Current Constraints and Future Directions in Estimating Coextinction

Authors


email mmoir@unimelb.edu.au

Abstract

Abstract: Coextinction is a poorly quantified phenomenon, but results of recent modeling suggest high losses to global biodiversity through the loss of dependent species when hosts go extinct. There are critical gaps in coextinction theory, and we outline these in a framework to direct future research toward more accurate estimates of coextinction rates. Specifically, the most critical priorities include acquisition of more accurate host data, including the threat status of host species; acquisition of data on the use of hosts by dependent species across a wide array of localities, habitats, and breadth of both hosts and dependents; development of models that incorporate correlates of nonrandom host and dependent extinctions, such as phylogeny and traits that increase extinction-proneness; and determination of whether dependents are being lost before their hosts and adjusting models accordingly. Without synergistic development of better empirical data and more realistic models to estimate the number of cothreatened species and coextinction rates, the contribution of coextinction to global declines in biodiversity will remain unknown and unmanaged.

Abstract

Resumen: La coextinción es un fenómeno poco cuantificado, pero los resultados de modelos recientes sugieren grandes pérdidas de biodiversidad local mediante la pérdida de especies dependientes cuando los hospederos se extinguen. Hay vacíos críticos en la teoría de coextincíón, y los delineamos en un marco de referencia para dirigir la investigación futura hacia estimaciones más precisas de las tasas de coextinción. Específicamente, las prioridades más críticas incluyen la obtención de datos más precisos de los hospederos, incluyendo el estatus de amenaza de las especie hospedera; obtención de datos sobre el uso de hospederos por especies dependientes en una amplia gama de localidades, hábitats y amplitud tanto de hospederos como dependientes; desarrollo de modelos que incorporen correlaciones de extinciones no aleatorias de hospederos y dependientes, como la filogenia y atributos que incrementan la susceptibilidad a la extinción; y determinar sí los dependientes se pierden antes que sus hospederos y consecuentemente ajustar los modelos. Sin el desarrollo sinérgico de mejores datos empíricos y modelos más realistas para estimar el número de especies coamenazadas y las tasas de coextinción, la contribución de coextinción a las declinaciones globales de biodiversidad permanecerá desconocida y no podrá ser manejada.

Introduction

The loss of species through extinction is the only truly irreversible global environmental change occurring today (Dirzo & Raven 2003). Of the many mechanisms of species extinction, the least understood is coextinction. Coextinction occurs when species go extinct because the host on which they depend becomes extinct (Pimm 1986), or at least undergoes significant decline. Our definition of coextinction is the loss of dependent species due to a change in their host population, such as reduced host abundance or removal of individual hosts from the wild. Extinction of a dependent species through a change in the host population, without immediate loss of the host species, is not traditionally regarded as coextinction. This host-induced extinction, however, should be recognized as a form of coextinction because it requires the same management measures as extinction of both host and dependent simultaneously to maintain a viable host population accessible to the dependent species.

The degree of species specialization “is arguably the most fundamental concept in the history of thought on extinction risk” (McKinney 1997). Host specificity is therefore a crucial factor in coextinction. Every animal species may host a range of external and internal parasites, including mites, lice, fleas, and nematodes. Many of these dependent species have restricted or unique host relationships and hence face coextinction if their hosts become extinct (Nunn et al. 2004; Hughes & Page 2007; Dobson et al. 2008). Similarly, many plant species are host to at least one host-specific herbivorous species of insect (Strong et al. 1984). Approximately 22,100 (7.4%) of the world's 300,000 plant species are considered threatened (Smith et al. 1993; Myers et al. 2000), so the potential for coextinction of numerous host-dependent insect species is high (Koh et al. 2004a). Despite the critical implications of coextinction for global biodiversity, the topic is poorly studied. More generally, species interactions are usually ignored when the extinction risks of species are assessed (Sabo 2008).

Current conservation strategies overlook most invertebrates and microorganisms, including pathogens (many of which depend on hosts). Inefficient use of scarce conservation resources and unforeseen extinctions may result, partly due to the losses of dependent taxa themselves and other taxa reliant on trophic processes in which dependent species are involved (the latter are often termed extinction cascades; Diamond 1989). Some conservation strategies for host taxa may even promote extinction risks for their dependent species either directly (actively removing the dependent species from hosts [e.g., Bevill et al. 1999]) or indirectly (removal of the host from the wild [e.g., Gompper & Williams 1998]). Although the loss of a dependent species, such as a parasite or herbivore, may be expected to benefit the short-term survival of a threatened host (Bevill et al. 1999), unexpected consequences for the host could result, such as increased abundances of other parasites or herbivores (González-Megías & Gómez 2003). Insects with parasitic larval stages that depend on plants or animals may also perform important ecosystem functions, for example as pollinators, during other stages of their life cycle (Hudson et al. 2006). In some circumstances, conservation of dependent species may benefit the host species because parasites can increase the genetic diversity of hosts (Nunn et al. 2004; Poulin & Morand 2004; Duffy et al. 2008) and thereby reduce susceptibility of host populations to extinction through epidemics of novel pathogens (e.g., Altizer et al. 2003). Thus, dependent species, themselves a large and important component of biodiversity, should be conserved to sustain an even wider array of species and ecological processes.

Because coextinction rates must be estimated with models of dependent species and their hosts, here we first examine current models. Second, we propose future directions for empirical studies that are needed to structure and parameterize realistic models of coextinction.

Linear versus Nonlinear Relationships in Models Estimating Coextinction

Invertebrates are the most speciose eukaryote group, and most of them are undescribed (Hammond 1995). Thus, estimating coextinction rates for invertebrates will require modeling to supplement available data and direct further data collection (Koh et al. 2004a). There is no standard model for estimating the potential loss of dependent species through coextinction of their hosts. The simplest approach assumes a linear extrapolation of extinction in which the number of threatened dependents reflects the number of hosts that are threatened (Stork & Lyal 1993; Poulin & Morand 2004; Thacker et al. 2006). The linear approach assumes unique host dependency (monophagy), and departure from this assumption adversely affects the accuracy of estimates of threatened dependents.

For most species the linear approach to estimating coextinction risk is overly simplistic for several reasons. First, dependent species may require several different hosts during their life cycle; thus, their extinction risk is compounded if any one host is threatened. For example, the endangered butterfly Maculinea arion depends on a host plant and a tending ant species, but extinction probability of the butterfly greatly increases if either the host–plant cover falls below 5% or host–ant density is <500 nests/ha (Griebeler & Seitz 2002). The assumption of a linear relationship between host and dependent extinctions is therefore likely to underestimate extinction rates of dependents because dependents with more complex life histories are likely to have higher coextinction rates (Koh et al. 2004a). Second, a dependent species may be capable of using several alternative host species, only one of which may be threatened. In this case linear extrapolation of coextinction risks overestimating the threat (Dobson et al. 2008). Moreover, alternative hosts may vary in suitability, such that dependent populations may have different rates of survival, growth, and reproduction on different host species. For example, if the endangered butterfly Maculinea rebeli is tended by ant species other than Myrmica schencki, it has 29 times higher larval mortality (Steiner et al. 2003). Such relationships suggest a departure from simple linear extrapolation.

To move beyond simple linear assumptions of dependent–host relationships, Koh et al. (2004a) estimated coextinction risks for a range of invertebrates and their hosts with probabilistic and nomographic models derived from host–dependent matrices that incorporated different levels of host specificity. They found that as the number of host extinctions increases, the number of dependent species being extinguished increases at an accelerating rate (Fig. 1). The exceptions were assemblages in which all dependent species were host specific to a single host species, which produced a linear relationship (e.g., primates and their nematodes; Fig. 1). From these models Koh et al. (2004a) estimated that on the basis of threat status of host taxa (Hilton-Taylor 2000), 4672 beetles, 598 fish parasites, 446 lice, 193 bird mites, 142 butterflies, 20 primate nematodes, nine primate fungi, and eight fig wasps were threatened globally with coextinction. They suggest that this mechanism of extinction has most likely already caused the loss of over 200 dependent species since 1500 AD and will become increasingly important as more host species are extinguished.

Figure 1.

Hypothetical examples of the proportion of dependent species expected to be extinguished through coextinction with increasing proportions of host extinctions derived from a probabilistic model approach (sensuKoh et al. 2004a) (line a); dependent species having a 1:1 ratio with hosts (i.e., monophagy) or a linear relationship (line b); and dependent species becoming extinct before hosts (line c).

Current Gaps in Coextinction Theory

Despite recent advances in estimating potential coextinction, particularly by incorporating some variability in host specificity (Koh et al. 2004a; Dobson et al. 2008), further refinements are required to improve the empirical basis for parameterizing coextinction models and to accommodate inevitable uncertainties that remain in relationships between dependents and their hosts. Coextinction estimates are influenced by the host data, dependent data, and the interactions between these two components (Fig. 2). Several (often interrelated) factors may alter hosts, dependents, or their interaction. For hosts these factors include the accuracy of the threat status of hosts (Fig. 2a), the breadth of host species examined (Fig. 2b), and the correlated pattern of host extinctions (Fig. 2c). For dependents, important factors are breadth of the dependent species assessed (Fig. 2d) and host specificity of the dependents (Fig. 2e). Important considerations influencing interactions between hosts and dependent species include differences in host–dependent interactions across regions (Fig. 2f) and whether hosts or their dependents go extinct first (Fig. 2g). We discuss each factor below in further detail.

Figure 2.

A conceptual framework for the components influencing coextinction estimates. Coextinction is primarily influenced by hosts, dependent species, and their interactions. These variables are in turn influenced by several factors. See text for an explanation of factors a through g.

Estimating Extinction Risks of Hosts

Estimating extinction risks of host taxa is crucial for estimating extinction risks of their dependents. Koh et al. (2004a) used data from the IUCN (International Union for Conservation of Nature) Red List (Hilton-Taylor 2000) to estimate the number of threatened species in various taxonomic groups of hosts. Coextinction is not listed as a threatening process by the IUCN (IUCN 2008); therefore, no direct values can be obtained regarding how many species are threatened by coextinction. The IUCN assessments of threat reflect relative extinction risks of mammals and birds well in 70–80% of cases (Keith et al. 2004), but assessments of plant and many other invertebrate hosts are notoriously incomplete, at least relative to mammals, birds, amphibians, and corals, for which there are recent global assessments. Because red lists underestimate the total number of threatened taxa for particular host groups (Cuarón 1993; Smith et al. 1993), coextinction rates derived from red-list data will also be severely underestimated. A more reliable method of determining the regional number of threatened taxa (both host and dependent species) is to use local sources (Rodriguez et al. 2000) such as government and nongovernment organizations, published literature, and experts (field ecologists and taxonomists) in the particular taxa of interest. In addition, such local sources may provide further critical information about threatened hosts and their associated dependents for models. For example, a host-specific species of plant louse (Psyllidae) has been recorded from only one of six populations of its threatened host and is therefore at greater risk of extinction than the host (Taylor & Moir 2009). Combining information from such a wide range of sources lends itself to Bayesian analysis (Clark 2007; McCarthy 2007) of the extinction risk of hosts and their dependents. Undoubtedly, use of more accurate data on the threat status of host species will increase the overall number of dependents regarded as being cothreatened.

Sampling Variation across Hosts

Estimates of coextinction rates have often been made across relatively few hosts. To gain insight into which hosts are more likely to be associated with high coextinction rates, the numbers of specialist dependents need to be examined across a wide range of host taxa. Koh et al. (2004a) initiated this research by compiling host–dependent matrices for fish, bird, primate, and plant hosts. Further detailed examination is now required to determine how coextinction risks vary across functional and taxonomic groups of hosts. Thacker et al. (2006) found the global risk of coextinction for aphids and scale insects on trees to be low. Nevertheless, there could be higher levels of coextinction risk on shrubs or lianas. Ødegaard (2000), for example, found more monophagous beetles on lianas than trees. Furthermore, host species from monospecific genera may have more monophagous dependents than hosts from more speciose genera or families because the dependents are isolated from potential hosts (Ødegaard et al. 2000; Dobson et al. 2008). These comparisons are vital because host extinction is not random and identifying the host taxa with higher proportions of host-specific dependents will help determine conservation priorities.

Correlated Host Extinctions

Extinction risks are not distributed randomly across host taxa (Vamosi & Wilson 2008). Particular groups of species are more prone to extinction than others because of shared phylogeny, life-history characters, habitat dependencies, and geographic distributions (McKinney 1997). Likewise, dependent species are often clustered on related hosts (Ødegaard et al. 2005; Poulin et al. 2006; Mouillot et al. 2008), with host occupancy further constrained by host life history, habitat, and distribution. Therefore, even if dependent taxa are associated with several alternative but related hosts, extinctions of related hosts may cause a cascade of coextinctions of related dependents (Rezende et al. 2007). For example, 62 bird species in New Zealand and 79 birds of the Hawaiian Islands became extinct after the arrival of humans (Duncan & Blackburn 2004; Boyer 2008). The birds most vulnerable to extinction were closely related; all species in the order Dinornithiformes went extinct in New Zealand (Duncan & Blackburn 2004). Bird parasites, especially ectoparasites, such as lice, can often use several bird hosts (e.g., Hughes & Page 2007). However in New Zealand and Hawaii this characteristic would not have saved the dependent species from coextinction, because the cascade of related bird extinctions would have resulted in loss of oligophagous or even polyphagous dependent species. Thus, the rates of coextinction will increase greatly with the loss of related hosts and will not be restricted to the monophagous dependents, a factor overlooked in linear approaches to modeling coextinction.

Correlated extinctions of host species will also be expected when groups of hosts are exposed to the same threatening processes. Examples include predator-related extinctions of mammals within particular weight ranges (Chisholm & Taylor 2007) and fire-driven extinctions of shrubs with certain life histories (Keith 1996). Patterns of host extinction may thus be related to features such as body size (Duncan & Blackburn 2004; Chisholm & Taylor 2007; Boyer 2008), growth form or life history (Keith 1996; Vamosi & Vamosi 2005), geographic distribution (McClean et al. 2005; Thuiller et al. 2005; Boyer 2008), and degree of specialization (Boyer 2008). Although phylogenetic relationships and other correlates of extinction risk are well-studied (e.g., Boyer 2008; Vamosi & Wilson 2008), investigation into the role that such influences have on coextinction rates has just begun (e.g., Rezende et al. 2007). Future work should consider the potential for hosts to be extinguished in a nonrandom way and identify those dependents most prone to coextinction. This may, in turn, reveal particular groups of dependents that are more prone to coextinction, regardless of whether they are monophagous, oligophagous, or polyphagous.

Sampling Variation across Dependents

To determine the coextinction proneness of dependents, a wide range of taxa must be examined because extinction risks may vary within and across a range of dependent taxa. These taxa must include even those dependents that rely on hosts that are themselves dependent on another organism, such as the parasitic nematodes of gall-inducing flies on plants (Taylor & Davies 2008). Furthermore, the distinctions between invertebrate species are commonly cryptic and only identifiable from examination of genitalia or molecular analysis, which typically uncovers more host-specific-dependent species within an assemblage (Poulin & Keeney 2008). Therefore, systematists are essential for accurate estimation of host-specific fauna. Other issues of sampling variation across dependent species are analogous to those described for hosts.

Measuring Host Specificity

Accurate measures of the host specificity of dependents are crucial for determining coextinction rates. Monophagous dependent taxa are likely to be more prone to coextinction than their oligophagous or polyphagous relatives (Thacker et al. 2006). The high specificity of Lycaenidae butterflies to their plant hosts is one reason they are considered one of the most extinction-prone families of butterflies (Koh et al. 2004b). Because host–dependent databases often contain many dependent species with ambiguous feeding preferences, host specificity indices need to incorporate uncertainty when assigning dependent species to hosts.

Although Koh et al. (2004a) substantially improved estimates of coextinction by incorporating variation in specificity of dependent species on their hosts, data sets representing entire assemblages of invertebrates solely from threatened hosts (distinct from, and in addition to, nonthreatened host data sets) are lacking. Although museum collections provide the most readily available host data for construction of host–dependent matrices, these data are typically patchy and biased in coverage and therefore unlikely to support accurate inferences about host specificity. Direct inventories of dependent assemblages on selected hosts offer a more reliable, rapid, and cost-effective basis for inferring host specificity. If sampling is suitably structured, probability distributions can be calculated for the number of hosts associated with each dependent (i.e., degree of polyphagy). Direct inventories are especially needed in regions where the invertebrate fauna is speciose and poorly described (e.g., Australia, Austin et al. 2004; Panama, Ødegaard 2003; New Zealand, Poulin 2004). Because dependent rarity is often correlated with rarity of its host (Hopkins et al. 2002), direct inventories of dependent assemblages on threatened hosts will provide crucial data for estimating how many dependent species are potentially cothreatened (e.g., Thacker et al. 2006). Nevertheless, direct inventories will be limited when the host is already extinct or has too few individuals remaining to support destructive sampling. Because sampling effort will largely determine the proportion of the dependent assemblage uncovered (Poulin 1998), we recommend appropriate sampling designs (e.g., collecting methods, number of individuals) with subsequent tests of inventory completeness (e.g., species accumulation curves).

Data sets from threatened hosts are important because invertebrate assemblages on common hosts may not be characteristic of those on threatened host species. Evolutionary, phenotypic, or behavioral changes that reduce dependence on a single host can allow dependent species to avoid coextinction (Nosil 2002; Poulin et al. 2006). Thus, the assemblage on a threatened host species may contain a higher proportion of oligophagous and polyphagous species as some originally monophagous species adapt to their shrinking food resource or become extinct. For example, native aphids on Antarctic beech (Nothofagus) trees in Australia have wider host ranges in regions where the trees have gone locally extinct (Austin et al. 2004). In such cases, the models of Koh et al. (2004a) would overestimate coextinction.

But how likely is it that dependent species have avoided coextinction through mechanisms such as host switching or, alternatively, have already gone extinct as their host population shrank? Host switching may involve phenotypic plasticity or evolutionary mechanisms and is most likely when the dependent can already feed (albeit perhaps suboptimally) on a closely related host species or if the dependent species is already polyphagous. Host switching will be less likely where there are no closely related hosts within the dependent's geographical range (e.g., on species-poor host genera or families [Vamosi & Wilson 2008]) and with strictly monophagous dependent species. In a historical context, slower extinctions from events such as past climate change may have allowed invertebrates to avoid coextinction through host switching. Nevertheless, because current host extinctions occur more rapidly in response to anthropogenic causes (e.g., introduced species, habitat fragmentation, anthropogenic climate change) (Williams et al. 2007; Brook et al. 2008), the potential for mass coextinction of invertebrates may now be substantially higher.

Biogeographical Implications

Biogeographical factors influence the accuracy of coextinction estimates in several respects, although principally through host specificity. Host specificity of dependents may rely on the biotic context in which the organisms occur and can vary between different ecosystems. The specificity of parasites on fishes, for example, varies between oceans, lakes, and streams (Dobson et al. 2008). Host specificity may also vary between tropical and temperate zones and result in over- or underestimation when dependent and host matrices are extrapolated globally from few localities. Justine (2007) points out that Koh et al. (2004a) may have underestimated the global number of cothreatened monogenean ectoparasites (593 taxa from 746 threatened fish hosts listed by IUCN) because the Canadian database of fish parasites from which their host–dependent matrix was derived failed to account for large numbers of coral fish species which, on average, are each hosts to more than 10 species of monogeneans. This discrepancy highlights how biogeographic variation and knowledge gaps in dependent species biodiversity (McKinney 1999; Poulin 2004; Justine 2007) may lead to biased estimates of coextinction risk (Windsor 1998; Dobson et al. 2008). Therefore, it is vital that global estimates of coextinction be based on host-specificity data stratified across a range of biomes.

Biogeographical issues other than host specificity also need to be assessed to provide more accurate global rates of coextinction. For example, degree of host specificity may be important in determining coextinction proneness of tropical Lycaenidae butterflies (Koh et al. 2004b), but is this the case in temperate zones where other factors such as dispersal potential may be more important (e.g., Hanski et al. 2006)? Greater host ranges often equate to a higher species richness of dependents. For example, due to their larger geographic ranges and population sizes, albatrosses and petrels have higher diversities of louse compared with other seabirds, possibly because the lice have access to greater numbers of sympatric bird species (Hughes & Page 2007). Therefore, host range size may also play a role in determining the number of specialist-dependent species. Furthermore, within the range of a single host species there may be more specialist-dependent species in different geographical localities (i.e., Sobhian & Zwölfer 1985). Alternatively, the host specificity of a single dependent species may vary within its own range, as has been shown for some flea species (Krasnov et al. 2004).

Contrasts of the host specificity of the same dependent groups (e.g., within all herbivorous beetles) across different biomes, ecosystems, habitats, and host species are specifically required to provide more accurate generalizations of global coextinction rates. Higher host specificity values in different habitats or across different host taxa would result in hotspots of coextinction proneness and increase current estimates of coextinction rates.

Order of Extinction of Host and Dependent Species

Current models (linear, probabilistic, and nomographic models) assume dependent taxa are extinguished when, or after, the hosts disappear. Nevertheless, there is strong evidence that dependents may go extinct before their host, including the extinction of insects before their host plants (Biedermann 2000; León-Cortés et al. 2003; Koh et al. 2004c); limpets before their seagrass hosts (Carlton et al. 1991); and microbial pathogens before their vertebrate hosts (de Castro & Bolker 2005). This early extinction of a dependent occurs when the host population becomes too small to sustain a viable population of dependents. The level at which an organism will go extinct due to a change in some required variable (e.g., number of habitat patches) has been termed the extinction threshold in studies of metapopulations (e.g., Benton 2003). Nevertheless, critical sizes of host populations and their dependent species remain unexplored for entire dependent species assemblages because few population viability models have addressed multispecies interactions (Sabo 2008). More empirical data are needed to address this deficiency (Benton 2003), and this may illuminate a need for coextinction models to incorporate declines (as well as extinctions) in host populations.

Incorporating early extinction of dependents into models of coextinction may have profound effects on the shape of curves predicting the rate at which dependents are lost as their hosts are extinguished. It is possible that the shape of the curve will be the inverse of that proposed by Koh et al. (2004a) (curve c on Fig. 1, rather than curve a). This new model would be consistent with predictions that extinction rates of terrestrial invertebrate species are high (e.g., between 7 and 30 species lost globally every week) (Mawdsley & Stork 1995). If Fig. 1c is correct in some situations, invertebrates may be facing higher extinction rates than plants or vertebrates (Thomas et al. 2004; Dunn 2005; Dobson et al. 2008). The concept that coextinction rates are higher for dependents than hosts urgently requires testing with empirical data.

Conclusions

Understanding and predicting global rates of coextinction are formidable but crucial challenges if the rate of extinction is to be slowed. Insufficient taxonomic and basic ecological knowledge of invertebrate species has inhibited documentation of extinctions and necessitated highly speculative estimates by extrapolation, which are very sensitive to the empirical data used and modeling assumptions. Therefore, it is critical that there be a rapid and synergistic development of empirical studies and modeling approaches to obtain more accurate estimates of coextinction. The most urgent challenges ahead are to (1) acquire more accurate host data including the threat status of host species, (2) acquire host–dependent matrices through targeted sampling (e.g., direct inventories) across a wide array of localities, habitats, and breadth of hosts and dependents, (3) develop models that incorporate correlates of nonrandom host and dependent extinctions, such as phylogeny, biogeographical, and species traits that increase extinction proneness, and (4) determine whether dependents are being lost before the hosts and adjust models accordingly.

Meeting the above challenges requires investment of more resources toward basic surveys, taxonomic work on key species, host specificity studies, and modeling of dependent population responses to changes in host populations resulting from anthropogenic actions. Understanding the process of coextinction will allow proactive conservation, rather than the often reactive and piecemeal methods thus far employed.

Acknowledgments

Grants from the Australian Research Council (DP0772057), Australia & Pacific Science Foundation (APSF 07/3), University of Melbourne Botany Foundation, Commonwealth Environment Research Facility (Applied Environmental Decision Analysis hub), and New South Wales National Parks & Wildlife Service supported this work. We thank three anonymous reviewers and the editors for their valuable suggestions that substantially improved this paper.

Ancillary