Hosts and their symbionts are involved in intimate physiological and ecological interactions. The impact of these interactions on the evolution of each partner depends on the time-scale considered. Short-term dynamics – ‘coevolution’ in the narrow sense – has been reviewed elsewhere. We focus here on the long-term evolutionary dynamics of cospeciation and speciation following host shifts. Whether hosts and their symbionts speciate in parallel, by cospeciation, or through host shifts, is a key issue in host–symbiont evolution. In this review, we first outline approaches to compare divergence between pairwise associated groups of species, their advantages and pitfalls. We then consider recent insights into the long-term evolution of host–parasite and host–mutualist associations by critically reviewing the literature. We show that convincing cases of cospeciation are rare (7%) and that cophylogenetic methods overestimate the occurrence of such events. Finally, we examine the relationships between short-term coevolutionary dynamics and long-term patterns of diversification in host–symbiont associations. We review theoretical and experimental studies showing that short-term dynamics can foster parasite specialization, but that these events can occur following host shifts and do not necessarily involve cospeciation. Overall, there is now substantial evidence to suggest that coevolutionary dynamics of hosts and parasites do not favor long-term cospeciation.
Interest in the reciprocal influences between hosts and symbionts has recently increased because of the need to control devastating diseases, to identify or develop biocontrol agents against invasive pests, to improve agricultural production and to decipher the processes of diversification in symbiosis as a widespread lifestyle (Poulin & Morand, 2004). Host–symbiont interactions occur over short time-scales, from a single disease cycle in the case of opportunistic and transient infections by parasites, to very long time-scales persisting over multiple host speciation events. Short time-scales have been associated with reciprocal selection pressure between host and parasite, leading to changes in allele frequencies over successive generations (i.e. ‘coevolution’ in the narrow sense, Clayton & Moore, 1997; see Box 1 for glossary of terms used in this review). By contrast, long time-scales may encompass several speciation events. The concomitant occurrence of speciation in hosts and their symbionts is referred to as ‘cospeciation’ (Page, 2003). However, the speciation of symbionts may occur independently of host speciation, often through host shifts as the symbiont comes to occupy a new host environment in isolation from the ancestral lineage. ‘Coevolution’ is used by some authors to describe long-term dynamics as a synonym for cospeciation but this usage may be misleading, as pointed out by some authors (Smith et al., 2008a). We will therefore use ‘coevolution’ in the narrow sense here: reciprocal selection pressure and resulting micro-evolutionary changes.
The often obligate and specialized interactions between hosts and symbionts suggest that any bifurcation of the host lineage is likely to result in the simultaneous isolation of its associated symbionts (Fig. 1a). Thus speciation in one lineage is then pegged to speciation in the other, and this process is referred to as cospeciation. Alternatively, new host–symbiont combinations may arise owing to movement and specialization of the symbiont to a new host, on which the symbiont's immediate ancestor did not occur. Symbiont speciation subsequent to such movement is often referred to as ‘host-shift speciation’ (Fig. 1b, Agosta et al., 2010; Giraud et al., 2010).
In this review, we aim to: outline the origin of the concept of cospeciation; provide a description of the various methods developed for determining whether cospeciation has actually occurred, together with their advantages and pitfalls; critically review recent inferences on the history of host–symbiont associations based on these methods; and examine the relationship between coevolution in its narrowest sense and symbiont speciation. We caution against the use of ‘coevolution’ as a synonym for cospeciation because of the implication that short-term dynamics contributes directly to cospeciation in the long term, although the rationale underlying this idea and its potential implications have never been fully articulated. Indeed, recent studies comparing host and parasite phylogenies and theoretical developments relating to parasite specialization and speciation seem to argue against cospeciation being the predominant mode of host and symbiont diversification, despite the occurrence of reciprocal selection over short time-scales.
II. Origin of the cospeciation concept
The idea of cospeciation was put forward in pioneering studies on avian parasites, such as those of Kellogg (1913) and Fahrenholz (1913), at the beginning of the 20th century. These authors noted that closely related avian parasites, with similar phenotypic features, were associated with closely related host species. They proposed the following hypothesis, known today as the Fahrenholz rule: ‘parasite phylogeny mirrors that of its host’ (1913). A similar principle was proposed by Szidat some years later (1940): ‘primitive hosts harbor primitive parasites’. The idea was that similarity between the parasites of related hosts results from cospeciation (i.e. concurrent and interdependent bifurcation of host and parasite lineages), leading, in turn, to congruent host and parasite phylogenies.
The first studies referring to the Fahrenholz rule did not actually test cospeciation as a hypothesis. Without DNA sequencing being possible at the time it was therefore very important to obtain other forms of phylogenetic information. The narrow host distribution of many animal parasites led researchers to use parasites as characters for inferring phylogenetic relationships between host taxa (Hoberg et al., 1997). Similar hypotheses were proposed for plant parasites (Savile, 1979). Conversely, host taxa were often used as taxonomic criteria for the classification of parasites (see for example Downey, 1962). In both cases, the phylogeny of one partner was used to build the phylogeny of the other, so the two phylogenies tended to be congruent. As congruence between host and parasite phylogenies was the most widely accepted criterion for inferring cospeciation, this led to the widespread belief that cospeciation was common.
However, this reasoning is clearly circular and the evidence put forward for cospeciation in host–parasite associations was for many years inadequate. It was not until the late 1980s that robust phylogenies, built independently for hosts and parasites, were used to test for cospeciation in a more specific manner (Hafner & Nadler, 1988).
III. Theoretical framework and methods for testing for cospeciation
Macro-evolutionary aspects of host–parasite associations cannot be observed within the lifespan of a researcher. Methods for inferring the effects of interactions have thus been developed based on comparisons of the phylogenies of the interacting species. These methods, which are described as ‘cophylogenetic methods’, are based on the idea that two interacting lineages will have completely congruent phylogenies if they have diversified exclusively by cospeciation (Fig. 1a). However, it is important to note that congruent topologies can also be obtained after host shifts to closely related hosts under certain realistic conditions of time-span between host-switch and subsequent speciation (Fig. 1b, see de Vienne et al., 2007b for details). Events that reduce the congruence between host and symbiont phylogenies include: (1) host-shift speciation (Fig. 1c), when a population of the symbiont species adapts to a new host followed by speciation (under certain conditions, see de Vienne et al., 2007b for details); (2) speciation of the symbiont without speciation of the host or host switching, also known as intrahost speciation or duplication; and (3) symbiont extinction (Fig. 1d).
Cophylogenetic methods can be divided into two main classes (Table 1). The first class encompasses methods aiming to reconstruct the evolutionary history of the host and parasite lineages, to infer the nature and frequency of different evolutionary scenarios by comparing phylogenetic trees (event-based methods). Diversification by cospeciation is generally inferred if the number of cospeciation events is significantly greater than the number of cospeciation events inferred when randomizing the associations, although this merely indicates topological congruence and not necessarily cospeciation. Significant congruence can indeed be obtained after repeated host shifts, as noted above (Fig. 1b). The second class of methods tests the overall congruence between the host and parasite phylogenies (i.e. topology or distance-based methods using the similarity and/or symmetry in the time of divergence between hosts and parasites) and it is generally considered that high levels of congruence provide evidence of frequent cospeciation – although this conclusion may be similarly unwarranted (Fig. 1b). We will explain these two approaches in more detail in the following text and provide a brief overview of the existing cophylogenetic tools (summarized in Table 1). Finally, we will discuss some of the limitations of these methods in the light of recent results on the likelihood of host and parasite trees congruence in the absence of cospeciation.
|Event-based methods Basic concept: consider cospeciation as the most parsimonious explanation for congruence between host and parasite trees|
|Method||Main feature||Software/method||Estimation of the best reconstruction||Advantages||Disadvantages||References||Availabilityb|
|Brooks Parsimony analysis||Considers parasites as character states of the hosts||BPA||Minimum number of character state changes in the host phylogeny (parsimony)||Can handle more than just 1 : 1 correspondence between hosts and parasite tips||Multiple equally parsimonious reconstructions for large phylogenies and/or for multiple associations between host and parasiteCospeciation considered the most parsimonious hypothesis||Brooks (1981); Brooks & McLennan (1991)||To be implemented by the user. Refer to Brooks et al. (2001) for details|
|Reconciliation analysis||Mapping of the parasite phylogeny onto the host phylogeny. The best scenario may be that with the minimum number of events inferred or the least costly||Component||Minimization of the number of extinctions and intrahost speciations and maximization of the number of cospeciations||No host shifts considered Cospeciation considered the most parsimonious hypothesisNeeds 1 : 1 correspondence between hosts and parasites||Page (1993)||http://taxonomy.zoology.gla.ac.uk/rod/cpw.html|
|treemap 1a||Minimization of the number of host shifts and maximization of the number of cospeciations||Host shifts are taken into account Gives a graphical representation of the history of the host-parasite associationIncludes a test to assess whether the number of cospeciation events is higher than for random phylogenies (thus also listed with topology-based methods)||Cospeciation considered the most parsimonious hypothesis The number of parasites infecting ancestral host species can be unreasonably high Can give a very large number of reconstructions Does not guarantee that reconstructions involving more than one host shift are realistic (i.e. there may be timing incompatibilities) Needs one-to-one correspondence between host and parasite tips||Page (1994)||http://taxonomy.zoology.gla.ac.uk/rod/treemap.html|
|treemap 2a||Minimization of the total cost of the reconstruction, a cost associated with each event||Cost is associated with each eventImplementation of the ‘jungles’ method (Charleston, 1998), an algorithm allowing the rapid identification of the optimal reconstructions taking costs into account and ensuring the feasibility of each reconstruction|| |
Cospeciation considered the most parsimonious hypothesis
Very slow for large trees
|tarzan||Possibility of defining the timing of nodes in the parasite phylogenyVery fast||Does not guarantee that the solution is optimal Cannot always find a solution even when there is one Cospeciation considered the most parsimonious hypothesis||Merkle & Middendorf (2005)||http://pacosy.informatik.uni-leipzig.de/146-0-Download.html|
Possibility of defining the timing of nodes in both the parasite and host phylogenies
Possibility of defining different host-switch costs independentlyInteractive graphical interfaceFaster than treemap 2
Possibility of defining the maximum permitted host-switch distance
|Slower than tarzan Cospeciation considered the most parsimonious hypothesis||Conow et al. (2010)||http://www.cs.hmc.edu/~hadas/jane/Jane1/index.html|
|Cost-based methods||Cost associated with each event, no graphical representation||treefitter||Minimization of the total cost of the reconstruction, a cost being associated with each event||Probability associated with each type of eventCosts of each event are set by the user||Cospeciation considered the most parsimonious hypothesisCospeciations cannot be more costly than host-shift speciations Possible timing incompatibilities leading to potentially erroneous conclusions||Ronquist (1995)||http://sourceforge.net/projects/treefitter/|
|Bayesian methods||Combination of two models, one estimating the probability of a given evolutionary scenario and one used to infer host and parasite phylogenies||Determines the most likely evolutionary scenario leading to the observed host and parasite DNA sequences, not their phylogenies||Does not consider the phylogenies of the host and the parasites to be known|| |
Cospeciation considered the most parsimonious hypothesis
Only considers host shift and cospeciation
Works only for a 1 : 1 correspondence between host and parasite tips
|Huelsenbeck et al. (2000, 2003)||Theoretically, upon request to author. But seems unavailable|
|Topology and distance-based methods Basic concept: does not consider any event. These are simple tests of independence or similarity between trees or alignments|
|Method||Main features||Software/method||Input data||Advantages||Disadvantages||References||Availabilityb|
|Test of independence||Looks at the probability of observing a certain level of congruence between two trees with respect to expectations if the trees were independent||Icong index||Trees. No branch lengths||No random trees need to be generated for testing for higher levels of congruence than expected by chance||Works only for a 1 : 1 correspondence between host and parasite tipsConsiders trees to be correct||de Vienne et al. (2007a)||http://max2.ese.u-psud.fr/bases/upresa/pages/devienne/|
|Methods based on Mantel test between-distance matrices||Sequence alignments (converted into distance matrices)||Does not account for phylogenetic nonindependence (Felsenstein, 1985)||Hafner et al. (1994)||To be implemented by the user. Refer to Hafner et al. (1994) for details|
|parafit||Trees or alignments (converted into distance matrices)||Not restricted to 1:1 correspondence between host and parasitesAllows testing of the contribution of each individual host-parasite link to the total congruence statistic (taking into account both topological congruence and branch lengths)||Does not account for phylogenetic nonindependence (Felsenstein, 1985)Considers trees to be correct (if trees used)||Legendre et al. (2002)|
|Method based on Pearson's correlation analysis between host distances and parasite distances||Trees or alignments (converted into distance matrices)||Not restricted to 1:1 correspondence between host and parasitesApparently more accurate estimation of the contribution of each individual host–parasite link to the total congruence than parafit||Considers trees to be correct (if trees used)||Hommola et al. (2009)||http://www1.maths.leeds.ac.uk/~kerstin/ and Hommola et al. (2009)|
|MRCAlink algorithm||Trees||Applicable to methods like parafit: making it possible to take phylogenetic nonindependence into account||Considers trees to be correct||Schardl et al. (2008)||http://cophylogeny.net/research.php|
|treemap 1||Trees||Considers trees to be correct||Page (1994)||http://taxonomy.zoology.gla.ac.uk/rod/treemap.html|
|treemaptreemap 2||Trees||Based on the ‘jungles’ method.Several randomization test statistics available||Considers trees to be correct||Charleston (1998)||http://sydney.edu.au/engineering/it/~mcharles/software/treemap/treemap.html|
|Test of similarity or identity||Estimates the probability of observing the actual host and parasite DNA sequence variation assuming their phylogenies are congruent||Maximum likelihood method||Sequence alignments||Does not consider the trees to be known||Only topologies are considered, not branch lengths||Huelsenbeck et al. (1997)||Theoretically, upon request to author. But seems unavailable|
|Bayesian method||Sequence alignments||Does not consider the trees to be known||Only topologies are considered, not branch lengths||Huelsenbeck et al. (1997)||Theoretically, upon request to author. But seems unavailable|
|Second maximum Likelihood method||Sequence alignments||Does not consider the trees to be knownTests for temporal congruence, the null hypothesis being that the speciations occurred at the same time||Huelsenbeck et al. (1997, 2003)||Theoretically, upon request to author. But seems unavailable|
1. Event-based methods
The first event-based method developed was Brooks' Parsimony Analysis (BPA; Brooks, 1981). It opened the way for such methods but considered parasites as character states of the hosts. The parasitic character states are assigned to each branch in the host phylogeny and the most parsimonious reconstruction, the one with smallest number of parasite presence vs absence state changes in the host phylogeny, is retained. If host and parasite phylogenies are topologically congruent, then each internal branch in the host phylogeny is assigned one ‘parasite’ state so that no ‘state’ transition is required and cospeciation is inferred along the whole phylogeny. Although BPA was widely used in the 1980s and early 1990s, it received heavy criticism, particularly because of its requirement for a large number of a posteriori interpretations (Page, 1994).
Another method, ‘reconciliation analysis’, proposed by Page (1990), considers parasites as evolutionary lineages rather than character states. Implemented in the component program (Page, 1993), it estimates the minimum number of extinctions and intrahost speciations required to reconcile the separate host and parasite phylogenies. Cospeciation is explicitly considered as the most parsimonious hypothesis. Page (1994) subsequently added host-shift speciation in the treemap 1 program. This method tries to reconcile host and parasite phylogenies by maximizing the number of cospeciations and minimizing the number of host-shift speciations. There are no constraints on the numbers of intrahost speciations or extinctions or on numbers of parasites present on internal nodes, so the number of parasites infecting ancestral host species or number of intrahost speciations may be assumed to be unreasonably high (Refrégier et al., 2008). A graphical representation of the history of the host–parasite association is provided, although this representation is most often unlikely to be correct as exact costs for the events are impossible to assess a priori. treemap 1 also determines whether the number of cospeciation events in the host and parasite trees compared is greater than that in random phylogenies. This is the most useful part of the program, but it is often taken as a test for cospeciation, while in fact it is a test of topological congruence. Indeed, 100% of inferred events will be cospeciations in cases of complete congruence, while this can result from host-shift speciation (Fig. 1b, de Vienne et al., 2007b). Overall, reconciliation analyses overestimate cospeciation events because (1) they assume, a priori, that cospeciation is more likely than host-shift speciation or other events – this assumption likely being unfounded (Ronquist, 1995) – and (2) they interpret congruence as evidence for cospeciation, while this is not necessarily the case (Fig. 1b).
The most recent version, treemap 2, more rapidly identifies optimal phylogenetic reconstructions and takes into account the temporal feasibility of each reconstruction (host shifts only occurring between hosts present at the same time; for details on the method and its implementation in treemap 2, see Charleston, 1998; Charleston & Perkins, 2003). Similar methods with faster computation have since been developed. tarzan (Merkle & Middendorf, 2005) handles phylogenies by allowing uncertainty on the age of parasite nodes (associating each node to a time zone) and selecting the cost of each event. jane (Conow et al., 2010) takes into account uncertainty in time for the host phylogeny without substantially increasing the computation time.
The first series of methods allowing the user to attribute a cost to each evolutionary event (cospeciation, host-shift speciation, intrahost speciation and extinction) was developed by Ronquist (1995). These ‘cost-based’ methods find the most parsimonious scenario by minimizing the total cost. The most popular cost-based method is that implemented in treefitter software (Ronquist, 1995). treefitter estimates the number of events of each type that could explain the observed congruence between the two phylogenies. It then associates each event with the probability that it arose by chance, calculated by permutations of the host and/or parasite leaves on the phylogeny. treefitter finds the optimal numbers of each type of event by minimizing the total cost of the reconstruction, but it does not allow cospeciations to be more costly than host-shift speciation.
All the methods presented consider the host and parasite phylogenies to be known and fully resolved trees, and therefore they are sensitive to the selection of different optimal trees. The Bayesian method developed by Huelsenbeck et al. (2000, 2003) overcomes this problem. This method aims to determine the most likely evolutionary scenario leading to the observed host and parasite DNA sequences, rather than their phylogenies. It is based on two simple stochastic models: one for host-shift speciations and the other for DNA substitutions. The two models are mixed and subjected to Bayesian analysis.
2. Topology- and distance-based methods
All the methods presented earlier and summarized at the top of Table 1 are based on the idea that host and parasite phylogenies should be identical (congruent) in the absence of host-shift speciation, extinction and intrahost speciation. This is a logical conclusion of the principles first formulated by Fahrenholz (1913) and Szidat (1940) (see Section II, 'Origin of the cospeciation concept'). Yet, host shifts can also lead to phylogenetic congruence under realistic conditions (Fig. 1b, see de Vienne et al., 2007b for details). Another set of methods is based on statistical tests for congruence between host and parasite phylogenies. These methods do not directly consider high levels of congruence to constitute proof of cospeciation. Instead, they compare the probability of observing a certain level of congruence between two trees, with expectations based on the independence between trees. By linking the results obtained with such methods to the common history of the interacting lineages it is possible to obtain an a posteriori interpretation that is not integral to the test. This approach may thus be considered less biased than event-based or event- and cost-based methods, such as those already presented.
These methods can be assigned to different classes according to the null hypothesis tested (similarity or independence, Huelsenbeck et al., 2003) and the data used for the test (trees, distance matrices or raw sequence alignments; Light & Hafner, 2008). Tests of independence are based on comparisons of the topological or genetic distances of the focal host–parasite association with a distribution of distances computed from a large number of randomly generated trees. If the distance of interest is significantly smaller than expected by chance, the association is considered to be significantly congruent. This principle is similar to that underlying the test implemented in treemap 1.
One of the weaknesses of these methods lies in the large number of random trees that must be generated de novo for each new comparison of trees. A test of tree independence has been proposed to overcome this problem, being based on the use of previously simulated associations (de Vienne et al., 2007a, 2009a; Kupczok & von Haeseler, 2009).
Tests of independence have also been used to evaluate temporal congruence in the speciation histories of hosts and parasites. Repeated cospeciation events imply the simultaneous occurrence of speciation events (i.e. temporal congruence) and thus proportional branch length and identical dates for the nodes in the phylogenies compared (Fig. 1a). One method (Hafner et al., 1994) tests whether the two species have accumulated similar numbers of genetic differences. Input data include host–parasite species associations and the alignment of one specific locus (or several concatenated loci) for hosts and parasites. These alignments are used to calculate distance matrices. The significance of the correlation between the two matrices is then assessed using a Mantel test (Hafner et al., 1994). A second method compares matrices of branch lengths from host and parasite trees in the same way (Hafner et al., 1994; Page, 1996). If molecular clocks are available for both host and parasite it is possible to compare the estimated absolute ages of the nodes in the two trees. The determination of identical ages for each node is actually the only way to establish cospeciation with confidence. Indeed, identical relative divergence times, as deduced from proportional branch lengths, may exist in some host–parasite associations in which speciation times are not identical. This can be the case when parasites jump preferentially onto closely related hosts and take a time to speciate that is proportional to the phylogenetic distance between initial and novel hosts (Charleston & Robertson, 2002). Furthermore, while Mantel tests account for statistical nonindependence in matrices, they do not account for phylogenetic nonindependence (Felsenstein, 1985; illustrated in Fig. 2), in that the data for divergence at ancient nodes include the same information as those for divergence at more recent nodes along the same branches (Felsenstein, 1985; Schardl et al., 2008). All the points used in the distance matrices are thus phylogenetically nonindependent, which should preclude the use of a Mantel test.
parafit (Legendre et al., 2002) tests the independence of host and symbiont genetic or patristic distances (patristic distances are calculated by summing the lengths of the branches in the estimated tree, joining each pair of taxa). This method is advantageous because it can (1) deal with cases in which multiple symbionts are associated with a single host, or where multiple hosts are associated with a single symbiont, and (2) be used to assess the contribution of each individual host–symbiont link to the total congruence statistics. The host sequences and/or tree and the symbiont sequences and/or tree are transformed into distance matrices. A sum of the squared distances gives a value for the overall similarity between trees (ParafitGlobal), which is compared with a distribution of ParafitGlobal values obtained by permutations to assess significance. The contribution of each individual link to the overall congruence between trees is assessed by removing the links one by one. However, the problem of nonindependence of phylogenies (Fig. 2, Felsenstein, 1985) also applies to this method.
Hommola et al. (2009) recently introduced a new permutation method for evaluating the independence of host and parasite phylogenies. This test is based on the calculation of Pearson's correlation coefficients between host distances and parasite distances, considering all pairs of interacting hosts and parasites. This correlation coefficient is then compared with that obtained after random permutations of the data, retaining the observed interaction links. This method is thus a generalization of the Mantel test, making it possible to test data in the absence of a one-to-one correspondence between hosts and parasites. This method seems to be more powerful than parafit, with more accurate estimated P-values, although this superior performance may be attributable to the larger number of permutations performed (100 000, vs only 99 for parafit).
Finally, Schardl et al. (2008) proposed a modification for programs such as parafit, taking into account the nonindependence of pairs of species from the same branch and using a method similar to the phylogenetic independent contrasts (PIC) method proposed by Felsenstein (1985). The algorithm, MRCAlink (MRCA for Most Recent Common Ancestor), identifies phylogenetically independent pairs between host and parasite trees and the reduced host and parasite matrices can then be compared.
3. Pitfalls in the theoretical framework when considering host–parasite associations
All the methods presented above and summarized in Table 1 have drawbacks (Nieberding et al., 2010). These problems include testing for congruence on the basis of estimated phylogenies without taking into account uncertainty in the inference (treemap, treefitter or Icong, which require fully resolved trees), phylogenetic nonindependence (treemap, treefitter), tests considering only topologies and thus ignoring branch lengths (Icong, Huelsenbeck's methods) or underestimation of the potentially high probability of host-shift speciations (treemap). A key issue that is rarely discussed (but see Hafner & Nadler, 1988; Hafner et al., 1994) is the common but potentially erroneous interpretation of these tests, specifically that congruence between host and parasite phylogenies results from frequent cospeciations between host and parasite phylogenies, whereas incongruence results from host-shift speciation, extinction, intrahost speciation and other evolutionary scenarios.
A good illustration of the limitations of reconciliation methods was provided by Lanterbecq et al. (2010), who reviewed studies based on the use of treemap to reconstruct the history of host–parasite associations. Most of the examples in their Table 5 (Lanterbecq et al. 2010) refer to studies in which host shifts were eventually identified because of asynchronous splitting events as the main mode of parasite speciation, whereas the number of host shifts suggested by treemap was smaller than numbers of cospeciation events. This was the case, for example, for legume-feeding insects and plants of the Genistae (Percy et al., 2004), for which 16 cospeciations and no host shifts were inferred and for algal and fungal mutualists (lichens, Piercey-Normore & DePriest, 2001), for which 10–11 cospeciations and 3–5 host shifts were inferred. This example also illustrates one of the greatest pitfalls of event-based methods (Fig. 3); the cospeciation events could only be inferred while assuming unreasonably large numbers of intrahost speciations and sorting events (29 intrahost speciations and 220 sorting events for the plant-insect interaction and 7–9 intrahost speciations and 65–81 sorting events for lichens). Similarly unlikely inferences were also made in a cophylogenetic study between neobatrachian frogs and their parasitic platyhelminthes, for which 22 cospeciations were estimated for 26 species pairs, but with 10 intrahost speciations and 16 extinction events (Badets et al., 2011). The parafit test was not significant and the tree node ages appeared to be inconsistent with cospeciations. The large number of cospeciation events inferred was thus clearly misleading. The default cost values for cospeciation, host-shift, intrahost speciation and sorting events in reconciliation methods thus bear little resemblance to the actual probabilities of these events (see Section IV 'Pitfalls in the theoretical framework when considering host–parasite associations'). For example, if parasite extinction occurs in a host lineage and this host lineage is then recolonized through host-shift speciation, reconstructions by event-based methods tend to suggest the occurrence of intrahost speciations in the distant past, followed by many extinction events (Fig. 1d). This tendency to avoid inferring host shift makes it necessary to include many more evolutionary steps to reconcile the two phylogenies than reconstructions involving a host shift.
Experimental and theoretical studies have shown that congruence between host and parasite phylogenies can be achieved in the absence of cospeciation if there is a preferential host shift towards closely related hosts (Charleston & Robertson, 2002) and under certain conditions of time lag between the switch and the following speciation (Charleston & Robertson, 2002; de Vienne et al., 2007b; Fig. 1b). Preferential host shifts towards related hosts have been found using experimental cross-inoculations in many host–parasite associations (Gilbert & Webb, 2007; de Vienne et al., 2009b) and the possibility of topological congruence without cospeciation highlights the importance of testing temporal congruence between host and parasite phylogenies, as only such tests can validate the occurrence of cospeciation events (Charleston & Robertson, 2002; Hirose et al., 2005; Mikheyev et al., 2010).
Another pitfall of cophylogenetic studies is the failure to delimit species correctly as this may lead several methods to artificially inflate congruence when generalist species are found on closely related hosts (Refrégier et al., 2008). Indeed, species delimitation in parasites is often difficult and generalist symbionts often infect closely related hosts; congruent intraspecific nodes then artificially increase the number of cospeciations inferred (Fig. 4). Multiple individuals per parasite species are often included in analyses, particularly when these species are generalists (Light & Hafner, 2007; Bruyndonckx et al., 2009), which can cause the same bias towards congruence.
A last issue in cophylogenetic studies is the frequent use of mtDNA phylogenies. It is increasingly recognized that a single marker cannot reliably be used to reconstruct species phylogenies, and this is particularly true for mtDNA, which is more prone to introgression than nuclear DNA (Coyne & Orr, 2004) and can be subject to strong selective pressures and low recombination rates (Balloux, 2010).
We present the approach we recommend to test for cospeciation in Fig. 3, to avoid as much as possible the different pitfalls discussed.
IV. Studies of natural associations reveal the prevalence of host shifts
The methods described earlier have been used in diverse host–parasite associations to test cospeciation hypotheses. After > 50 yr of research, convincing examples of cospeciation between host and symbiont seem to be the exception rather than the rule. We have performed an extensive search in ISI Web of Knowledge, and we summarize in Table 2 and Fig. 5 the studies reporting cophylogeny analyses. We include the system and its type of symbiosis, the conclusion inferred by authors, the type of phylogenetic data, the results of cophylogenetic analyses, the results of the test for temporal congruence (when available) and our own conclusions. Convincing cospeciation between host and symbiont trees is seldom found except for a few mutualist associations, most often involving vertically transmitted symbionts. Host-shift speciation has been recognized for some time as the main mode of speciation in many systems involving plant viruses, plant fungi, plant parasitoids and animal viruses (Table 2, Fig. 5). Host shifts are also frequent in phytophagous insects (for an extensive review see Nyman, 2010). In addition, we show here that even in associations where cospeciation has been claimed to occur together with other events, host shifts may be the only convincingly demonstrated mode of speciation. Indeed, in all these cases where absolute dates could be obtained, they indicated more recent speciation by symbionts, even when cophylogenetic analyses suggested cospeciation as the major mode of diversification. Furthermore, the number of duplications inferred is more often unrealistically high, casting doubt on the conclusion of cospeciation (Table 2, Fig. 5). Indeed, when host-shifts are considered costly, they will be replaced in most reconstructions by duplications and extinctions (Fig. 1d).
|Typea||System||Inferred conclusion by authors: cospeciation vs host shifts||Type of symbiont||Number of taxa||Methods for testing codivergence||% cospeciation events inferred||Markers for phylogenies||Congruence in time of divergence||References|
|1a||Devescovinid flagellates (Devescovina spp.) and Bacteroidales ectosymbionts||Codivergence||Mutualistic termite gut flagellates and their bacterial symbionts||7 pairs of Devescovina flagellates and Bacteroidales ectosymbionts||treemaptreemap||100% codivergence||SSU rRNA for flagellates. For bacteria : 16S||Very good correlation of the host and symbiont coalescent times (r2 = 0.98), but no absolute calibration||Desai et al. (2010)|
|1a||Trichonympha termite gut flagellates and Candidatus bacteria||Codivergence||Mutualistic termite gut flagellates and their bacterial endosymbionts||Flagellate and bacteria from 11 termite species||treemaptreemap||7/11 cospeciation events||For both: SSU rRNA genes||Not tested||Ikeda-Ohtsubo & Brune (2009)|
|1a||Brachycaudus aphids and Buchnera aphidicola bacteria||Codivergence||Vertically transmitted mutualistic bacteria of aphids||56 specimens of the host Brachycaudus, representing 27 species||treemaptreemap and parafit||34 cospeciation events, 1 host shift; parafit, also indicated significant codivergence||For the bacteria: TrpB and two intergenic regions For the host: CytB, COI and ITS2||Strong correlation between the divergences in the two lineages, (R = 0.9455), the y-intercept was not significantly different from 0||Jousselin et al. (2009)|
|1a||Leafhoppers (Cicadellinae) and their two main symbionts: Sulcia (Bacteroidetes) and Baumannia (Proteobacteria)||Codivergence||Leafhoppers and two endosymbiont species providing nutrients||29 leafhoppers species and their symbionts||Parsimony-based ILD test, Shimodaira–Hasegawa test and treefitter||The results of all tests suggest that the diversification of both endosymbionts was largely or entirely dependent on the phylogenetic history of their host leafhoppers||Host: COI, COII, 16S rDNA and H3. For the symbionts: 16S rDNA||Likelihood-ratio test to assess whether the 16S rDNA of Baumannia and Sulcia were evolving with a constant rate across different host- associated lineages||Takiya et al. (2006)|
|1a||Plataspidae Stinkbugs and γ-Proteobacteria||Strict cospeciation||Stinkbugs of the family Plataspidae, and their highly specific mutualistic gut endocellular γ-Proteobacteria. Bacteria vertically transmitted||Three genera, seven species, and 12 populations of stinkbugs and their bacteria||treemaptreemap||Strict congruence (6 codivergence events)||For the host: mitochondrial 16S rRNA gene For the bacteria: 16S rRNA gene||Not tested||Hosokawa et al. (2006)|
|1a||Cockroaches (Polyphagidae, Cryptocercidae and Blattidae) and their Blattabacterium bacteria||Cospeciation||Blattabacterium vertically transmitted intracellular mutualists (that presumably participate in the recycling of uric acid) that are located in specialized cells of cockroaches||Four cockroach species and their Blattabacterium bacteria||Component Lite, Templeton test and Shimodaira and Hasegawa test||Host and symbiont topologies were found to be highly similar, and tests indicated that they were not statistically different||For the bacteria: 16S rDNA. For the host: 18S rDNA and mitochondrial COII, 12S rDNA, and 16S rDNA combined with morphological data already published||Congruence of divergence times||Lo (2003)|
|1a||Deep sea clams (Vesicomya, Calyptogena, and Ectenagena) and bacteria||Cospeciation||Vesicomyid clams depends entirely on their sulfur-oxidizing endosymbiotic bacteria||16 clam species and their associated bacteria||Kishino–Hasegawa criteria||The topologies are not significantly different||Bacteria: 16S rDNA; Clams: 16S and mtDNA COI||Congruent dates based on fossils||Peek et al. (1998)|
|1b||Crematogaster ants and Macaranga plants||Cospeciation||Highly species specific mutualistic interaction between Crematogaster ants and Macaranga plants, but two ant species have multiple hosts||Nine Macaranga plant species and four species of Crematogaster ants||Tree Mapping in Component||The congruence of the two phylogenies is statistically significant although there is a major disagreement||For the plant: phylogeny already published based on morphology and the nuclear ITS. For the ants: COI||Tertiary climate and the restriction of Macaranga to seasonal forests suggest that this plant clade diversified in the late Tertiary, which corresponds to the diversification period of the ants||Itino et al. (2001)|
|1b||Camponotus Ants and their bacteria (Candidatus Blochmannia)||Cospeciation||Mutualism between ants and their bacterial associates, that are located within bacteriocytes and are transmitted vertically although some horizontal transmission has been suggested||16 host species and their bacteria||Shimodaira–Hasegawa test||No conflict on well-resolved nodes||For the bacteria: 16S ribosomal DNA [rDNA], groEL, gidA, and rpsB. For the host: the nuclear EF-1αF2 and mitochondrial COI and COII||Correlated rates of synonymous substitution (dS) in the two phylogenies||Degnan et al. (2004)|
|2||Tephritinae fruit flies and bacteria (Candidatus spp.)||62.5% of nodes with codivergence inferred||Mutualistic relationships between fruit flies and their extracellular bacterial symbionts (some vertically transmitted)||33 Tephritinae flies species in 17 different genera||treemap, parafit and Shimodaira–Hasegawa likelihood-based test||A maximum of 20 codivergence events (= 10 cospeciations), from 6 to 17 losses, 1 to 6 switches and 12 to 14 duplication events||For the host: 16S rDNA and COI-tRNALeu-COII; for the symbiont: 16S rDNA||Not tested||Mazzon et al. (2010)|
|2||Makialgine mites (Acari, Psoroptidae, Makialginae) and Galagalges primates||Mainly cospeciations and duplications||Permanent and highly specialized ectoparasite mites||For the parasite : 9 taxa||treefitter and treemap||4/5 cospeciation, but at least as many duplication events as cospeciation events||Morphological traits||Not tested||Bochkov et al. (2011)|
|2||Crinoids (Echinodermata) and myzostomids (Myzostomida, Annelida)||Mainly cospeciation and losses||Obligate and highly specific commensal marine worms||16 species of crinoids (belonging to 6 different families) and their 16 associated myzostomids (belonging to 15 species)||treemap, parafit and KH and SH tests||8 or 9 cospeciations, but 7–10 losses and 3–4 host shifts||For crinoids: 18S rDNA, and COI; for myzostomid: 18S rDNA, 16S rDNA, and COI||Not tested||Lanterbecq et al. (2010)|
|2||Rodents (Muridae: Sigmodontinae) and their hoplopleurid sucking lice (Phthiraptera: Anoplura)||Cospeciation but with prevalent host switching||Generalist parasitic sucking lice of rodents||15 distinct louse species and 19 rodent species||treemap and treefitter||treemap: 12–20 codivergences, 10–14 duplications, 12–15 extinctions, 3–4 host switchings. treefitter: 6–9 codivergences, 0 duplications, 0–3 extinctions, 6–10 host switchings||For the parasite: CO I and EF1a||Not tested||Smith et al. (2008b)|
|2||Fig trees (Moraceae, Ficus) and fig wasps||Significant cospeciation but with host shifts and duplications||Pollinating and nonpollinating fig wasps and Ficus||23 fig species||treemap and parafit||Pollinators: no significant cospeciation in the tree with all species, but significant cospeciation in the combined tree with fewer species. Non pollinators: significant cospeciation, but with almost as many duplications needed as cospeciation events||Figs: two nuclear DNA fragments (ETS and ITS). Wasps: 28S and ITS2||Significant correlation of MRCA, with intercept at 0 but slope < 1||Jousselin et al. (2008)|
|2||Geomydoecus lice on Cratogeomys pocket gophers||Codivergence||Chewing parasite lice and their pocket gopher hosts||For the parasite: 41 specimens of chewing lice from seven species. Gophers: 16 individuals from 3 species||treemap, parafit, KH and SH tests,||treemap: significant cophylogeny between host and parasites, 16 codivergence events, 6–8 duplications, 3–4 extinctions, 3–4 host switches||Louse: COI and EF-1a For the host, COI||Regression analyses of estimated branch lengths in gophers and lice showed intercepts that were not significantly different from zero||Light & Hafner (2007)|
|2||Figs (Ficus spp., Moraceae) and wasps (Hymenoptera, Agaonidae, Chalcidoidea)||‘Diffuse coevolution’||Host specific mutualistic pollinator and nonpollinator wasps of figs||411 individuals from 69 pollinating and nonpollinating fig wasp species, 17 species of Urostigma figs||treemap and parafit||Significant congruence. Host-switching and multiple wasp species per host are however ubiquitous; 1–6 cospeciations, 1–10 duplications, 4–68 sorting events, 0–1 host switch||Wasp phylogeny based on COI||Not tested||Marussich & Machado (2007)|
|2||Pelecaniform birds and Pectinopygus lice||Significant congruence but with host shifts||Host-specific parasitic lice that infect a single order of birds (Pelecaniform)||17 Pectinopygus species and their pelecaniform host||treefitter, treemap, ILD, and parafit||Significant overall congruence. However, without invoking any host switching, treemap had to introduce 10–11 cospeciation events, 5–6 duplications, and 19–24 sorting events. Allowing host shifts: 10–11 cospeciations, 5–6 duplications, 3–19 losses, and 0–6 switches||For the parasite: mitochondrial 12S rRNA, 16S rRNA, COI, and nuclear wingless and EFl-a gene. For the host: mitochondrial 12S rRNA, COI, and ATPases 8 and 6 genes||Significant correlation between coalescence times (r = 0.94). The intercept of the slope is positive but not significantly different from zero||Hughes et al. (2007)|
|2||Wing lice of the genus Anaticola (Ischnocera) and several genera of flamingoes and ducks||Cospeciations and host shifts||Parasitic lice infecting flamingoes and ducks||43 genera of avian lice||treemap||Codivergences = 4–5, duplications = 5–6, losses = 1–32, host switches = 0–6||For the parasite: nuclear EF-1a, mitochondrial 12S and cytochrome oxidase I (COI). Avian phylogeny already published||Not tested||Johnson et al. (2006)|
|2||Polyomaviridae (polyomaviruses) and vertebrates (avian and mammals)||Codivergence||Parasitic double-stranded DNA viruses, which are widely distributed among vertebrates; avian viruses infect a broader host range than the highly specific mammalian polyomaviruses||72 full genomes: nine mammalian (67 strains) and two avian (5 strains) polyomavirus||treemap||Codivergences = 12, duplications = 8, losses = 2–13, host switches = 0–4||For the virus: the main five genes of the genome (VP1, VP2, VP3, large T antigen, and small T antigen)||Not tested||Perez-Losada et al. (2006)|
|2||Mealybugs Hemiptera (Subfamily Pseudococcinae) and endosymbiont bacteria||Codivergence and sorting events||Hemipterans, mealybugs and their obligate intracellular bacterial symbionts, thought to be strictly vertically inherited||21 host mealybug taxa and their bacterial symbionts||treemap and SOWH test||treemap: 14 codivergences, 0–3 duplications, 7–12 sorting events and 2–5 host shifts. Significantly congruent||For the mealybugs: EF-1a, 28S and 18S. For the endosymbionts: 16S and 23S rDNA||Strong correlation between branch lengths in host and symbiont trees (r = 0.785, P < 0.001)||Downie & Gullan (2005)|
|2||Plants (Fabaceae, Asteraceae, Rosaceae, Cyperaceae) and gall-forming nematodes (Tylenchida: Anguinidae)||Cospeciation||Gall-forming nematodes, obligate specialized parasites of plants||58 nematode samples from 53 populations||treemap||12 cospeciations, 4–6 duplications, 1–4 host switches. The level of cospeciation was estimated as 60%||For the parasitic nematode: ITS1, 5.8S and ITS2. For the plant: ITS1 and ITS2||Not tested||Subbotin et al. (2004)|
|2||Doves and pigeons (Aves: Columbiformes) and feather lice in the genus Columbicola (Insecta: Phthiraptera)||Cospeciation, but also significant level of incongruence and host switches||Vertically transmitted parasitic lice of pigeons and doves. Some species are host specific, other are found on multiple host species||27 host species and their associated 15 lice species||treemap and treefitter||9 cospeciation events, 11 duplications and 61 sorting events. Up to 3 host switches under certain costs. Number of cospeciation events higher than expected by chance||For the parasite: COI and the nuclear EF-1α. For the host: mitochondrial cyt b, COI and the nuclear FIB7||Not tested||Johnson et al. (2003)|
|2||Feather mites (Subfamily Avenzoariinae) and birds (Charadriiformes, Procellariiformes, Pelecaniformes, Ciconiiformes, and Falconiformes)||Cospeciation||Mostly commensal and some parasitic mites of birds from the Subfamily Avenzoariinae||26 mite species||treemap||12–13 cospeciation events, 6–7 duplications, 2 host shifts, and 26–29 sorting events||Mite phylogeny based on 41 morphological characters and mtDNA. For birds, phylogeny constructed from several published phylogenies based on morphological and molecular data||Not tested||Dabert (2001)|
|2||Seabirds (Procellariiformes and Sphenisciformes) and lice (Phthiraptera)||Cospeciation||Seabirds and their parasitic lice||11 species of seabirds from the sphenisciform genera and 14 species of lice from six genera||treemap||One host-switching, 9 cospeciation, 3–4 intrahost speciation, and 11–14 sorting events||For the parasite: 12S rRNA. For the hosts: 12S ribosomal RNA, isoenzyme, and behavioral data||Not tested||Paterson et al. (2000)|
|3||Three trophic levels: geometrid moths (Eois), braconid parasitoids (Parapanteles) and plants in the genus Piper||Host shifts and host conservatism (shifts to closely related hosts) in Eois||Herbivore moths, specialist moth parasitoid wasp||N = 94 (> 13 spp.) for Eois, N = 38 (> 10 spp.) for Parapanteles N = 52 for Piper||Permutation test of Hommola (nonrandom association of matrices)||NASignificant correlation between the branch lengths, but due to host conservationism||COI and Ef1-a for Eois; ITS1 and ITS2 for Piper; COI and two nuclear genes for Parapanteles||Fossil calibration for the Piper and Eois trees, molecular clock estimate for the Parapanteles tree: lack of temporal congruence||Wilson et al. (2012)|
|3||Neobatrachian anurans (frogs and toads) and Platyhelminthes (Monogenea)||Host shifts||Parasitic relationship: flatworm and anurian||26 parasite species, 23 anuran species||treemap, parafit, DIVA analysis||4 host shifts, 22 codivergences, 10 duplications, and 16 extinction events; Parafit test nonsignificant||For the parasite: 18S and 28S. For the host: Rhodopsin and mitochondrial (12S and 16S)||No : Inferred datations inconsistent with codivergence||Badets et al. (2011)|
|3||Chewing lice (Pappogeomys) and Geomydoecus pocket gophers||Prevalent cospeciation||Highly host-specific parasitic chewing lice on pocket gophers occurring on a single pocket gopher species or subspecies||57 individuals from the Geomydoecus bulleri species group||treemap and parafit||12 cospeciation events, 4 duplications, 1 loss, and 2 host switches||COI for chewing lice. Phylogeny of the host previously published based on mtDNA Cytb and CoI and 1 nuclear gene (b-fib)||Absolute time congruence not tested, but the estimated molecular substitution rate is fourfold higher in lice than in hosts under assumed codivergence||Demastes et al. (2012)|
|3||Cyttaria fungi on southern beech trees (Nothofagus)||Codivergence, but also host shifts and extinction events||Obligate Ascomycete fungi parasites of trees||12 species of Cyttaria and their hosts||parafit||Significant cophylogenetic structure with Parafit; reconstruction of the history by hand with 7–8 codivergence, 1–2 duplications, 1–2 host shifts||Cyttaria phylogenies already published. For Nothofagus: cpDNA, rbcL, nucITS, rRNA, cpDNA atpB-rbcL intergenic spacer and morphological data||BEAST calibrated with fossils inferred a more ancient divergence of the fungus than Nothofagus||Peterson et al. (2010)|
|3||Nosema (Microsporidia: Nosematidae) and bees (Hymenoptera: Apidae)||Cospeciation and host shifts||Microsporidian parasites in bees||4 host species and 4 parasite species||treemap and treefitter||0–1 cospeciation, 1–2 host shifts||For the parasite: LS and SS rRNA. For the host: cytochrome b||Not tested||Shafer et al. (2009)|
|3||Wheat, barley and oat (Poaceae) and Wheat dwarf viruses (WDV) (Mastrevirus)||Codivergence for some viruses but not for others||Parasitic DNA viruses||Full genomes of 46 isolates of Wheat dwarf virus||treemap||6 codivergences and 2 host jumps||For viruses: Phylogenetic trees constructed using full genomes. Host: rbcL||Correlation between host lineage and WDV divergence estimates. However, assuming codivergence, the inferred rate of substitutions implied stronger constraints against change than by other methods||Wu et al. (2008)|
|3||Heteromyid Rodents (Rodentia: Heteromyidae) and Fahrenholzia sucking lice (Phthiraptera: Anoplura)||Codivergence||Rodents and their permanent and obligate ectoparasitic sucking lice||43 heteromyid specimens and their lice||parafit, treemap||parafit: 39 of the 44 host-parasite pairs were significant. treemap: 26 codivergences, 14 duplications, 23 extinctions, 1 host switching||Host and parasite phylogenies: COI||Correlation between branch lengths, but r is weak (r = 0.7) and the slope is 2.8, interpreted as different rates of substitutions in lice; intercept significantly < 0, indicating delayed divergence in lice relative to host divergence||Light & Hafner (2008)|
|3||Lice (Pediculus, Pedicinus, Pthirus) and primates (Homo, Pan, Gorilla)||Significant cospeciation, but also parasite duplication, extinction, and host switching||Highly specialized and permanent obligate ectoparasites of primates||5 species of lice from primates and one species from rodents as outgroup||treemap||treemap: 5 cospeciation events and one host switch 1 duplication and 2 losses. Significantly greater similarity between the host and parasite trees than expected by chance||For the lice: mitochondrial Cox1 and elongation factor 1 alpha (EF-1α) gene||Divergence date estimates show that the nodes in the host and parasite trees are not contemporaneous||Reed et al. (2007)|
|3||Simian foamy viruses and primates (Hominoidea and Cercopithecoidea)||Cospeciation||Non-pathogenic RNA retroviruses infecting mammals||55 primate species and viruses isolated from 44 primate species||treemap||Significant support for overall cospeciation (22 events/44), with some obvious cases of some instances of cross-species infections||For the virus: polymerase gene (pol). For the host: mitochondrial (mtDNA) cytochrome oxidase subunit II (COII)||Significant linear relationship (r = 0.8486) between branch lengths. However, the molecular clock calibrations under cospeciation hypothesis infers an extremely low rate of SFV evolution, that would make it the slowest-evolving RNA virus documented so far||Switzer et al. (2005)|
|3||Gyrodactylus flatworms and Pomatoschistus Gobies fishes||Host switches||Two types of platyhelminth parasites: a monophyletic group of host-specific species, mainly infecting gills and a second group with lower specificity, dominantly found on fin and skin||15 Gyrodactylus taxa||treefitter, treemap and parafit||The overall fit between trees was significant according to treemap and treefitter, but not according to the timed analysis in treemap or to the parafit analysis||For the parasite: the V4 region of the 18S rRNA and the complete ITS rDNA region. For the host: the 12S and 16S mtDNA fragments||An absolute timing of speciation events in host and parasite ruled out the possibility of synchronous speciation for the gill parasites||Huyse et al. (2005)|
|3||Primate lentiviruses (PLV) and primates||Host switches||Parasitic retroviruses that have been cited as evidence for codivergence||12 primate taxa (including outgroup) and their lentiviruses: 11 events||treemap||8 codivergences events of a possible 11 events for perfectly matched trees, but simulated phylogenies based on the hypothesis of preferential shifts between closely related hosts were mostly congruent, and cospeciation was inferred||Host and parasite phylogenies based on a number of published studies||Divergence time incompatible||Charleston & Robertson (2002)|
|3||Brood parasitic finches (Vidua spp.) and their finch hosts (Estrildidae)||Host shifts inferred from dates while cophylogeny tests pointed to cospeciations||Host specific African brood parasitic finches (Vidua spp.) that mimic the songs and nestling mouth markings of their finch hosts (family Estrildidae)||74 estrildids, 21 parasitic finches, and nine ploceid finches as the outgroup||treemap and parafit||Basal divergences among Vidua species are more recent than those among host species, allowing cospeciation to be rejected, while tests for cospeciation indicated significant congruence between host and parasite tree topologies||For host and parasites: most of the analyses were done using mtDNA data set, although some nuclear sequences were also used in some clades||More recent divergence of parasites than hosts||Sorenson et al. (2004)|
|3||Malaria parasites (Plasmodium and Haemoproteus) and Haemoproteus birds||Cospeciation||Plasmodium parasites and Haemoproteus birds. Individual parasite species are thought to be restricted to host taxonomic families||68 lineages of Plasmodium/Haemoproteus recovered from 79 species of birds in 20 avian families||treefitter||Significantly more cospeciation events (9–16) than in randomized trees; however, they required up to 52 switching events or 366 extinction events||For the parasite: Cytochrome b. For the host: phylogenies already published based on the DNA–DNA hybridization studies||Assuming codivergence, the mitochondrial DNA nucleotide substitution appears to occur about three times faster in hosts than in parasites||Ricklefs & Fallon (2002)|
|3||Frankia bacteria and angiosperm plants (Actinorhizae)||Significant tree congruence but incongruent dates||Actinorhizae, mutualistic relation between angiosperm roots and nitrogen fixing Frankia bacteria||19 Actinorhizal angiosperms||treemap, Component||8 events of codivergence and 9 duplication events. The probability of eight coevolutionary events occurring by chance was about 0.23 when 1000 host and symbiont trees were randomly associated||For the bacteria: nifH and 16S rDNA. For actinorhizal plants: rbcL||Estimated divergence times among Frankia and plant clades indicated that Frankia clades diverged more recently than plant clades||Jeong et al. (1999)|
|4||Sigma viruses (Rhabdoviruses) and Drosophila fruit flies||Host shifts||Parasite vertically transmitted RNA virus||4 species of Diptera||Shimodaira–Hasegawa test and Robinson–Foulds distance||4/7||RNA polymerase gene for viruses||Not tested||Longdon et al. (2011)|
|4||Papillomavirus and mammals||Host shifts||Parasitic double-stranded DNA viruses||207 PV genomes||treemap, treefitter, and parafit||1/3||3 genes for Papilloma; 68-genes for the hosts||Not tested||Gottschling et al. (2011)|
|4||Gammaretroviruses and bats (Chiroptera)||Host shifts||Exogenous parasitic retroviruses transmitted horizontally||11 bat species||treemap||2/7||Viruses: Gag and Pol proteins. Host tree from the tree of life||Not tested||Cui et al. (2012)|
|4||Lymphocystis viruses and fishes (Paralichthyidae)||Independent divergence||Parasitic DNA viruses causing lymphocystis disease in fish||25 virus isolates, 8 fish species||treemap||3 codivergences, 11 duplications and 19 sorting events||Cytochrome b for the fishes, mcp gene for Lymphocystis||Not tested||Yan et al. (2011)|
|4||Maculinea butterfly and Myrmica ants||Independent divergence||Parasitic relationship: caterpillars need to be adopted and nursed by ants||32 Maculinea specimens (8 species including outgroup), 14 species of Myrmica||parafit, treefitter||Random association between the host and the parasite||COI, tRNA-Leu, trnL, COII and Elongation Factor for Maculinea. For Myrmica: COI, Cytb, 28S ArgK, EF 1 alpha and LwRh||Not tested||Jansen et al. (2011)|
|4||Tobamovirus and plants (monocotyledonous and dicotyledonous)||Independent evolution||Parasitic relationship: plant RNA viruses||31 species of Tobamovirus||treemap||Lack of congruence between the host and the parasite phylogenies||Genes for the virus: CP (ORF4). For the plants: rbcL||More recent divergence of viruses than of their hosts (BEAST estimations for viruses)||Pagán et al. (2010)|
|4||Fig trees (Ficus) and fig wasps (Elisabethiella, Courtella, Alfonsiella)||Host shifts||Mutualistic relation between Ficus and extreme host specific African fig wasps||42 wasp taxa and 26 Ficus species||parafit, treefitter||Nonsignificant Parafit test; A least twice as many host shifts as cospeciation events, even with high costs to host shifts||EF-1a and Cytb for Ficus and CO1 for wasps||Host shifts occurred later than host diversification events, although overall confidence intervals overlap||Mcleish & Noort (2012)|
|4||Steinernema nematodes and γ-Proteobacteria (Xenorhabdus)||Host shifts||Mutualistic relationship between nematodes and their associated c-Proteobacteria||30 host species and their associated bacteria||Tarzan||12 cospeciation events, 17 host-switches and 7 occurrences of sorting||For the nematode: 28S, 12S, and COI. For the bacteria: 16S, RecA and SerC genes||Not tested||Lee & Stock (2010)|
|4||Picornaviruses and animals (Aves and mammals: Primates, Rodentia, Carnivora, Perissodactyla, Certatiodactyla)||Host shifts||Parasitic RNA viruses causing a broad spectrum of diseases in several orders of birds and mammals||752 complete genome sequences of piconaviruses||parafit||Lack of congruence||2C, 3Cpro, and 3Dpol||Not tested||Lewis-Rogers & Crandall (2010)|
|4||Malaria (Plasmodium) and primates||Independent evolution||Parasitic Plasmodium and their primate hosts||18 Plasmodium species||treefitter and parafit||0–5 cospeciations, but assuming either up to 93 sorting events or up to 12 duplications or up to 11 host shifts||For Plasmodium: 18S rRNA, β-tubulin, cell division cycle 2, EF, cyt b, merozoite surface; Host phylogeny previously published||Not tested||Garamszegi (2009)|
|4||Hantavirus and Rodents (Arvicolinae, Murinae, and Sigmodontinae subfamilies)||Mainly host shifts||Parasitic single-stranded RNA viruses||For the parasite: 65 taxa. For the host: 95 sequences||treemap||13–14 codivergence, 20–23 host shifts, 5–7 duplications and 4–10 sorting events; Parafit test nonsignificant||For the virus: S, M, and L segments. For the host: cyt b||Overlap of the mean node ages||Ramsden et al. (2008)|
|4||Candidatus endobugula bacteria and their Bugula bryozoan host||No support for a history of strict cospeciation||Mutualistic vertically transmitted bacteria of bryozoan||Five host species and their associated symbionts||treemap||3 cospeciation events and 1 host switch, but this was not significantly more congruent than expected by chance||Host: 16S LSU rRNA and COI; Symbiont: 16S SSU rRNA||Not tested||Lim-Fong et al. (2008)|
|4||Grasses (Pooideae) and Epichloë fungal endophytes||Overall non-significant congruence, but early codivergence suggested||Symbiont (from mutualist to parasites) fungal Endophytes in grasses, mostly vertically transmitted||26 grass species-Epichloë species||parafit and MRCALink||Analysis of the 26 associations did not reject random association. When five obvious host jumps were removed, the analysis significantly rejected random association and supported grass–endophyte codivergence||For the plant: a trnL intron and two intergenic spacers (trnT-trnL, trnL-trnF) from cpDNA. For the fungus: tubB (formerly tub2) and tefA (formerly tef1)||No correlation between MRCA ages in the 26 species tree||Schardl et al. (2008)|
|4||Mussels (Mytilidae: Bathymod-iolinae) and endosymbiotic bacteria||Incongruence||Bathymodiolin mussels and their associated thiotrophic (sulfur-oxidizing) bacterial endosymbiont||For the host, 25 OTU||parafit||Host and symbiont tree topologies were not congruent||For the host: ND4, COI and 28S. For the parasite: 16S rRNA||Inferred time-depths of the gene trees were inconsistent (Mantel's test)||Won et al. (2008)|
|4||Fig trees (Ficus) and their associated fig wasps||Incongruence||Figs and their mutualistic pollinators||For the host: 18 neotropical fig species||treemap||No significant codivergence. Reconciliation of phylogenies inferred 3–5 cospeciations. If switching events are excluded, reconciliation required 40–45 losses||For the host: g3pdh, tpi and the ITS. For the pollinator: Phylogeny based on data already published||Not tested||Jackson et al. (2008)|
|4||Chaetodactylid mites and long-tongued bees (Apidae and Megachilidae)||Infrequent host shifts at a higher taxonomic level, and frequent shifts at a lower level||Mites of bees including mutualists (feeding on nest waste), parasitoids (killing the bee egg or larvae), commensals or cleptoparasites||230 mite species from 1500 museum specimens of long-tongued bees||parafit, DistPCoA, treefitter||0–3 cospeciation, 5–8 duplications, 0–6 host shifts, 0–35 extinctions||Mite phylogeny: 51 morphological characters. Host phylogenies already published||Not tested||Klimov et al. (2007)|
|4||Polyomavirus in human populations (Homo sapiens sapiens)||No evidence for codivergence||Double-stranded DNA viruses transmitted in a quasi-vertical manner (from parent to child postnatally)||333 viral genomes and 158 human mitochondrial sequences||treemap||< 10 codivergence events||Viral genomes and mitochondrial human sequences||The analysis suggests that this virus may evolve nearly two orders of magnitude faster than predicted under the codivergence hypothesis||Shackelton et al. (2006)|
|4||Penguins (Sphenisciformes) and chewing lice (Phthiraptera: Philopteridae)||Incongruence interpreted as caused by failure to speciate (parasites not speciating in response to their hosts speciating)||Multihost parasites, all species of chewing lice are parasites of an entire host order||15 species of chewing lice parasitizing all 17 species of penguins||treefitter, treemap and parafit||No evidence of extensive cospeciation but support for significant congruence between the phylogenies interpreted as possible failure to speciate events||For the parasitic lice: mitochondrial 12S and COI regions. Host phylogeny based on 70 integumentary and breeding characters||Not tested||Banks et al. (2006)|
|4||Urophora insects (Diptera: Tephritidae) and plants (Centaureinae)||No evidence for overall congruence||Herbivorous insects fruit fly genus||11 European Urophora taxa||treemap||The number of cospeciation events (3 and 4) did not differ from random expectation||For the herbivore: allozyme frequency data from 20 loci. Host phylogeny already published based on allozymes||Inferred divergence times indicated that the split of insect taxa lagged behind the split of their hosts||Brändle et al. (2005)|
|4||Anther smut fungi (Microbotryum) and their host plants (Caryophyllaceae)||Host shifts between relatively closely related species||Microbotryum complex: Parasitic sexually transmitted and species-specific fungi of the Caryophyllaceae||21 host plants and their fungal parasites||treemap, treefitter, Maximum Agreement Subtrees (Icong index), parafit||Overall, results suggest that cospeciation is not the rule in the Microbotryum–Caryophyllaceae system, that host shifts were pervasive, but that fungal species could not shift to too distant host species||For the host plant: ITS and cpDNA (trnL and trnF). For the parasite: β-tubulin, γ-tubulin and Elongation factor 1 α||Not tested||Refrégier et al. (2008)|
|4||Achrysocharoides parasitoid wasps, Lepidoptera insects and plants (Rosales, Sapindales and Fabaceae)||Incongruence between the three phylogenies||Achrysocharoides parasitoid wasps, highly host-specific and attack leaf-mining Lepidoptera and the plant host of Lepidoptera larvae||15 Achrysocharoides species||treemap to compare the three phylogenies pairwise||No evidence that the phylogenies were more congruent than expected by chance||For the parasitoid: cyt b sequences and 28S. For the Lepidoptera and the plant host, phylogenies already published||Not tested||Lopez-Vaamonde et al. (2005)|
|4||Glochidion trees and Epicephala moths||No perfect congruence||Obligate species-specific pollination mutualism between plants and their seed-parasitic pollinators||18 Glochidion species. For the pollinator a single individual from each of the 18 morphologically delimited species||treefitter, treemap and parafit||Greater congruence between the phylogenies than expected in a random association. Perfect congruence between phylogenies is not found, which likely resulted from host shift by the moths||For the plant: the entire ITS-1, 5.8S rDNA, and ITS-2 regions and the entire intergenic spacer region between 28S and 18S rDNA including ETS For the moth: CO1, ArgK and EF-lox||Not tested||Kawakita (2010)|
|4||New World arenaviruses (NWA) and rodents (subfamilies Sigmodontinae and Neotominae)||Host switches||Single-stranded parasitic RNA viruses. One-quarter of them infect multiple hosts and one-third of the host species can be infected by more than one NWA virus||21 host taxa and 22 viral taxa||Parafit||22 of 31 host–virus associations were not significantly congruent||For the virus: complete coding region sequences of GP, NP, L and Z proteins. For the host: mitochondrial cytochrome b||Not tested||Irwin et al. (2012)|
|4||Seabirds (Procellariidae) and lice (Phthiraptera: Ischnocera)||Codivergence and host switches||Parasitic lice from seabirds (petrels, albatrosses, and their relatives) with a high degree of lineage specificity||39 lice species from diverse hosts. The louse tree was broken into four subtrees and analysed separately||treemap||Mixture of cospeciation and host switching, with some clades of lice showing close fidelity to their hosts (high codivergence) and other clades showing higher levels of host switching||For the parasite 12S rRNA and COI. Previously published elongation factor 1a. For the host, phylogeny constructed using a published dataset based on cytochrome b||Correlation between sequence divergences||Page et al. (2004)|
|4||Decacrema ants and Macaranga trees||Lack of overall phylogenetic congruence||Highly specific mutualistic ants that inhabits and defends trees in Southeast Asia||Decacrema ants from 262 trees corresponding to 22 Macaranga species||treemap and parafit||The Parafit analysis suggests only partial congruence between ants and plants. No cospeciation events were inferred by treemap||For the ant phylogeny based on COI. Macaranga phylogeny based on morphological characters and nuclear ITS already published||Not tested||Quek et al. (2004)|
|4||Avian malaria parasites (Plasmodium) and birds (Aves)||Host shifts||Bird parasites vector-transmitted parasites from the genus Plasmodium and Haemoproteus||65 parasite lineages, 44 host species, and 121 host–parasite links||Component, treefitter and parafit||Lack of significant congruence||For the parasite and the host: cytochrome b||Not tested||Ricklefs et al. (2004)|
|4||Austrophilopterus chewing lice and Ramphastos toucans||Host switches||Chewing lice, parasites of toucans, considered to be host specific||26 Austrophil-opterus lice collected from 10 Ramphastos toucans and 7 Pteroglossus toucans||treemap and treefitter||Overall, treemap indicated lack of cospeciation. Analyses identified one potential cospeciation event but then required 3 duplications and 17–22 sorting events. treefitter: 0–2 cospeciation events and 6 host switches||For parasitic lice: COI and EF-1α. For the toucans: phylogeny already published based on different sequences such as mitochondrial COI and Cyt b||Not tested||Weckstein (2004)|
|4||Drosophila fruit flies and Howardula nematodes||Host shifts||Howardula nematodes, horizontally transmitted parasites of Drosophila||Almost all known Drosophila hosts of Howardula||treemap||Host and parasite phylogenies are not congruent. The reconstruction with the fewest steps yielded 3 cospeciation events, 5 host switches, 0 duplication events and 25 sorting events||For the parasite: rDNA: 18S, ITS1 and COI. For the host: COI, COII, COIII||Not tested||Perlman et al. (2003)|
|4||Deep sea vestimentiferan tubeworms and bacteria||No evidence for cospeciation||Vestimentiferan tubeworm relying on intracellular sulfide-oxidizing bacteria located in specialized tissues||15 Vestimentiferan taxa and their symbionts||treemap||No evidence for cospeciation||For the symbiont: 16S ribosomal gene. For the host: COI||Not tested||McMullin et al. (2003)|
|4||Fishes (Sparidae) and monogenean parasites Lamellodiscus||Associations considered to be due more to ecological factors than to cospeciation||Fish hosts (Sparidae) and their highly host specific monogenean parasites (Lamellodiscus)||20 described Lamellodiscus species and 16 Sparidae||treefitter, treemap and parafit||All methods agreed on the absence of widespread cospeciation if the cost of a host switch is not assumed to be very high||For the parasite: 18S rDNA. For the host: mitochondrial cyt b and previously published 16S mtDNA sequences||Not tested||Desdevises et al. (2002)|
|4||Wolbachia and fig wasps (Hymenop-tera)||Incongruent phylogenies||Mainly vertically (and pervasive horizontally) transmitted Wolbachia bacteria in fig wasps||70 individuals representing 22 wasp species and their 23 species of associated Wolbachia||treemap||The total number of matches between the two cladograms (7 cospeciation events) was not signicantly different from random expectation||For the parasite: wsp gene. For the host: phylogeny already published based on partial COI and COII sequences||Not tested||Dewayne Shoemaker et al. (2002)|
|4||Brueelia lice and birds (Passeriformes, Trogoniformes, Piciformes, Coraciiformes, Psittaciformes, Caprimulgiformes, Charadriiformes and Columbiformes)||Inconruent phylogenies||Brueelia parasitic lice considered to be highly host-specific, infecting birds||15 species of Brueelia collected from 21 host species||treemap||7 cospeciation events not beyond that expected by chance||For the parasite: nuclear EF-1α and mitochondrial COI. For the host: phylogenies already published based on the DNA-DNA hybridization studies||Not tested||Johnson et al. (2002)|
|4||Fig trees (Malvanthera) and fig wasps (Pleisto-dontes, Sycoscapter)||Partial codiver-gence; Host plant switching less constrained in parasites than in pollinators||Figs, obligated mutualistic pollinating Pleistodontes wasps and parasitic nonpollinating Sycoscapter wasps. Each Ficus species is typically host to one pollinating and many different nonpollinating wasp species||20 species of Pleistodontes and 16 species of Sycoscapter associated with Ficus species in the section Malvanthera||treemap, SH tests, ILD||The level of cospeciation is significantly greater than that expected by chance. However, the maximum level of cospeciation was only 50–64% of nodes||For the mutualistic wasp Pleistodontes: cyt b, 28S, and ITS2. For the parasitic Sycoscapter: cyt b and 28S||The greater genetic distances between Sycoscapter species than between their associated pollina-tors suggest that Sycoscapter may have the higher rate of molecular evolu-tion. Another possibility is that Sycoscapter species are older||Lopez-Vaamonde et al. (2001)|
|4||Lichens (Trebouxia): algae and fungi||Switching of algal genotypes occurred repeatedly among these symbiotic lichen associations||Long-term mutualism between of photosynthetic algae or cyanobacteria and heterotrophic fungi. Low algal specificity||33 natural lichen associations: 46 fungal species are associated with only 36 genotypes, representing four or fewer species of algae||treemap||10– 11 cospeciations. However, this required 7–9 duplications, 3–5 switches and 65–81 sorting events||For both symbionts: ITS||Not tested||Piercey-Normore & DePriest (2001)|
|4||Primates and Oxyuridae nematodes||Host-switching and codivergence||Enterobiinae oxyurid, nematodes parasites of primates. In most of the cases, one parasite species per host species||48 species of Enterobiinae analysed (46 species of the subfamily and 2 outgroup species) and their hosts||treemap||6–8 cospeciation events, 1 duplication, 1–3 host switching, 1–4 sorting events||For the parasite: 45 morphological characters from various organ systems. For the host, modified from a previously published phylogeny||Not tested||Hugot (1999)|
|4||Puccinia rust fungi and Brassicaceae plants||Host shifts more common than codivergence||Crucifers and their flower-mimicking fungal pathogens||17 Brassicaceae species and 3 rust species (multiple individuals of each)||Partition homogeneity test||Incongruent phylogenies||For the host: cp trnL-F and ITS; for the fungi: ITS and 5.8S||Not tested||Roy (2001)|
|4||Ascomycete mycangial (Ophiostom-ataceae) fungi and Dendroctonus bark beetles||No widespread codivergence||Mutualist and specific relationship: beetles carry mycangia, tegument invagination for fungal dissemination||11 fungal species and 6 beetle species||treemap||4 cospeciations, 3 duplications, 4 sorting events and 1 host shift; more cospeciations than expected by chance||Isoenzymes||Not tested||Six & Paine (1999)|
|4 (8 cases) and 5 7 cases)||15 Plant–fungal symbioses||A continuum of cophylo-genetic patterns ranging from mostly codivergence to mostly switching||Different plant-fungal associations, ranging from parasitism to mutualism||Symbioses from 5 Orders and 10 families||POpt and treemap||Seven associations showed significant congruence while eight were incongruent. Even the association inferred as significantly congruent exhibited a number of losses or duplication and/or host shifts||Phylogenies already published and bases on different molecules depending on the symbiosis. In general, for the fungal symbiont: ITS or nuclear rRNA. Different molecules used for the host phylogeny||Not tested||Jackson (2004)|
|5||Fungal Pneumocystis and mammals||Codivergence||Parasitic fungus||19 species of mammals||treemap||14 cospeciation out of 18 events (number of other events inferred not indicated)||For the parasite: mtLSU rDNA, mtSSU rDNA and DHPS. Phylogeny of the mammals previously published||Not tested||Chabé et al. (2012)|
|5||Spinturnix mites and bats (Rhinolophus, Myotis, Nyctalus, Plecotus, Miniopterus and Barbastellus)||Cospeciation and host shifts||European bats and their ectoparasitic mites||78 Spinturnix mites (11 morphospecies) from 20 European bat species||parafit, mesquite||Significant cophylogenetic structure, but at least five host switch events||For mites: two mitochondrial genes (16S–COI). For bats, published phylogenies plus cyt b||Not tested||Bruyndonckx et al. (2009)|
|5||α-Proteobacteria and Ishikawaella stinkbugs||Mainly codivergence||Vertically transmitted gut mutualistic bacteria of stinkbugs||14 host species and their symbiotic bacteria||treemap and treefitter||10–11 codivergence events, 2–3 host shifts, 2–3 duplications, 2–3 sorting events||For the bacteria: 16S rRNA and groEL t. For the host, COI||Not tested||Kikuchi et al. (2009)|
|5||Fungal Pneumocystis and Primates||Cospeciation||Highly specific fungal parasites||20 primate species||treemap||61–77% of the nodes interpreted as resulting from codivergence events, but the numbers of other events then required are not reported||For the parasite: DHPS, mtSSU-rRNA, and mtLSU-rRNA. For the host: phylogenies already published based on several mitochondrial and nuclear sequences and morphological characters||Not tested||Hugot et al. (2003)|
|5||Cryptocercus cockroaches and their bacteria Blattabacterium cuenoti||Cospeciation||Cryptocercus subsocial, xylophagous cockroaches and their endosymbiotic and vertically transmitted bacteria Blattabacterium cuenoti||Six out of the seven Cryptocercus species and their endosymbionts||component Lite||Significant similarity between phylogenies||For the bacteria: 16S rRNA and 23S rRNA. For the host: portions of the 28S rRNA and 5.8S rRNA genes and the entire ITS2||Not tested||Clark et al. (2001)|
|5||Uroleucon aphids and endosymbiotic Buchnera bacteria||Cospeciation||Aphids and their mutualistic vertically transmitted endobacteria, required for host reproduction||14 representative species of Uroleucon and their bacteria||treemap, Kishino–Hasegawa test, likelihood-ratio test||Highly significant levels of similarity between the trees: 8–9 cospeciation out of 14 possible||For the mutualist: partial sequences of trpB. For the host: tree based on mitochondrial and nuclear sequences already published||Not tested||Clark et al. (2000)|
|4 and 2||Viruses (Partitiviridae), plants (Viridiplantae) and fungi (Ascomycetes and Basidiomycetes)||Two virus families with codivergence inferred and two families without codivergence||Parasitic relationship: Vertically and horizontally transmitted RNA virus||175 viral genomes||parafit as implemented in axparafit, treefitter, treemap||Many duplication and switching events inferred even for the families where codivergence is suggested||Complete genomes for viruses||Not tested||Göker et al. (2011)|
|4 and 5||Doves (Aves: Columbiformes) and lice (Columbicola and Physconelloides)||Cospeciation in body lice but not in wing lice||Dove body lice (Physconelloides) and dove wing lice (Columbicola) parasitizing pigeons and doves, dove body lice being more host-specific than dove wing lice||13 species of doves and their associated wing and body lice||treemap||For dove wing lice: 4/12 cospeciation events, which is not more than expected by chance. For body lice: 8/12 cospeciation events, congruence being inferred as significant, but the numbers of other events assumed are not reported||For the parasite: mitochondrial COI and 12S rRNA and the nuclear EF-1α For the host: mitochondrial cyt b and the nuclear FIB7||Not tested||Clayton & Johnson (2003)|
|4 and 5||Fig trees (Sycomorus) and fig wasps (Ceratosolen and Apocryptophagus)||Cospeciation for mutualists and host shift for parasites||Different types of symbionts of figs: Mutualist pollinator Ceratosolen wasps and parasite Apocryptophagus wasps||19 species of Sycomorus figs. 19 Ceratosolen species and 18 species of Apocryp-tophagus||treemap||9–10 cospeciation (significant) for mutualists and 7–8 for the parasites (not significant)||For the symbiotic wasps: mitochondrial COI. For the host fig: ITS||Not tested||Weiblen & Bush (2002)|
|4 and 5||Chondracanthid copepods and fishes (Ophidiiformes, Pleuronectiformes, Scorpaeniformes, Zeiformes and Gadiformes)||Cospeciation in one fish order but not in the second||Chondracanthid copepods parasitic on fish considered to be host specific although this has been debated||26 Chondr-acanthus spp. and their teleost host genera from five orders||treemap||Support for cospeciation of copepods and their fish hosts in the orders Ophidiiformes, Pleuronectiformes, Scorpaeniformes and Zeiformes, but no support for cospeciation in the Gadiformes||Phylogenies already published||Not tested||Paterson & Poulin (1999)|
Examples in the literature are also found illustrating that significant congruence between host and symbiont phylogenies may occur without cospeciation, by the preferential occurrence of host shifts between closely related hosts under certain conditions of time lag between host shift and subsequent speciation. Indeed, most of the few studies in which absolute node dates were inferred have shown the dates of speciation to be incongruent for the interacting host and parasite species, despite the inference of cospeciation events by topology-based analyses (Charleston & Robertson, 2002; Sorenson et al., 2004; Huyse & Volckaert, 2005). Good illustrations are also found for our claim that mere correlations between branch lengths without absolute calibrations based on fossils are not sufficient to show temporal congruence. In a study analysing codivergence in a tritrophic association between Piper plants, Eois moths and their Parapanteles parasitoids (Wilson et al., 2012), the branch lengths of the phylogenies were found to be significantly correlated, but dating analyses revealed that the correlation resulted from host conservationism (i.e. the moth radiated preferentially on closely related hosts after host shifts or closely related moths radiated on the same hosts) rather than codivergence. Another study has shown that correlations between branch lengths of the phylogenies of Caryophyllaceous plants and their anther smut fungi most likely result from host shifts occurring preferentially between closely related hosts (Refrégier et al., 2008).
The well-known association between pocket gophers and their chewing lice (Hafner et al., 1994, 2003) remains the ‘textbook example’ of cospeciation, and it played a central role in the development of the methods presented here. Interestingly, the high level of cospeciation in this system may be linked to the life history and ecology of these parasites and their hosts: pocket gophers (Rodentia: Geomyidae) are herbivorous rodents that spend most of their life in tunnels that they do not share with other individuals. Species of pocket gophers are mostly allopatric, decreasing the likelihood of their parasites shifting to other hosts. Moreover, the chewing lice (family Trichodectidae) are obligate parasites that spend their entire life on the host, with no dispersal stage (Reed & Hafner, 1997; Clayton et al., 2004). Experimental studies have shown that lice can colonize new gopher species, suggesting that limited dispersal is the main constraint preventing host shifts. The combination of the solitary and allopatric host lifestyle and the limited dispersal ability of the parasite may account for the rarity of host-shift speciation in this system (Clayton & Johnson, 2003; Clayton et al., 2003). Other ecological factors that may influence the probability of codivergence include the abundance of the main host, the community of parasites, the degree of specialization, the population sizes and generation times of hosts and symbionts (Whiteman et al., 2007; Gibson et al., 2010; Nieberding et al., 2010).
Notwithstanding the exemplary nature of the case of pocket gophers and their chewing lice, analyses of their association have assumed multiple host shifts and intrahost speciation events to reconcile phylogenies, even with the great costs assumed for these events (Light & Hafner, 2007). Lice species other than those of the pocket gopher have been investigated for codivergence. The heteromyid gophers, which are more social than pocket gophers, display lower levels of tree congruence with their sucking lice (Light & Hafner, 2008). Furthermore, the intercept of the regression line between the gopher and lice divergence times was significantly < 0, indicating that lice divergence occurred after host divergence (Light & Hafner, 2008). Similarly, the estimated dates of divergence between lice and primates shows that the nodes in the host and parasite trees did not coincide temporally (Reed et al., 2007). Nevertheless, event-based methods analyses misleadingly inferred ‘significant cospeciation’ (Page, 1996).
The most convincing examples of cospeciation appear to concern mutualist associations in which the symbiont is transmitted vertically (Table 2, Fig. 5), as could be expected (Nieberding & Olivieri, 2007). A few host shifts have, nevertheless, been detected in associations of mutualists with vertical transmission (Table 2, cases Fig. 1b).
Important conclusions from this literature review and theoretical considerations are that symbiont speciation by host shift appears to be more common than cospeciation – even more than is currently recognized (Fig. 5, convincing examples of cospeciation represent only 7% of the cases) – and that the results of cophylogenetic tests are often overinterpreted to suggest cospeciation. A key question thus concerns the short-term ecological and genetic mechanisms promoting host-shift speciation rather than cospeciation. Nieberding et al. (2010) put forward a list of ecological traits that might influence the degree of cospeciation. In the next section, we consider the evolutionary mechanisms affecting the likelihood of symbiont specialization and speciation in relation to short-term coevolution with hosts.
V. Relationship between host–symbiont coevolution and symbiont speciation
We aim here to review the processes by which coevolutionary mechanisms can promote symbiont diversification. For this to occur, coevolution must first foster the specialization of symbionts, which could then lead to speciation. We thus review studies (1) showing how coevolution can promote symbiont specialization and (2) providing experimental and theoretical evidence for symbiont specialization leading to speciation. We argue that divergence as a result of specialization may occur, but that it occurs more frequently through host-shift speciation than cospeciation.
1. Coevolution: short-term host–parasite interaction
Host–parasite coevolution is a process of prolonged reciprocal selection, for better recognition of the parasite by its host, and for greater infectious ability of the parasites and the prevision of parasitism by the host. In the simplest systems, this selection involves a single locus in each partner. Two outcomes for the dynamics of host and pathogen allele frequencies are commonly distinguished under frequency-dependent selection (Holub, 2001; Woolhouse et al., 2002). The ‘arms race’ model describes allele frequency dynamics where advantageous new variants go to fixation. By contrast, the ‘trench warfare’ model depicts allele frequencies in oscillating dynamically over time or converging to equilibrium frequencies, resulting in the maintenance of several host and pathogen alleles (Brown & Tellier, 2011).
Another classification considers the dynamics of phenotype shifts caused by selection. When the phenotype values always shift in the same direction, as in predator–prey systems with density-dependent selection, the interaction has been termed ‘phenotype difference’ (Dawkins & Krebs, 1979), whereas when the system oscillates depending on the phenotypic value of the interacting species, as in most self/nonself recognition systems with frequency-dependent selection, the interaction has been called ‘phenotype matching’ (Lahti, 2005).
Such dynamical systems led Van Valen (1973) to refer to the coevolutionary processes between hosts and parasites as ‘Red Queen’ dynamics, in reference to Lewis Carroll's tale Through the Looking Glass (the Red Queen character explains to Alice that in her world that ‘it takes all the running you can do, to keep in the same place’). His paper was the first to connect short-term coevolutionary dynamics with macroevolution, including the long-term persistence of species in particular. The question here is whether coevolution, regardless of the prevailing mechanism (arms race, trench warfare, etc.), can actually directly promote parasite specialization.
2. From coevolution to specialization, models and observations
A priori, we might expect all species to be selected for the exploitation of broad ecological niches. Becoming a generalist decreases the spatial and temporal risks and efforts required for food collection and ensures survival in conditions in which the availability of particular resources may reveal unreliable. Generalism is common in plant viruses (Garcia-Arenal et al., 2003) and in animal viruses (Pedersen et al., 2005). However, specialization seems to be far more common than generalism in various parasite species ranging from phytophagous insects (Dres & Mallet, 2002; Nyman, 2010) to fungi (Giraud et al., 2008) and avian parasites (Proctor & Owens, 2000).
The relative paucity of generalist parasites may result from trade-offs between the ability to infect a broad range of host species and optimized rates of exploitation for any particular host type. Such trade-offs have been observed in serial passage experiments, in which propagating a microorganism on a host species different from its original host species consistently leads to a decrease in fitness on the original host (Ebert, 1998). By contrast, the instability of host abundance proposed as a factor explaining the evolution of generalists in natural systems (Jaenike, 1990; Norton & Carpenter, 1998) has received some experimental support (Soler et al., 2009). A combination of these selection pressures may occur, as both specialists and generalists have emerged in several experimental evolution studies (Little et al., 2006; Poullain et al., 2008).
Box 1. Glossary
|Codivergence||Process whereby a symbiont population or species splits at the same time as that of its host population or species. This is a pattern and does not assume causal relationships.|
|Coevolution (to be distinguished from cospeciation)||Process of never-ending reciprocal selection for improvements in parasite recognition in the host, and for improvements in recognition escape mechanisms in the parasite.|
|Congruence||Phylogenetic trees are said to be congruent when their topologies are highly similar; temporal congruence also implies that the corresponding nodes are of similar ages in the two phylogenies.|
|Cospeciation||Process whereby a symbiont speciates at the same time as another species (this may result from vicarious events or from narrow host specificity). This is a pattern and does not assume causal relationships.|
|Generalist||Symbiont able to take resources from different host species.|
|Host||Organism from which another smaller organism (the symbiont), from another species, takes resources; the symbiont may be either a parasite or a mutualist. Mutualists also provide the host with resources.|
|Host-shift speciation||Speciation of the symbiont by specialization of a daughter species on a new host.|
|Intrahost speciation (called ‘duplication’ in some papers and cophylogeny software)||Speciation of the symbiont without speciation of the host or host shift: both daughter symbiont species continue to parasitize the same host species. This may be because of vicarious events affecting only the symbiont or specialization on different organs of the host.|
|Mutualist||Organism both taking and resources from and providing resources to another larger organism (the host), from another species, resulting in an overall increase in host fitness.|
|Parasite||Organism taking resources from another larger organism (the host), from another species, decreasing host fitness.|
|Specialist||Symbiont able to take resources from a single host species.|
|Symbiont||Organism taking resources from another larger organism (the host) from another species. The symbiont is either a parasite or a mutualist. Mutualists also provide the host with resources.|
Factors favoring specialization even in the absence of fitness trade-offs and in the presence of stable host populations have been investigated in theoretical studies. In particular, parasite specialization may also evolve because of the more rapid adaptation of specialists than generalists to each host species (Whitlock, 1996; Kawecki, 1998) as assumed in the ‘Red Queen dynamics’ theory (Whitlock, 1996). According to the model developed by Kawecki (1998), if recurrent selection for new alleles at the loci controlling infectivity occurs because of coevolution, then specialization will be selected for because specialist parasites adapt more rapidly than generalists. Indeed, selection for a greater ability to infect a given host operates at every generation in specialized parasites, but only occasionally in generalists distributed between several host species. The chances of specialist parasites to persist are thus increased. In addition, once a species specializes in a narrow niche, the other species suffer less competition in the alternative niches, indirectly promoting specialization on these other niches (Whitlock, 1996). In summary, specialization (i.e. the formation of host races in parasites) can be promoted directly by coevolution because of the improbability of success on several different hosts and/or a higher rate of adaptation of specialists, and indirectly through competition with other specialist species.
Theoretical models have shown that the type of interaction may determine whether coevolution promotes or hinders specialization in both hosts and parasites (Yoder & Nuismer, 2010). Interactions mediated by phenotype matching promote specialization of the species experiencing a cost of phenotype matching, for example, pathogens being recognized by hosts and prevented from infecting. By contrast, they inhibit specialization of the interacting species that benefit from phenotypic matching, for example, the host being able to detect a pathogen and thereby impair infection (Yoder & Nuismer, 2010).
Cospeciation or host-shift speciation thus requires host speciation by independent mechanisms, such as geographic isolation, or parasite specialization by the mechanisms described earlier followed by parasite speciation. In the next section, we present theoretical considerations concerning the effects of specialization on parasite speciation.
3. Specialization and parasite speciation, theoretical considerations
The evolution of host-specific genotypes leads to the emergence of specialist parasite species only if reproductive isolation also occurs (Giraud et al., 2008). This corresponds to ecological speciation, in which parasite species occupying different niches (i.e. different host species) become reproductively isolated one from another (Giraud et al., 2010). The possibility of ecological speciation has been supported by many different studies on systems as diverse as herbivorous insects, vertebrates and plants (for reviews, see Hendry et al., 2007; Nyman, 2010).
Two factors promote the evolution of reproductive isolation in populations adapted to different ecological niches. First, there should be low levels of dispersal among populations (Hendry et al., 2007). Second, mating should occur only among individuals specialized for the niche (Rice, 1984), by means of adapted behavior (Funk, 1998), specific life-history traits, such as the mating of microbial parasites within hosts after infection (Giraud et al., 2006, 2010), or physical linkage between the loci controlling niche choice and mate choice (Slatkin, 1996). For example, pea aphids harbor tightly linked loci controlling host preference and mating preference, potentially facilitating the observed divergence between species (Hawthorne & Via, 2001). Phytophagous insects experience selection against mating with congeners feeding on a different plant species, potentially contributing to future divergence (Johnson et al., 1996; Nosil et al., 2002; Egan et al., 2008). Fungal ascomycete plant parasites that mate within their host plants display high rates of divergence without selection for strong intersterility, possibly because the genes responsible for adaptation to the host pleiotropically cause reproductive isolation (Peever, 2007; Le Gac & Giraud, 2008; Giraud et al., 2010). As a result, parasite specialization seems to contribute to diversification through speciation in various systems. The speed at which this speciation occurs depends on many factors, including parasite and host generation time, dispersal rates and effective population size (Huyse et al., 2005).
Coevolution thus clearly fosters parasite speciation by specialization to particular hosts (for a review, see Summers et al., 2003) such that specialization of two parasite lineages on sister host species may result in a cospeciation event. However, is cospeciation the most likely outcome in the long term, as is often implicitly assumed? The reasons for disruption of a host–parasite association are numerous, and such disruption may interfere with long-term parallel evolution between hosts and parasites, even in highly specialized lineages. Parasites may go extinct or may have a low incidence in host populations or small population sizes, such that a host speciation event may be missed. This becomes highly probable if, for example, a new host species originates by founding a population in allopatry from only a few individuals that are free of parasites. Many examples are known of biological invasions in which a population of hosts invading a new continent have undergone ‘enemy release’ (Keane & Crawley, 2002; Genton et al., 2005). Extinctions are also quite frequent in parasites owing to, for example, the evolution of resistance in host, decreasing niche size (Thrall et al., 1993; Ricklefs, 2010), or to a decline in host population size (de Castro & Bolker, 2005). Indeed, endangered plant and animal species, with their smaller and more fragmented population structures, have been shown to harbor a lower diversity of parasites than hosts with larger population sizes (Altizer et al., 2007; Gibson et al., 2010). Small host population sizes may not be compatible with the persistence of specialist pathogens (de Castro & Bolker, 2005), and this may be another reason for which coevolution does not promote cospeciation: incipient host species often have small populations and therefore cannot sustain specialist parasites evolving with them. If coevolution hinders the persistence of generalist parasites, as argued earlier, it would even decrease the probability of cospeciation in cases in which the new host species is initially present as small populations. In the few cases in which the dates of divergence events have been estimated, plant speciation has been shown to be followed by rapid host shifts of parasites, as reported for Eios moths on Piper plants (Wilson et al., 2012).
The converse question of whether parasites can trigger host speciation has been less explored. Cophylogenetic analyses show that speciation occurs at a higher rate in primate lineages harboring larger numbers of parasites (Nunn et al., 2004), so there may well be reciprocal influences on speciation of hosts and parasites (but see also Pedersen & Davies, 2009). By contrast, some experimental studies have suggested that coevolution with parasites may hinder host diversification (Buckling & Rainey, 2002).
Overall, theoretical evidence and natural observations of complexes of sibling species of parasites suggest that coevolution may promote parasite speciation via specialization on different hosts. As a consequence of specialization, parasites may thus be expected to form two different species as a host lineage splits, and this is termed cospeciation. However, this leads to cospeciation patterns only if parasites remain associated with the same host lineages throughout host speciation events. This assumption of continuity of host–parasite associations during speciation is rarely made explicitly or tested directly. In addition, we argued earlier that there may be reasons why coevolution could impede cospeciation. By reviewing cophylogenetic analyses, we have shown that host-shift speciations seem to be much more prevalent than cospeciation in host–parasite associations, even noting the predominant influence of coevolution over short time-scales. In any case, the rare instances of convincing codivergence relate to vertically inherited mutualists, where host shifts can still be observed even in these systems (Table 2).
Several important conclusions can be drawn from our review on the theoretical advances and available data concerning long-term host and parasite coevolutionary dynamics:
- Parasite speciation was long expected to follow the Fahrenholz rule of cospeciation (‘parasite phylogeny mirrors that of the host’), but we have seen that speciation following host shifts (host-shift speciation) is at least as likely as cospeciation. The early studies suggesting a predominance of cospeciation are now subject to some doubt with the use of larger samples and the advent of more reliable and powerful tools for comparing phylogenies. In many instances, parasites have been shown to diverge more recently than their hosts, mostly by host-shift speciation. In the rare cases where cospeciation seems to have occurred, the synchronous divergence of host and parasite lineages seems to result primarily from strict vertical inheritance, rather than the reciprocal selection pressures exerted by the partners.
- As the reciprocal selection pressures between hosts and parasites do not prevent speciation mechanisms other than cospeciation, coevolution does not imply widespread cospeciation. We argue that the term ‘coevolution’ should be used only to mean reciprocal selection pressure in host and parasite systems, as already advocated by other authors (Smith et al., 2008a), and that this term should not refer simply to patterns of diversification.
- The concept of cospeciation has fostered the development of very useful tools for comparing phylogenies, based on systems with interesting ecological features (such as the pocket gophers and their chewing lice). Although the basis of the cospeciation concept – that tight physiological interaction leads to parallel speciation – has now largely been invalidated, the methods developed so far have help us to understand the extent to which the partners in a host–symbiont system influence their own diversification. For example, do host-shift speciations occur more frequently between more closely related hosts or between hosts with similar ecological traits? We argue that the results obtained with any of cophylogenetic methods should be interpreted with caution because many of these methods overestimate the probability of cospeciation. Most importantly, evaluating the temporal coincidence of speciation events in symbionts and hosts, with calibrated phylogenies, is required to distinguish between cospeciation and host-shift speciations on closely related host species.
- Further methodological developments would be also welcome in the field. For example, Nieberding et al. (2010) proposed a method that could be used to identify ecological traits (e.g. number of host species, abundance of main host, degree of specialization, dispersal ability, population sizes of hosts and symbionts, sex ratio and generation time) influencing the cophylogenetic pattern. Such cophylogenetic analyses of ecological traits have revealed, for example, that dispersal, rather than an ability to colonize new hosts, seems to be the main factor affecting codivergence in the louse–pocket gopher system (Reed & Hafner, 1997; Clayton et al., 2004). Further developments would also be welcome for the analysis of biological networks, the neutral theory of tree diversity and phylogenetic community structure models.
This work was funded by grants ANR 06-BLAN-0201 and ANR 07-BDIV-003. A.T. thanks the Volkswagen Stiftung (grant I/82752) and DFG (grant HU1776/1 to S. Hutter) for financial support. M.E.H. received funding from grant NSF-DEB 0747222. We thank Nova Science Publishers, Inc. for permission to use some of the text from Tellier et al. (2010). We thank the anonymous referees for their helpful comments and we apologize to all those colleagues whose work we have omitted to cite in this article.