The structure and robustness of ecological networks with two interaction types

Virginia Domínguez-García; Sonia Kéfi

doi:10.1371/journal.pcbi.1011770

. 2024 Jan 19;20(1):e1011770. doi: 10.1371/journal.pcbi.1011770

The structure and robustness of ecological networks with two interaction types

Virginia Domínguez-García ^1,^2,^*, Sonia Kéfi ^1,³

Editor: Mercedes Pascual⁴

PMCID: PMC10830016 PMID: 38241353

Abstract

Until recently, most ecological network analyses investigating the effects of species’ declines and extinctions have focused on a single type of interaction (e.g. feeding). In nature, however, diverse interactions co-occur, each of them forming a layer of a ‘multilayer’ network. Data including information on multiple interaction types has recently started to emerge, giving us the opportunity to have a first glance at possible commonalities in the structure of these networks. We studied the structural features of 44 tripartite ecological networks from the literature, each composed of two layers of interactions (e.g. herbivory and pollination), and investigated their robustness to species losses. Considering two interactions simultaneously, we found that the robustness of the whole community is a combination of the robustness of the two ecological networks composing it. The way in which the layers of interactions are connected to each other affects the interdependence of their robustness. In many networks, this interdependence is low, suggesting that restoration efforts would not automatically propagate through the whole community. Our results highlight the importance of considering multiple interactions simultaneously to better gauge the robustness of ecological communities to species loss and to more reliably identify key species that are important for the persistence of ecological communities.

Author summary

In the face of the current biodiversity crisis, predicting how species loss will affect ecological communities is becoming increasingly relevant. Previous studies including only one type of ecological interactions (e.g. feeding or pollination) revealed the relevance of the structure of ecological networks for the persistence of ecological communities. However, there is mounting evidence that considering multiple interactions simultaneously can alter the results based on a single interaction. Here, we study the robustness of ecological networks with two interaction types to the loss of plant species, and we show that it is a combination of the robustness of the two bipartite ecological networks composing the ecological community. By analyzing networks from multiple communities, we are able to identify commonalities across interaction types, as well as singularities specific to a given interaction type, caused by underlying biological constraints. Our results highlight that a multi-interaction approach is crucial to better gauge the overall robustness of ecological communities, and to correctly determine the relative importance of different plants species at the whole community level, which can be key for biodiversity conservation.

Introduction

The rate of decline of many species populations is accelerating [1], and species extinctions are seriously threatening the functioning of ecological communities worldwide. Understanding how species interact and how this affects the robustness of ecological communities to species loss is essential to anticipate the consequences of biodiversity losses and extinction cascades as well as to design protection and restoration plans. The study of ecological networks—where species are represented by nodes and the ecological interactions by links between these nodes—have contributed significantly to the understanding of how ecological interactions are structured and have unveiled important relationships between network structure and their robustness to species loss [2–5]. However, while the ecological network literature has long been dominated by studies of networks containing a single interaction type, it has become increasingly clear that species in nature are connected by a myriad of interaction types simultaneously and that considering networks which include this diversity of interaction types could greatly improve our knowledge of the structure and dynamics of ecological communities [6–13].

A number of previous studies have investigated the effect of including multiple interaction types on the functioning of ecological communities, especially on their stability [14–20]. Yet the vast majority of these studies have so far remained theoretical. With the publication of the first multi-interaction empirical networks, we begin to know more about their structure [6, 10, 13, 21–27], and how this structure affects their persistence [6, 10] and robustness [21, 24, 27, 28]. In particular, studies on multi-interaction networks have provided new insights on whether the inclusion of several interactions can significantly alter their robustness to species loss [24] and how extinctions propagate through such networks [21]. However, in spite of these pioneering studies, there is currently no consensus about the structure of multi-interaction networks and its consequences for the robustness of ecological communities, in part due to the lack of data sets, whose amount has only recently started to increase.

A key question, of relevance given the current biodiversity crisis, is how robustness varies across network types, and what we can learn from including multiple interactions simultaneously. With this in mind, we gathered ecological networks with multiple interaction types currently available in the literature. More specifically, we focused on tripartite networks because they were the most abundant in the literature, allowing us to compare a wide variety of ecological systems. Tripartite ecological networks are composed of two interaction layers (e.g. pollination and herbivory), each of the bipartite kind [29]. They therefore contain three different species sets (e.g. plant, pollinator and herbivore guilds in a pollination-herbivory network), one of which is shared between the two interaction layers (e.g. plant species can interact with both pollinators and herbivores in a pollination-herbivory network). We call the set of nodes that can have interactions in both interaction layers the shared set, and the subset of nodes in the shared set that have interactions in both interaction layers the connector nodes (see Fig 1A and 1B).

Fig 1 — A) An Herbivory(h)—Pollination(p) tripartite network, where plants (P) are the shared set of species. B) An Herbivory(h)—Parasitism(pa) tripartite network, where herbivores (H) are the shared set of species. Link colours represent the two interaction layers, and node colours the three sets of species. Connector nodes in the shared set of species are highlighted in black. C) Extinction curve showing the fraction of surviving animal species as a function of plant loss for a given plant extinction sequence in network A. The robustness to plant loss, R, is the area under the curve. Extinction protocol: plants (green nodes) are progressively removed from the community in the prescribed order, their corresponding links are erased (colored in red) and animal species are declared extinct (colored in red) whenever they lose all their feeding links. D) Pairwise correlation in the robustness of the two animal sets—interdependence, I—resulting from 3.000 simulations of random sequential loss of plant taxa in network A.

Our data set consists of 44 tripartite networks from 6 different studies, in which the interaction layers include mutualistic (pollination, seed-dispersal and ant-mutualism) and antagonistic (herbivory and parasitism) interactions (see Methods). To identify possible generalities across interaction types as well as singularities specific to a given interaction type, we divided the networks in three types according to the signs of the interactions involved: mutualism-mutualism (MM) if both interactions were positive, antagonism-antagonism (AA) if both interactions were negative, and mutualism-antagonism (MA) if one interaction was positive and the other negative, given that interaction type can determine network architecture through the underlying biological constraints [30].

Using this data set, we investigated how the two interaction layers are connected and the consequences for the robustness of these networks to plant loss. Robustness was assessed by sequentially removing plants in a random order and estimating secondary extinctions (Fig 1C and Methods). Although this approach lacks realism (since there are no underlying temporal dynamics), it has proven useful in understanding the threat that biodiversity loss poses to ecosystem services and functioning [3, 21, 31, 32]. Furthermore, it provides a lower bound on the damages that may be caused to an ecological community since it relies on the conservative hypothesis that secondary extinctions happen only when an animal species has lost all its links. We focused on the extinctions of plants because they are the only group of species, whose disappearance can potentially harm all other species groups, and also because plants can be managed more directly [21]. Note that while plants are not the shared set of species in all networks (see Fig 1), it is still possible to quantify robustness to plant loss in all the networks of our data set (see Methods). Extending the study of robustness to include multiple interactions simultaneously allowed us to study the interdependence of the robustness of animal species sets (Fig 1D), which is relevant to know how cascading extinctions will propagate through a multi-interaction network [21], and to better identify keystone plant species [13, 21], of importance when designing protection and restoration interventions. We used four null models with increasing constraints (see Methods) to study how different structural properties could determine the interdependence and robustness in the tripartite networks.

Taken together, our results suggest that considering multiple ecological interactions simultaneously does not have a dramatic impact on the robustness of tripartite networks to plant losses. However, a multi-interaction approach is crucial to better gauge the overall robustness of ecological communities, to know the interdependence of the robustness of the different animal sets, and to correctly determine the relative importance of different plants species at the whole community level, which can be key for biodiversity conservation.

Results

Different ways of connecting the interaction layers

We gathered a total of 44 ecological networks, each containing two types of ecological interactions, including mutualistic (pollination, seed-dispersal and ant-mutualism, corresponding to respectively 19, 3 and 1 networks) and antagonistic (herbivory and parasitism, corresponding to respectively 41 and 24 networks) interactions (See Tables A and B in S1 Text, and Figs A-D in S1 Text for more details). We divided these networks in three types according to the signs of their interactions: mutualistic-mutualistc, mutualistic-antagonistic, and antagonistic-antagonistic (see Methods).

To study how the interaction layers are connected, we focused our attention on the shared set of species between the two interaction layers. We measured three structural properties of the shared species: the proportion of the shared species that are connector nodes, i.e. that have links in both interaction layers (C); the proportion of shared species hubs, i.e. 20% of the shared species with the most connections, that are connectors nodes (H_C); and the participation coefficient of the connector nodes between the two interaction layers, i.e how well split between the two interaction layers are their links (PC_C) (see Methods).

This revealed fundamental differences across the three types of tripartite networks (Fig 2). In antagonistic-antagonistic networks, ∼35% of the shared species (herbivore hosts) are involved in both parasitic and herbivory interactions (i.e. are connector nodes). Moreover, most of the shared species hubs (∼96%) are acting as connectors between interaction layers, and they have their links equally split among the two interaction layers (average PC_C of 0.89). We found a very different pattern in mutualistic-mutualistic networks, for which only ∼10% of the shared species (plants in this case) are involved simultaneously in the two types of mutualistic interactions, and only 32% of shared species hubs act as connector nodes. Also, the connector nodes have their links less equally split among the two interaction layers (average PC_C of 0.59). Mutualistic-antagonistic networks are not significantly different from mutualistic-mutualistic networks and tend to have values intermediate between those of antagonistic-antagonistic and mutualistic-mutualistic networks (Fig 2A–2C). About ∼22% of the shared species are involved simultaneously in the two types of mutualistic interactions, ∼ 56% of shared species hubs act as connector nodes and the average PC_C is ∼0.59. An example of this contrasting structure is visible at a glance in the way the connector nodes link the interaction layers differently in the two networks in Fig 1.

Interdependence of the robustness of animal species

We expected these differences in structure to affect the correlation of the robustness of the two animal species sets. Following recent studies [21, 28], for each network, we measured the robustness of the two animal species sets following the extinction of plants, and we investigated whether they were correlated, i.e. if they were interdependent (Fig 1D and Methods). When driving plants to extinction, a ‘high’ correlation between the robustness of the two animal species sets implies that the same plants that are important for one of the species set are also important in the other species set [21], (e.g. the plants whose extinctions lead to a relatively high number of secondary extinctions of pollinators also do so for herbivores).

We found that, in general, when plants are driven to extinction in a random order, interdependence (I) is either positive or null (Fig 3A), with, again, fundamental differences between antagonistic-antagonistic networks and the two other types of networks. The value of interdependence found in antagonistic-antagonistic networks is on average significantly higher from that found in the other two network types, which is consistent with our results on hubs and connectors, suggesting that the two layers in antagonistic-antagonistic networks tend to be strongly interconnected. Note that data collection for parasitoids relies on their sampling on herbivores found on leaves. This sampling difference (compared to the two other types of networks where species sets can be collected independently from each other) could potentially introduce a positive correlation. However, the correlation we found is not significantly higher from what is expected in the null models. The presence of this positive correlation in the four null models considered (Fig E in S1 Text) suggests that it is due to the particular layout of these networks, more specifically, to the cascading extinction process characteristic of these tripartite antagonistic-antagonistic networks, in which plants are not the shared set of species, meaning that their extinctions sequentially spread from plants to herbivores and to parasites.

In mutualistic-mutualistic networks, the interdependence is close to null, meaning that the robustness of the two species sets seem largely decoupled from each other (but more correlated than expected by chance if we do not control for degree heterogeneity, i.e. the heterogeneity of the number of links each species has (panels B and C in Fig E in S1 Text).

Mutualistic-antagonistic networks exhibit a range of values going from moderate correlations (I ∼ 0.5) to weak negative correlations (I ∼ −0.2), and comparisons with null models showed a similar trend as in mutualistic-mutualistic networks, with empirical networks being more correlated than their randomized counterparts without taking degree heterogeneity into account (panels B and C in Fig E in S1 Text.).

Studying how interdependence relates to the three structural features we measured revealed differences among network types as well. More specifically, in antagonistic-antagonistic networks, interdependence is correlated (albeit weakly) with the proportion of connectors (C), while in the other two network types it varies with the proportion of hubs that are connectors (H_C) and their (un)balanced participation in the two interaction layers (PC_C) (Table 1, and see Fig F and Table C in S1 Text for more details).

Table 1. Table of regression of interdependence (I) and robustness (R) on the structural features we studied: Degree heterogeneity (σ_k/ < k >), proportion of connector nodes (C), proportion of shared species hubs that are connectors (H_C), and (un)even split of interactions among interaction layers (PC_C).

	I		R
	AA	MA & MM	AA	MA & MM
σ_k/ < k >		0.30	0.68^***	0.38^*
C	0.40^*		0.24^**	-0.51^**
H _C		0.70^***
PC _C		0.50^**	-0.25^*
Observations	24	20	24	20
R ²	0.16	0.70	0.76	0.38
Adjusted R²	0.12	0.64	0.73	0.31
F Statistic	4.20^*	12.41^***	21.39^***	5.18^**

Open in a new tab

Note:

*p<0.1;

**p<0.05;

***p<0.01

Tripartite networks’ robustness

The robustness of antagonistic-antagonistic networks was found to be lower than that of Mutualistic-Mutualistic networks when plants were randomly driven to extinction (Fig 3B), although differences among the three types of networks are overall not significant. Surprisingly, this suggests that even if the different ways in which the tripartite networks are connected seem to have a significant effect on interdependence, this difference does not translate into significant differences in the global robustness of the tripartite networks. In other terms, a higher interdependence between the interaction layers does not cause a lower overall robustness. As expected, all tripartite ecological networks were most fragile when plants were selectively attacked targeting the most connected plants first, and the least fragile when plants were attacked selecting the specialists plants first, as previously reported in networks with only one interaction type [2, 3, 33–35] (Fig G in S1 Text).

The structural features that most determine the robustness of the networks are the degree heterogeneity and the proportion of connector nodes in mutualistic-antagonistic and in mutualistic-mutualistic networks, as well as the even split of links between the two interaction layers in antagonistic-antagonistic networks (Table 1, and see Figs H and I, and Table D in S1 Text for more details). We included the degree heterogeneity of nodes in the analysis (i.e. the variance of the interaction degree divided by the average degree) because broad degree distributions are known to make ecological networks with one interaction type more robust to random deletion of species [3, 36], a result we recover here in the case of tripartite networks. Comparison with the null models further corroborates this result, since the robustness of mutualistic-mutualistic and mutualistic-antagonistic networks was not significantly different from that of their randomized counterparts when degree heterogeneity is conserved (panels C to E in Fig J in S1 Text).

Furthermore, the robustness of the tripartite networks could be predicted by the robustness of the two bipartite networks composing it (Fig 3C). The estimated overall robustness, a combination of the robustness of the two bipartite networks (Methods), is in very good agreement (R² = 0.96) with the robustness of the tripartite networks. When the robustness of only one bipartite network was used, R² was at most 0.8 (Fig K in S1 Text). While in the main text we only consider the classical co-extinction algorithm in unweighted networks because it is the more parsimonious and offers a lower bound to the damage the community can suffer, we show that the results hold when using a stochastic version in weighted networks [37] (Figs L and M in S1 Text).

Plant importance for robustness

The results on interdependence suggest that the important plants for one set of animal species may not always be as important for the other species set (e.g. important plants for pollinators may not be important for herbivores and vice versa). We investigated this point further and asked which plants were more important for the survival of the whole ecological community, and to what extent those plants were the same for the two animal species sets. We therefore built three rankings of plant importance—one for each animal species set and one for the whole community—in which a plant is considered to be more important if robustness is lower when that plant is attacked earlier in the extinction sequence [21] (Methods). For example, a plant can be considered important based on the pollinator and whole community rankings (e.g. plant 1, Fig 4A and 4B), but not so based on the herbivore ranking (Fig 4C). Other plants can be important based on the three rankings (e.g. plant 2, Fig 4D–4F). Comparing the three rankings in the example shows that plant importance when the two interaction layers are considered simultaneously (whole community) is not just a simple combination of the ranking of plant importance for each set of animal species (Fig 4G). While it is more difficult to differentiate between the less important plants (those with lower values of importance), the ranking is well defined, as can be seen from the correlation values between the importance and the ranking based on importance (Fig N in S1 Text). Interestingly, it becomes better defined when the two interaction layers are considered simultaneously (Fig O in S1 Text).

Fig 4 — A-F: Scatter plot of the robustness of pollinators (R_P), of the tripartite network (R), and of herbivores (R_H) vs the order of two plants (plant 1 and plant 2, chosen as an illustrative example) in the extinction sequence. The correlation coefficients are used to determine the ranking of importance of plant species. G: Ranking of plant importance for pollinators (right), for the whole community (center) and for herbivores (left). Each plant is represented by a disk whose number reflects its order in the ranking of importance of the whole community (in the tripartite network). The height of the disk represents its order in each of the three different rankings (i.e. the higher the position, the more important). Lines between balls are a visual help to track changes in the rankings. H: Classification of the tripartite networks in our database according to S_set (similarity between the ranking of plant importance in the whole community and in the animal sets), illustrating if the ranking of plant importance in the whole community is mainly determined by only one animal set, is a mixture of the rankings of importance in the two animal sets (mixed), or does not resemble any of the rankings of importance in the two animal sets (emergent).

We studied to what extent the importance of a given plant at the whole community level was driven by its importance for the two animal species sets (Fig 4H). In the majority of networks (∼63%), the importance of a plant for the whole community is a mixture between its importance for the two animal sets (i.e. the similarity between the ranking in the whole community and in the animal sets (S_set) is between 0.5 and 0.9; Methods), while in ∼25% of the networks it is mostly driven by its ranking in one of the animal species sets (i.e S_set of one animal species set is above 0.9). This was especially relevant in mutualistic-mutualistic networks, where 2 out of the 3 networks lie in this category, probably because of the high dissimilarity between the sizes of the two animal sets (180 pollinators vs 27 seed-dispersers and 173 pollinators vs 30 ants). In a few cases (∼12%), the ranking of plant importance for the whole community did not resemble any of the rankings for the animal sets (i.e. both S_set were below 0.5), meaning that the importance of a plant when the two interactions are considered simultaneously changes dramatically compared to its importance when the interactions are considered separately. While in the main text we only consider the classical co-extinction algorithm in unweighted networks because it is the more parsimonious and offers a lower bound to the damage the community can suffer, we show that these results hold when using a stochastic version in weighted networks (Figs L and M in S1 Text).

Discussion

We gathered 44 tripartite ecological networks composed by two types of ecological interactions (including herbivory, parasitism, pollination, seed dispersal, and ant-mutualism) to investigate how different interaction types were connected to each other in tripartite ecological networks and to study how considering multiple interactions simultaneously changed our knowledge of their robustness to plant loss. While multi-interaction network data sets have been gradually appearing in the literature in the last years, only a few studies have compared several of them [13]. Such comparison allows us to reveal possible commonalities of network properties (or particularities) across the different types of tripartite networks, categorized based on the sign of the ecological interactions composing them. The rationale behind this categorization is that previous studies showed that the structure of mutualistic and antagonistic ecological networks was clearly different [30].

We found fundamental differences in the way the two interaction layers are connected in the different types of tripartite networks (Fig 2A–2C), possibly as a consequence of underlying biological constraints. In antagonistic-antagonistic networks, the shared species hubs are almost all connectors (meaning that generalist herbivores tend to have more parasitoids, maybe because they tend to be more abundant too, or maybe due to the sampling procedure in which parasitoids can only be reared out of the sampled herbivores), while in mutualistic-mutualistic networks most shared species hubs are not connectors (meaning that generalist plants tend not to be involved in two types of mutualism simultaneously, which hints at trade-offs in the type of interactions a given species can invest in, making it unlikely that a species can e.g. invest in attracting both pollinators and ant bodyguards [38]). The more varied behaviour of mutualistic-antagonistic networks may be related to highly complex trade-offs between herbivory and pollination [39].

Intuitively, we expected these differences in the connection patterns to affect the correlation between the robustness of the animal species sets in the different types of tripartite networks. These correlations (which we named ‘interdependence’) suggest that in antagonistic-antagonistic networks the same plant species are important for both animal sets (in terms of secondary extinctions), whereas this is not the case in mutualistic-mutualistic networks. Our results add to previous evidence showing that the benefits of an intervention are not always expected to propagate throughout the whole network [21], which has implications for biodiversity conservation. They highlight the relevance of knowing the type of ecological interactions involved in an ecological community before planning restoration efforts, since, in the analysed networks containing mutualistic interactions, positive cascading effects could only be expected if the generalist plants acted as connector nodes and were the focus of the restoration plan.

Surprisingly, we found that more interdependent communities are not necessarily less robust to plant losses. Rather, robustness of the overall tripartite network is determined by the particular organization of each network, with degree heterogeneity playing an important role, especially in antagonistic-antagonistic networks. The positive effect of degree heterogeneity on the robustness of food webs and bipartite mutualistic networks was already reported in [36] and in [3, 40] (in mutualistic networks through nestedness, but it was also shown that nestedness is a consequence of degree heterogeneity [41]). It is worth noting that the robustness of mutualistic-mutualistic and mutualistic-antagonistic tripartite networks was found to be a combination of the robustness of the two bipartite networks composing them, stressing the relevance of knowing the structure of connections in both interaction layers to better quantify the robustness of the whole tripartite network. This is good news for ecologists, because it means that when measuring overall robustness to plant loss it is still possible to use multiple bipartite networks (with only one interaction type) and assume their effects are additive, as long as we know how plants connect them. Interestingly, looking at the two interaction layers simultaneously did not result in a dramatic change in the robustness of the whole community, as already reported for one of the networks in the database [24]. Nonetheless, considering the two interactions simultaneously improved the quantification of the overall robustness and is crucial to identify the most important plants in a given community.

The approach we used to study robustness also allowed us to identify keystone species in the whole community. In most tripartite networks, the ranking of plant importance in the whole community is determined by the importance of plants for both animal sets (with the exception of mutualistic-mutualistic networks, that are mostly driven by one interaction layer, probably because of their disproportionate size and low connection among interaction layers). In a few cases, considering the whole community could even alter the picture considerably, since the ranking of plant importance in the whole community is emergent, i.e. it is not similar to the ranking of importance for neither of the animal sets. This evidences that considering multiple interactions simultaneously can be crucial for correctly identifying keystone species in a community.

The results we present here advance our knowledge of how different interactions connect ecological communities, and how that affects the robustness of tripartite networks to plant losses. Taken together, our results suggest that considering multiple ecological interactions simultaneously does not have a dramatic impact on the overall robustness of multi-interaction networks to plant losses. However, a multi-interaction approach is crucial to know the interdependence of the robustness of the different animal sets, to better gauge the overall robustness, and to correctly determine the importance of the plants at the whole community level.

Methods

Data set

We gathered from the literature ecological networks which included different types of interactions. Because most studies only provided two interactions simultaneously, we decided to study networks with two interaction layers. Also, we only considered unweighted networks because not all studies provided interaction strengths. From all the networks we found, we only kept those which had at least 5 connector nodes. In the end, our data set contains 44 unweighted networks from 6 studies (see Table 2). Each network is composed of two ecological bipartite layers including mutualistic (pollination, seed-dispersal and ant-mutualism) and antagonistic interactions (herbivory and parasitism). In the cases where multiple types of herbivory were present, all interactions were combined in a single herbivory layer. See Tables A and B in S1 Text, and Fig A in S1 Text for more details.

Table 2. Tripartite networks included in our analyses, indicating the sign of the interactions (i.e. if the tripartite network has both mutualistic and antagonistic interactions (MA), only antagonistic interactions (AA), or only mutualistic interactions (MM)), the two ecological interactions composing the tripartite network, the number of network of each type, and the reference.

Sign	Interactions (Acronym)	Number of networks	references
MA	herbivory-pollination (H-P)	16	[6] [42] [21] [43]
MA	herbivory-seed dispersal (H-SD)	1	[6]
AA	herbivory-parasitism (H-Pa)	24	[44] [43]
MM	pollination-seed dispersal (P-SD)	2	[24] [6]
MM	pollination-ant mutualism (P-A)	1	[24]

Open in a new tab

Structural metrics of the connector nodes

We were interested in studying how the two different interactions of the tripartite networks were interconnected through the connector nodes. We used three metrics to quantify this:

The proportion of connectors nodes in the shared set of species (C), i.e. the proportion of shared species that have links simultaneously in the two interaction layers [25].
The proportion of shared species hubs that are connectors (H_C), i.e. the 20% of the species in the shared set of species with the highest degree that are connector nodes. Note that the degree of a node is the number of links it has with other species. We used a threshold of 20% to ensure that all networks had at least 1 “most connected” node, but the results are robust to that choice (Fig P in S1 Text).
The participation coefficient. This species-level metric quantifies whether the links of node i are primarily concentrated in one interaction layer or if they are well distributed among the two interaction layers [45, 46]. We quantified it as two times the ratio between the lowest degree in both interaction layers divided by the total degree of the node ( $2 \frac{k_{m i n}}{k_{t o t}}$ ). Hence PC = 1 if the links are perfectly split among the two interaction layers, and it approaches 0 as the split grows more uneven. We obtained the participation coefficient of the connector nodes (PC_C) by computing the average value over the connector nodes.

Quantifying robustness

We simulated plant loss following an established method [2, 3] and assuming bottom-up control of the animals, as justified by [21, 47]. To quantify robustness to plant loss we sequentially removed plants in a given order (the ‘extinction sequence’) keeping track of the number of secondary extinctions of animal species at each step. We considered that an animal species undergoes extinction when it has lost all its links. Note that secondary extinctions work differently in mutualistic-antagonistic and mutualistic-mutualistic networks compared to antagonistic-antagonistic networks. In the former, after removing a plant, all herbivores that no longer have resources go extinct and so do all pollinators without any resources, which means that erasing a plant may generate simultaneous secondary extinctions in the two animal species set (Fig 1A). In antagonistic-antagonistic networks herbivores are the shared set of species, so when a plant disappears all herbivores without resources go extinct, which may subsequently trigger extinctions of parasitoids. In this case, removal of a plant will generate cascading extinctions (Fig 1B). By plotting the proportion of remaining animal species as a function of the proportion of deleted plant species and measuring the area under the curve, we obtained the ‘robustness’ (R) (Fig 1C). This is a standard way of measuring the efficiency of a given extinction sequence to tear down an ecological community [48, 49]: as R → 0 the most impact a given extinction sequence has on the community, indicating that it targets the species following the ‘correct’ order of importance for the maintenance of the community.

Working with multipartite networks such as those in [21, 24], several robustness metrics can be measured depending on the species set on which secondary extinctions are considered.

Here, we measured:

the robustness of the tripartite network (R): we kept track of the proportion of remaining animal species as a function of the proportion of deleted plant species, where the proportions are measured with respect to the total number of animals (irrespective of their species set) and plants.
the robustness of the two animal species sets (R_P, R_H): we measured the proportion of remaining animal species with respect to the total number of animals in each species set (e.g. how many pollinators remain from the original number of pollinators), and the proportion of deleted plants is measured with respect to the total number of plants in the tripartite network.
the robustness of the two bipartite networks: In this case, the tripartite network is split in two bipartite networks, on which the same protocol as above is performed. These two networks are not identical to the two interaction layers because the shared set of species that are not connected in a given layer are not considered in the bipartite network, which affects the calculation of the robustness. We thereby obtain two robustness (R_L and R_S, respectively for the smaller and larger networks, in terms of species number). Note that in antagonistic-antagonistic networks, the protocol can be performed only on the herbivory network since there is no direct link between plants and parasitoids.

We applied 3000 random extinction sequences of plants to each of the tripartite networks in the data set, and for each extinction sequence we measured the different robustness measures above. Here, results are presented for random extinction sequences but results for other extinction scenarios (increasing or decreasing degree of plants) are presented in Figs G, H and I in S1 Text.

We also measured the robustness of the mutualistic-antagonistic network using an stochastic version of the co-extinction algorithm [37] and weighted networks (when available) to compare with the results of the classic co-extinction algorithm (see Figs L and M in S1 Text). In this stochastic version, a species i will undergo a secondary extinction following the extinction of plant j with a probability P_ij = R_i.d_ij, where d_ij is the dependency of species i on j (interaction weight), and R_i represents the intrinsic demographic dependence of species i on mutualism (we considered R_i = 1 for animals and R_i = 0 for plants, to keep the bottom-up control of animals).

Interdependence

We measured the correlation between the robustness of the two different species sets (other than plants) in the tripartite networks, hereafter called ‘interdepence’ (I) (Fig 1D). When driving plants to extinction, a ‘high’ correlation between the robustness of pollinators and herbivores implies that the same plants that are relevant for one of the species set will also be relevant in the other species set [21], (i.e. sequences of plant loss that were relatively benign for pollinators were also benign for herbivores). If, on the other hand, the relevant plants are not the same in the two species sets, we expect a low correlation in robustness.

Plant importance rankings

The importance of each plant species for the different animal sets and for the whole community (i.e. for the tripartite network) was quantified based on the correlation coefficient between robustness and the position of the plant in the extinction sequence [21]. The rationale is that the ‘importance’ of a plant cannot be directly assessed from the number of secondary extinctions caused by its loss because if lost at the start (rather than at the end) of the extinction sequence, fewer secondary extinctions are expected; however, if a plant is ‘important’, then robustness is expected to be lower when it is lost earlier in the sequence than when it is lost later. Hence, the lower the robustness to an extinction sequence, the better that extinction sequence actually resembles the importance of plants for the survival of the community. To obtain the plant importance rankings (three in total: one for each of the two interaction layers and one for the whole community), we ranked each plant species by increasing correlation between its order of appearance in extinction sequences and the corresponding robustness (i.e. plants that have a larger negative correlation are considered more important; Fig 4A–4G).

To asses to what extent one of the two interaction layers was driving the robustness of the whole community we measured the similarity between the importance of a plant for one animal set (for example for pollinators or herbivores) and the importance of a plant in the whole community, namely S_P or S_H. We quantified S_set as the square of the correlation coefficient between the ranking of the plant in each species set and the ranking in the whole community.

We then classified the networks in three categories: those where one interaction layer was driving the process (one S_set was above 0.9, meaning that 90% of the variance in the importance ranking in the whole community can be traced to one of the two animal rankings), those where the ranking in the whole community was a mixture of the rankings in both animal sets (both S_set were between 0.5 and 0.9) and finally those where the importance ranking was emergent (both S_set were below 0.5, meaning that no animal set ranking was able to explain at least 50% of the ranking of importance in the whole community).

Estimating tripartite robustness from networks with one interaction

We also tested whether one can express the robustness of the whole community (R) as a combination of the robustness of the two independent bipartite network composing the tripartite network. To do that we performed the following linear regression:

R^{(e s t)} = a . R_{L} + b . R_{S}

where R_L and R_S are the robustness of the two bipartite networks composing the tripartite network under study, respectively the larger (i.e. with more species) and the smaller one.

Multiple regressions

We performed a multiple regression of interdependence and robustness based on the structural features we measured in the tripartite networks using the package statsmodel in Python. We selected the structural features that were more relevant for interdependence or robustness by choosing the model with a lowest AIC.

Null models

To assess the importance of network structure in determining a certain network feature, we compared measurements of that feature performed on empirical networks with measurements performed on randomized versions of those networks keeping some properties fixed. We used four different null-models, represented in Fig 5, which—going from the least to the most constraining—are as follows: “1” keeps the number of species constant in each species set and the number of links constant in each interaction layer, “2” adds the constraint of keeping the degree distribution of the animal nodes constant, “3” keeps the degree distribution of animals and plants but not the total degree of the shared set species (i.e. it breaks the correlation between the degree of the shared set of species in the two interaction layers), and “4” keeps the degree of each node constant while links are reshuffled within a layer (see S1 Text for more details).

Fig 5 — Each figure represents what is kept fixed in each null model, going from the less restrictive on the left, to the more restrictive on the right. N_x is the number of nodes in the species set, L_x the number of links in the interaction layer, the color of the nodes represent the different species set, the colour of the link the two different ecological interactions, the size of the node is proportional to its degree (when kept), and connector nodes are highlighted in red.

Supporting information

S1 Text. Supplementary information file, including supplementary figures A-P and supplementary tables A-D.

(PDF)

Click here for additional data file.^{(608.4KB, pdf)}

Data Availability

The data and code supporting the results is available in Zenodo: https://doi.org/10.5281/zenodo.10198613.

Funding Statement

Authors acknowledge funding from ANR-18-CE02-0010, EcoNet (Advanced statistical modelling of ecological networks) of the French National Research Agency (ANR) awarded to SK, that funded VDG. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

1.IPBES. Global assessment report on biodiversity and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services; 2019. Available from: https://zenodo.org/record/3831673.
2. Dunne JA, Williams RJ, Martinez ND. Network structure and biodiversity loss in food webs: robustness increases with connectance. Ecology Letters. 2002;5(4):558–567. doi: 10.1046/j.1461-0248.2002.00354.x [DOI] [Google Scholar]
3. Memmott J, Waser NM, Price MV. Tolerance of pollination networks to species extinctions. Proceedings of the Royal Society of London Series B: Biological Sciences. 2004;271(1557):2605–2611. doi: 10.1098/rspb.2004.2909 [DOI] [PMC free article] [PubMed] [Google Scholar]
4. Montoya JM, Pimm SL, Solé RV. Ecological networks and their fragility. Nature. 2006;442(7100):259–264. doi: 10.1038/nature04927 [DOI] [PubMed] [Google Scholar]
5. Kaiser-Bunbury CN, Muff S, Memmott J, Müller CB, Caflisch A. The robustness of pollination networks to the loss of species and interactions: a quantitative approach incorporating pollinator behaviour. Ecology Letters. 2010;13(4):442–452. doi: 10.1111/j.1461-0248.2009.01437.x [DOI] [PubMed] [Google Scholar]
6. Melián CJ, Bascompte J, Jordano P, Krivan V. Diversity in a complex ecological network with two interaction types. Oikos. 2009;118(1):122–130. doi: 10.1111/j.1600-0706.2008.16751.x [DOI] [Google Scholar]
7. Fontaine C, Guimaraes PR, Kéfi S, Loeuille N, Memmott J, van der Putten W, et al. The ecological and evolutionary implications of merging different types of networks. Ecology Letters. 2011;14(11):1170–1181. [DOI] [PubMed] [Google Scholar]
8. Kéfi S, Berlow EL, Wieters EA, Navarrete SA, Petchey OL, Wood SA, et al. More than a meal… integrating non-feeding interactions into food webs. Ecology Letters. 2012;15(4):291–300. [DOI] [PubMed] [Google Scholar]
9. Kéfi S, Berlow EL, Wieters EA, Joppa LN, Wood SA, Brose U, et al. Network structure beyond food webs: mapping non-trophic and trophic interactions on Chilean rocky shores. Ecology. 2015;96(1):291–303. doi: 10.1890/13-1424.1 [DOI] [PubMed] [Google Scholar]
10. Kéfi S, Miele V, Wieters EA, Navarrete SA, Berlow EL. How Structured Is the Entangled Bank? The Surprisingly Simple Organization of Multiplex Ecological Networks Leads to Increased Persistence and Resilience. PLOS Biology. 2016;14(8):e1002527. doi: 10.1371/journal.pbio.1002527 [DOI] [PMC free article] [PubMed] [Google Scholar]
11. García-Callejas D, Molowny-Horas R, Araújo MB. Multiple interactions networks: towards more realistic descriptions of the web of life. Oikos. 2017;127(1):5–22. doi: 10.1111/oik.04428 [DOI] [Google Scholar]
12. Kéfi S. Ecological Networks: from structure to dynamics. In: Theoretical ecology: Concepts and applications. Oxford University Press; 2020. p. 143–160. [Google Scholar]
13. Timóteo S, Albrecht J, Rumeu B, Norte AC, Traveset A, Frost CM, et al. Tripartite networks show that keystone species can multitask. Functional Ecology. 2022. [Google Scholar]
14. May RM. Will a Large Complex System be Stable? Nature. 1972;238(5364):413–414. doi: 10.1038/238413a0 [DOI] [PubMed] [Google Scholar]
15. Mougi A, Kondoh M. Diversity of Interaction Types and Ecological Community Stability. Science. 2012;337(6092):349–351. doi: 10.1126/science.1220529 [DOI] [PubMed] [Google Scholar]
16. Allesina S, Tang S. Stability criteria for complex ecosystems. Nature. 2012;483(7388):205–208. doi: 10.1038/nature10832 [DOI] [PubMed] [Google Scholar]
17. Sauve AMC, Fontaine C, Thébault E. Structure-stability relationships in networks combining mutualistic and antagonistic interactions. Oikos. 2013;123(3):378–384. [Google Scholar]
18. Lurgi M, Montoya D, Montoya JM. The effects of space and diversity of interaction types on the stability of complex ecological networks. Theoretical Ecology. 2015;9(1):3–13. [Google Scholar]
19. McWilliams C, Lurgi M, Montoya JM, Sauve A, Montoya D. The stability of multitrophic communities under habitat loss. Nature Communications. 2019;10(1). doi: 10.1038/s41467-019-10370-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
20. Hale KRS, Valdovinos FS, Martinez ND. Mutualism increases diversity, stability, and function of multiplex networks that integrate pollinators into food webs. Nature Communications. 2020;11(1). doi: 10.1038/s41467-020-15688-w [DOI] [PMC free article] [PubMed] [Google Scholar]
21. Pocock MJO, Evans DM, Memmott J. The Robustness and Restoration of a Network of Ecological Networks. Science. 2012;335(6071):973–977. doi: 10.1126/science.1214915 [DOI] [PubMed] [Google Scholar]
22. Sauve AMC, Thébault E, Pocock MJO, Fontaine C. How plants connect pollination and herbivory networks and their contribution to community stability. Ecology. 2016;97(4):908–917. doi: 10.1890/15-0132.1 [DOI] [PubMed] [Google Scholar]
23. Genrich CM, Mello MAR, Silverira FAO, Bronstein JL, Paglia AP. Duality of interaction outcomes in a plant-frugivore multilayer network. Oikos. 2016;126:361–368. [Google Scholar]
24. Dáttilo W, Lara-Rodríguez N, Jordano P, Guimarães PR, Thompson JN, Marquis RJ, et al. Unravelling Darwin’s entangled bank: architecture and robustness of mutualistic networks with multiple interaction types. Proceedings of the Royal Society B: Biological Sciences. 2016;283(1843):20161564. doi: 10.1098/rspb.2016.1564 [DOI] [PMC free article] [PubMed] [Google Scholar]
25. Astegiano J, Altermatt F, Massol F. Disentangling the co-structure of multilayer interaction networks: degree distribution and module composition in two-layer bipartite networks. Scientific Reports. 2017;7(1). doi: 10.1038/s41598-017-15811-w [DOI] [PMC free article] [PubMed] [Google Scholar]
26. Mello MAR, Felix GM, Pinheiro RBP, Muylaert RL, Geiselman C, Santana SE, et al. Insights into the assembly rules of a continent-wide multilayer network. Nature Ecology & Evolution. 2019;3(11):1525–1532. [DOI] [PubMed] [Google Scholar]
27. Morrison BML, Brosi BJ, Dirzo R. Agricultural intensification drives changes in hybrid network robustness by modifying network structure. Ecology Letters. 2019;23(2):359–369. doi: 10.1111/ele.13440 [DOI] [PubMed] [Google Scholar]
28. Evans DM, Pocock MJO, Memmott J. The robustness of a network of ecological networks to habitat loss. Ecology Letters. 2013;16(7):844–852. doi: 10.1111/ele.12117 [DOI] [PubMed] [Google Scholar]
29. Kivela M, Arenas A, Barthelemy M, Gleeson JP, Moreno Y, Porter MA. Multilayer networks. Journal of Complex Networks. 2014;2(3):203–271. doi: 10.1093/comnet/cnu016 [DOI] [Google Scholar]
30. Thébault E, Fontaine C. Stability of Ecological Communities and the Architecture of Mutualistic and Trophic Networks. Science. 2010;329(5993):853–856. doi: 10.1126/science.1188321 [DOI] [PubMed] [Google Scholar]
31. Dunne JA, Williams RJ, Martinez ND. Food-web structure and network theory: The role of connectance and size. Proceedings of the National Academy of Sciences. 2002;99(20):12917–12922. doi: 10.1073/pnas.192407699 [DOI] [PMC free article] [PubMed] [Google Scholar]
32. Staniczenko PPA, Lewis OT, Jones NS, Reed-Tsochas F. Structural dynamics and robustness of food webs. Ecology Letters. 2010;13(7):891–899. doi: 10.1111/j.1461-0248.2010.01485.x [DOI] [PubMed] [Google Scholar]
33. Bascompte J, Stouffer DB. The assembly and disassembly of ecological networks. Philosophical Transactions of the Royal Society B: Biological Sciences. 2009;364(1524):1781–1787. doi: 10.1098/rstb.2008.0226 [DOI] [PMC free article] [PubMed] [Google Scholar]
34. Dallas T, Cornelius E. Co-extinction in a host-parasite network: identifying key hosts for network stability. Scientific Reports. 2015;5(1). doi: 10.1038/srep13185 [DOI] [PMC free article] [PubMed] [Google Scholar]
35. Strona G, Lafferty KD. Environmental change makes robust ecological networks fragile. Nature Communications. 2016;7(1). doi: 10.1038/ncomms12462 [DOI] [PMC free article] [PubMed] [Google Scholar]
36. Solé RV, Montoya M. Complexity and fragility in ecological networks. Proceedings of the Royal Society of London Series B: Biological Sciences. 2001;268(1480):2039–2045. doi: 10.1098/rspb.2001.1767 [DOI] [PMC free article] [PubMed] [Google Scholar]
37. Vieira MC, Almeida-Neto M. A simple stochastic model for complex coextinctions in mutualistic networks: robustness decreases with connectance. Ecology Letters. 2014;18(2):144–152. doi: 10.1111/ele.12394 [DOI] [PubMed] [Google Scholar]
38. Dutton EM, Luo EY, Cembrowski AR, Shore JS, Frederickson ME. Three’s a Crowd: Trade-Offs between Attracting Pollinators and Ant Bodyguards with Nectar Rewards in Turnera. The American Naturalist. 2016;188(1):38–51. doi: 10.1086/686766 [DOI] [PubMed] [Google Scholar]
39. Jacobsen DJ, Raguso RA. Lingering Effects of Herbivory and Plant Defenses on Pollinators. Current Biology. 2018;28(19):R1164–R1169. doi: 10.1016/j.cub.2018.08.010 [DOI] [PubMed] [Google Scholar]
40. Burgos E, Ceva H, Perazzo RPJ, Devoto M, Medan D, Zimmermann M, et al. Why nestedness in mutualistic networks? Journal of Theoretical Biology. 2007;249(2):307–313. doi: 10.1016/j.jtbi.2007.07.030 [DOI] [PubMed] [Google Scholar]
41. Jonhson S, Domínguez-García V, Muñoz MA. Factors Determining Nestedness in Complex Networks. PLoS ONE. 2013;8(9):e74025. doi: 10.1371/journal.pone.0074025 [DOI] [PMC free article] [PubMed] [Google Scholar]
42. Shinohara N, Uchida K, Yoshida T. Contrasting effects of land-use changes on herbivory and pollination networks. Ecology and Evolution. 2019;9(23):13585–13595. doi: 10.1002/ece3.5814 [DOI] [PMC free article] [PubMed] [Google Scholar]
43. Hackett TD, Sauve AMC, Davies N, Montoya D, Tylianakis JM, Memmott J. Reshaping our understanding of species’ roles in landscape-scale networks. Ecology Letters. 2019;22(9):1367–1377. doi: 10.1111/ele.13292 [DOI] [PubMed] [Google Scholar]
44. Macfadyen S, Gibson R, Polaszek A, Morris RJ, Craze PG, Planqué R, et al. Do differences in food web structure between organic and conventional farms affect the ecosystem service of pest control? Ecology Letters. 2009;12(3):229–238. doi: 10.1111/j.1461-0248.2008.01279.x [DOI] [PubMed] [Google Scholar]
45. Guimerà R, Amaral LAN. Cartography of complex networks: modules and universal roles. Journal of Statistical Mechanics: Theory and Experiment. 2005;2005(02):P02001. doi: 10.1088/1742-5468/2005/02/P02001 [DOI] [PMC free article] [PubMed] [Google Scholar]
46. Battiston F, Nicosia V, Latora V. Structural measures for multiplex networks. Physical Review E. 2014;89(3). doi: 10.1103/PhysRevE.89.032804 [DOI] [PubMed] [Google Scholar]
47. Scherber C, Eisenhauer N, Weisser WW, Schmid B, Voigt W, Fischer M, et al. Bottom-up effects of plant diversity on multitrophic interactions in a biodiversity experiment. Nature. 2010;468(7323):553–556. doi: 10.1038/nature09492 [DOI] [PubMed] [Google Scholar]
48. Allesina S, Pascual M. Googling Food Webs: Can an Eigenvector Measure Species’ Importance for Coextinctions? PLoS Computational Biology. 2009;5(9):e1000494. doi: 10.1371/journal.pcbi.1000494 [DOI] [PMC free article] [PubMed] [Google Scholar]
49. Domínguez-García V, Muñoz MA. Ranking species in mutualistic networks. Scientific Reports. 2015;5(1). doi: 10.1038/srep08182 [DOI] [PMC free article] [PubMed] [Google Scholar]

PLoS Comput Biol. doi: 10.1371/journal.pcbi.1011770.r001

Decision Letter 0

Mercedes Pascual, James O'Dwyer

Transfer Alert

This paper was transferred from another journal. As a result, its full editorial history (including decision letters, peer reviews and author responses) may not be present.

24 Mar 2023

Dear Dr Domínguez-García,

Thank you very much for submitting your manuscript "The structure and robustness of ecological networks with two interaction types." for consideration at PLOS Computational Biology.

As with all papers reviewed by the journal, your manuscript was reviewed by members of the editorial board and by independent reviewers. The three reviewers agree on the relevance and quality of the work, and have provided a series of constructive comments with questions on terminology and approach, in particular on the null models and on specifics of the robustness analysis. In light of the reviews (below this email), we would like to invite resubmission of a revised version that takes into account the reviewers' comments.

We cannot make any decision about publication until we have seen the revised manuscript and your response to the reviewers' comments. Your revised manuscript is likely to be sent to reviewers for further evaluation.

When you are ready to resubmit, please upload the following:

[1] A letter containing a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

[2] Two versions of the revised manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file).

Important additional instructions are given below your reviewer comments.

Please prepare and submit your revised manuscript within 60 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email. Please note that revised manuscripts received after the 60-day due date may require evaluation and peer review similar to newly submitted manuscripts.

Thank you again for your submission. We hope that our editorial process has been constructive so far, and we welcome your feedback at any time. Please don't hesitate to contact us if you have any questions or comments.

Sincerely,

Mercedes Pascual

Academic Editor

PLOS Computational Biology

James O'Dwyer

Section Editor

PLOS Computational Biology

***********************

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: In this paper the authors investigate the robustness of tripartite networks with two interaction types. They take the most naive approach of plant removal and consequent cascading or simultaneous co-extinction of animals in the two layers. This approach has many assumptions and limitations (e.g., no rewiring, no weights). However, as the authors claim, it serves as a starting point for basic insights into community robustness. I agree with that notion. The robustness of mulitpartite networks has already been investigated. However, this is the first work to do so in a systematic manner, and comparing three types of networks. In that sense this work is novel and is a necessary contribution to the knowledge on community robustness.

There are several conceptual insights arising from this work. First, that the robustness of the tripartite networks can be predicted from that of its two constituting layers. Second, the proportion and role of connector species in the different kinds of networks. Third, that interdependent communities were not less robust to plant losses. The analysis is strong and covers multiple aspects of the structure-stability (or structure-robustness) link. The comparison to null models makes the analysis more robust and insightful, although this part is not as emphasized in the main text (see my comments). Overall, I find this paper merits publication in PLOS Comp Biol.

I have just a few comments and suggestions.

Thanks for the efforts put into that work.

The results presented in the main text are for random removal order. This is very null-model-like. It would be good to include some of the DD and ID results in the main text. Fig S9 states quite the obvious, so maybe not this one. But are there any meaningful insights regarding the effect of structural properties on robustness for nonrandom removal?

The null models provide important insights but are hidden in the SI. I really liked the approach of increasing constraints and the fiure that describes the null models. Consider moving them to the main text. I especially liked the results of interdependence that are in the SI (paragraph starting with: “Comparison with the four null models indicates that, for AA networks…”). Also Fig. S7, and S11.

Plant importance ranking: What are the actual differences between the rankings? A plant can rank higher than another but the actual difference in importance value can be very small. What is the distribution of importance values?

The inherent difference between AA and MA/MM in the role of plants will cause a "slower" (i.e., more removal events needed) extinction in AA networks. Generally, if you take a network and remove nodes from the linking set it will collapse faster than if you remove from any of the other two sets. Hence, given an AA and an MM/MA network with a similar number of nodes in corresponsing sets and similar degree distributions, AA network will be more robust than MM and MA networks. Is this the underlying cause for the differences between AA/MM-MA networks? In Fig 3B you find a pattern opposite to what I expected, whereby AA networks are less robust than MM. Why? I somehow feel that you might have explained all that in the text and I missed it…

Related — L 134: Isn’t the robustness of AA is lower than MM? At least as far as I can tell from Fig 3B.

It was difficult to understand the results without referring to the methods to see the definitions of the structural properties. Maybe move some text from the methods to the results.

How sensitive are the results for the thresholds? E.g., the 20% of the .

L108: change ‘behaviour’ to ‘structure’

Fig. 3: What are the gray data points in panels A-B? What are the green and blue points in C?

Reviewer #2: General assessment:

This is a study on the robustness of multilayer networks, using as a model different kinds of tripartite networks that combine mutualisms and antagonisms. Overall, its rationale is very interesting and the topic, timely. As claimed by the authors, multilayer networks are a new frontier in Ecology, as this approach finally provided us with a tool set to address two or more interaction types together in the same system. This is a great leap forward, as interaction types do not occur isolated in nature, but rather interact with one another, generating conditional outcomes. Tripartite networks are indeed a good starting point to advance this approach also to the assess the topic of stability. The study is well designed and well presented. Its language is strongly focused on the analyses and math, which would be problematic for a strictly biological journal, but is not the case in a more computational venue. After a few adjustments this study can become a highly relevant contribution to the fields of network ecology and species interactions.

Specific comments:

1. Be careful with typos and grammar errors, especially errors of number agreement. A review by a professional proofreader (native speaker or not) is always a good idea.

2. Avoid using so many acronyms and abbreviations in the text, especially custom-made ones that are not regularly used in the field. They may shorten the text, but severely hinder its readability.

3. A central point when studying multilayer networks is the definition of the interlayer edges. Paying attention to them is the main difference between using a monolayer or multilayer approach. In the case of tripartite networks, which fit the category of "diagonally coupled", interlayer edges occur between shared species, which are present in both layers (plants, for that matter). It is important to explain those definitions more clearly and how they affect the interdependence between layers. See:

Mucha, P. J., Richardson, T., Macon, K., Porter, M. A., & Onnela, J.-P. (2010). Community Structure in Time-Dependent, Multiscale, and Multiplex Networks. Science, 328(5980), 876 LP – 878. https://doi.org/10.1126/science.1184819

Pilosof, S., Porter, M. A., Pascual, M., & Kéfi, S. (2017). The multilayer nature of ecological networks. Nature Ecology & Evolution, 1(4), 0101. https://doi.org/10.1038/s41559-017-0101

4. Please explain more clearly how you defined the "connector nodes" and "connector hubs". There is much confusion in terminology between graph theory and network theory, also when it comes to centrality. When network ecology, this confusion becomes a real mess. For instance, in mathematical studies about multilayer networks, a node that connects two or more layers is usually named a "state node". When two nodes are connected to one another in two or more layers, their connection is named a "multilink". In the field of ecological networks, those kinds of nodes and links receive other nicknames. Furthermore, connectors and hubs are usually defined in ecological studies according to two centrality metrics: participation coefficient and within-module degree. See:

Guimerà, R., & Amaral, L. A. N. (2005). Cartography of complex networks: modules and universal roles. Journal of Statistical Mechanics: Theory and Experiment, P02001. https://doi.org/doi:10.1088/1742-5468/2005/02/P02001

Olesen, J. M., Bascompte, J., Dupont, Y. L., & Jordano, P. (2007). The modularity of pollination networks. Proceedings of the National Academy of Sciences, 104(50), 19891–19896. https://doi.org/10.1073/pnas.0706375104

Boccaletti, S., Bianconi, G., Criado, R., del Genio, C. I., Gómez-Gardeñes, J., Romance, M., Sendiña-Nadal, I., Wang, Z., & Zanin, M. (2014). The structure and dynamics of multilayer networks. Physics Reports, 544(1), 1–122. https://doi.org/10.1016/j.physrep.2014.07.001

Kivela, M., Arenas, A., Barthelemy, M., Gleeson, J. P., Moreno, Y., & Porter, M. A. (2014). Multilayer networks. Journal of Complex Networks, 2(3), 203–271. https://doi.org/10.1093/comnet/cnu016

5. In this study, the classical “Memmott approach” (or Barabasi-Albert) was used to assess secondary extinctions. As commented by the authors themselves, this approach is very simplistic and unrealistic, although it was a good first step in the early 2000s, when studies on the robustness of ecological networks began to appear. Nevertheless, much more realistic approaches have been proposed in recent years, which should be also considered. See:

Vieira, M. C., & Almeida-Neto, M. (2015). A simple stochastic model for complex coextinctions in mutualistic networks: robustness decreases with connectance. Ecology Letters, 18(2), 144–152. https://doi.org/10.1111/ele.12394

6. The authors proposed some new null models to investigate the topology of tripartite networks. This is a very nice step forward in the field of multilayer networks, especially in Ecology, where null model analysis is virtually considered mandatory. However, caution is advised. Recent studies have pointed out that some null models have been used wrongly in ecological studies. In spite of removing some undesired processes from network assembly, they might be actually simulating those same processes. This is the case of some of the null models used in this study, which were derived from classical null models proposed by Bascompte and Vázquez, among others. See:

Dormann, C. F., Frund, J., Bluthgen, N., & Gruber, B. (2009). Indices, Graphs and Null Models: Analyzing Bipartite Ecological Networks. The Open Ecology Journal, 2(1), 7–24. https://doi.org/10.2174/1874213000902010007

Farine, D. R. (2017). A guide to null models for animal social network analysis. Methods in Ecology and Evolution, 8(10), 1309–1320. https://doi.org/10.1111/2041-210X.12772

Pinheiro, R. B. P., Dormann, C. F., Felix, G. M., & Mello, M. A. R. (2021). A novel perspective on the meaning of nestedness with conceptual and methodological solutions. BioRxiv, 2021.04.05.438470. https://doi.org/10.1101/2021.04.05.438470

7. Isn’t there a better way to visualize tripartite networks? Yes, on the one hand, they are traditionally plotted as two bipartite graphs connected to one another. On the other hand, those “bipartite drawings” are quite poor in depicting topology and centrality. They serve only the purpose of visualizing nestedness. Consider more efficient, imaginative designs. See:

Pocock, M. J. O., Evans, D. M., Fontaine, C., Harvey, M., Julliard, R., McLaughlin, Ó., Silvertown, J., Tamaddoni-Nezhad, A., White, P. C. L., & Bohan, D. A. (2016). The Visualisation of Ecological Networks, and Their Use as a Tool for Engagement, Advocacy and Management. In G. Woodward & D. A. Bohan (Eds.), Advances in Ecological Research (1st ed., pp. 41–85). Academic Press. https://doi.org/10.1016/bs.aecr.2015.10.006

Marai, G. E., Pinaud, B., Bühler, K., Lex, A., & Morris, J. H. (2019). Ten simple rules to create biological network figures for communication. PLOS Computational Biology, 15(9), e1007244. https://doi.org/10.1371/journal.pcbi.1007244

Reviewer #3: Overview: I reviewed this manuscript with great interest. As the authors point out, it is only recently that datasets of networks comprised of multiple interaction types have been published in high enough numbers to start drawing conclusions across them. Here, the authors examine networks consisting of multiple antagonistic interactions, multiple mutualistic interactions and a combination of mutualistic and antagonistic interactions. They investigate structural differences and similarities as well as how robustness changes among these types of networks, and the possibility of emergent properties in the whole network not captured in the component webs. The analyses are for the most part sound and conclusions are broadly well-supported. However, I do have some questions about the effects of some analytical decisions (detailed below). For instance, it would be helpful for the authors to comment on how collection/rearing methods might be contributing to the conclusions. Also, how the decision to use unweighted networks might affect the outcome; a lot of information is lost by removing interaction strength so perhaps the authors can comment on how they might expect their conclusions to change with weighted networks? Some more technical aspects of the network metrics and descriptions could be explained earlier and in more detail (e.g. degree, connectors and hubs) to help non-specialist understanding. It is generally well written (although I have included some typos and suggestions for rewording a few awkward phrases at the end). I detail some further questions and suggestions for improvement below, but in general, I found it to be an interesting read that presents novel cross-study conclusions.

General comments:

I tend to think of tripartite networks to mean three sequentially linked trophic levels (e.g. plant-herbivore-parasitoid, or plant-herbivore-consumer), rather than combining networks of 2 interaction types that share a common resource (e.g. plant-pollinator + plant-herbivore). There’s nothing inherently incorrect with your definition, but to me it confuses the terminology a bit in a field that already has multiple terms for the same thing. Please consider using another term to describe combining networks of different interaction types. Annoyingly, the field is far from unanimous in what to call this type of network, but perhaps consider something like: network of networks, network of multiple interaction types, multilayer network, combined network, hybrid network...

A paper that might have been missed that also examined robustness across networks of different interaction types is Morrison et al. (2020). I am not suggesting this be added to the analyses per say, but it might be of interest to the authors.

Specific Comments

Line 76: In what order were species removed? Random? By degree? Can you clarify here?

Line 85: Your methods would only highlight keystone plant species, correct? Not a particularly well connected/central insect?

Lines 87-91: It reads a bit strangely to have the conclusion of the results in the introduction before having presented the results. Consider moving this to the discussion, and rather in this paragraph, outlining specifically which hypotheses you were testing/investigating and predictions.

Line 92: It would really help with readability to first have a paragraph summarising the datasets used and key properties of them (interaction types, size etc.).

Line 95: The Connector is a node that exists in both networks, correct? Not necessarily one with a particular property (e.g. high betweeness)? This could be made clearer at this point to ease understanding

Line 100: How many linking set hubs were there?

Line 119-120: How much of the interconnection in AA networks is due to it being parasitoids reared out of herbivores? Leaves with herbivores are collected and parasitoids are coming from them so they will, by necessity, be strongly interconnected. It is not necessarily the antagonism but the specific mechanism or sampling design.

Line 165: How much does this trend for dissimilarity between the sizes of networks affect your other findings? I suspect some of the MA networks were also pollinator dominated, no? Could this have skewed more of your findings?

Line 183: Can you test for the effect of abundance? Or control for it in your analyses?

Line 206-7: Your analyses don't account for any species that might exist in multiple networks (e.g. butterflies). Can you comment on how this might affect your conclusions? Could the impact of looking at robustness in combined networks be stronger if this were included?

Line 224: Using unweighted networks means potentially important information is lost when removing interaction strength. Can you comment on how this might have affected your conclusions? Was there any attempt to normalise across networks of different sizes?

Line 235: Non-specialists might not know what degree is. Probably worth defining on first use.

Line 243: How was the order of removal determined? From least connected/lowest degree to most? randomly? another way?

Line 256: Multipartite networks is inconsistent terminology. I say above about my preference to not use tripartite here, but whatever term is used, it should be consistent

Line 271: So the extinction sequence is based on random removal? Can you clarify this earlier

Line 281-2: This is a neat approach. Does it correlate at all with degree? Did you get the same/similar importance when species were removed in order of degree?

SI table 1: How did you handle multiple types of herbivory (e.g. leaf miners, caterpillars)? Were they combined into 1 network? Or did you not use networks with combined herbivory (although I think some of the studies did have multiple herbivores, no?)

Figures:

Fig. 2 caption: Please define PR in the caption for clarity.

The asterisks seems 1 too many. I tend to think of * for < 0.05, ** for < 0.01, *** for < 0.001

Minor issues (e.g. typos, syntax etc.):

Line 27: “trough” should be “through”

Line 100: “the ones doing the connection” is awkward wording. Consider something along the lines of “…are acting as connectors” or “are forming connections”

SI 2: 1st paragraph “less” and “more” should be “least” and “most”, respectively

In NL2 paragraph (and other places): “allows to study” is not correct wording. Replace with “allows the study of” or “allows us to study” or similar.

References:

Morrison, B. M. L., Brosi, B. J. & Dirzo, R. Agricultural intensification drives changes in hybrid network robustness by modifying network structure. Ecology Letters 23, 359–369 (2020).

**********

Have the authors made all data and (if applicable) computational code underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data and code underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data and code should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data or code —e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: None

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

Figure Files:

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

Data Requirements:

Please note that, as a condition of publication, PLOS' data policy requires that you make available all data used to draw the conclusions outlined in your manuscript. Data must be deposited in an appropriate repository, included within the body of the manuscript, or uploaded as supporting information. This includes all numerical values that were used to generate graphs, histograms etc.. For an example in PLOS Biology see here: http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1001908#s5.

Reproducibility:

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

PLoS Comput Biol. 2024 Jan 19;20(1):e1011770. doi: 10.1371/journal.pcbi.1011770.r002

Author response to Decision Letter 0

4 Aug 2023

Attachment

Submitted filename: Response_PlosCBiolTripartite.pdf

Click here for additional data file.^{(248.8KB, pdf)}

PLoS Comput Biol. doi: 10.1371/journal.pcbi.1011770.r003

Decision Letter 1

Mercedes Pascual, James O'Dwyer

26 Oct 2023

Dear Dr Domínguez-García,

Thank you very much for submitting your manuscript "The structure and robustness of ecological networks with two interaction types." for consideration at PLOS Computational Biology. As with all papers reviewed by the journal, your manuscript was reviewed by members of the editorial board and by several independent reviewers.

All reviewers found the revised manuscript clearer and more compelling. Based on the reviews, we are very likely to accept this manuscript for publication, providing that you modify the manuscript to incorporate the discussion point suggested by reviewer 3 and address the point on the data made by reviewer 1.

Please prepare and submit your revised manuscript within 30 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email.

When you are ready to resubmit, please upload the following:

[1] A letter containing a detailed list of your responses to all review comments, and a description of the changes you have made in the manuscript. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out

[2] Two versions of the revised manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file).

Important additional instructions are given below your reviewer comments.

Thank you again for your submission to our journal. We hope that our editorial process has been constructive so far, and we welcome your feedback at any time. Please don't hesitate to contact us if you have any questions or comments.

Sincerely,

Mercedes Pascual

Academic Editor

PLOS Computational Biology

James O'Dwyer

Section Editor

PLOS Computational Biology

***********************

A link appears below if there are any accompanying review attachments. If you believe any reviews to be missing, please contact ploscompbiol@plos.org immediately:

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: The authors have addressed my comments. My only comment is that they should remove line 69-73 (starting with: "In the network science literature..."). This statement is wrong. Diagonally-coupled networks have interlayer links. The tripartite networks are not multilayer and do not have interlayer links. They are multipartite (tripartite in this case).

Reviewer #2: The authors did a terrific job revising their manuscript. This new version is much better than the previous. I have no further suggestions to make.

Reviewer #3: The authors have done a great job of addressing the comments and this revised version is much improved. It is clearer to read and the questions about the methodology have been fully addressed and the terminology and approach is much clearer throughout.

The main point that remains, to my mind, is with regard to the relevance of the comparisons between different types of tipartite networks. Because the original data sets were not necessarily collected to compare between the different bipartite networks that comprise them, some of the detected differences might be due to sampling differences rather than structural ones. In particular, the antagonistic-antagonistic networks are all plant-herbivore-parasitoid networks which is a specific kind of A-A network (as supposed to say, herbivory and insectivory combinations that could be collected independently). These are typically collected by sampling a herbivore on a plant and then rearing a parasitoid out of it. Whereas a plant-pollinator/plant-herbivore or plant-pollinator/plant-seed disperser network would have 2 independent measures for each interaction. This means that the herbivore linking to a parasitoid is the only way a parasitoid could be in the network. The authors now note that this is relevant for the robustness analyses as any plant removal simulation would necessarily propagate up the A-A network due to the sampling design (but also biologically) rather than any network property (a parasitoid has no ability to interact with a plant outside of the sampled herbivores). But I think it is important for the “fundamental differences” reported as well.

I’m not sure there is a way to disentangle this confounding effect, and the conclusions are still relevant and interesting, but it is a consideration that needs to be acknowledged, particularly in the discussion.

Specifically, in lines 210-212: the conclusion that the hubs being connectors is because generalist herbivores might be more abundant, could also be because the parasitoid can only be reared out of one of the sampled herbivores. I would be very cautious about the interpretation here being about an ecological or structural difference when it could be an unavoidable sampling effect.

The same is true in the respective results (lines 113-114) regarding the equal splitting of links. Sampling methods mean the herbivore hub would necessarily be equally split between the 2 layers. If the links were collected independently (as in plant-pollinator/plant-herbivore networks), then you might expect fewer shared links. It is important to distinguish which of the results are due to collection method differences, especially since the data were typically not collected to test for these sorts of differences so the variable collection methods are likely to be an important confounding effects.

Otherwise, the conclusions are well supported by the results and the approach including multiple null models and now including weighted networks, where possible, is sound and much improved!

Very minor typo/syntax suggestions:

Line 74: “consists in 44…” should be “consists of 44…”

**********

Have the authors made all data and (if applicable) computational code underlying the findings in their manuscript fully available?

Reviewer #1: No: I did not see a link to the data the authors used. I might have missed it, but it must appear. Referring to the original papers is not good enough because sometimes there are changes when using other peoples data. The authors should store their data online in a permanent data repository (Figshare, dryad, etc).

Reviewer #2: Yes

Reviewer #3: None

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

Figure Files:

Data Requirements:

Reproducibility:

References:

Review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript.

If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

PLoS Comput Biol. 2024 Jan 19;20(1):e1011770. doi: 10.1371/journal.pcbi.1011770.r004

Author response to Decision Letter 1

24 Nov 2023

Attachment

Submitted filename: Response_2.pdf

Click here for additional data file.^{(58.3KB, pdf)}

PLoS Comput Biol. doi: 10.1371/journal.pcbi.1011770.r005

Decision Letter 2

Mercedes Pascual, James O'Dwyer

18 Dec 2023

Dear Dr Domínguez-García,

We are pleased to inform you that your manuscript 'The structure and robustness of ecological networks with two interaction types' has been provisionally accepted for publication in PLOS Computational Biology.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. A member of our team will be in touch with a set of requests.

Please note that your manuscript will not be scheduled for publication until you have made the required changes, so a swift response is appreciated.

IMPORTANT: The editorial review process is now complete. PLOS will only permit corrections to spelling, formatting or significant scientific errors from this point onwards. Requests for major changes, or any which affect the scientific understanding of your work, will cause delays to the publication date of your manuscript.

Should you, your institution's press office or the journal office choose to press release your paper, you will automatically be opted out of early publication. We ask that you notify us now if you or your institution is planning to press release the article. All press must be co-ordinated with PLOS.

Thank you again for supporting Open Access publishing; we are looking forward to publishing your work in PLOS Computational Biology.

Best regards,

Mercedes Pascual

Academic Editor

PLOS Computational Biology

James O'Dwyer

Section Editor

PLOS Computational Biology

***********************************************************

PLoS Comput Biol. doi: 10.1371/journal.pcbi.1011770.r006

Acceptance letter

Mercedes Pascual, James O'Dwyer

15 Jan 2024

PCOMPBIOL-D-23-00025R2

The structure and robustness of ecological networks with two interaction types

Dear Dr Domínguez-García,

I am pleased to inform you that your manuscript has been formally accepted for publication in PLOS Computational Biology. Your manuscript is now with our production department and you will be notified of the publication date in due course.

The corresponding author will soon be receiving a typeset proof for review, to ensure errors have not been introduced during production. Please review the PDF proof of your manuscript carefully, as this is the last chance to correct any errors. Please note that major changes, or those which affect the scientific understanding of the work, will likely cause delays to the publication date of your manuscript.

Soon after your final files are uploaded, unless you have opted out, the early version of your manuscript will be published online. The date of the early version will be your article's publication date. The final article will be published to the same URL, and all versions of the paper will be accessible to readers.

Thank you again for supporting PLOS Computational Biology and open-access publishing. We are looking forward to publishing your work!

With kind regards,

Zsofi Zombor

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 Text. Supplementary information file, including supplementary figures A-P and supplementary tables A-D.

(PDF)

Click here for additional data file.^{(608.4KB, pdf)}

Attachment

Submitted filename: Response_PlosCBiolTripartite.pdf

Click here for additional data file.^{(248.8KB, pdf)}

Attachment

Submitted filename: Response_2.pdf

Click here for additional data file.^{(58.3KB, pdf)}

Data Availability Statement

The data and code supporting the results is available in Zenodo: https://doi.org/10.5281/zenodo.10198613.

[pcbi.1011770.ref001] 1.IPBES. Global assessment report on biodiversity and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services; 2019. Available from: https://zenodo.org/record/3831673.

[pcbi.1011770.ref002] 2. Dunne JA, Williams RJ, Martinez ND. Network structure and biodiversity loss in food webs: robustness increases with connectance. Ecology Letters. 2002;5(4):558–567. doi: 10.1046/j.1461-0248.2002.00354.x [DOI] [Google Scholar]

[pcbi.1011770.ref003] 3. Memmott J, Waser NM, Price MV. Tolerance of pollination networks to species extinctions. Proceedings of the Royal Society of London Series B: Biological Sciences. 2004;271(1557):2605–2611. doi: 10.1098/rspb.2004.2909 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref004] 4. Montoya JM, Pimm SL, Solé RV. Ecological networks and their fragility. Nature. 2006;442(7100):259–264. doi: 10.1038/nature04927 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref005] 5. Kaiser-Bunbury CN, Muff S, Memmott J, Müller CB, Caflisch A. The robustness of pollination networks to the loss of species and interactions: a quantitative approach incorporating pollinator behaviour. Ecology Letters. 2010;13(4):442–452. doi: 10.1111/j.1461-0248.2009.01437.x [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref006] 6. Melián CJ, Bascompte J, Jordano P, Krivan V. Diversity in a complex ecological network with two interaction types. Oikos. 2009;118(1):122–130. doi: 10.1111/j.1600-0706.2008.16751.x [DOI] [Google Scholar]

[pcbi.1011770.ref007] 7. Fontaine C, Guimaraes PR, Kéfi S, Loeuille N, Memmott J, van der Putten W, et al. The ecological and evolutionary implications of merging different types of networks. Ecology Letters. 2011;14(11):1170–1181. [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref008] 8. Kéfi S, Berlow EL, Wieters EA, Navarrete SA, Petchey OL, Wood SA, et al. More than a meal… integrating non-feeding interactions into food webs. Ecology Letters. 2012;15(4):291–300. [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref009] 9. Kéfi S, Berlow EL, Wieters EA, Joppa LN, Wood SA, Brose U, et al. Network structure beyond food webs: mapping non-trophic and trophic interactions on Chilean rocky shores. Ecology. 2015;96(1):291–303. doi: 10.1890/13-1424.1 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref010] 10. Kéfi S, Miele V, Wieters EA, Navarrete SA, Berlow EL. How Structured Is the Entangled Bank? The Surprisingly Simple Organization of Multiplex Ecological Networks Leads to Increased Persistence and Resilience. PLOS Biology. 2016;14(8):e1002527. doi: 10.1371/journal.pbio.1002527 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref011] 11. García-Callejas D, Molowny-Horas R, Araújo MB. Multiple interactions networks: towards more realistic descriptions of the web of life. Oikos. 2017;127(1):5–22. doi: 10.1111/oik.04428 [DOI] [Google Scholar]

[pcbi.1011770.ref012] 12. Kéfi S. Ecological Networks: from structure to dynamics. In: Theoretical ecology: Concepts and applications. Oxford University Press; 2020. p. 143–160. [Google Scholar]

[pcbi.1011770.ref013] 13. Timóteo S, Albrecht J, Rumeu B, Norte AC, Traveset A, Frost CM, et al. Tripartite networks show that keystone species can multitask. Functional Ecology. 2022. [Google Scholar]

[pcbi.1011770.ref014] 14. May RM. Will a Large Complex System be Stable? Nature. 1972;238(5364):413–414. doi: 10.1038/238413a0 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref015] 15. Mougi A, Kondoh M. Diversity of Interaction Types and Ecological Community Stability. Science. 2012;337(6092):349–351. doi: 10.1126/science.1220529 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref016] 16. Allesina S, Tang S. Stability criteria for complex ecosystems. Nature. 2012;483(7388):205–208. doi: 10.1038/nature10832 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref017] 17. Sauve AMC, Fontaine C, Thébault E. Structure-stability relationships in networks combining mutualistic and antagonistic interactions. Oikos. 2013;123(3):378–384. [Google Scholar]

[pcbi.1011770.ref018] 18. Lurgi M, Montoya D, Montoya JM. The effects of space and diversity of interaction types on the stability of complex ecological networks. Theoretical Ecology. 2015;9(1):3–13. [Google Scholar]

[pcbi.1011770.ref019] 19. McWilliams C, Lurgi M, Montoya JM, Sauve A, Montoya D. The stability of multitrophic communities under habitat loss. Nature Communications. 2019;10(1). doi: 10.1038/s41467-019-10370-2 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref020] 20. Hale KRS, Valdovinos FS, Martinez ND. Mutualism increases diversity, stability, and function of multiplex networks that integrate pollinators into food webs. Nature Communications. 2020;11(1). doi: 10.1038/s41467-020-15688-w [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref021] 21. Pocock MJO, Evans DM, Memmott J. The Robustness and Restoration of a Network of Ecological Networks. Science. 2012;335(6071):973–977. doi: 10.1126/science.1214915 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref022] 22. Sauve AMC, Thébault E, Pocock MJO, Fontaine C. How plants connect pollination and herbivory networks and their contribution to community stability. Ecology. 2016;97(4):908–917. doi: 10.1890/15-0132.1 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref023] 23. Genrich CM, Mello MAR, Silverira FAO, Bronstein JL, Paglia AP. Duality of interaction outcomes in a plant-frugivore multilayer network. Oikos. 2016;126:361–368. [Google Scholar]

[pcbi.1011770.ref024] 24. Dáttilo W, Lara-Rodríguez N, Jordano P, Guimarães PR, Thompson JN, Marquis RJ, et al. Unravelling Darwin’s entangled bank: architecture and robustness of mutualistic networks with multiple interaction types. Proceedings of the Royal Society B: Biological Sciences. 2016;283(1843):20161564. doi: 10.1098/rspb.2016.1564 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref025] 25. Astegiano J, Altermatt F, Massol F. Disentangling the co-structure of multilayer interaction networks: degree distribution and module composition in two-layer bipartite networks. Scientific Reports. 2017;7(1). doi: 10.1038/s41598-017-15811-w [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref026] 26. Mello MAR, Felix GM, Pinheiro RBP, Muylaert RL, Geiselman C, Santana SE, et al. Insights into the assembly rules of a continent-wide multilayer network. Nature Ecology & Evolution. 2019;3(11):1525–1532. [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref027] 27. Morrison BML, Brosi BJ, Dirzo R. Agricultural intensification drives changes in hybrid network robustness by modifying network structure. Ecology Letters. 2019;23(2):359–369. doi: 10.1111/ele.13440 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref028] 28. Evans DM, Pocock MJO, Memmott J. The robustness of a network of ecological networks to habitat loss. Ecology Letters. 2013;16(7):844–852. doi: 10.1111/ele.12117 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref029] 29. Kivela M, Arenas A, Barthelemy M, Gleeson JP, Moreno Y, Porter MA. Multilayer networks. Journal of Complex Networks. 2014;2(3):203–271. doi: 10.1093/comnet/cnu016 [DOI] [Google Scholar]

[pcbi.1011770.ref030] 30. Thébault E, Fontaine C. Stability of Ecological Communities and the Architecture of Mutualistic and Trophic Networks. Science. 2010;329(5993):853–856. doi: 10.1126/science.1188321 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref031] 31. Dunne JA, Williams RJ, Martinez ND. Food-web structure and network theory: The role of connectance and size. Proceedings of the National Academy of Sciences. 2002;99(20):12917–12922. doi: 10.1073/pnas.192407699 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref032] 32. Staniczenko PPA, Lewis OT, Jones NS, Reed-Tsochas F. Structural dynamics and robustness of food webs. Ecology Letters. 2010;13(7):891–899. doi: 10.1111/j.1461-0248.2010.01485.x [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref033] 33. Bascompte J, Stouffer DB. The assembly and disassembly of ecological networks. Philosophical Transactions of the Royal Society B: Biological Sciences. 2009;364(1524):1781–1787. doi: 10.1098/rstb.2008.0226 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref034] 34. Dallas T, Cornelius E. Co-extinction in a host-parasite network: identifying key hosts for network stability. Scientific Reports. 2015;5(1). doi: 10.1038/srep13185 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref035] 35. Strona G, Lafferty KD. Environmental change makes robust ecological networks fragile. Nature Communications. 2016;7(1). doi: 10.1038/ncomms12462 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref036] 36. Solé RV, Montoya M. Complexity and fragility in ecological networks. Proceedings of the Royal Society of London Series B: Biological Sciences. 2001;268(1480):2039–2045. doi: 10.1098/rspb.2001.1767 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref037] 37. Vieira MC, Almeida-Neto M. A simple stochastic model for complex coextinctions in mutualistic networks: robustness decreases with connectance. Ecology Letters. 2014;18(2):144–152. doi: 10.1111/ele.12394 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref038] 38. Dutton EM, Luo EY, Cembrowski AR, Shore JS, Frederickson ME. Three’s a Crowd: Trade-Offs between Attracting Pollinators and Ant Bodyguards with Nectar Rewards in Turnera. The American Naturalist. 2016;188(1):38–51. doi: 10.1086/686766 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref039] 39. Jacobsen DJ, Raguso RA. Lingering Effects of Herbivory and Plant Defenses on Pollinators. Current Biology. 2018;28(19):R1164–R1169. doi: 10.1016/j.cub.2018.08.010 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref040] 40. Burgos E, Ceva H, Perazzo RPJ, Devoto M, Medan D, Zimmermann M, et al. Why nestedness in mutualistic networks? Journal of Theoretical Biology. 2007;249(2):307–313. doi: 10.1016/j.jtbi.2007.07.030 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref041] 41. Jonhson S, Domínguez-García V, Muñoz MA. Factors Determining Nestedness in Complex Networks. PLoS ONE. 2013;8(9):e74025. doi: 10.1371/journal.pone.0074025 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref042] 42. Shinohara N, Uchida K, Yoshida T. Contrasting effects of land-use changes on herbivory and pollination networks. Ecology and Evolution. 2019;9(23):13585–13595. doi: 10.1002/ece3.5814 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref043] 43. Hackett TD, Sauve AMC, Davies N, Montoya D, Tylianakis JM, Memmott J. Reshaping our understanding of species’ roles in landscape-scale networks. Ecology Letters. 2019;22(9):1367–1377. doi: 10.1111/ele.13292 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref044] 44. Macfadyen S, Gibson R, Polaszek A, Morris RJ, Craze PG, Planqué R, et al. Do differences in food web structure between organic and conventional farms affect the ecosystem service of pest control? Ecology Letters. 2009;12(3):229–238. doi: 10.1111/j.1461-0248.2008.01279.x [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref045] 45. Guimerà R, Amaral LAN. Cartography of complex networks: modules and universal roles. Journal of Statistical Mechanics: Theory and Experiment. 2005;2005(02):P02001. doi: 10.1088/1742-5468/2005/02/P02001 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref046] 46. Battiston F, Nicosia V, Latora V. Structural measures for multiplex networks. Physical Review E. 2014;89(3). doi: 10.1103/PhysRevE.89.032804 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref047] 47. Scherber C, Eisenhauer N, Weisser WW, Schmid B, Voigt W, Fischer M, et al. Bottom-up effects of plant diversity on multitrophic interactions in a biodiversity experiment. Nature. 2010;468(7323):553–556. doi: 10.1038/nature09492 [DOI] [PubMed] [Google Scholar]

[pcbi.1011770.ref048] 48. Allesina S, Pascual M. Googling Food Webs: Can an Eigenvector Measure Species’ Importance for Coextinctions? PLoS Computational Biology. 2009;5(9):e1000494. doi: 10.1371/journal.pcbi.1000494 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pcbi.1011770.ref049] 49. Domínguez-García V, Muñoz MA. Ranking species in mutualistic networks. Scientific Reports. 2015;5(1). doi: 10.1038/srep08182 [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

The structure and robustness of ecological networks with two interaction types

Virginia Domínguez-García

Sonia Kéfi

Roles

Abstract

Author summary

Introduction

Fig 1. Tripartite networks, robustness and interdependence.

Results

Different ways of connecting the interaction layers

Fig 2. How does the shared set of nodes connect the network?

Interdependence of the robustness of animal species

Fig 3. Interdependence and robustness of tripartite networks.

Tripartite networks’ robustness

Plant importance for robustness

Fig 4. Plant importance rankings.

Discussion

Methods

Data set

Structural metrics of the connector nodes

Quantifying robustness

Interdependence

Plant importance rankings

Estimating tripartite robustness from networks with one interaction

Multiple regressions

Null models

Fig 5. The 4 different null models used in this study.

Supporting information

Data Availability

Funding Statement

References

Decision Letter 0

Mercedes Pascual

James O'Dwyer

Roles

Transfer Alert

Author response to Decision Letter 0

Decision Letter 1

Mercedes Pascual

James O'Dwyer

Roles

Author response to Decision Letter 1

Decision Letter 2

Mercedes Pascual

James O'Dwyer

Roles

Acceptance letter

Mercedes Pascual

James O'Dwyer

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases