La Fuente J.
A global set of Fourier-transformed remotely sensed
covariates for the description of abiotic niche in epi-
demiological studies of tick vector species. Parasit
Vectors. 2014;7(1):302.
[24] Hijmans RJ, Phillips S, Leathwick J, et al. (2020).
dismo: species distribution modeling. R package ver-
sion 1.3-3. Available from: https://CRAN.R-project.
[25] R Core Team (2020). R: a language and environment
for statistical computing. R Foundation for Statistical
Computing, Vienna Austria. Available from: https://
[26] Di Cola, V., Broennimann, O., Petitpierre, B., Breiner,
F. T., d'Amen, M., Randin, C., and Guisan, A. (2017).
ecospat: an R package to support spatial analyses and
modeling of species niches and distributions.
Ecography, 40(6), 774–787.
[27] Estrada-Peña A, Estrada-Sánchez A, Estrada-Sánchez
D. Methodological caveats in the environmental mod-
elling and projections of climate niche for ticks, with
examples for Ixodes ricinus (Ixodidae). Vet Parasitol.
[28] Rotllan-Puig, X., & Traveset, A. (2021). Determining
the minimal background area for species distribution
models: MinBAR package. Ecological Modelling, 439,
[29] Heibl C, Calenge C (2018). phyloclim: integrating
phylogenetics and climatic Niche modeling.
R package version 0.9.5. Available from: https://
[30] Evans MEK, Smith SA, Flynn RS, et al. Climate, Niche
evolution, and diversification of the “Bird-Cage” eve-
ning primroses (Oenothera, Sections Anogra and
Kleinia). Am Natur. 2009;173(2):225–240.
[31] Caliński, T., & Harabasz, J. (1974). A dendrite method
for cluster analysis. Communications in Statistics-the-
ory and Methods, 3(1), 1–27.
[32] Michonneau F, Brown JW, Winter DJ. rotl: an
R package to interact with the open tree of life data.
Meth Ecol Evolut. 2016;7(12):1476–1481.
[33] Kembel SW, Cowan PD, Helmus MR, et al. Picante:
r tools for integrating phylogenies and ecology.
Bioinformatics. 2010;26(11):1463–1464.
[34] Fernández-Ruiz, N., & Estrada-Peña, A. (2020). Could
climate trends disrupt the contact rates between
Ixodes ricinus (Acari, Ixodidae) and the reservoirs of
Borrelia burgdorferi sl?. PloS one, 15(5), e0233771.
[35] Krogh A. What are artificial neural networks? Nat
Biotechnol. 2008;26(2):195–197.
[36] Breiman L. Random forests. Mach Learn. 2001;45
[37] Afanador NL, Smolinska A, Tran TN, et al.
Unsupervised random forest: a tutorial with case
studies. Journal of Chemometrics. 2016;30
[38] Friedman JH. Greedy function approximation:
a gradient boosting machine. Ann Stat. 2001;29
[39] Freund Y, Schapire RE. A decision-theoretic general-
ization of on-line learning and an application to
boosting. J Comput Syst Sci. 1997;55(1):119–139.
[40] Kira K, Rendell LA. The feature selection problem:
traditional methods and a new algorithm.
Association for the Advancement of Artificial
Inteligence. 1992;2:129–134.
[41] Heylen D, Tijsse E, Fonville M, et al. Transmission
dynamics of Borrelia burgdorferi s.l. in a bird tick
community. Environ Microbiol. 2013;15(2):663–673.
[42] Mysterud A, Stigum VM, Linløkken H. How general
are generalist parasites? The small mammal part of
the Lyme disease transmission cycle in two ecosys-
tems in Northern Europe. Oecologia. 2019;190
[43] Fabri ND, Sprong H, Hofmeester TR. Wild ungulate
species differ in their contribution to the transmission
of Ixodes ricinus-borne pathogens. Parasit Vectors.
[44] Mysterud A, Stigum VM, Jaarsma RI, et al.
Genospecies of Borrelia burgdorferi sensu lato detected
in 16 mammal species and questing ticks from
Northern Europe. Sci Rep. 2019;9(1):1–8.
[45] Ostfeld, R. S., Levi, T., Jolles, A. E., Martin, L. B.,
Hosseini, P. R., & Keesing, F. (2014). Life history
and demographic drivers of reservoir competence for
three tick-borne zoonotic pathogens. PloS one, 9(9),
[46] Brunner JL, LoGiudice K, Ostfeld RS. Estimating
reservoir competence of Borrelia burgdorferi hosts:
prevalence and infectivity, sensitivity, and specificity.
J Med Entomol. 2008;45(1):139–147.
[47] Baquero RA, Tellería JL. Species richness, rarity and
endemicity of European mammals: a biogeographical
approach. Biodiversity & Conservation. 2000;10
[48] Mouchet M, Levers C, Zupan L, et al. Testing the effec-
tiveness of environmental variables to explain European
terrestrial vertebrate species richness across biogeographi-
cal scales. PLoS One. 2015;10(7):e0131924.
[49] Webb CO, Ackerly DD, McPeek MA, et al.
Phylogenies and community ecology. Annu Rev Ecol
Syst. 2002;33(1):475–505.
[50] Cadotte MW, Dinnage R, Tilman D. Phylogenetic
diversity promotes ecosystem stability. Ecology.
[51] Legendre P, Mi X, Ren H, et al. Partitioning beta
diversity in a subtropical broad-leaved forest of
China. Ecology. 2009;90(3):663–674.
[52] Legendre P, De Cáceres M. Beta diversity as the var-
iance of community data: dissimilarity coefficients
and partitioning. Ecol Lett. 2013;16(8):951–963.
[53] De Cáceres M, Legendre P, Wiser SK, et al. Using
species combinations in indicator value analyses.
Meth Ecol Evolut. 2012;3(6):973–982.
[54] Maiorano L, Amori G, Capula M, et al. Threats from
climate change to terrestrial vertebrate hotspots in
Europe. PloS One. 2013;8(9):e74989.
[55] Newbold T. Future effects of climate and land-use
change on terrestrial vertebrate community diversity
under different scenarios. Proceedings of the Royal
Society, Series B. 2018;285:20180792.
[56] Estrada-Peña A, Ortega C, Sánchez N, et al.
Correlation of Borrelia burgdorferi sensu lato preva-
lence in questing Ixodes ricinus ticks with specific
abiotic traits in the western Palearctic. Appl Environ
Microbiol. 2011;77(11):3838–3845.
[57] Hovius JW, van Dam AP, Fikrig E. Tick–host–patho-
gen interactions in Lyme borreliosis. Trends Parasitol.
[58] Jordán F, Liu WC, Mike Á. Trophic field overlap:
a new approach to quantify keystone species. Ecol
Modell. 2009;220(21):2899–2907.
[59] Kurtenbach, K., De Michelis, S., Etti, S., Schäfer, S. M.,
Sewell, H. S., Brade, V., & Kraiczy, P. (2002). Host associa-
tion of Borrelia burgdorferi sensu lato–the key role of host
complement. Trends in microbiology, 10(2), 74–79.
[60] Adams B, Walter KS, Diuk-Wasser MA. Host specia-
lisation, immune cross-reaction and the composition
of communities of co-circulating Borrelia strains. Bull
Math Biol. 2021;83(6):1–24.
[61] Huang CI, Kay SC, Davis S, et al. High burdens of
Ixodes scapularis larval ticks on white-tailed deer may
limit Lyme disease risk in a low biodiversity setting.
Ticks Tick Borne Dis. 2019;10(2):258–268.
[62] Ehrmann S, Ruyts SC, Scherer-Lorenzen M, et al.
Habitat properties are key drivers of Borrelia burgdorferi
(sl.) prevalence in Ixodes ricinus populations of decid-
uous forest fragments. Parasit Vectors. 2018;11(1):1–15.
[63] Mechai S, Margos G, Feil EJ, et al. Evidence for an
effect of landscape connectivity on Borrelia burgdorferi
sensu stricto dispersion in a zone of range expansion.
Ticks Tick Borne Dis. 2018;9(6):1407–1415.
[64] Millins C, Dickinson ER, Isakovic P, et al. Landscape
structure affects the prevalence and distribution of a
tick-borne zoonotic pathogen. Parasit Vectors.
[65] Pettersson JH, Golovljova I, Vene S, et al. Prevalence
of tick-borne encephalitis virus in Ixodes ricinus ticks
in Northern Europe with particular reference to
Southern Sweden. Parasit Vectors. 2014;7(1):1–11.
[66] Faith DP. Conservation evaluation and phylogenetic
diversity. Biol Conserv. 1992;61(1):1–10.
[67] Mysterud A, Easterday WR, Stigum VM, et al.
Contrasting emergence of Lyme disease across
ecosystems. Nat Commun. 2016;7(1):1–11.
[68] Mysterud A, Jore S, Østerås O, et al. Emergence of
tick-borne diseases at northern latitudes in Europe:
a comparative approach. Sci Rep. 2017;7(1):1–12.
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Background Several ungulate species are feeding and propagation hosts for the tick Ixodes ricinus as well as hosts to a wide range of zoonotic pathogens. Here, we focus on Anaplasma phagocytophilum and Borrelia burgdorferi ( s.l. ), two important pathogens for which ungulates are amplifying and dilution hosts, respectively. Ungulate management is one of the main tools to mitigate human health risks associated with these tick-borne pathogens. Across Europe, different species of ungulates are expanding their ranges and increasing in numbers. It is currently unclear if and how the relative contribution to the life-cycle of I. ricinus and the transmission cycles of tick-borne pathogens differ among these species. In this study, we aimed to identify these relative contributions for five European ungulate species. Methods We quantified the tick load and collected ticks and spleen samples from hunted fallow deer ( Dama dama , n = 131), moose ( Alces alces , n = 15), red deer ( Cervus elaphus , n = 61), roe deer ( Capreolus capreolus , n = 30) and wild boar ( Sus scrofa , n = 87) in south-central Sweden. We investigated the presence of tick-borne pathogens in ticks and spleen samples using real-time PCR. We determined if ungulate species differed in tick load (prevalence and intensity) and in infection prevalence in their tissue as well as in the ticks feeding on them. Results Wild boar hosted fewer adult female ticks than any of the deer species, indicating that deer are more important as propagation hosts. Among the deer species, moose had the lowest number of female ticks, while there was no difference among the other deer species. Given the low number of infected nymphs, the relative contribution of all ungulate species to the transmission of B. burgdorferi ( s.l. ) was low. Fallow deer, red deer and roe deer contributed more to the transmission of A. phagocytophilum than wild boar. Conclusions The ungulate species clearly differed in their role as a propagation host and in the transmission of B. burgdorferi and A. phagocytophilum . This study provides crucial information for ungulate management as a tool to mitigate zoonotic disease risk and argues for adapting management approaches to the local ungulate species composition and the pathogen(s) of concern. Graphic abstract
Full-text available
We use mathematical modelling to examine how microbial strain communities are structured by the host specialisation traits and antigenic relationships of their members. The model is quite general and broadly applicable, but we focus on Borrelia burgdorferi , the Lyme disease bacterium, transmitted by ticks to mice and birds. In this system, host specialisation driven by the evasion of innate immunity has been linked to multiple niche polymorphism, while antigenic differentiation driven by the evasion of adaptive immunity has been linked to negative frequency dependence. Our model is composed of two host species, one vector, and multiple co-circulating pathogen strains that vary in their host specificity and their antigenic distances from one another. We explore the conditions required to maintain pathogen diversity. We show that the combination of host specificity and antigenic differentiation creates an intricate niche structure. Unequivocal rules that relate the stability of a strain community directly to the trait composition of its members are elusive. However, broad patterns are evident. When antigenic differentiation is weak, stable communities are typically composed entirely of generalists that can exploit either host species equally well. As antigenic differentiation increases, more diverse stable communities emerge, typically around trait compositions of generalists, generalists and very similar specialists, and specialists roughly balanced between the two host species.
Full-text available
Background The density of Ixodes ricinus nymphs infected with Anaplasma phagocytophilum is one of the parameters that determines the risk for humans and domesticated animals to contract anaplasmosis. For this, I. ricinus larvae need to take a bloodmeal from free-ranging ungulates, which are competent hosts for A. phagocytophilum. Methods Here, we compared the contribution of four free-ranging ungulate species, red deer ( Cervus elaphus ), fallow deer ( Dama dama ), roe deer ( Capreolus capreolus ), and wild boar ( Sus scrofa ), to A. phagocytophilum infections in nymphs. We used a combination of camera and live trapping to quantify the relative availability of vertebrate hosts to questing ticks in 19 Dutch forest sites. Additionally, we collected questing I. ricinus nymphs and tested these for the presence of A. phagocytophilum. Furthermore, we explored two potential mechanisms that could explain differences between species: (i) differences in larval burden, which we based on data from published studies, and (ii) differences in associations with other, non-competent hosts. Results Principal component analysis indicated that the density of A. phagocytophilum -infected nymphs (DIN) was higher in forest sites with high availability of red and fallow deer, and to a lesser degree roe deer. Initial results suggest that these differences are not a result of differences in larval burden, but rather differences in associations with other species or other ecological factors. Conclusions These results indicate that the risk for contracting anaplasmosis in The Netherlands is likely highest in the few areas where red and fallow deer are present. Future studies are needed to explore the mechanisms behind this association.
Full-text available
This study addresses the modifications that future climate conditions could impose on the transmission cycles of Borrelia burgdorferi s.l. by the tick Ixodes ricinus in Europe. Tracking the distribution of foci of a zoonotic agent transmitted by vectors as climate change shapes its spatial niche is necessary to issue self-protection measures for the human population. We modeled the current distribution of the tick and its predicted contact rates with 18 species of vertebrates known to act as reservoirs of the pathogen. We approached an innovative way for estimating the possibility of permanent foci of Borrelia afzelii or Borrelia garinii tracking separately the expected spatial overlap among ticks and reservoirs for these pathogens in Europe. Environmental traits were obtained from MODIS satellite images for the years 2002–2017 (baseline) and projected on scenarios for the years 2030 and 2050. The ratio between MODIS baseline/current interpolated climatologies (WorldClim), and the ratio between MODIS-projected year 2050 with five climate change scenarios for that year (WorldClim) revealed no significant differences, meaning that projections from MODIS are reliable. Models predict that contact rates between the tick and reservoirs of either B. garinii or B. afzelii are spatially different because those have different habitats overlap. This is expected to promote different distribution patterns because of the different responses of both groups of reservoirs to environmental variables. Models for 2030 predict an increase in latitude, mainly in the circulation of B. garinii, with large areas of expected permanent contact between vector and reservoirs in Nordic countries and central Europe. However, climate projections for the year 2050 predict an unexpected scenario of contact disruption. Though large areas in Europe would be suitable for circulation of the pathogens, the predicted lack of niche overlap among ticks and reservoirs could promote a decrease in permanent foci. This development represents a proof-of-concept for the power of jointly modeling both the vector and reservoirs in a common framework. A deeper understanding of the unanticipated result regarding the year 2050 is needed.
Full-text available
The pathogens causing Lyme disease are all vectored by generalist tick species found on a wide range of vertebrates, but spatial and annual variation in host use has rarely been quantified. We here compare the load of Ixodes ricinus (the vector) on small mammals and investigate the infection prevalence of Borrelia burgdorferi s.l. (the pathogen) involved in the enzootic transmission cycle of Lyme disease in two contrasting ecosystems in Norway from 2014 to 2016. The most common larval tick host in the eastern region was the bank vole, while the common shrew dominated in the western region of Norway. However, the wood mouse and the bank vole had consistently higher larval tick loads than the common shrew in both ecosystems. Hence, the evidence indicated that species are differently suitable as hosts, regardless of their abundances. The pathogen infection prevalence was similar among small mammal species, but markedly higher in the region with larger small mammal populations and higher tick loads, while the seasonal and annual variation was less marked. Our study indicated that the generalist I. ricinus shows consistent patterns of load on species of small vertebrate hosts, while B. burgdorferi s.l. (B. afzelii) was a true generalist. The similar roles of host species across regions suggest that disease dynamics can be predicted from host community composition, but predicting the role of host community composition for disease dynamics requires a detailed understanding of the different species population limitations under global change.
Full-text available
This Forum article synthesizes the current evidence on the links between predator-prey interactions, protected areas and spatial variations in Lyme disease risk in Fennoscandia. I suggest key research directions to better understand the role of protected areas in promoting the persistence of diverse predator guilds. Conserving predators could help reducing host populations and Lyme disease risk in northern Europe. There is an urgent need to find possible win-win solutions for biodiversity conservation and human health in ecosystems facing rapid global environmental change.
Full-text available
Lyme borreliosis is the most common vector-borne zoonosis in the northern hemisphere, and the pathogens causing Lyme borreliosis have distinct, incompletely described transmission cycles involving multiple host groups. The mammal community in Fennoscandia differs from continental Europe, and we have limited data on potential competent and incompetent hosts of the different genospecies of Borrelia burgdorferi sensu lato (sl) at the northern distribution ranges where Lyme borreliosis is emerging. We used qPCR to determine presence of B. burgdorferi sl in tissue samples (ear) from 16 mammalian species and questing ticks from Norway, and we sequenced the 5S–23 S rDNA intergenic spacer region to determine genospecies from 1449 qPCR-positive isolates obtaining 423 sequences. All infections coming from small rodents and shrews were linked to the genospecies B. afzelii, while B. burgdorferi sensu stricto (ss) was only found in red squirrels (Sciurus vulgaris). Red squirrels were also infected with B. afzelii and B. garinii. There was no evidence of B. burgdorferi sl infection in moose (Alces alces), red deer (Cervus elaphus) or roe deer (Capreolus capreolus), confirming the role of cervids as incompetent hosts. In infected questing ticks in the two western counties, B. afzelii (67% and 75%) dominated over B. garinii (27% and 21%) and with only a few recorded B. burgdorferi ss and B. valaisiana. B. burgdorferi ss were more common in adult ticks than in nymphs, consistent with a reservoir in squirrels. Our study identifies potential competent hosts for the different genospecies, which is key to understand transmission cycles at high latitudes of Europe.
Full-text available
Background Landscape structure can affect pathogen prevalence and persistence with consequences for human and animal health. Few studies have examined how reservoir host species traits may interact with landscape structure to alter pathogen communities and dynamics. Using a landscape of islands and mainland sites we investigated how natural landscape fragmentation affects the prevalence and persistence of the zoonotic tick-borne pathogen complex Borrelia burgdorferi (sensu lato), which causes Lyme borreliosis. We hypothesized that the prevalence of B. burgdorferi (s.l.) would be lower on islands compared to the mainland and B. afzelii, a small mammal specialist genospecies, would be more affected by isolation than bird-associated B. garinii and B. valaisiana and the generalist B. burgdorferi (sensu stricto). Methods Questing (host-seeking) nymphal I. ricinus ticks (n = 6567) were collected from 12 island and 6 mainland sites in 2011, 2013 and 2015 and tested for B. burgdorferi (s.l.). Deer abundance was estimated using dung transects. Results The prevalence of B. burgdorferi (s.l.) was significantly higher on the mainland (2.5%, 47/1891) compared to island sites (0.9%, 44/4673) (P < 0.01). While all four genospecies of B. burgdorferi (s.l.) were detected on the mainland, bird-associated species B. garinii and B. valaisiana and the generalist genospecies B. burgdorferi (s.s.) predominated on islands. Conclusion We found that landscape structure influenced the prevalence of a zoonotic pathogen, with a lower prevalence detected among island sites compared to the mainland. This was mainly due to the significantly lower prevalence of small mammal-associated B. afzelii. Deer abundance was not related to pathogen prevalence, suggesting that the structure and dynamics of the reservoir host community underpins the observed prevalence patterns, with the higher mobility of bird hosts compared to small mammal hosts leading to a relative predominance of the bird-associated genospecies B. garinii and generalist genospecies B. burgdorferi (s.s.) on islands. In contrast, the lower prevalence of B. afzelii on islands may be due to small mammal populations there exhibiting lower densities, less immigration and stronger population fluctuations. This study suggests that landscape fragmentation can influence the prevalence of a zoonotic pathogen, dependent on the biology of the reservoir host. Electronic supplementary material The online version of this article (10.1186/s13071-018-3200-2) contains supplementary material, which is available to authorized users.
One of the crucial choices when modelling species distributions using pseudo-absences and background approaches is the delineation of the background area to fit the model. We hypothesise that there is a minimum background area around the geographical centre of the species distribution that characterises well enough the range of environmental conditions needed by the species to survive. Thus, fitting the model within this geographical area should be the optimal solution in terms of both quality of the model and execution time. MinBAR is an R package that calculates the optimal background area by means of sequentially fitting several concentric species distribution models (SDMs) until a satisfactory model in terms of the included metrics is reached. The version 1.1.2 is implemented for MaxEnt (using either maxnet or the original java program) and uses Boyce Index as a metric to assess models performance. Three case studies are presented to test the hypothesis and assess package's functionality. We show how partial models trained with part of the species distribution often perform equal or better than those fitted on the entire extent. MinBAR is a versatile tool that helps modellers to objectively define the optimal solution.
Areas that contain ecologically distinct biological content, called bioregions, are a central component to spatial and ecosystem‐based management. We review and describe a variety of commonly‐used and newly‐developed statistical approaches for quantitatively determining bioregions. Statistical approaches to bioregionalisation can broadly be classified as two‐stage approaches that either ‘Group First, then Predict’ or ‘Predict First, then Group’, or a newer class of one‐stage approaches that simultaneously analyse biological data with reference to environmental data to generate bioregions. We demonstrate these approaches using a selection of methods applied to simulated data and real data on demersal fish. The methods are assessed against their ability to answer several common scientific or management questions. The true number of simulated bioregions was only identified by both of the one‐stage methods and one two‐stage method. When the number of bioregions was known, many of the methods, but not all, could adequately infer the species, environmental, and spatial characteristics of bioregions. One‐stage approaches however, do so directly via a single model without the need for separate post‐hoc analyses and additionally provide an appropriate characterisation of uncertainty. One‐stage approaches provide a comprehensive and consistent method for objectively identifying and characterising bioregions using both biological and environmental data. Potential avenues of future development in one‐stage methods include incorporating presence‐only and multiple data types as well as considering functional aspects of bioregions.