Specialization and Rarity Predict Nonrandom Loss of Interactions from Mutualist Networks

See allHide authors and affiliations

Science  23 Mar 2012:
Vol. 335, Issue 6075, pp. 1486-1489
DOI: 10.1126/science.1215320


The loss of interactions from mutualistic networks could foreshadow both plant and animal species extinctions. Yet, the characteristics of interactions that predispose them to disruption are largely unknown. We analyzed 12 pollination webs from isolated hills (“sierras”), in Argentina, ranging from tens to thousands of hectares. We found evidence of nonrandom loss of interactions with decreasing sierra size. Low interaction frequency and high specialization between interacting partners contributed additively to increase the vulnerability of interactions to disruption. Interactions between generalists in the largest sierras were ubiquitous across sierras, but many of them lost their central structural role in the smallest sierras. Thus, particular configurations of interaction networks, along with unique ecological relations and evolutionary pathways, could be lost forever after habitat reduction.

Interspecific interactions link species within complex trophic and nontrophic webs (13). Disruption of individual interaction links can compromise both the survival of formerly interacting species pairs and of other species with whom they are directly or indirectly connected (4, 5). For mutually beneficial interactions, such as those between plants and pollinators, the loss of interactions from a pollination web can jeopardize plant sexual reproduction directly through pollen limitation (6, 7) and can reduce pollinator fitness by decreasing the availability of floral resources (8, 9). Mutualists can persist to different extents after link disruption, depending on individual longevity, initial population abundance, generalization in the use of mutualistic partners, and importance of the pollination mutualism itself for species survivorship (10, 11). Consequently, loss of mutualistic interactions from a pollination web usually precedes species loss (12), as has been observed after habitat fragmentation (9, 13) and species invasion (14, 15). This extinction lag suggests that interactions, rather than species statistics, should be the main focus of studies of web dynamics and stability under different environmental change scenarios, and justifies the management of interspecific interactions as target activities of conservation and restoration programs (16).

Despite much progress in understanding the structure and dynamics of mutualistic webs, we still have a limited ability to predict species extinctions. This ability would improve if we could identify those interactions most susceptible to disruption. However, increasing predictive ability rests on two untested assumptions: (i) interactions are lost nonrandomly from webs following disturbance; and (ii), analogous to the “response traits” of species (17), particular traits that characterize mutualistic interactions increase their chance of disruption. Here, we explore these two hypotheses using 12 pollination webs from untilled hills or “sierras” that rise from the Pampas of Argentina (18). Ranging from tens to thousands of hectares, these sierras were once connected by a matrix of natural grassland, but are nowadays completely isolated by an intensively managed surrounding agricultural matrix. Therefore, they can be viewed as representing a gradient of habitat reduction. In addition to containing several endemic species of Gondwanan origin, these sierras still preserve many floristic elements that were formerly common in the surrounding plains and elsewhere in southern South America (19). Previous work revealed that the number of plant and pollinator species and interaction links between them increase with area of the sierras and that the rate of increase was half as great for species as it was for the number of links (13). However, why specific links are lost in smaller sierras, whereas others persist, remains unexplained.

Across all 12 pollination webs, we recorded 1170 distinct interactions (links) among 96 and 172 species of plants and flower visitors, respectively (Fig. 1 and fig. S1). When sierras were ordered by decreasing size, we found that interactions present in each sierra tended to be proper (i.e., nested) subsets (20) of those recorded in the next-larger sierra (Z = 6.80 and Z = 5.43 based on the completely randomized and marginal-conditioned null models, respectively; P < 10−6 in both cases). This result is consistent with the hypothesis that mutualistic interactions are lost nonrandomly as habitat size decreases. Furthermore, interactions were more nested than plant and pollinator species themselves (fig. S2), which probably indicates their greater and more proximate susceptibility to habitat reduction (13). Thus, some mutualistic species could persist despite the disruption of some of their interactions, potentially because of mutualism redundancy and other buffering life-history traits (10) or simply as part of an extinction debt (21).

Fig. 1

Combined plant-animal pollinator interaction matrix depicting the 1170 distinct interactions among 96 and 172 species of plants and flower visitors, respectively, recorded across the 12 sierras. Species are ranked according to decreasing number of interactions per species. A colored cell specifies an observed interaction. Different colors and color hues indicate the number of sierras in which each interaction was found (from 1 to 12). Interactions occurring in most sierras, both large and small, are mostly restricted to the upper left corner of the matrix. The interaction matrix of each sierra is provided as Supporting Online Material (fig. S1).

This pattern of nonrandom losses prompted the question of which traits of plant-pollinator interactions make them most susceptible to disruption. We analyzed two traits, interaction frequency and degree of generalization (estimated here as the average number of species with which the plant and pollinator interact), which required no detailed information about the species involved, beyond knowing with how many species they interacted and how frequently (22). We chose these traits because, first, locally rare plant-pollinator interactions should be particularly susceptible to habitat reduction because any further decrease in interaction frequency, perhaps related to declining species abundance, could trigger complete disruption (22, 23). The second reason was that interactions between plant and pollinator species with limited numbers of alternative partners (i.e., interactions of low degree) should also be particularly susceptible beyond any confounding effect of interaction frequency, because they cannot be “subsidized” or “rescued” by third parties when, for instance, interacting species become spatially or phenologically isolated from each other (4, 24). Thus, low-frequency interactions and/or interactions between specialists should be restricted to continuous habitat or large habitat fragments, whereas frequent interactions and/or interactions between generalists should be more resistant to habitat reduction and, therefore, be more ubiquitous (i.e., occur in habitat fragments of all sizes).

For each sierra, we characterized the ubiquity of each plant-pollinator interaction as the proportion of other sampled sierras in which it also occurred. Specifically, we predicted that interactions from a large sierra with a high frequency and/or degree (i.e., involving generalist species) should be more ubiquitous than interactions with a low frequency and/or degree, which are expected to be disrupted by habitat reduction and thus absent from the small sierras. Therefore, the positive relation between interaction ubiquity and the two interaction traits, frequency or degree, which we predicted for large sierras should weaken in the small sierras that have already been mostly depleted of fragmentation-susceptible pollination interactions.

Following our expectation, the relation between interaction ubiquity and its two predictors, local interaction frequency and degree of generalization, became increasingly positive with increasing sierra size (Fig. 2). Particularly, these relations were strongest among interactions recorded in sierras >100 ha (fig. S1, A to H) and became weaker or disappeared for interactions in sierras <100 ha (fig. S1, I to L). For example, on Volcan, one of the largest sierras (>2000 ha), expected ubiquity increased from 0.15 to 0.82 and from 0.09 to 0.76 over the range of interaction frequencies and degree of generalization, respectively (fig. S1B). In contrast, on Difuntito, one of the smallest sierras (13 ha), expected ubiquity increased only from 0.12 to 0.38 over the range of interaction frequencies and remained fairly constant (~0.15) over the range of interaction generalization (fig. S1J). The results from this small sierra also illustrate that the nonrandom loss of vulnerable interactions is, to some extent, unrelated to changes in interaction diversity, because the pollination web of Difuntito (the only fenced sierra protected from grazing and fire) was unexpectedly rich in species and interactions (13). Nevertheless, its position within the general pattern depicted in Fig. 2 was in no way anomalous, which suggested that this sierra lacked most of the vulnerable interactions recorded in the largest sierras. This result further stresses the importance of an area-per-se effect on the selective loss of interactions.

Fig. 2

The dependence on sierra size of the relation between interaction ubiquity and interaction (A) frequency and (B) degree of generalization. Dependence is represented by regression coefficients (β ± 95% confidence intervals) from binomial generalized linear models conducted for each of the 12 sierras. Individual coefficients whose confidence intervals do not overlap the dotted line differ significantly from zero. Solid lines and summary statistics indicate that the linear relation between ubiquity and each interaction trait increases significantly with sierra area. Specific results for Difuntito (D), a small sierra, and Volcan (V), a large sierra, discussed in the text are shown in fig. S1, B and J, respectively.

Interaction frequency and degree of generalization had largely independent effects on interaction loss. First, these two traits of interactions were correlated positively, but generally weakly within sierras (r < 0.55 in all cases), with the strength of this correlation increasing only marginally with sierra size (fig. S3). Second, and more important, the increasingly positive relation between interaction ubiquity and interaction frequency or degree of generalization with increasing sierra size (Fig. 2) persisted after accounting for any collinearity between the predictors by using partial model coefficients (fig. S4). Thus, particular traits of plant-pollinator interactions—specifically, low frequency and high specialization—contribute systematically and additively to their vulnerability to habitat reduction. Consequently, disruption of rare mutualistic interactions and those between reciprocal specialists may signal future biodiversity loss, and so they should be the focus of biodiversity monitoring and restoration programs. In particular, specialized interactions should be of primary concern, as their disruption could lead to the rapid loss of species that lack alternative efficient mutualists. Based on our results, a low interaction frequency would further increase the vulnerability of such interaction links.

Frequent interactions between generalist plants and pollinators establish the architectural core of pollination networks (25), which provides stability and resilience to the entire web (1, 2, 25, 26). This core also governs coevolutionary dynamics of generalists engaged in strong interactions with other generalists and asymmetrically with most specialists (27, 28). The differential loss of relatively specialized interactions in particular would accentuate this intrinsic asymmetry of networks (29, 30) after habitat reduction, which was evidenced here by a weak but increasingly negative association between the specialization of plants and that of their interacting animal partners with decreasing sierra size (fig. S5). This result suggests that many specialists persist in fragmented landscapes by interacting with locally and regionally resilient generalists, around which interactions become increasingly concentrated. Such “supergeneralists,” also described for pollination webs on islands and in communities with many invaders (14, 31), should represent strong novel demographic and selection pressures for persisting specialists.

Our results also hint at subtle, but important, qualitative changes in the structure of the web core. Increasingly positive relations between interaction ubiquity and the two predictive interaction traits, frequency and degree of generalization (Fig. 2), indicate that the core in the largest sierras included a set of regionally widespread, robust interactions that were present in both large and small sierras (fig. S1). However, a trend toward decreasing frequency and degree of generalization of many of the most ubiquitous interactions (fig. S6) indicates their displacement from the inner core to relatively more marginal positions within the web as sierra size decreases. Even though some interactions [e.g., between species coded 32 and 108 (table S4)] remain part of this core, irrespective of the size of the sierra (fig. S1), the central structural role played previously by some of these ubiquitous interactions could remain vacant or be replaced by more facultative interactions present in one or a few small sierras [e.g., interaction between species coded 56 and 259 in Difuntito (fig. S1J and table S4)]. Thus, because of this core shift, species surviving in small habitat fragments could be subject to more variable ecological and evolutionary dynamics in space and perhaps time.

Functional redundancy in mutualistic interaction networks provides relative stability to minor or moderate random losses of species and interactions (4, 32), but nonrandom disruption can affect species survival and adaptation more immediately and profoundly. Particularly, infrequently occurring and geographically restricted specialized interactions that involve efficient pollination for the plant and/or some critical floral resource for the pollinator can be highly relevant at both ecological and evolutionary time scales (33, 34), and their disruption could lead to time-lagged species decline (35). Using a comparative interaction-network approach, we provide evidence that these particular interactions, occurring at low frequency and between species that lack alternative mutualists, are the most likely to be lost, which could accelerate the rate of species extinctions. In combination, our results suggest that nonrandom interaction disruption after habitat fragmentation and other anthropogenic disturbances will affect the most codependent and rare mutualisms and alter configurations of interaction networks, along with unique ecological relations and evolutionary pathways.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S6

Tables S1 to S4

References (36–50)

References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. Acknowledgments: The authors thank J. M. Gómez, L. D. Harder, D. P. Vázquez, M. Verdú, N. M. Waser, and two anonymous referees for useful comments and suggestions; A. Saez and D. Porrini for field assistance; V. Izpizua and M. Nuciari for help in plant identification; and J. Farina and A. Roig-Alsina for help in identifying insects. Partial funding by the National Institute of Agricultural Technology (INTA), Balcarce (PNECO1302), the Argentina National Research Council (CONICET) (PIP 01623), the National Fund for Research (PICT 01300), and the National University of Comahue (B152/04) is acknowledged. M.A.A. is a career researcher and M.S. a fellow of CONICET. J.M.T. is funded by a Rutherford Discovery Fellowship administered by the Royal Society of New Zealand. Data used in the analyses are available in the Supporting Online Material.
View Abstract

Stay Connected to Science

Navigate This Article