The Multilayer Nature of Ecological Networks

Shai Pilosof, Mason A. Porter, Mercedes Pascual, Sonia Kéfi

arXiv (arXiv: 1511.04453v3)

Generated on April 19, 2025

 The Multilayer Nature of Ecological Networks

Abstract
Although networks provide a powerful approach to study a large variety of
ecological systems, their formulation does not typically account for multiple
interaction types, interactions that vary in space and time, and interconnected
systems such as networks of networks. The emergent field of `multilayer
networks' provides a natural framework for extending analyses of ecological
systems to include such multiple layers of complexity, as it specifically
allows one to differentiate and model `intralayer' and `interlayer'
connectivity. The framework provides a set of concepts and tools that can be
adapted and applied to ecology, facilitating research on high-dimensional,
heterogeneous systems in nature. Here, we formally define ecological multilayer
networks based on a review of previous and related approaches, illustrate their
application and potential with analyses of existing data, and discuss
limitations, challenges, and future applications. The integration of multilayer
network theory into ecology offers largely untapped potential to further
address ecological complexity, to ultimately provide new theoretical and
empirical insights into the architecture and dynamics of ecological systems.

The Multilayer Nature of Ecological Networks

Shai Pilosof*1, Mason A. Porter2,3,4, Mercedes Pascual1, Sonia K´ efi5
1Department of Ecology and Evolution, University of Chicago, 1103 E 57
St., Chicago, 60637, USA
2Oxford Centre for Industrial and Applied Mathematics, Mathematical
Institute, Oxford OX2 6GG, UK
3CABDyN Complexity Centre, University of Oxford, Oxford OX1 1HP, UK
4Department of Mathematics, University of California, Los Angeles, Los
Angeles, California 90095, USA
5Institut des Sciences de l’Evolution, Universit´ e de Montpellier, CNRS,
IRD, EPHE, CC 065, Place Eug` ene Bataillon, 34095 Montpellier Cedex 05,
France
Corresponding author: Pilosof, S. ( [email protected] )
Key words : Network ecology; Multilayer networks; Metacommunities;
Network of networks; Multilevel networks; Multiplex networks; Temporal
networks
1
arXiv:1511.04453v3 [q-bio.QM] 25 Nov 2016
Preface
Ecological systems are complex and multifaceted in the nature of their in-
teractions. For example, species interact in diverse ways with multiple part-
ners, whose identity can change in space and time. Although networks have
provided fundamental representations of ecological complexity, they require
extension to systematically and simultaneously capture these multifaceted
interactions. We build on recent advances in network science to provide
theoretical and practical guidelines on how to tackle multidimensionality in
ecological networks with multiple “layers”. Multilayer networks hold valu-
able but largely unrealized potential to complement current representations
and provide novel insights into the structure, functioning and dynamics of
ecological systems.
Abstract
Although networks provide a powerful approach to study a large variety
of ecological systems, their formulation does not typically account for mul-
tiple interaction types, interactions that vary in space and time, and in-
terconnected systems such as networks of networks. The emergent field of
‘multilayer networks’ provides a natural framework for extending analyses
of ecological systems to include such multiple layers of complexity, as it
specifically allows one to differentiate and model ‘intralayer’ and ‘interlayer’
connectivity. The framework provides a set of concepts and tools that can
be adapted and applied to ecology, facilitating research on high-dimensional,
heterogeneous systems in nature. Here, we formally define ecological mul-
tilayer networks based on a review of previous and related approaches, il-
lustrate their application and potential with analyses of existing data, and
discuss limitations, challenges, and future applications. The integration of
multilayer network theory into ecology offers largely untapped potential to
further address ecological complexity, to ultimately provide new theoreti-
cal and empirical insights into the architecture and dynamics of ecological
systems.
2

Introduction
Networks provide a powerful approach to explore ecological complexity and
have generated numerous insights into the understanding of the structure,
function, and dynamics of ecological systems1–7. Although ecological net-
works have become fundamental to ecological theory, they have for the most
part been studied in ‘isolation’, as disconnected from other networks, defined
at a single point in space and time, and/or aggregated over multiple spatial
locations and times. However, natural systems typically exhibit multiple
facets of complexity, such as pollinators interacting with flowers in one sea-
son but not in another,8and the same plant species interacting with both
pollinators and herbivores9,10. Despite the recognized need to extend the
investigation of monolayer networks10,11, there are challenges to doing so
explicitly and within a unified framework. Specific examples include the
detection of community structure in networks with edges representing dif-
ferent interaction types10,12and the analyses of resource flow in temporal
networks13.
Recent advances in the theory of multilayer networks14,15provide a
promising approach. A mathematical framework for the analysis of mul-
tilayer networks has been developed only recently, although multilayer net-
work structures, which encode different types of interactions and/or entities
as a single mathematical object, have a long history in subjects like sociol-
ogy and engineering14,16. The study of multilayer networks is providing new
insights into diverse scientific areas14,15, including a better understanding
of the properties of multimodal transportation networks17, effects of cou-
pled infection processes on disease dynamics18, and changes of functional
brain networks during the learning of a motor task19, among others. These
advances, in concert with the growing availability of large ecological data
sets, provide an exciting opportunity for their theoretical and practical inte-
gration into network ecology. In this paper, we define ecological multilayer
networks, give examples of the kinds of insights they can enable, and discuss
challenges and future applications.
Ecological Multilayer Networks
Multilayer networks encompass two or more ‘layers’, which can represent dif-
ferent types of interactions, different communities of species, different points
in time, and so on (Fig. 1). Dependency across layers results from ecological
processes that affect multiple layers. For example, dispersal of individuals
between two patches affects the network structure of both patches20. A mul-
tilayer network consists of (i) a set of physical nodes representing entities
(e.g., species); (ii) a set of layers , which can include multiple aspects of lay-
ering (e.g., both time-dependence and multiple types of relationships); (iii) a
3

set of state nodes , each of which corresponds to the manifestation of a given

physical node on a specific layer; and (iv) a set of (weighted or unweighted)
edges to connect the state nodes to each other in a pairwise fashion. The
edge set includes both the familiar intralayer edges and the interlayer ones,
which connect state nodes across layers. We provide a formal definition with
a detailed example in Box 1.
Previous studies of ecological multilayer networks (Supplementary Ta-
ble 1) have predominantly used multiple but independent networks of the
same system, with interlayer edges formally absent. In such cases, network
diagnostics are calculated independently for each layer. For instance, Olesen
et al.8reported—using a plant–pollinator system sampled over 12 years and
represented with 12 individual networks—that connectance (i.e., edge den-
sity) exhibits little variation over time despite significant turnover of species
and interactions. By contrast, networks with explicit interlayer connectivity
enable one to address questions about interactions between the processes
that operate within and among layers. To set the stage for ecological devel-
opment, we first identify the major types of layering that are relevant for
ecological systems (Fig. 1 and Supplementary Table 1).
Layers Defined Across Space or Time
Early work on spatial and temporal networks focused primarily on how the
composition of species changes in time (e.g., across seasons) or over envi-
ronmental gradients, especially in food webs21–26. More recent studies have
focused on studying spatial and temporal dissimilarities in species and in-
teractions27–29. This variability finds a natural representation in multilayer
networks: one can define a monolayer network at each point in space or time,
and then use interlayer edges to connect each node to its counterparts in dif-
ferent layers. Layers in temporal multilayer networks are typically ordered
(‘ordinal coupling’; Fig. 1a), but the order of the layers is not important
for spatial networks (‘categorical coupling’; Fig. 1b). Another approach is
to use a network of networks. For example, Gilarranz et al.20defined a
spatial network of plant–pollinator networks in which each community of
plants and pollinators is a layer, and in which interlayer edges represent
species extinction and colonization. By considering a network of layers (see
the inset of Fig. 1f), in which each layer is construed as a node, the au-
thors demonstrated an association between the importance of communities
(quantified with a measure of node betweenness centrality) and their archi-
tecture: communities with higher betweenness in the network of layers are
also more nested, with potential consequences for the local stability of the
communities30,31.
4

Multilevel f
Interconnected
(node-colored)e
Shared species
(diagonal coupling)c
Different interaction types
(node-aligned, multiplex)d
Patch 1 Patch 2 Patch 3
Spatial b
Time 1 Time 2 Time 3
Temporal a
N
NFigure 1: Multilayer networks in ecology . In these toy examples, lay-
ers are squares, solid black lines are intralayer interactions, and dashed blue
arcs are interlayer edges. a, Temporal food webs. Nodes are connected to
themselves across layers in an ordinal way (one layer follows another). b,
Spatial food webs. Nodes are connected to themselves across all layers (i.e.,
categorical interconnections). In panels aandb, it is permissible for species
or intralayer interactions to appear in one layer but not in another. c, Net-
works with different interaction types that are connected through shared
species. Layers are ‘diagonally-coupled’, so interlayer edges occur only be-
tween shared species. d, One can represent different interaction types among
a given set of species (e.g., trophic and facilitative) using a node-aligned mul-
tiplex network, in which all nodes appear in all layers and each layer corre-
sponds to a different interaction type. Nodes are connected to all of their
counterparts across layers. e, Two interacting populations of different hosts.
Nodes are individuals, and intralayer and interlayer edges are, respectively,
social ties within and between populations. In this example, each node ap-
pears in one layer. f, A multilevel network representing a metacommunity.
Layers are communities, and nodes are species. Intralayer edges are trophic
interactions, and interlayer edges represent species dispersal among commu-
nities. A species can also disperse to a new community (represented by the
yellow node). Communities are often associated with some space (e.g., with
different habitats or patches). In the inset, we illustrate that such networks
are sometimes represented without explicit specification of interlayer edges.
Interactions among species at a lower level automatically impose interactions
among the communities.5

Layers Defined by Interaction Type

Because the stability and function of ecological networks can depend on the
way in which different interaction types are combined in communities, con-
sidering only a single interaction type can give an incomplete picture of sys-
tem properties10–12,31–35. For instance, Bastolla et al.32illustrated that both
the structure of mutualistic networks and competition for common resources
can determine the number of co-existing species in a system. In another ex-
ample, addition of facilitative interactions to a resource–consumer model
affects total system biomass and hence the way an ecosystem functions11,36.
Finally, Rudolf and Lafferty37showed that food webs that include edges
representing ontogenetic shifts in addition to trophic interactions change
the robustness of the system to extinctions.
Ideally, it is desirable to simultaneously consider the structure of multiple
interaction types, and this is achievable with multilayer networks. Some eco-
logical networks have a common set of species (e.g., a set of plants connected
to their pollinators and herbivores9). One can represent such data9,38,39
in a multilayer formalism using a ‘diagonally coupled’ multilayer network
(Fig. 1c). Each ecological interaction occurs in a different layer, and inter-
layer edges connect common species to their counterparts in other layers.
Another approach is to examine different interaction types between all
species in a system using ‘node-aligned’ multilayer networks, in which all
entities exist on all layers12,35,40,41(Fig. 1d). K´ efi et al.12,35used a highly-
resolved ecological community from the central intertidal coast of Chile to
construct a multilayer network in which each layer includes all species of the
community but represents different interaction types: trophic, non-trophic
positive (e.g., refuge provisioning), and non-trophic negative (e.g., preda-
tor interference). They found that the distribution of non-trophic edges
throughout the food web was different from what would be expected by
chance (by shuffling the non-trophic edges while fixing the trophic web),
suggesting that there is a strong association between the different layers
of the network35. Such structural patterns suggest the possibility of im-
portant dynamic constraints on the combined architecture of trophic and
non-trophic interactions12,35.
Layers Defined by Different Group Identity
An intuitive way to describe and examine variation in individual-based inter-
actions between populations of the same or different species is with an inter-
connected network in which each node appears only in one layer (Fig. 1e). In
disease ecology, this representation models interpopulation or interspecific
disease transmission (when each layer is a population of a different species)
at the same time that it considers underlying social networks42. For exam-
ple, intralayer edges can represent the social structure of bat groups, and
6

interlayer edges, transmission of a vector-borne disease among these pop-

ulations. Moreover, interconnected networks are not limited to individual
organisms. For example, one can define a multilayer network in which each
layer represents a food web (with its own trophic interactions), and inter-
layer edges represent trophic interactions among species from different food
webs43,44.
Layers Defined by Levels of Organization
Biological processes at any given level of organization (e.g., genes, individ-
uals, populations, etc.) can depend on processes at other levels45. For
example, changes in species’ biomass can affect the stability of food webs
in dynamical models based on allometry46. When the layers in a multilayer
network represent different levels of organization, one has a ‘multilevel’ net-
work, and interactions among nodes at a lower level automatically entail
interactions at upper levels14. For example, a trophic interaction between
two species from two different patches implies that there is an interaction
between the patches. The simplest example is a 2-level multilevel network,
which can also be construed as a network of networks (Fig. 1f). In an anal-
ysis of a 3-level multilevel network (population, community, and metacom-
munity), Scotti et al.47illustrated that the metacommunity was sensitive
to population-level processes (e.g., social dynamics) that cascaded through
different levels. The identification of such dependencies is one of the values
of a multilayer approach.
Analyses of Ecological Multilayer Networks
To illustrate the kinds of insights that can be gained from taking a multilayer
approach, we analyse examples of ecological multilayer networks in which
layers are explicitly connected. We consider (i) maximum modularity, a
structural property that is commonly studied in monolayer networks; and (ii)
extinction cascades, a consequence of structure that is common in robustness
analyses of networks. We use both synthetic networks and those constructed
from empirical data.
7

Box 1. Mathematical Definition of a Multilayer Network

Amultilayer network is a quadruplet M= (VM,EM,V,L ). Multilayer
networks can have several ‘aspects’ of layering, and an ‘elementary layer’ is
a single element in one aspect of layering. A ‘layer’ encompasses one choice
of elementary layer for each type of aspect (see the figure for an example).
We include such relationships using sequences L={La}d
a=1of setsLaof
elementary layers, where aindexes the ddifferent aspects. Note that d= 0
for a monolayer network, d= 1 when there is one type of layering, and d= 2
when there are two types of layering (as in the figure). The set of entities
(i.e., physical nodes) is V. The setVM⊆V×L1×···×Ldof node-layer
tuples (i.e., state nodes) encodes the manifestations of an entity v∈Von a
particular layer l∈ˆL=L1×···×Ld.
The edge set EM⊆VM×VM, which includes both intralayer and in-
terlayer edges, encodes the connections between pairs of state nodes. In
a given layer, the intralayer edges encode connections of a specified type
(e.g., a certain type of interaction at a given point in time). A function
w:EM→Rencodes weights on edges. A pair of node-layer tuples, ( u,α)
and (v,β), are ‘adjacent’ if and only if there is an edge between them. One
places a 1 in the associated entry in an adjacency tensor (a generalization of
a matrix that consists of a higher-dimensional array of numbers)14,16if and
only if ((u,α),(v,β)) = 1. Otherwise, one places a 0 in the corresponding
entry. One can ‘flatten’ such an adjacency tensor into a matrix, called a
‘supra-adjacency matrix’, with intralayer edges on the diagonal blocks and
interlayer edges on the off-diagonal blocks (see Supplementary Fig. 1b).
Constraints on the above general definition restrict the structure of a
multilayer network14. For example, ‘diagonal coupling’ (see Fig. 1c) is a con-
straint in which the only permissible type of interlayer edge is one between
counterpart entities on different layers. See Kivel¨ a et al.14for additional
definitions and important types of constraints on Mthat produce common
types of multilayer networks.
XAB
Y12
31
2
55
1
21 (2,A,X)
(1,A,X)(3,A,X)(2,B,X)
(4,B,X)(1,B,X)(4,B,Y)
(5,B,Y)
(5,A,Y)
(2,A,Y)
(1,A,Y)b
446
(6,B,Y)a
8

Toy example of a multilayer network .a, The network has d= 2

aspects: (i) different types of ecological interactions, with trophic interac-
tions are in the blue layer ( A) and host–parasite interactions in the orange
layer (B); and (ii) space, where XandYrepresent different patches. The
elementary-layer set for ecological interaction types is L1={A,B}), and
the one for patches is L2={X,Y}). A layer consists of a tuple of elemen-
tary layers. For example, the layer ( A,X ) encodes trophic interactions at
patchX. We show intralayer edges using solid arrows. We depict diago-
nal interlayer edges (e.g., between node 2 on layer ( A,X ) and node 2 on
layer (B,X )) with dotted lines; such edges encode the extent to which a
parasitized species is more susceptible to predation than a non-parasitized
one. Interlayer edges between patches represent dispersal; we show them
with dashed arcs. b, The ‘supra-graph’ that corresponds to the multilayer
network in panel a. Each node in this graph is a node-layer tuple (i.e., a
state node) in the corresponding multilayer network. See Supplementary
Fig. 1 for an example of how to represent a similar multilayer network as
a supra-adjacency matrix. [This figure is modelled after Fig. 2 of Kivel¨ a et
al.14.]
Modularity
In monolayer networks, maximizing modularity can help quantify the ex-
tent to which a network is organized into groups (modules) of species that
interact more strongly with each other than with other species48,49. To
our knowledge, modularity has not yet been explored in ecological networks
that incorporate interlayer connectivity. To illustrate the distinction be-
tween studying a multilayer network and studying a collection of networks,
we start with a synthetic example from Fontaine et al.10. In this example,
a plant–herbivore network and a plant–root-parasite network interconnect
via a common set of species, which are the plants (Fig. 2a). Interlayer edges
connect each plant species to its counterpart in the other layer (Figs. 1c
and 2c,d) and represent the extent to which parasitism affects herbivory.
Thus, each plant appears in both layers and has two instances corresponding
to different ‘state nodes’, which can be assigned to different modules. See
Supplementary Note 1 for details on how we calculate multilayer modularity
and assign state nodes to modules.
The propensity of a state node to belong to distinct modules depends
on the relative weights of interlayer and intralayer edges50,51. Ecologically,
it depends on the extent to which processes in one layer affect those in
the other layer. To develop our argument, we consider three conceptual
scenarios:
1. When the interlayer edge weights are 0, the two networks are indepen-
9

dent entities. The instances of plants in different layers must belong

to different modules, and modules are defined separately for herbivory
and parasitism. Hence, herbivory has no effect on parasitism (and
vice versa), and no perturbation can pass from one layer to another
(Fig. 2b).
2. At the other extreme, interlayer edge weights are much larger than in-
tralayer ones . (They are infinite in the limiting case.) In this scenario,
herbivory and parasitism always affect each other. This implies that
herbivory always renders plants more susceptible to parasites, and vice
versa. (Note that it does not imply that each plant is always para-
sitized and preyed upon.) Consequently, each of the two instances of
a plant always belongs to the same module. Note that modules can
contain species from any of the three guilds and from either interac-
tion type (Fig. 2d). We also remark that considering infinite interlayer
edge weights produces in general different results from aggregating a
multilayer network into a monolayer network.
3.Interlayer and intralayer edge weights have comparable values (i.e.,
they are on similar scales). In this scenario, herbivory has some effect
on the propensity of a plant to be parasitized, and vice versa. A
plant can therefore interact strongly with a given set of herbivores in
one layer and with a given set of parasites in another. Each of the
two instances of a plant can belong to different modules. Modules
can contain species from any of the three guilds and either interaction
type, but the identity of the modules can be rather different from those
in the previous two cases (Fig. 2c).
This example illustrates that defining community organization in multi-
layer networks depends strongly on the extent to which the ecological pro-
cesses that operate in the different layers affect each other (in our example,
the relationship between parasitism and herbivory). Another insight is that
considering intermediate values of interlayer edge weights provides a pos-
sible means to identify the plants that can buffer perturbations: the two
instances of such plants would be assigned to different modules (Fig. 2c).
One challenge is to quantify the interlayer edges. In this example, one
way to measure the extent to which herbivory and parasitism affect each
other is to conduct a series of experiments in which one group of plants of
a given species is exposed to herbivores while a second group of the same
species (control) is not. For example, if a plant exposed to a given herbivore
species is infected by twice the number of parasites compared to the control
plant, then the value of the interlayer edge is 2. One can then average
across a series of experiments of different herbivore–parasite combinations
to obtain the values of the interlayer edges for that plant species.
10

=1000=0
=0.5Figure 2: Modularity maximization in a diagonally-coupled mul-
tilayer network .a, An example network (from10) that does not have
interlayer edges. In panels b,c, and d, the network has two layers, which
interconnect via a common set of nodes (with thick borders). Interlayer
edges (in blue) connect the two instances of a node. In panels b,c, and d,
we test three different scenarios. In each panel, nodes of the same colour
belong to the same module. We construe the left set of nodes as the root-
parasite guild, the middle set with thick borders as plants, and the right
set as herbivores. For our calculations, we use ω= 1000 to approximate
interlayer edge weights of ∞. In panel b, we obtain a maximum modularity
ofQB≈0.667, each layer has 3 modules, and no plant state nodes appear
in the same module. In panel c,QB≈0.651, and the network is partitioned
into a mean of n≈3.04 modules. On average, 4.6 (i.e., approximately 51%)
of the plants are assigned to more than one module. In panel d,QB≈0.99,
and we obtain a mean of n≈3.38 modules, and state nodes of each plant
belong exclusively to a single module. See Supplementary Notes 1 and 2 for
the details of our calculations.
11

Several studies outside of ecology have illustrated that modular (and

other mesoscale) network structures can change over time19,50,52. In ecol-
ogy, such network variation was considered in a study based on the analysis
of multiple disconnected networks53. Time-dependent modular structure,
including changes in module composition over time, can also be examined
by studying a multilayer network. As an example, consider a network repre-
senting the infection of 22 small mammalian host species by 56 ectoparasite
species during six consecutive summers in Siberia (1982–1987)54,55, yielding
a multilayer network with 6 layers. We quantify intralayer edge weights as
the prevalence of a given parasite on a given host. Interlayer edges connect
instances of the same species across consecutive time points (Fig. 1a), repre-
senting the relative changes in abundance between two consecutive summers.
For example, if a host has an abundance of 10 in one year and 5 in the next,
then the value of the interlayer edge is 5 /10 = 0.5 (Supplementary Note 3).
The idea behind this choice of interlayer edge values is that temporal fluctu-
ations in abundance affect the availability of hosts to parasites and parasite
pressure on hosts, and these factors in turn affect host–parasite interactions
in any given time point (i.e., the intralayer edges; see Eq. (3) in Supplemen-
tary Note 3). This choice results in (nonuniform) interlayer edge weights
that are on a similar scale than that of intralayer edges.
In a temporal network, a given species can interact strongly with some
species at one time and with other species at other times. Consequently,
each state node can belong to a different module at different times, and
modules can vary in size over time50,51(Fig. 3). We define ‘host adjusta-
bility’ and ‘parasite adjustability’ as the proportion of hosts and parasites,
respectively, that change module affiliation at least once. We observe non-
negligible values of this measure: about 47% of the hosts and about 35% of
the parasites change their module affiliation at least once, and a module’s
size also changes over time (Supplementary Figs. 2,3). One interpretation
of this pattern is that the same species is functionally different at different
times. For example, one ecological characteristic of a host is the extent
to which it supports populations of different parasite species in a commu-
nity. Fluctuations in the abundance of different host species (encoded in
the interlayer edges) can lead to fluctuations in the availability of hosts to
parasites. Additionally, there are temporal changes in interaction patterns
among other species in the network (encoded in the intralayer edges). These
mechanisms lead to a time-dependent distribution of parasites in hosts in
which the same host species supports populations of different parasites at
different times. This variation is expressed as the assignment of the same
host species to different modules in different layers.
Constructing and analysing a temporal network also allows one to con-
sider hypotheses on the effect of intralayer and/or interlayer connectivity
on community structure. This approach dominates studies of monolayer
ecological networks in which the structure of an observed network is com-
12

01234Module size
Layer1 2 3 1 2 3 1 2 3654321a
time 1
time 2
time 3bcFigure 3: A toy example of temporal modularity maximization.
The figure illustrates that species can change their module affiliation across
time points. We represent six species with numbers and indicate three dif-
ferent modules using different colours. a, The toy temporal network has
3 layers, and each layer has a bipartite structure. All species occur in all
layers. For clarity, we represent interlayer edges using blue dashed lines for
two species. b, Modules can change in size (i.e., the number of species that
they include ) across layers. For example, the green module does not exist
at time point 3, and the purple module does not exist at time point 1. c,
Representation of the module affiliation of each species in different layers.
For example, species 1 does not switch modules, whereas the other species
switch once each.
13

pared to that of networks within an ensemble that have been shuffled in

particular ways4,56(or to those generated with a generative random-graph
model48,49). The added value of the generalization of this approach to mul-
tilayer networks is that it allows one to test hypotheses that relate directly
to the temporal structure of a community. For example, we can shuffle
the interlayer edges between each pair of consecutive layers (separately for
hosts and parasites) to test the hypothesis that the modular structure is
a result of random temporal changes in species abundance (Supplementary
Note 3). The observed network has higher maximum modularity than the
shuffled networks ( Qobserved
B≈0.55 versusQshuffled
B≈0.21; the p-value is
P < 0.001). Additionally, it has about 6 (15) times fewer modules than
in networks where the order of interlayer host edges (parasite edges) have
been shuffled. All hosts and parasites switch modules at least once in the
shuffled networks. These observations lead us to reject the above hypothe-
sis. One can also hypothesize that the modular structure of the community
is: (i) a result of random associations between hosts and parasites in any
given layer, and (ii) independent of the temporal order in which hosts and
parasites are observed (Supplementary Note 3). We reject these hypotheses
as well (Supplementary Tables 4,5). The rejection of the three hypotheses
improves understanding of the functional groupings of species, by demon-
strating that host–parasite interactions are structured nonrandomly in time,
depending both on how species interact within a given time period and on
their persistence in time as measured by changes in species abundance. Con-
sequently, altering host–parasite interactions or the survival probability of
species (e.g., by applying parasite control programs) would strongly affect
temporal community organization.
Any assessment of such questions by studying each layer separately
would be necessarily incomplete. Modules in independently-examined layers
of a network are disconnected and independent from each other. Hence, one
cannot directly address the effect of interlayer phenomena (e.g., temporal
changes in species abundance or in local network architecture) on modular
structure (Supplementary Notes 1, 3). In the host–parasite network without
interlayer edges, species assignment to modules in any given layer contains,
on average, only about 35% of the information on species assignment to
modules in the interconnected temporal network (Supplementary Note 3.3).
As an alternative to using a multilayer network, ecologists can aggregate
species and interactions (e.g., across space or time). Aggregation may be
necessary when species interactions (intralayer edges) are sampled sparsely
in time. However, data aggregation entails a set of (usually implicit) assump-
tions14,57, and different aggregation methods can lead to qualitatively differ-
ent conclusions. For instance, by calculating ‘reducibility’58(Supplementary
Table 2) in the host–parasite network, we can quantify whether some layers
contain overlapping information, and can therefore be aggregated. We find
14
that all six layers are necessary to describe the entire complexity of the sys-
tem (Supplementary Note 3.4). Consistent with this finding, the affiliation
of species to modules in the aggregated network provides only about 52%
of the information on their affiliation to modules in the multilayer network
(Supplementary Note 3.3).
Taken together, our computations illustrate that multilayer networks
lend themselves naturally to asking questions about time-dependent phe-
nomena13,19,50,51. Temporal variations in the size and composition of mod-
ules may be relevant to phenomena such as species coevolution, coexistence,
and community stability4,10,59. For instance, hosts assigned to more than
one module may provide important bridges for transmission of ectopara-
sites across years and/or groups of strongly connected hosts. These hosts
can change their ecological function in a system55, and such flexibility in
community structure may contribute to system robustness in the face of
perturbations.
Robustness to Perturbations
Network structure can affect the stability of ecological communities7,30,31.
To illustrate how multilayer networks can contribute to studies of stability,
we use a network with two layers, plant–flower-visitors and plant–leaf-miner
parasitoids38, that are interconnected via the same set of plants (Fig. 1c,
Supplementary Note 4). We investigate the patterns of parasitoid extinc-
tions in two scenarios: (i) direct secondary extinctions due to plant removal;
and (ii) tertiary extinctions due to the removal of pollinators, which causes
plant secondary extinctions that, in turn, result in parasitoid tertiary ex-
tinctions (Supplementary Table 6). A plant (respectively, a parasitoid) goes
extinct when it becomes completely disconnected from flower visitors (re-
spectively, flowers). We find that parasitoid extinctions occur more slowly in
the multilayer network than in the plant–parasitoid monolayer network. Ad-
ditionally, allowing plant extinctions in addition to flower visitors extinctions
in scenario (ii) leads to nontrivial extinction patterns (Fig. 4). Therefore,
considering the multilayer nature of the network changes the qualitative
conclusions about the robustness of parasitoids to extinctions.
There are many ways to extend this kind of analysis (e.g., Supplementary
Note 4). For example, we considered above the values of the interlayer edges
to be infinity, to enforce perfect correspondence between the plant states in
the two layers. By relaxing this assumption, one can model for instance,
changes in plants abundance as a function of changes in flower visitors or
leaf-miners abundance. This can be done, for example, using a dynamical
system (described by a set of ordinary differential equations, as has been
done recently in the study of a meta-foodweb model60) on each layer, and
by using the values of the interlayer edges to couple these processes. Analysis
of such scenarios should be valuable for understanding the interplay among
15

different interaction types and their effect on system robustness.

Limitations and Challenges
Whether a multilayer approach is more appropriate than a monolayer one
obviously depends on the specific research question. Collecting the necessary
data for questions requiring multilayer networks can be resource-intensive, as
the data needs to be gathered from multiple places, at multiple times, and/or
with different observational methods to capture different types of interac-
tions. Measuring interlayer edge weights may require additional sampling
efforts that are different from those used for collecting data on intralayer
edges. Fortunately, data sets are already becoming available (Supplemen-
tary Table 1), but they are scattered in the literature and would need to be
curated.
One challenge is to define the meaning, and measure the values, of inter-
layer edges, and the choice of definition can itself play a significant role in
the analyses. For example, interlayer edges that connect species in two dif-
ferent communities may relate to species dispersal or changes in a species’
state (e.g., abundance). Furthermore, intralayer and interlayer edges can
represent ecological processes at different scales, and it is not always clear
how to define the relative weight of interlayer edges with respect to intralayer
edges. How to choose appropriate values for interlayer edges remains a topic
of active research in the study of multilayer networks. For some applications
(e.g., transportation), there already exist principled ways to choose values,
and for others (e.g. social networks) progress is underway. In ecology, this
issue remains completely uncharted territory. Where possible, it would be
best to measure interlayer edges directly61, and we illustrated one possibil-
ity in our analysis of a temporal network. An ad hoc approach, which has
been very insightful for several applications outside of ecology19,51,52,62, is
to systematically sweep through a set of values, and assess the robustness of
qualitative conclusions in different regions of parameter space, as well as how
they change as a function of interlayer edge weights. For example, one can
ask how the relative relationship between two interaction types affects the
dynamics of an ecological system. For this theoretical question, one would
measure a quantity of interest (e.g., some network diagnostic) across differ-
ent interlayer edge weights. We also note that different types of interactions
can also involve different ‘currencies’. For example, pollination is measured
differently than dispersal, and it is important to consider discrepancies in
the scales of the two edge types.
As with previous advances in network ecology, one can borrow and adapt
techniques from the more general science of networks2,4,59,63,64. Although a
natural first step is to adopt (and adapt) diagnostics and algorithms from
the many that are available in applications of multilayer networks to other
16

a b Parasitoids Plants
Proportion of species removedProportion of species survivingScenario 1
Scenario 2
Scenario 3 (0.3)
Scenario 3 (0.8)Figure 4: Network robustness to species removal in a multilayer
network of plant–flower-visitors and plant–leaf-miner parasitoids.
The proportion of surviving ( a) leaf-miner parasitoids and ( b) plants as a
function of the proportion of species removed in three scenarios. Scenario
(i) simulates a monolayer case, where we remove plants from the plant–
leaf-miner parasitoid layer (orange curves). This results in a linear decrease
in the proportion of plants surviving (in panel b). In scenario (2), we re-
move flower visitors and quantify secondary extinctions (the proportion of
surviving species) in plants and concomitant tertiary extinctions of leaf-
miner parasitoids (brown curves). Scenario (3) is the same as (2), but with
additional plant removal (which is independent of the secondary plant ex-
tinctions) with probability 0.3 (purple curves) or 0.8 (green curves). We
detail the algorithms for each of the removal scenarios in Supplementary
Table 6. The is overlap between the curves for scenarios 2 and 3 with low
probability, so we separated them slightly for clarity. In panel a, we note a
‘transition’ in the proportion of species that survive in scenario 3. Although
discerning whether this is a general feature of extinction processes in eco-
logical multilayer networks requires further research, a plausible explanation
is that when there is a high probability of random independent extinctions
of plants, the system reaches a threshold in which too many plants have
become extinct and thus can no longer sustain the community.
17

disciplines14,15(Supplementary Table 2), it is important to determine which

existing methods are best suited for ecological applications, and to develop
methods and diagnostics with ecology specifically in mind. These methods
include multilayer versions of common network diagnostics (e.g., vulnera-
bility and generality), null models to test hypotheses about network struc-
ture62,65, and mathematical models for dynamical processes. Availability of
software59,66,67has played an important role in the accessibility of monolayer
network analyses, and the same should become the norm for the analysis and
visualization of multilayer networks with the development of well-described
and easy-to-use tools. Several software packages are already available (Sup-
plementary Table 3), even though these are not yet as mature as those for
monolayer ecological networks.
Future Directions and Applications
We have argued that multilayer networks provide a versatile and powerful
framework for investigations in community ecology. This is also true for bio-
geography, where networks that include species dispersal or colonization/ex-
tinction dynamics as interlayer edges can help one understand how spatial
dynamics affect community structure and stability20. A multilayer formal-
ism can address not only correlations between geographic distance and beta
diversity of species and interactions (which to date have been studied using
multiple disconnected networks; see, e.g., Trøjelsgaard et al.29), but also the
mechanisms behind such correlations. Another open area concerns the effect
of space on the stability of ecological communities68. To this end, one could
use a multilayer network that explicitly incorporates different instances of
the same species at different locations instead of a matrix of interactions for
a single community in space30,34. Habitat alteration is a growing concern,
and it would be interesting to explore the stability of ecological networks
with different spatial connectivity patterns (encoded in the interlayer edges).
Ecological multilayer networks is a new framework, and the development
of theoretical models is critical as they provide a basis for comparison with
data11. Indeed, a multilayer ecological framework should stimulate further
theoretical developments. For instance, it provides a way to consider the
structure underlying multiple interaction types in dynamical models, which
thus far have been modelled on top of a single web describing a single in-
teraction type32,34. Interlayer edges in such models encode the amount of
coupling between different dynamical processes and provide a way to encode
the relative importances of different interaction types in determining pro-
cesses such as species’ population dynamics. For example, a key question
in disease ecology is assessing the roles that different types of host–parasite
interactions play in disease transmission (e.g., the contact patterns among
individuals69or the trophic web within which parasites and hosts are em-
18

bedded70). Using a multilayer network with three layers—contact network,

trophic web, and host–vector interactions—Stella et al.41modelled the rel-
ative importances of vectors for diseases that can be transmitted both via
trophic interactions and vectors by varying the coupling between the trophic
and the vector layers. They reported that transmission that spreads only
on the trophic layer hinders the infection of host populations.
Multilayer networks can also advance metacommunity theory, where in-
terlayer edges provide a way to develop spatially-explicit models to inves-
tigate how species move among local communities, thereby creating spatial
structure in the regional species’ pool. For example, one can study spatial
structure in resource and species flow in interconnected food webs, such as
those of lakes or ponds that are interconnected via common sets of ani-
mals. In such examples, interlayer edge values can represent dispersal20or
biomass flows among patches. One can also use a temporal food web to
explore bioenergetic flows across food webs—an area that remains largely
unexplored13. Layers can represent temporal instances of a given food web,
and interlayer edges can represent changes in species biomass with time.
Another key question in food-web theory is the effect of parasitism on
food-web structure and stability70–72. For example, parasitism may have dif-
ferent effects than trophic interactions on a given species. Parasitized hosts
can be more susceptible to predation, but it is unclear how the structure of
host–parasite networks affects the trophic interactions between hosts. One
way to model such systems is by coupling a host–parasite network and a
food web (Box 1 and Supplementary Fig. 1). Multilayer networks also pro-
vide a possible approach for analysing disease transmission when there are
multiple hosts and parasites18,42, which to date has been difficult to study.
In animal behaviour, one can explore networks in which intralayer edges
between individuals represent reproduction and interlayer edges represent
movement (and hence gene flow), to study genetic relatedness among indi-
viduals as a function of both dispersal and within-group social behaviour.
In movement ecology, multilayer networks can help model relationships be-
tween a network of social interactions and a network of movement patterns
to understand moving decisions. Understanding movement has consequences
for conservation biology, where one can represent different connectivity sce-
narios using a multilayer network in which each layer describes the movement
patterns of a different species. Such models would inform decision-makers
on which land-use designs are best-suited for the movement of a diverse set
of species. Multilayer networks can also help improve the identification of
keystone species. A simple notion of a keystone species arises from calcu-
lating a “centrality” measure of species in a food web (i.e., keystone species
have the highest values of that centrality)73, but a more nuanced definition
of centrality—and hence of keystone species—can incorporate participation
in several interaction types61, as well as temporal and spatial dependencies.
Finally, multilayer networks can be used to study reciprocal effects be-
19

tween ecological and non-ecological systems. For example, Baggio et al.74

represented three indigenous Alaskan communities using a multiplex net-
work in which each layer is a unique combination of ecological resources and
social relations. They found that changes to the social relations have greater
impact on the robustness of the networks (and hence the human communi-
ties) than depletion of ecological resources (e.g., removal of marine species
used as food).
Conclusions
The simultaneous expansion in the availability of ecological data and the
tools to analyse multilayer networks14,15,18,75, provides a timely and valu-
able opportunity for ecologists to explore the multilayer nature of ecological
networks. The strength of a multilayer approach lies in its ability to for-
mulate and analyse complex systems in a way that explicitly incorporates
processes operating both within and across layers (as well as the interaction
between these processes). Formulating systems as multilayer networks also
allows one to address questions that are off-limit based on monolayer repre-
sentations. Working within the same framework facilitates comparisons of
results across ecological systems and network types, because of consistency
in technical terms and methodology. A unified framework should further
encourage collaboration with scientists from other disciplines. Other fields
of study, in turn, would benefit from methodology and theory developed
for ecological multilayer networks, as has been the case for monolayer eco-
logical networks76. In closing, the integration of multilayer network theory
into ecology offers an additional perspective, with the potential to provide
new theoretical and empirical insights, into the architecture and dynamics
of ecological systems.
Data and Code
We provide the raw data for the example temporal network in the Supple-
mentary Information (Data set 1). The code for general procedures to pre-
pare data, manipulate networks, post-process modularity-maximization cal-
culations, and analyse network robustness is written in R, and it is available
at 10.6084/m9.figshare.3472664. The code for examination of modular struc-
ture is written in Matlab , and it is available at 10.6084/m9.figshare.3472679.
Acknowledgements
SP was supported by a James S. McDonnell foundation postdoctoral fel-
lowship for the study of complex systems and by a Fulbright postdoctoral
20

fellowship from the US Department of State. MAP was supported by FET-

Proactive project PLEXMATH (FP7-ICT-2011-8; grant #317614) funded
by the European Commission. We thank Lucas Jeub for help with the anal-
ysis of multilayer modularity, Manlio De Domenico for insightful discussions
on multilayer networks, and Boris R. Krasnov for help with the temporal
network data. MAP thanks Mikko Kivel¨ a and other collaborators for help-
ing to shape his view of multilayer networks and the participants at the
first GCEE workshop on Resilience and Recovery of Biological Networks to
Environmental Fluctuations for illuminating discussions.
Author contributions
SP conceived the idea, performed numerical simulations, and analysed the
data; MAP contributed insights about the mathematics of multilayer net-
works and data analysis; and MP and SK contributed insights on the use of
multilayer networks in ecology and the interpretation of results. All authors
wrote the manuscript.
Competing financial interests
The authors declare no competing financial interests.
References
1. Bascompte, J., Jordano, P., Meli´ an, C. J. & Olesen, J. M. The nested assembly of
plant–animal mutualistic networks. Proc. Natl. Acad. Sci. U. S. A. 100, 9383–9387
(2003).
2. Jordano, P., Bascompte, J. & Olesen, J. M. Invariant properties in coevolutionary
networks of plant–animal interactions. Ecol. Lett. 6,69–81 (2003).
3. Bascompte, J. & Jordano, P. Plant-animal mutualistic networks: The architecture
of biodiversity. Annu. Rev. Ecol. Evol. Syst. 38,567–593 (2007).
4. Olesen, J. M., Bascompte, J., Dupont, Y. L. & Jordano, P. The modularity of
pollination networks. Proc. Natl. Acad. Sci. U. S. A. 104, 19891–198916 (2007).
5. Bascompte, J. Disentangling the web of life. Science 325, 416–419 (2009).
6. Ings, T. C. et al. Ecological networks–beyond food webs. J. Anim. Ecol. 78,253–
269 (2009).
7. Th´ ebault, E. & Fontaine, C. Stability of ecological communities and the architecture
of mutualistic and trophic networks. Science 329, 853–856 (2010).
8. Olesen, J. M., Stefanescu, C. & Traveset, A. Strong, long-term temporal dynamics
of an ecological network. PLoS One 6,e26455 (2011).
9. Meli´ an, C. J., Bascompte, J., Jordano, P. & Krivan, V. Diversity in a complex
ecological network with two interaction types. Oikos 118, 122–130 (2009).
10. Fontaine, C. et al. The ecological and evolutionary implications of merging di fferent
types of networks. Ecol. Lett. 14,1170–1181 (2011).
21

11. K´ efi, S. et al. More than a meal: integrating non-feeding interactions into food webs.
Ecol. Lett. 15,291–300 (2012).
12. K´ efi, S., Miele, V., Wieters, E. A., Navarrete, S. A. & Berlow, E. L. How Struc-
tured Is the Entangled Bank? The Surprisingly Simple Organization of Multiplex
Ecological Networks Leads to Increased Persistence and Resilience. PLoS Biol. 14,
e1002527 (2016).
13. Blonder, B., Wey, T. W., Dornhaus, A., James, R. & Sih, A. Temporal dynamics
and network analysis. Methods Ecol. Evol. 3,958–972 (2012).
14. Kivel¨ a, M. et al. Multilayer networks. J. Complex Networks 2,203–271 (2014).
15. Boccaletti, S et al. The structure and dynamics of multilayer networks. Phys. Rep.
544, 1–122 (2014).
16. De Domenico, M. et al. Mathematical formulation of multilayer networks. Phys.
Rev. X 3,041022 (2013).
17. Gallotti, R. & Barthelemy, M. Anatomy and efficiency of urban multimodal mobil-
ity.Sci. Rep. 4,6911 (2014).
18. Salehi, M et al. Spreading processes in multilayer networks. IEEE Transactions on
Network Science and Engineering 2,65–83 (2015).
19. Bassett, D. S. et al. Dynamic reconfiguration of human brain networks during
learning. Proc. Natl. Acad. Sci. U. S. A. 118, 7641–7646 (2011).
20. Gilarranz, L. J., Sabatino, M., Aizen, M. & Bascompte, J. Hot spots of mutualistic
networks. J. Anim. Ecol. 84,407–413 (2014).
21. Kitching, R. L. Spatial and Temporal Variation in Food Webs in Water-Filled
Treeholes. Oikos 48,280–288 (1987).
22. Moore, J. C. & de Ruiter, P. C. Temporal and spatial heterogeneity of trophic
interactions within below-ground food webs. Agric. Ecosyst. Environ. 34,371–397
(1991).
23. Schoenly, K. & Cohen, J. E. Temporal variation in food web structure: 16 empirical
cases. Ecol. Monogr. 61,267–298 (1991).
24. Closs, G. P. & Lake, P. S. Spatial and temporal variation in the structure of an
intermittentstream food web. Ecol. Monogr. (1994).
25. Winemiller, K. O. in Food Webs (eds Polis, G. A. & Winemiller, K. O.) 298–312
(Springer US, 1996).
26. Winemiller, K. O. & Jepsen, D. B. Effects of seasonality and fish movement on
tropical river food webs. J. Fish Biol. 53,267–296 (1998).
27. Poisot, T. et al. The dissimilarity of species interaction networks. Ecol. Lett. 15,
1353–1361 (2012).
28. Carstensen, D. W., Sabatino, M., Trøjelsgaard, K. & Morellato, L. P. C. Beta diver-
sity of plant-pollinator networks and the spatial turnover of pairwise interactions.
PLoS One 9,e112903 (2014).
29. Trøjelsgaard, K., Jordano, P., Carstensen, D. W. & Olesen, J. M. Geographical
variation in mutualistic networks: similarity, turnover and partner fidelity. Proc. R.
Soc. B 282, 20142925 (2015).
30. Allesina, S. & Tang, S. Stability criteria for complex ecosystems. Nature 483, 205–
208 (2012).
31. Rohr, R. P., Saavedra, S. & Bascompte, J. On the structural stability of mutualistic
systems. Science 345, 1253497 (2014).
22
32. Bastolla, U. et al. The architecture of mutualistic networks minimizes competition
and increases biodiversity. Nature 458, 1018–1020 (2009).
33. Olff, H. et al. Parallel ecological networks in ecosystems. Philos. Trans. R. Soc.
Lond. B Biol. Sci. 364, 1755–1779 (2009).
34. Mougi, A & Kondoh, M. Diversity of interaction types and ecological community
stability. Science 337, 349–351 (2012).
35. K´ efi, S. et al. Network structure beyond food webs: mapping non-trophic and
trophic interactions on Chilean rocky shores. Ecology 96,291–303 (2015).
36. Gross, K. Positive interactions among competitors can produce species-rich com-
munities. Ecol. Lett. 11,929–936 (2008).
37. Rudolf, V. H. W. & Lafferty, K. D. Stage structure alters how complexity affects
stability of ecological networks. Ecol. Lett. 14,75–79 (2011).
38. Pocock, M. J. O., Evans, D. M. & Memmott, J. The robustness and restoration of
a network of ecological networks. Science 335, 973–977 (2012).
39. Pilosof, S. et al. Host-parasite network structure is associated with community-level
immunogenetic diversity. Nat. Commun. 5,5172 (2014).
40. Bauer, S & Hoye, B. J. Migratory animals couple biodiversity and ecosystem func-
tioning worldwide. Science 344, 1242552 (1–8) (2014).
41. Stella, M., Andreazzi, C. S., Selakovic, S., Goudarzi, A. & Antonioni, A. Parasite
spreading in spatial ecological multiplex networks. arXiv:1602. 06785 [physics, q-
bio](2016).
42. Pilosof, S., Greenbaum, G., Krasnov, B. R. & Zelnik, Y. R. Asymmetric disease
dynamics in multi-host interconnected networks. arXiv:1512. 09178 [q-bio] (2016).
43. Nakano, S. & Murakami, M. Reciprocal subsidies: dynamic interdependence be-
tween terrestrial and aquatic food webs. Proc. Natl. Acad. Sci. U. S. A. 98,166–
170 (2001).
44. Knight, T. M., McCoy, M. W., Chase, J. M., McCoy, K. A. & Holt, R. D. Trophic
cascades across ecosystems. Nature 437, 880–883 (2005).
45. O’Neill, R. V. A Hierarchical Concept of Ecosystems (Princeton University Press,
1986).
46. Brose, U. Body-mass constraints on foraging behaviour determine population and
food-web dynamics. Funct. Ecol. 24,28–34 (2010).
47. Scotti, M, Ciocchetta, F & Jord´ an, F. Social and landscape effects on food webs: a
multi-level network simulation model. J. Complex Networks 1,160–182 (2013).
48. Porter, M. A., Onnela, J.-P. & Mucha, P. J. Communities in networks. Not Am
Math Soc 56,1082–1097 (2009).
49. Fortunato, S. & Hric, D. Community detection in networks: A user guide. Phys.
Rep. 659, 1–44 (2016).
50. Mucha, P. J., Richardson, T., Macon, K., Porter, M. A. & Onnela, J.-P. Community
structure in time-dependent, multiscale, and multiplex networks. Science 328, 876–
878 (2010).
51. Bazzi, M. et al. Community detection in temporal multilayer networks, with an
application to correlation networks. Multiscale Model. Simul. 14,1–41 (2016).
52. Sarzynska, M., Leicht, E. A., Chowell, G. & Porter, M. A. Null models for com-
munity detection in spatially embedded, temporal networks. Journal of Complex
Networks 4,363–406 (2015).
23

53. Dupont, Y. L. & Olesen, J. M. Stability of modular structure in temporal cumula-

tive plant–flower-visitor networks. Ecol. Complex. 11,84–90 (2012).
54. Krasnov, B. R., Matthee, S., Lareschi, M., Korallo-Vinarskaya, N. P. & Vinarski,
M. V. Co-occurrence of ectoparasites on rodent hosts: null model analyses of data
from three continents. Oikos 119, 120–128 (2010).
55. Pilosof, S., Fortuna, M. A., Vinarski, M. V., Korallo-Vinarskaya, N. P. & Krasnov,
B. R. Temporal dynamics of direct reciprocal and indirect effects in a host-parasite
network. J. Anim. Ecol. 82,987–996 (2013).
56. Fortuna, M. A. et al. Nestedness versus modularity in ecological networks: two sides
of the same coin? J. Anim. Ecol. 79,811–817 (2010).
57. Holme, P. Modern temporal network theory: A colloquium. Eur. Phys. J. B 1508.01303,
1–30 (2015).
58. De Domenico, M., Nicosia, V., Arenas, A. & Latora, V. Structural reducibility of
multilayer networks. Nat. Commun. 6(2015).
59. Th´ ebault, E. Identifying compartments in presence-absence matrices and bipartite
networks: insights into modularity measures. J. Biogeogr. 40(ed Veech, J.) 759–768
(2013).
60. Brechtel, A., Gramlich, P., Ritterskamp, D., Drossel, B. & Gross, T. Master stability
functions reveal diffusion-driven instabilities in multi-layer networks (2016).
61. Genrich, C. M., Mello, M. A. R., Silveira, F. A. O., Bronstein, J. L. & Paglia, A.
P. Duality of interaction outcomes in a plant–frugivore multilayer network. Oikos
(2016).
62. Bassett, D. S. et al. Robust detection of dynamic community structure in networks.
Chaos 23,013142 (2013).
63. Newman, M. E. J. Networks: An Introduction (Oxford University Press, Oxford,
UK, 2010).
64. Mart´ ın Gonz´ alez, A. M., Dalsgaard, B. & Olesen, J. M. Centrality measures and
the importance of generalist species in pollination networks. Ecol. Complex. 7,36–
43 (2010).
65. Ulrich, W. & Gotelli, N. J. Null model analysis of species associations using abun-
dance data. Ecology 91,3384–3397 (2010).
66. Dormann, C. F., Fr¨ und, J., Bl¨ uthgen, N. & Gruber, B. Indices, graphs and null
models: analyzing bipartite ecological networks. The Open Ecology Journal 2,7–24
(2009).
67. Flores, C. O., Poisot, T., Valverde, S. & Weitz, J. S. BiMat : a MATLAB package to
facilitate the analysis of bipartite networks. Methods Ecol. Evol. 7,127–132 (2016).
68. Lurgi, M., Montoya, D. & Montoya, J. M. The effects of space and diversity of
interaction types on the stability of complex ecological networks. Theor. Ecol. 1–11
(2015).
69. Craft, M. E. & Caillaud, D. Network models: an underutilized tool in wildlife epi-
demiology? Interdiscip. Perspect. Infect. Dis. 676949 (2011).
70. Lafferty, K. D. et al. Parasites in food webs: the ultimate missing links. Ecol. Lett.
11,533–546 (2008).
71. Dunne, J. et al. Parasites affect food web structure primarily through increased
diversity and complexity. PLoS Biol. 11,e1001579 (2013).
72. Selakovic, S., de Ruiter, P. C. & Heesterbeek, H. Infectious disease agents mediate
interaction in food webs and ecosystems. Proc. R. Soc. B 281, 20132709 (2014).
24

73. Jord´ an, F. Keystone species and food webs. Philos. Trans. R. Soc. Lond. B Biol.
Sci.364, 1733–1741 (2009).
74. Baggio, J. A. et al. Multiplex social ecological network analysis reveals how social
changes affect community robustness more than resource depletion. Proc. Natl.
Acad. Sci. U. S. A. (2016).
75. Wang, Z., Andrews, M. A., Wu, Z.-X., Wang, L. & Bauch, C. T. Coupled disease–
behavior dynamics on complex networks: A review. Phys. Life Rev. 15,1–29 (2015).
76. Saavedra, S., Reed-Tsochas, F. & Uzzi, B. A simple model of bipartite cooperation
for ecological and organizational networks. Nature 457, 463–466 (2009).
25

Supplementary Information
26

a
bHost–parasite Food webSupplementary Figure 1: An ecological multilayer network with
two interaction types .a, The network depicts (light blue layer) an
aquatic food web and (light orange layer) a host–parasite network. Species
that occur in both layers are interconnected with dashed blue interlayer
edges. Note that one parasite appears in both layers and can therefore play
a role as a prey and/or a parasite for the pelican. b, A supra-adjacency
representation of the network in panel a. Coloured blocks correspond to the
network layers. We show intralayer and interlayer edges, respectively, with
black and blue matrix cells. Species that appear in two layers also appear
twice in the matrix. These are different state nodes of the same physical
node.
27
Module
sizeModuleSupplementary Figure 2: Yearly variation in module size . We ex-
amine modular structure in a network that represents the infection of small
mammalian hosts by fleas and mites. Because species can switch modules
across years (i.e., across layers), we observe variation in module size. For
example, module 1 maintains a relatively large size across years, whereas
module 4 exists only in four of the six layers. This figure depicts a particu-
lar example among 100 instantiations of the generalized Louvain algorithm77
for multilayer modularity maximization of the host–parasite network). This
instantiation is the one with the maximum value of the maximum modular-
ity. See Supplementary Note 1 for details about our calculations.
28

ModuleHosts Parasites
ParasitesSupplementary Figure 3: Module affiliation of species across years
for the host–parasite network examined in the main text. We rep-
resent different modules using different colours. We show community assign-
ments for a particular example (the one with the highest value of maximum
modularity QB) among 100 instantiations of applying the generalized Lou-
vain algorithm77for multilayer modularity maximization. The figure illus-
trates that species can change their module affiliation across years. Missing
data points indicate that a species is not present in a particular year.
29

Interlayer edge valueCountLayer 1 Layer 2 Layer 2 Layer 3 Layer 3 Layer 4

Layer 4 Layer 5 Layer 5 Layer 6Supplementary Figure 4: Distribution of the values of interlayer
edges of a host–parasite temporal network . We calculate interlayer
edge weights as the relative changes in abundance of a species during the
time window t→t+ 1. Each histogram gives the distribution of values for
a given time window of the network (there are 6 layers and hence 5 time
windows). See the main text and Supplementary Note 3 for details on the
calculation of interlayer edge weights.
30

Proportion of flower visitors removedProportion of species remaining

0.00 0.25 0.5 0.75 1.000.000.250.50.751.00Supplementary Figure 5: Network robustness to
species re-
moval for the plant–flower-visitors and plant–leaf-miner para-
sitoids diagonally-coupled multilayer network. The curves indicate
the proportion of (blue curve) surviving plants and (orange curve) surviv-
ing plants and parasitoids as a function of the proportion of flower visitors
that have been removed. In each scenario, we remove flower visitors in the
same order, which is by increasing intralayer node degree (that is, we first
remove the lowest-degree flower visitors, and we proceed accordingly). See
Supplementary Note 4 for more details.
31

quality function (q)

Step00.20.40.60.8
1 2 3 4 51983
1984
1982
1986
1985
1987
1984
1983
1982
1986
1985
1987
1984
1983
1982
1986
1985
19870.1
0.3
0.4
0.50a
bcSupplementary Figure 6: Reducibility of a host–parasite tempo-
ral network. a , Following De Domenico et al.58, we compute the Jensen–
Shannon distance Djsbetween each pair of layers as a proxy for layer re-
dundancy. b, We perform hierarchical clustering on the resulting distance
matrix. The output is a dendrogram whose leaves represent the initial layers
and whose internal nodes indicate layer merging. c, At each step, we aggre-
gate the clustered layers corresponding to the smallest value of Djs, and we
quantify the quality of the resulting network in terms of distinguishibility
from the aggregated graph (i.e., a graph which includes all physical nodes
and in which edges are the sum of intralayer edges across all layers) by a
global quality function q. (See De Domenico et al.58for details on how to
calculateq.) An ‘optimal’ partition (which need not be unique) is one for
whichqis maximal. In this example, we obtain maximal qfrom the partition
(indicated by the dashed red line) in which the 6 layers are separate.
32

Supplementary Table 1: Types of ecological multilayer networks that have been examined and
some questions that can
be addressed by using them. In our list, we include studies that used multiple networks that were
not connected explicitly to
each other via interlayer edges.
Layer Nodes Interlayer Intralayer Structural Primary research Reference Figure
edges edges characteristics themes/questions
Time S Each node
to itself
(ordinal)M, P, T A set of networks defined separately at
different time points.Temporal variation in network or node
diagnostics (e.g., centrality, connectance).
Species and/or interaction turnover. Network-
assembly dynamics. Temporal variation in
beta diversity of species and/or interactions.8,13,21–23,27,55,78–97Fig. 1 a
Space S Each node
to itself
(categori-
cal)M A set of networks defined separately at
different sites.Spatial variation in network or node diagnos-
tics (e.g., number of species, connectance).
Spatial variation in beta diversity of species
and their interactions as a function of geo-
graphic distance.21,22,27,29,79,98–100Fig. 1 b
Interaction
typeS, FA Each node
to itselfM, P, T, G Networks interconnected via shared
species (diagonal coupling).How do different interaction types change un-
derstanding of ecological and evolutionary dy-
namics? How do different interaction types af-
fect the stability and robustness of a system?
What is the association between host immuno-
genetics and host–parasite interactions?9,10,38,39,68,101,102Fig. 1 c
Interaction
typeS Each node
to itselfT, P, NT+,
NT−Layers with the same set of species and
different interaction types (e.g., par-
asitism, trophic, facilitation, mutual-
ism) or signed networks.Quantify architecture of trophic and non-
trophic interactions, and quantify the associa-
tion between these different interaction types.
How does including different interaction types
affect functional diversity and species roles?
How are parasite transmission dynamics af-
fected by different routes of transmission?
How do invasive or migratory species affect the
communities into which they arrive?12,37,40,61,103Fig. 1 d
Community S Each node
to itselfT Food webs interconnected via common
species.How does interaction between food webs (via
common species) affect the dynamics of each
food web separately?26,43,104Fig. 1 c
Community S Dispersal M, T Networks of communities (e.g., food
webs, mutualistic bipartite networks)
connected by species dispersal. The
communities can be spatially embed-
ded.How are dispersal and colonization–extinction
dynamics associated with network architecture
in each community?20,105–107Fig. 1 f
Multiple I, P, C Each node
to a node in
an adjacent
layer.Sc,T, D Network layers describe different hier-
archical levels of a system.How do processes at one level affect those in
other levels?47Fig. 1 f
Population I Each node
to itselfSc Network layers defined separately for
each behavioural interaction.How are the function and dynamics of ani-
mal social networks defined by different be-
haviours?108Fig. 1 d
Population I Between
nodes from
different
layersSc Layers interconnected through one or
more nodes.Disease dynamics in interconnected popula-
tions of different hosts.42Fig. 1 e
Nodes: Species (S), Individuals (I), Communities (C), Metacommunities (MC), Functional
alleles (FA). Intralayer edges: Mu-
tualistic (M), Parasitism (P), Trophic (T), Non-trophic positive (NT+), Non-trophic negative (NT
−), Genetic (G), Dispersal
(D), Social (Sc). References are to studies that explored either networks with multiple but
independent layers or multilayer
networks with interlayer edges. The ‘Interlayer’ column is relevant for studies that explicitly
used interlayer edges, and it also
indicates how the different layers in studies of multiple disconnected networks can be linked to
each other. We have listed
several types of layering, and it likely is also useful to explore other varieties.
33

Supplementary Table 2: We list a selection of diagnostics for, and other features of, multilayer
networks; and we give
example applications of these diagnostics and features for ecological multilayer networks
(EMLNs).
Diagnostic Definition (words) Example application(s) in EMLNs Reference
MultiedgeA vector (whose length is equal to the number of lay-
ers) that consists of the set of edges that connect a
given pair of nodes in different layers of a multilayer
network.Represent a set of ecological interactions between two
species.109
Multidegree Total number of multiedges incident to a given node.Quantify generality and/or
vulnerability of a species
across different interaction types, different places, etc.109
VersatilityA generalisation of centrality measures that considers
a node’s importance across multiple layers.Identify species that are important (i.e., versatile), be-
cause they bridge different interaction types, commu-
nities, and/or locations110
Clustering coefficientsMeasure extent to which nodes in a network tend to
form transitive triples.Evaluate the tendency of species to form interaction
triples across sites or when there are multiple types of
interactions.111
Network community structureSets of nodes (in the same or different layers) that are
connected to each other more densely than to other
nodes in a network; network communities (i.e., mod-
ules) are usually identified as the output of an algo-
rithm.Identify modular structure across multiple times, spa-
tial locations, and/or interaction types.50,112,113
Stochastic block modelsGenerative statistical models to which one fits net-
work data to find mesoscale features (such as modular
structure, core–periphery structure, or others).Identify sets of species across multiple times,
spatial
locations, and/or interaction types.114,115
Node degree correlation across
layersIndicates relationship between intralayer degrees in
different layers. (For example, are high-degree nodes
in one layer typically also high-degree nodes in other
layers?)Are species generalists or specialists with respect to
various interaction types or in various patches?116
ReducibilityFind the layers needed to represent an ecological sys-
tem in a way that is distinguishable from an aggre-
gated network.(1) Guide for temporal and/or spatial resolution at
which data should be collected. (2) In a network
in which each layer is a different community, indi-
cates which of the communities contain unique species
and/or interactions. (3) Informs researchers about ag-
gregation of data.58
Participation coefficient and
overlapping degreeStructural role that a node plays in the distribution of
edges across layers (in the spirit of the roles of species
in interconnecting communities4).Explore the functional role that species play in a com-
munity while considering several interaction types.117
Node/interaction turnoverMean proportion of node/edge appearance and disap-
pearance in different layers.Understand the time-dependence of network structure
and the temporal turnover rates of species.8,118
Measures of beta diversitySimilarity between a pair of network layers due to dif-
ferences in species and/or interactions.Investigate how networks vary in space or time, and
connect such variation to ecological factors such as dis-
tance between patches or species’ dispersal abilities.27
See the references for explicit mathematical definitions. Many other diagnostics for, and other
features of, multilayer networks
are discussed in Refs.14,15
34

Supplementary Table 3: Available software for analyzing multilayer

networks
Software Interface Uses Link
MuxViz119Graphical user
interface (but
may require
some knowl-
edge of Rand
Octave )Visualization, interlayer cor-
relations, network centrality,
reducibility, modularity, node
and edge statistics, motifs.http://muxviz.net
Multilayer
Networks
LibraryPythonProvides data structures for
multilayer networks and basic
methods for analysing them.http://www.plexmath.
eu/wp-content/
uploads/2013/11/
multilayer-networks-library_
html_documentation/
GenLouvain77MatlabCode for examining network
community structure in mul-
tilayer networks with diagonal
coupling.http://netwiki.amath.unc.
edu/GenLouvain/GenLouvain ;
in the SI, we include code
(modified from the original)
for the case in which layers
have a bipartite structure.
LocalCommunityMatlabLocal community-detection
method used in ref.112.https://github.com/LJeub/
LocalCommunities
Infomap113 Stand-alone
web interfaceCommunity detection using
flow-based ideas.http://www.mapequation.
org/code.html
Betalink
package27,120 RQuantify dissimilarity between
ecological networks.https://github.com/
PoisotLab/betalink
timeordered
package13,121 R Analyse temporal networks.https://cran.r-project.
org/web/packages/
timeordered/index.html
Multiplex
package122 RAlgebraic procedures for the
analysis of multiplex networks.
Create and manipulate multi-
variate network data with dif-
ferent formats.http://cran.r-project.org/
web/packages/multiplex/
index.html
We specify websites to download the software. When relevant, we also in-
clude citations to associated references.
35

Supplementary Table 4: Community structure of the host–parasite

temporal network for three examples of interlayer edge weights.
Ecological
Communitynetwork typeInterlayer
edge weightsQBNumber
of Mod-
ules (n)Host adjustability,
parasite adjustabil-
ity
Layer 1 monolayer 0 0.46 6.47 N/A
Layer 2 monolayer 0 0.43 5.63 N/A
Layer 3 monolayer 0 0.40 6.22 N/A
Layer 4 monolayer 0 0.49 4.16 N/A
Layer 5 monolayer 0 0.35 4.49 N/A
Layer 6 monolayer 0 0.56 5.88 N/A
Whole network multilayer 1000 0.99 2.96 0 , 0
Whole network multilayerRelative
change in
abundance0.55 6.30 0.474, 0.351
Each of the values of maximized modularity ( QB), the number of modules
(n), and the fraction of hosts and parasites that are assigned to more than
one module (i.e., host and parasite ‘adjustability’, respectively) is a mean
over 100 instantiations of the generalized Louvain algorithm50,77. We adjust
the null model in the multilayer modularity objective function from Mucha
et al.50to make it suitable for layers with a bipartite structure. In mono-
layer networks, it is not possible to compare modules across layers without
resorting to an ad hoc approach, so we place a value of N/A in the table.
We use the interlayer edge weight ω= 1000 for our calculations near the
extreme case ω=∞.
36

Supplementary Table 5: Comparison of the observed host–parasite net-

work with weighted intralayer and interlayer edges to three null models.
Shuffled QB Number of
modules (n)Host adjustability parasite adjustabil-
ity
None (observed) 0.55 6.3 0.474 0.351
Intralayer edges 0.41
(P < 0.001)4.9 (P < 0.001) 0.372 ( P < 0.001) 0.448 ( P < 0.001)
Interlayer edges: hosts 0.21
(P < 0.001)38.6
(P < 0.001)1 (P < 0.001) 0.928 ( P < 0.001)
Interlayer edges: parasites 0.21
(P < 0.001)94.6
(P < 0.001)1 (P < 0.001) 0.928 ( P < 0.001)
Permuted order of layers 0.55 ( P≈0.52) 17.1
(P < 0.001)0.446 ( P < 0.001) 0.395 ( P < 0.001)
Each of the values of maximized modularity ( QB), the number of modules
(n), and the fraction of hosts and parasites that are assigned to more than
one module (i.e., host and parasite adjustability, respectively) is a mean over
100 instantiations of the generalized Louvain algorithm50,77. We adjust the
null model in the multilayer modularity objective function from Mucha et
al.50to make it suitable for layers with a bipartite structure. We calculate
the p-values P(in parentheses) by comparing the mean of 100 instantiations
of the observed network to 100 shuffled networks (see Supplementary Note 3
for details on how we shuffled networks).
37

Supplementary Table 6: Algorithms for examining network robustness

to species extinctions.
Removal scenario Algorithm Ecological reasoning/com-
ments
Monolayer: Remove
plants and follow par-
asitoids’ secondary ex-
tinctions.1.Remove the plant with the lowest de-
gree in the plant–parasitoid layer (in case
of ties we order extinctions by species
name). 2.Remove all intralayer edges
of the plant. 3.Remove any parasitoids
in the layer that remain disconnected (i.e.,
have degree 0). 4.Record the proportion
of surviving parasitoids in the layer as a
function of the proportion of plants that
have been removed. 5.Repeat until all
parasitoids are extinct in the layers.1. Work only on the plant–
parasitoid layer. 2.We use pro-
portions, because it allows us to
compare the different scenarios.
Multilayer: Remove
flower visitors and fol-
low parasitoids’ ter-
tiary extinctions.1.Remove the flower visitor with the low-
est degree (in case of ties we order extinc-
tions by species name). 2.Remove all in-
tralayer edges of the flower visitor. 3.Re-
move any plants in the layer that remain
disconnected (i.e., have degree 0). 4.Re-
move any parasitoids that remain discon-
nected. 5.Record the proportion of sur-
viving plants and the proportion of surviv-
ing parasitoids (separately) as a function of
the proportion of flower visitors that have
been removed. 6.Repeat until all para-
sitoids are extinct in the layers.1.A removal of a plant from the
flower-visitor layer entails immedi-
ate removal of a plant from the par-
asitoid layer. 2.We use propor-
tions, because it allows us to com-
pare the different scenarios.
Multilayer: Remove
flower visitors and fol-
low parasitoids’ ter-
tiary extinctions; we
also include uniformly
random plant removal1.Remove the flower visitor with the low-
est degree (in case of ties we order extinc-
tions by species name). 2.Remove all
intralayer edges of the flower visitor. 3.
Remove any plants in the layer that re-
main disconnected (i.e., have degree 0). 4.
Select a plant from the surviving plants
uniformly at random and remove it with
a probability of prand.5.Remove any
parasitoids that remain disconnected. 6.
Record the proportion of surviving plants
and the proportion of surviving parasitoids
(separately) as a function of the proportion
of flower visitors that have been removed.
7.Repeat until all parasitoids are extinct
in the layers.1.Same as in the previous scenario
but incorporates a uniformly ran-
dom plant extinction with proba-
bility prand (we consider examples
with prand = 0.3 and prand = 0.8)
to acknowledge the fact that ex-
tinctions can happen not only from
removal of flower visitors but also
from other processes that operate
directly on plants. 2.Because of
the random plant extinctions, we
average results over 100 instantia-
tions.
38

1 Supplementary Note 1: Multilayer Modularity

Maximization
We detect communities using a multilayer modularity quality function Qmultislice50.
Consider a (weighted or unweighted) unipartite multilayer network whose
only interlayer edges are diagonal (i.e., interlayer edges only connect nodes
to their counterparts across layers). With the standard (Newman–Girvan)
null model, multilayer modularity is50
Qmultislice =1
2µ/summationdisplay
ijsr/bracketleftbigg/parenleftbigg
Aijs−γskiskjs
2ms/parenrightbigg
δsr+δijωjsr/bracketrightbigg
δ(gis,gjr), (1)
whereAijsis the weight of the intralayer edge between nodes iandjon layer
s(for an unweighted network, the weight is exactly 1 if there is an edge), the
tensor element ωjsrgives the weight of the interlayer edge between node j
on layerrand nodejon layers, the resolution-parameter value on layer sis
γs, the quantity kisis the intralayer strength (i.e., weighted degree) of node
ion layers(andkjsis defined analogously), 2 msis the total edge weight
in layers, the setgisis the community that includes node-layer ( i,s) (and
gjsis defined analogously), we denote the Kronecker delta between indices
xandybyδxy(which is equal to 1 for x=yand equal to 0 for x/negationslash=y),
and 2µ=/summationtext
ijsAijs. In an ordinal multilayer network (e.g., for the usual
multilayer representation of a temporal network), ωjsrcan be nonzero only
whensandrare consecutive layers.
For our calculations, we need to adjust the null-model contribution
Pijs=γskiskjs
2ms, which gives the expected number of interactions between
nodesiandjin layers, to consider the bipartite structure of a host–parasite
network. That is, hosts are only allowed to be adjacent to parasites, and
parasites are only allowed to be adjacent to hosts. (See refs.51,52,123and
references therein for discussions of null models for multilayer modularity
maximization.) To do this, we adapt Eq. (15) from Barber et. al124and
write a bipartite multilayer modularity as
QB=1
2µ/summationdisplay
ijsr/bracketleftbigg/parenleftbigg
Aijs−γskisdjs
ms/parenrightbigg
δsr+δijωjsr/bracketrightbigg
δ(gis,gjr), (2)
wherekisis the strength of host ion layersanddjsis the strength of parasite
jon layers. We modify the GenLouvain code77from http://netwiki.
amath.unc.edu/GenLouvain/GenLouvain (see Supplementary Table 2 and
theMatlab code in the Supplementary Information) by changing the null
modelPijs. We use the default resolution-parameter value of γ= 150,51.
We maximize the quality function (2) using a generalized Louvain-like
locally greedy algorithm77(see51for details on the generalized Louvain al-
gorithm and125for the original Louvain algorithm), which also determines
39

module affiliation50,51. As with all other algorithms for maximizing modu-

larity in monolayer networks48,49,126, the Louvain algorithm is a heuristic,
and it yields a network partition that corresponds to a local maximum of
modularity (and, importantly, we note that the modularity landscape has
a very large number of “good” local maxima)127. It is also a stochastic al-
gorithm, so different results can (and generally do) yield different module
assignments and somewhat different modularity values. We thus average
the values of the maximized modularity QBand the number nof modules
across 100 instantiations of the algorithm and present results for the mean
modularity QBand mean number nof modules.
The generalized Louvain algorithm77receives as input a modularity ma-
trix, which is calculated according to the supra-adjacency matrix of a net-
work. (See Eq. (2), and Supplementary Fig. 1b for an example of a supra-
adjacency matrix.) Given any supra-adjacency matrix that represents a
diagonally-coupled multilayer network, we have a viable input for optimiz-
ing multilayer modularity of that network. The output of one instantiation
of the generalized Louvain algorithm77algorithm is a value for QBand an
assignment of each state node to a module.
2 Supplementary Note 2: Analysis of a Synthetic
Network with Two Interaction Types
The network in Fig. 2 has a ‘planted’ modular structure of 3 modules in
each layer. We use a binary (i.e., unweighted) form of the network, in which
intralayer edges have a value of 1 if they exist and 0 if they do not10. For sim-
plicity, we assume that all interlayer edges have the same weight. To present
the results of the optimization procedure, we report mean values ( QB,n),
which we obtain by averaging over 100 instantiations of the modularity-
maximization algorithm, for multilayer modularity QBand the number nof
modules.
3 Supplementary Note 3: Analysis of a Temporal
Network
3.1 Interlayer Edge Weights on a Comparable Scale as In-
tralayer Edges.
We examine a temporal network that encodes the infection of 22 small mam-
malian host species by 56 ectoparasite species (mites and fleas) during six
consecutive summers in Siberia (1982–1987), giving 6 different layers. (See
Refs.54,55for more details on data.) Not all species occur in all years. The
data set also includes the number of host specimens collected in the field
and the number of parasites recovered from each host specimen. We use
40

these individual-level estimates of host and parasite abundances determine,

respectively, the weights of intralayer and interlayer edges. The weight of an
intralayer edge between host iand parasite jat timetisAijt=mt
ij=pt
ij/at
i,
whereAijtdenotes an intralayer edge weight in layer t, the quantity pt
ijis the
number of parasite specimens of species jrecovered from all of the specimens
of hostiat timet, andat
iis the number of specimens of host iat timet.
The interlayer edge that connects host species iat timetto its counterpart
at timet+ 1 has a weight ωit,t+1=nt→t+1
i =at+1
i/at
igiven by the relative
change in host abundance between the time points. An interlayer edge from
nodeiin layerrto nodeiin layershas weightωirs, where we have included
a comma in the subscript between r=tands=t+ 1 for clarity and where
we takeωirs=ω= constant in the synthetic network example. Analogously,
the interlayer edge that connects parasite species jat timetto its counter-
part at time t+ 1 has a weight ωjt,t+1=nt→t+1
j =pt+1
j/pt
jgiven by the
relative change in the total abundance of the parasite (across all hosts). We
do not include an interlayer edge between corresponding species in layers t
andt+ 1 when (i) a species occurs in layer tbut not in layer t+ 1 or (ii)
when a species does not occur in layer tbut occurs in layer t+ 1.
The above formulation of edge weights is ecologically meaningful, as the
weights reflect temporal dynamics in species abundances. It also allows one
to model how changes in species abundances across time affect host–parasite
interactions in each time point. For example, the future infection of a host
iwith parasite jdepends on the interlayer edges and abundance values as
follows:
mt,t+1
ij =pt+1
ij
at+1
i=pt+1
ij
nt→t+1
iat
i=nt→t+1
jpt
j
nt→t+1
iat
i. (3)
Although we do not use the information in Eq. (3), it may be helpful for re-
searchers who are interested in various questions. For example, one can ask
how does reducing the abundance of a given host (e.g., by control programs)
will affect future interactions of the host with parasites, or, more generally,
network architecture. The formulation of edge weights gives nonzero weights
to non-diagonal interlayer edges, which was done previously (and very effec-
tively) for temporal networks of disease propagation in Valdano et al.128.
It will also be instructive, for the sake of comparison with the two ex-
treme scenarios ω= 0 andω=∞(see Section 3.2), to indicate the numerical
values of the weights of interlayer edges, which vary between 0 .02 and 10.33.
We show the distribution of the interlayer edge values in Supplementary
Fig. 4.
For simplicity, in all of our calculations of the host–parasite temporal
network, we follow previous studies and assume that the intralayer and
interlayer50,51,56edges are undirected. It will be interesting to relax this
assumption in future studies of multilayer temporal networks with interlayer
connectivity.
41

3.1.1 Testing Hypotheses About the Modular Structure

We now compare the modular structure in the observed network to that ob-
tained from three different null models19,62. For each null model, we test the
hypothesis that the most modular partition of the observed temporal net-
work is more modular than the most modular partition of shuffled networks,
and we calculate the significance of the modularity QOBS
B of the observed
network’s most modular partition as the proportion of the 100 shuffled net-
works that have a lower maximized modularity than that of the observed
network56. We also calculate the number of modules and the ‘adjustabil-
ity’ of hosts and parasites as the proportion of nodes of a given type that
change modules at least once. For comparative applications of adjustability,
we take into account the total number of temporal layers. We compare our
results with shuffled networks to the values that we obtain for the observed
network using one-sample t-tests19.
In the test with the first null model, we shuffle the intralayer edges in
each layer to test the hypothesis that modular structure over time depends
on temporally-local interaction patterns. In each layer, we first shuffle the
total number of parasites that are on a given host. We use the r0both
algorithm from the vegan package129inR. The algorithm shuffles the val-
ues and position of the matrix entries within each row, while preserving
the total sum of the matrix. It therefore conserves both the number of
parasite species that infect a given host and the mean abundance of those
species. We then divide each matrix entry of each shuffled matrix by the
host abundance in that particular layer (i.e., year) to transform the abun-
dance to prevalence values. We find statistically significant support for the
hypothesis (see Supplementary Table 5). Additionally, the observed net-
work differs in a statistically significant way from the shuffled networks with
respect to both number of modules and host and parasite adjustability (see
Supplementary Table 5).
In the second test, we shuffle the interlayer edges between each pair of
consecutive layers to test the hypothesis that the modular structure depends
on the identity of hosts and parasites that change in abundance. To con-
struct this null model, which is equivalent to the ‘nodal’ null model from
Bassett et al.19, we separately permute the orders of node labels of hosts and
parasites. We permute the node labels separately in each layer so that node
identity is not preserved. In the bipartite structure of a given layer, this is
akin to permuting the order of rows (for hosts) or columns (for parasites)
in each layer. We find that (i) the most modular partition of the observed
network is more modular than the most modular partition of the shuffled
networks (see Supplementary Table 5) and (ii) their temporal structures
differ. Specifically, the most modular partition of the observed network has
approximately 6 times fewer modules than when shuffling the order of hosts
and approximately 15 times fewer modules than when shuffling the order
42

of parasites. Additionally, in the shuffled networks, almost all hosts and

parasites switch modules at least once (see Supplementary Table 5).
In the third test, we permute the order of the layers to test the hypothesis
that the order of time affects the network’s modular structure. To construct
this null model, which is equivalent to the ‘temporal’ null model of Bassett
et al.19, we draw a uniformly random permutation of the sequence {1,..., 6}
(e.g., (4,3,2,5,1,6)) to give a new ordering of the layers. We thereby destroy
the temporal identity of the layers. Note that the interlayer edges are not
shuffled. We find that the most modular partition of the observed network is
not more modular than the most modular partition of the randomized net-
works (see Supplementary Table 5), but that the observed and randomized
networks have different temporal structures. The most modular partition
of the observed network has about 2 .5 times fewer modules than those of
the randomized networks, and it also has different proportions of hosts and
parasites that switch modules (see Supplementary Table 5).
3.2 Interlayer Edge Weights of 0or Infinity
We compare the modular structure of the network with weighted intralayer
and interlayer edges to the modular structure of two versions of the same
network but in which interlayer edge values are set to either 0 or are very
large. We perform these computations in the same manner as we did for the
interlayer-weighted network that we discussed in detail in Section 3.1. We
again use the generalized Louvain algorithm from Jutla et al.77and Mucha
et al.50. As a proxy for the scenario with infinitely large interlayer edge
weights using numerical calculations, we set ω= 1000. (Technically, we are
thus approximate in our exploration of this extreme situation.) To explore
the scenario with 0 interlayer edge weights, we examine 6 independent mod-
ularity matrices (corresponding to the 6 layers). We also provide the code
for these computations.
Very large interlayer edge weights. When the weight of all interlayer
edges is very large, processes that govern interactions across contiguous lay-
ers (e.g., species persistence, survival, or changes in abundance) have a much
larger effect on community structure than those that govern interactions
within layers. Because the effect of species connectivity across time is much
stronger than that of the intralayer processes, we expect species to remain
in the same module over time. In the limit of infinite-weight interlayer edges
for all species, all state nodes that correspond to a given species must al-
ways be assigned to the same module. In our computation with interlayer
edge weights of ω= 1000, we obtain QB≈1, and the maximum-modularity
partition of the network has n= 3 modules. By construction, no species
changes its module affiliation across time (see Supplementary Table 4).
43

Interlayer edge weights of 0.At the other extreme is a multilayer net-

work in which all of the interlayer edges have a weight of 0, which is akin
to maximizing modularity separately in each of the layers. In this case, the
modules that one finds in different layers are independent of each other,
as they reflect only the internal structure of each layer50,51. This is the
same situation as calculating modularity for two disconnected networks in
the toy example (see Fig. 2b). From an ecological perspective, one cannot
understand the effect of interlayer phenomena (e.g., temporal changes in
species abundance or in network architecture) on modular structure in a di-
rect manner if one has exclusively 0 interlayer edge weights. Although some
quantification of module evolution may be possible53, such a quantification
does not simultaneously encompass processes that occur within and between
layers. For example, if a species interacts strongly with a given set of species
in layertbut with another set of species in layer t+ 1, one can argue that
this is a change in the modular structure over time. However, even though
it is possible, it need not be true, because the modules have been calculated
independently for two distinct networks. Moreover, if a community splits
in time, including nonzero interlayer edges provides a natural way to track
which nodes leave to form a new community and which remain part of the
original community.
When all interlayer edge weights are 0, we find that both QBandnvary
among the 6 layers. (See Supplementary Table 4 for the exact values.) It
is not possible to calculate changes in module composition (including host
and parasite adjustability) in an unambiguous way, because modules are
defined separately in a layer. One can, of course, attempt ad hoc analyses
of modular structure changes in time even when interlayer edge weights are
0.
3.3 Calculation of Normalized Mutual Information
To support the claim that modularity maximization in a multilayer network
provides additional information about the assignment of species to modules
compared to independent computations in all individual layers, we quan-
tify the amount of novel information from the addition of interlayer edges
for the assignment of species to modules. To do this, we use a measure of
normalized mutual information (NMI) based on information theory. This
quantity has been used previously to evaluate methods that assign nodes
to modules130,131. For each of the six layers, the value of NMI is 1 if the
composition of modules is identical in the monolayer and the multilayer net-
works, and it is 0 if they are completely uncorrelated. That is, a value of
1 means that one has complete information on the assignment of species to
modules in the temporal network from independent computations with the
corresponding layer. In that case, the interlayer edges have not given any
additional information. In contrast, a value of 0 means that observing a sin-
44

gle layer at a time provides no information whatsoever on the assignment of

species to modules in the multilayer representation of the temporal network.
Naturally, the only way to know if one gets the same answer is to perform
the multilayer calculation in the first place. We find that across layers the
mean NMI (±standard deviation) is 0 .35±0.12. (Thus, the range of values
is 0.19−−0.51, and we recall that there are M= 6 layers.) Hence, the calcu-
lation of modularity of each layer by itself contains, on average, only about
35% of the information on species assignment to modules in the multilayer
network with the interlayer edges. Combined with the fact that we can use
the interlayer edges at time t−1 to calculate the intralayer edges at time
t(see Eq. 3), this gives compelling evidence that including interlayer edges
provides new information on the temporal organization of the network.
We also calculate the amount of information that is contained in the
assignment of species to modules in an aggregated network compared to
module assignments in any individual layer. We use two versions of an
aggregated multilayer network: (i) we sum the intralayer edge weights (i.e.,
interaction values) across the 6 layers; and (ii) we take the mean of the edge
weights across the 6 layers. We find on average that the correlation in the
information is about 0 .521±0.029 and 0.517±0.032 for the two types of
aggregated networks, respectively. Consequently, the affiliation of species
to modules in either of these aggregated networks provides only about 52%
of the information on their affiliation to modules in the multilayer network
(which includes interlayer edges).
3.4 Reducibility of the Host–Parasite Network
To examine if we need all 6 layers in the host–parasite network to accurately
describe structure in the host–parasite network, we calculate reducibility of
the network58. Briefly, reducibility is a measure of the structural redundancy
of different layers with respect to an aggregated network G=G[1]+G[2]+
···+G[M], where{G[1],G[2],...,G[M]}is the set of adjacency matrices from
theMindividual layers of a multilayer network. Intuitively, if all of the
layers are the same, then there is complete redundancy, and a single layer
includes all of the information about intralayer structure. However, if each
layer includes different intralayer edges, then we need to retain all layers to
fully describe structure in a multilayer network, and aggregation leads to
loss of information.
From our computation, we see for our temporal network that we need
all 6 layers to accurately describe its structure (see Supplementary Fig. 6).
More generally, temporal interaction dynamics are accompanied by a tem-
poral ordering of interactions, which are also often bursty, so aggregating
temporal layers into a time-independent network (especially if done naively,
as is common) removes all information about the ordering of interactions
and can lead to qualitatively incorrect conclusions57,132.
45

4 Supplementary Note 4: Further Details on Ro-

bustness Analysis
We use empirical data from Pocock et al.38, who assembled a network com-
posed of 10 layers, which are connected to each other mostly via plants.
Layers in the original network include interactions of plants with granivo-
rous birds, rodents, butterflies and other flower visitors, aphids, granivorous
insects, and leaf-miner parasitoids. For simplicity, we select a subset of the
10 layers and form a diagonally-coupled multilayer network with two layers.
Our network describes two interconnected mutualistic interactions, although
the mutualism in the leaf-miner parasitoid layer is indirect. The network
includes 189 flower visitors, 60 leaf-miner parasitoids, and 17 plants. For
simplicity, we use a binary version of the original data.
Pocock et al.38studied the effect of plant removal on the simultaneous
co-extinction of the species that are interconnected to them. However, they
did not study how disturbance percolates through plants, and they thus did
not explore how the multilayer structure of their network affects cascading
perturbations. In the context of our example, extinction of pollinators en-
tails co-extinction of plants, and extinction of plants entails co-extinction
of leaf-miner parasitoids. Because an extinction of plants in one layer au-
tomatically implies their extinction in the second layer, we assume that the
weights of interlayer edges are infinity. We modify an established approach
for examining co-extinctions38,133so that we can use it with multilayer net-
works (see Supplementary Table 6).
We explore another option to compare the robustness of a plant–parasitoid
monolayer network (which is essentially one layer of the multilayer network)
to that of a flower visitor–plant–parasitoid multilayer network. Unlike in
the main text, we remove (in the same order) the same set of nodes (flower
visitors) in the monolayer network and in the multilayer network, and we
track the total proportion of species that remain in each of these networks.
Specifically, in the monolayer network, we remove flower visitors in order
of increasing degree and then compute the proportion of plants that re-
main. In the multilayer network, we remove the same flower visitors in the
same order and then track the proportions of both plants and parasitoids
that remain. In the multilayer network, we find that the total proportion
of surviving species is larger and that the pace of extinction is slower (see
Supplementary Fig. 5).
Like standard simple approaches for studying network robustness63, our
calculations have the flavour of a percolation process. See Nagler et al.134for
an example of a percolation-based robustness study of an EMLN. Multilayer
networks allow one to examine a much richer variety of percolation processes
than in monolayer networks14,15,116, and that in turn allows increasingly
nuanced analyses of species robustness in ecological networks in a wealth of
46

scenarios.
References
4. Olesen, J. M., Bascompte, J., Dupont, Y. L. & Jordano, P. The modularity of
pollination networks. Proc. Natl. Acad. Sci. U. S. A. 104, 19891–198916 (2007).
8. Olesen, J. M., Stefanescu, C. & Traveset, A. Strong, long-term temporal dynamics
of an ecological network. PLoS One 6,e26455 (2011).
9. Meli´ an, C. J., Bascompte, J., Jordano, P. & Krivan, V. Diversity in a complex
ecological network with two interaction types. Oikos 118, 122–130 (2009).
10. Fontaine, C. et al. The ecological and evolutionary implications of merging di fferent
types of networks. Ecol. Lett. 14,1170–1181 (2011).
12. K´ efi, S., Miele, V., Wieters, E. A., Navarrete, S. A. & Berlow, E. L. How Struc-
tured Is the Entangled Bank? The Surprisingly Simple Organization of Multiplex
Ecological Networks Leads to Increased Persistence and Resilience. PLoS Biol. 14,
e1002527 (2016).
13. Blonder, B., Wey, T. W., Dornhaus, A., James, R. & Sih, A. Temporal dynamics
and network analysis. Methods Ecol. Evol. 3,958–972 (2012).
14. Kivel¨ a, M. et al. Multilayer networks. J. Complex Networks 2,203–271 (2014).
15. Boccaletti, S et al. The structure and dynamics of multilayer networks. Phys. Rep.
544, 1–122 (2014).
19. Bassett, D. S. et al. Dynamic reconfiguration of human brain networks during
learning. Proc. Natl. Acad. Sci. U. S. A. 118, 7641–7646 (2011).
20. Gilarranz, L. J., Sabatino, M., Aizen, M. & Bascompte, J. Hot spots of mutualistic
networks. J. Anim. Ecol. 84,407–413 (2014).
21. Kitching, R. L. Spatial and Temporal Variation in Food Webs in Water-Filled
Treeholes. Oikos 48,280–288 (1987).
22. Moore, J. C. & de Ruiter, P. C. Temporal and spatial heterogeneity of trophic
interactions within below-ground food webs. Agric. Ecosyst. Environ. 34,371–397
(1991).
23. Schoenly, K. & Cohen, J. E. Temporal variation in food web structure: 16 empirical
cases. Ecol. Monogr. 61,267–298 (1991).
26. Winemiller, K. O. & Jepsen, D. B. Effects of seasonality and fish movement on
tropical river food webs. J. Fish Biol. 53,267–296 (1998).
27. Poisot, T. et al. The dissimilarity of species interaction networks. Ecol. Lett. 15,
1353–1361 (2012).
29. Trøjelsgaard, K., Jordano, P., Carstensen, D. W. & Olesen, J. M. Geographical
variation in mutualistic networks: similarity, turnover and partner fidelity. Proc. R.
Soc. B 282, 20142925 (2015).
37. Rudolf, V. H. W. & Lafferty, K. D. Stage structure alters how complexity affects
stability of ecological networks. Ecol. Lett. 14,75–79 (2011).
38. Pocock, M. J. O., Evans, D. M. & Memmott, J. The robustness and restoration of
a network of ecological networks. Science 335, 973–977 (2012).
39. Pilosof, S. et al. Host-parasite network structure is associated with community-level
immunogenetic diversity. Nat. Commun. 5,5172 (2014).
47

40. Bauer, S & Hoye, B. J. Migratory animals couple biodiversity and ecosystem func-
tioning worldwide. Science 344, 1242552 (1–8) (2014).
42. Pilosof, S., Greenbaum, G., Krasnov, B. R. & Zelnik, Y. R. Asymmetric disease
dynamics in multi-host interconnected networks. arXiv:1512. 09178 [q-bio] (2016).
43. Nakano, S. & Murakami, M. Reciprocal subsidies: dynamic interdependence be-
tween terrestrial and aquatic food webs. Proc. Natl. Acad. Sci. U. S. A. 98,166–
170 (2001).
47. Scotti, M, Ciocchetta, F & Jord´ an, F. Social and landscape effects on food webs: a
multi-level network simulation model. J. Complex Networks 1,160–182 (2013).
48. Porter, M. A., Onnela, J.-P. & Mucha, P. J. Communities in networks. Not Am
Math Soc 56,1082–1097 (2009).
49. Fortunato, S. & Hric, D. Community detection in networks: A user guide. Phys.
Rep. 659, 1–44 (2016).
50. Mucha, P. J., Richardson, T., Macon, K., Porter, M. A. & Onnela, J.-P. Community
structure in time-dependent, multiscale, and multiplex networks. Science 328, 876–
878 (2010).
51. Bazzi, M. et al. Community detection in temporal multilayer networks, with an
application to correlation networks. Multiscale Model. Simul. 14,1–41 (2016).
52. Sarzynska, M., Leicht, E. A., Chowell, G. & Porter, M. A. Null models for com-
munity detection in spatially embedded, temporal networks. Journal of Complex
Networks 4,363–406 (2015).
53. Dupont, Y. L. & Olesen, J. M. Stability of modular structure in temporal cumula-
tive plant–flower-visitor networks. Ecol. Complex. 11,84–90 (2012).
54. Krasnov, B. R., Matthee, S., Lareschi, M., Korallo-Vinarskaya, N. P. & Vinarski,
M. V. Co-occurrence of ectoparasites on rodent hosts: null model analyses of data
from three continents. Oikos 119, 120–128 (2010).
55. Pilosof, S., Fortuna, M. A., Vinarski, M. V., Korallo-Vinarskaya, N. P. & Krasnov,
B. R. Temporal dynamics of direct reciprocal and indirect effects in a host-parasite
network. J. Anim. Ecol. 82,987–996 (2013).
56. Fortuna, M. A. et al. Nestedness versus modularity in ecological networks: two sides
of the same coin? J. Anim. Ecol. 79,811–817 (2010).
57. Holme, P. Modern temporal network theory: A colloquium. Eur. Phys. J. B 1508.01303,
1–30 (2015).
58. De Domenico, M., Nicosia, V., Arenas, A. & Latora, V. Structural reducibility of
multilayer networks. Nat. Commun. 6(2015).
61. Genrich, C. M., Mello, M. A. R., Silveira, F. A. O., Bronstein, J. L. & Paglia, A.
P. Duality of interaction outcomes in a plant–frugivore multilayer network. Oikos
(2016).
62. Bassett, D. S. et al. Robust detection of dynamic community structure in networks.
Chaos 23,013142 (2013).
63. Newman, M. E. J. Networks: An Introduction (Oxford University Press, Oxford,
UK, 2010).
68. Lurgi, M., Montoya, D. & Montoya, J. M. The effects of space and diversity of
interaction types on the stability of complex ecological networks. Theor. Ecol. 1–11
(2015).
48
77. Jutla, I. S., Jeub, L. G. S. & Mucha, P. J. A generalized Louvain method for commu-
nity detection implemented in MATLAB, version 2.0.
http://netwiki.amath.unc.edu/GenLouvain/GenLouvain
(2014).
78. Carpenter, S. R. et al. Regulation of Lake Primary Productivity by Food Web
Structure. Ecology 68,1863–1876 (1987).
79. Warren, P. H. Spatial and temporal variation in the structure of a freshwater food
web. Oikos, 299–311 (1989).
80. Winemiller, K. O. Spatial and temporal variation in tropical fish trophic networks.
Ecol. Monogr. 60,331–367 (1990).
81. Berninger, U.-G., Wickham, S. A. & Finlay, B. J. Trophic coupling within the
microbial food web: a study with fine temporal resolution in a eutrophic freshwater
ecosystem. Freshw. Biol. 30,419–432 (1993).
82. Yoshioka, T., Wada, E. & Hayashi, H. A Stable Isotope Study on Seasonal Food
Web Dynamics in a Eutrophic Lake. Ecology 75,835–846 (1994).
83. Tavares-Cromar, A. F. & Williams, D. D. The importance of temporal resolution
in food web analysis: evidence from a detritus-based stream. Ecol. Monogr. 66,
91–113 (1996).
84. Thompson, R. M. & Townsend, C. R. The effect of seasonal variation on the com-
munity structure and food-web attributes of two streams: implications for food-web
science. Oikos, 75–88 (1999).
85. Fisher, S. J., Brown, M. L. & Willis, D. W. Temporal food web variability in an up-
per Missouri River backwater: energy origination points and transfer mechanisms.
Ecol. Freshw. Fish 10,154–167 (2001).
86. Alarc´ on, R., Waser, N. M. & Ollerton, J. Year-to-year variation in the topology of
a plant–pollinator interaction network. Oikos 117, 1796–1807 (2008).
87. Basilio, A. M., Medan, D., Torretta, J. P. & Bartoloni, N. J. A year-long plant-
pollinator network. Austral Ecol. 31,975–983 (2006).
88. Petanidou, T., Kallimanis, A. S., Tzanopoulos, J., Sgardelis, S. P. & Pantis, J. D.
Long-term observation of a pollination network: fluctuation in species and inter-
actions, relative invariance of network structure and implications for estimates of
specialization. Ecol. Lett. 11,564–575 (2008).
89. Olesen, J. M. et al. Temporal dynamics in a pollination network. Ecology 89,1573–
1582 (2008).
90. Dupont, Y. L., Padr´ on, B., Olesen, J. M. & Petanidou, T. Spatio-temporal variation
in the structure of pollination networks. Oikos 118, 1261–1269 (2009).
91. Henson, K. S. E., Craze, P. G. & Memmott, J. The restoration of parasites, para-
sitoids, and pathogens to heathland communities. Ecology 90,1840–1851 (2009).
92. Olesen, J. M. et al. inAdvances in Ecological Research (ed Woodward, G.) 1–69
(Academic Press, 2010).
93. Fang, Q. & Huang, S.-Q. Relative stability of core groups in pollination networks
in a biodiversity hotspot over four years. PLoS One 7,e32663 (2012).
94. D´ ıaz-Castelazo, C., S´ anchez-Galv´ an, I. R., Guimar˜ aes, P. R., Raimundo, R. L. G.
& Rico-Gray, V. Long-term temporal variation in the organization of an ant–plant
network. Ann. Bot. 111, 1285–1293 (2013).
95. Rasmussen, C., Dupont, Y. L., Mosbacher, J. B., Trøjelsgaard, K. & Olesen, J. M.
Strong impact of temporal resolution on the structure of an ecological network.
PLoS One 8,e81694 (2013).
49

96. Saavedra, S., Rohr, R. P., Fortuna, M., Selva, N. & Bascompte, J. Seasonal species
interactions minimize the impact of species turnover on the likelihood of community
persistence. Ecology (2015).
97. Ramos-Robles, M., Andresen, E. & D´ ıaz-Castelazo, C. Temporal changes in the
structure of a plant-frugivore network are influenced by bird migration and fruit
availability. PeerJ 4,e2048 (2016).
98. Deegan, L. A. & Garritt, R. H. Evidence for spatial variability in estuarine food
webs. Mar. Ecol. Prog. Ser. 147, 31–47 (1997).
99. Polis, G. A., Anderson, W. B. & Holt, R. D. Toward an integration of landscape
and food web ecology: the dynamics of spatially subsidized food webs. Annu. Rev.
Ecol. Syst. 289–316 (1997).
100. Richardson, K et al. Spatial heterogeneity in the structure of the planktonic food
web in the North Sea. Mar. Ecol. Prog. Ser. 168, 197–211 (1998).
101. Evans, D. M., Pocock, M. J. O. & Memmott, J. The robustness of a network of
ecological networks to habitat loss. Ecol. Lett. 16,844–852 (2013).
102. Sauve, A. M. C., Fontaine, C. & Th´ ebault, E. Structure-stability relationships in
networks combining mutualistic and antagonistic interactions. Oikos 123, 378–384
(2014).
103. Sander, E. L., Wootton, J. T. & Allesina, S. What can interaction webs tell us
about species roles? PLoS Comput. Biol. 11,e1004330 (2015).
104. Nakano, S., Miyasaka, H. & Kuhara, N. TERRESTRIAL–AQUATIC LINKAGES:
RIPARIAN ARTHROPOD INPUTS ALTER TROPHIC CASCADES IN A STREAM
FOOD WEB. Ecology 80,2435–2441 (1999).
105. Meli´ an, C. J., Bascompte, J. & Jordano, P. in Complexity in Aquatic Food Webs: An
Ecosystem Approach (eds Belgrano, A., Scharler, U. M., Dunne, J. A. & Ulanowicz,
R. E.) 19–24 (Oxford University Press, 2005).
106. Hagen, M. et al. inAdvances in Ecological Research (ed Ute Jacob And) 89–210
(Academic Press, 2012).
107. Barter, E. & Gross, T. Meta-food-chains as a many-layer epidemic process on net-
works. Phys Rev E 93,022303 (2016).
108. Barrett, L., Henzi, S. P. & Lusseau, D. Taking sociality seriously: the structure of
multi-dimensional social networks as a source of information for individuals. Philos.
Trans. R. Soc. Lond. B Biol. Sci. 367, 2108–2118 (2012).
109. Bianconi, G. Statistical mechanics of multiplex networks: entropy and overlap.
Physical Review E 87,062806 (2013).
110. De Domenico, M., Sol´ e-Ribalta, A., Omodei, E., G´ omez, S. & Arenas, A. Ranking in
interconnected multilayer networks reveals versatile nodes. Nat. Commun. 6(2015).
111. Cozzo, E. et al. Structure of triadic relations in multiplex networks. New J. Phys.
17,073029 (2015).
112. Jeub, L. G. S., Balachandran, P., Porter, M. A., Mucha, P. J. & Mahoney, M. W.
Think locally, act locally: detection of small, medium-sized, and large communities
in large networks. Phys. Rev. E 91,012821 (2015).
113. De Domenico, M., Lancichinetti, A., Arenas, A. & Rosvall, M. Identifying modular
flows on multilayer networks reveals highly overlapping organization in intercon-
nected systems. Phys. Rev. X 5,011027 (2015).
114. Peixoto, T. P. Hierarchical block structures and high-resolution model selection in
large networks. Phys. Rev. X 4,011047 (2014).
50

115. Peixoto, T. P. Inferring the mesoscale structure of layered, edge-valued and time-
varying networks. Phys. Rev. E 92,042807 (2015).
116. Lee, K.-M., Min, B. & Goh, K.-I. Towards real-world complexity: an introduction
to multiplex networks. Eur. Phys. J. B 88,1–20 (2015).
117. Battiston, F., Nicosia, V. & Latora, V. Structural measures for multiplex networks.
Phys. Rev. E 89,032804 (2014).
118. Holme, P. & Saram¨ aki, J. Temporal networks. Phys. Rep. Temporal Networks 519,
97–125 (2012).
119. De Domenico, M. D., Porter, M. A. & Arenas, A. MuxViz: a tool for multilayer
analysis and visualization of networks. J. Complex Networks 3,159–176 (2015).
120. Poisot, T. betalink: Beta-Diversity of Species Interactions 2016.
121. Blonder, B. timeordered: Time-ordered and time-aggregated network analyses.
https://CRAN.R-
project.org/package=timeordered (2015).
122. Rivero Ostoic, A. multiplex: Algebraic Tools for the Analysis of Multiple Social
Networks. https://CRAN.R-project.org/package=multiplex (2016).
123. Danchev, V. & Porter, M. A. Heterogeneity of global and local connectivity in
spatial network structures of world migration. arXiv 1603.09313 (2016).
124. Barber, M. Modularity and community detection in bipartite networks. Physical
Review E 76,1–9 (2007).
125. Blondel, V. D., Guillaume, J.-L., Lambiotte, R. & Lefebvre, E. Fast unfolding of
communities in large networks. J. Stat. Mech. 2008, P10008 (2008).
126. Fortunato, S. Community detection in graphs. Phys. Rep. 486, 75–174 (2010).
127. Good, B. H., de Montjoye, Y.-A. & Clauset, A. Performance of modularity maxi-
mization in practical contexts. Phys. Rev. E 81,046106 (2010).
128. Valdano, E., Ferreri, L., Poletto, C. & Colizza, V. Analytical computation of the
epidemic threshold on temporal networks. Phys. Rev. X 5(2015).
129. Oksanen, J. et al. vegan: Community Ecology Package (version 2.3-1). https://CRAN.R-
project.org/package=vegan (2015).
130. Danon, L., D´ ıaz-Guilera, A., Duch, J. & Arenas, A. Comparing community struc-
ture identification. J. Stat. Mech. 2005, P09008 (2005).
131. Guimer` a, R., Sales-Pardo, M. & Amaral, L. A. N. Module identification in bipartite
and directed networks. Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 76,036102
(2007).
132. Hoffmann, T., Porter, M. A. & Lambiotte, R. Generalized master equations for
non-Poisson dynamics on networks. Phys. Rev. E Stat. Nonlin. Soft Matter Phys.
86,046102 (2012).
133. Memmott, J., Waser, N. M. & Price, M. V. Tolerance of pollination networks to
species extinctions. Proc. R. Soc. B 271, 2605–2611 (2004).
134. Stollmeier, F., Geisel, T. & Nagler, J. Possible origin of stagnation and variability
of Earths biodiversity. Phys. Rev. Lett. 112(2014).
51