By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Damien de Vienne Laboratoire Ecologie, Systématique et Evolution Département Génétique et Ecologie Evolutives Université Paris Sud, Bâtiment 360, 91405 Orsay Cedex, France. Tel.: +33 1 69 15 56 64; fax: +33 169 154697; e-mail: email@example.com
Congruence between host and parasite phylogenies is often taken as evidence for cospeciation. However, ‘pseudocospeciation’, resulting from host-switches followed by parasite speciation, may also generate congruent trees. To investigate this process and the conditions favouring its appearance, we here simulated the adaptive radiation of a parasite onto a new range of hosts. A very high congruence between the host tree and the resulting parasite trees was obtained when parasites switched between closely related hosts. Setting a shorter time lag for speciation after switches between distantly related hosts further increased the degree of congruence. The shape of the host tree, however, had a strong impact, as no congruence could be obtained when starting with highly unbalanced host trees. The strong congruences obtained were erroneously interpreted as the result of cospeciations by commonly used phylogenetic software packages despite the fact that all speciations resulted from host-switches in our model. These results highlight the importance of estimating the age of nodes in host and parasite phylogenies when testing for cospeciation and also demonstrate that the results obtained with software packages simulating evolutionary events must be interpreted with caution.
Host–parasite interactions now occupy a central place in studies of evolutionary ecology, thanks to, among others, the seminal work of Price (1980) and the ground breaking papers of Hamilton (1980) and Hamilton & Zuk (1982). With the development of molecular biology, and hence molecular phylogenies, these interactions have been studied in a new way: the joint analysis of host and parasite phylogenies, or cophylogenetic analysis. These analyses of molecular data revealed that phylogenies of interacting taxa sometimes have very similar and even identical topologies (for a review, see Page, 2003), referred to as congruence. This congruence is considered to be generated by multiple cospeciation events of host and parasite.
We explore here an alternative mechanism that can give rise to topological congruence between host and parasite phylogenies: an adaptive radiation of a parasite on a range of host species, by multiple host-switches at the tip of the host phylogeny followed by speciation. This idea has been proposed by some authors (Hafner & Nadler, 1988; Hafner et al., 1994; Page, 1996; Roy, 2001) and clearly explained by Paterson & Banks (2001): ‘Congruence, in itself, means little, as congruence may be generated by parasites undergoing a series of host-switches that mirror the host phylogeny[…]’. This mechanism, called ‘pseudocospeciation’ (Hafner & Nadler, 1988), is, however, rarely taken into account in cophylogenetic studies and many authors still consider congruence to result from cospeciation and incongruence to result from host-switches (Brooks & McLennan, 1991).
The idea that congruence between host and parasite trees can arise following preferential host-switching has been verified by Charleston & Robertson (2002) in a particular case. They observed that the phylogenies of primates and their lentiviruses were more congruent than expected by chance. Simulations of parasites switching hosts at the tips of the primate phylogeny generated similar levels of congruence between host and parasite phylogeny when parasites were more likely to switch between close relatives. However, the conditions favouring pseudocospeciation have not been examined in a systematic and general way to date. To fill this large gap in our understanding of host–parasite evolutionary interactions, we used a simulation approach of the adaptive radiation of a parasite on a set of pre-existing host species to determine what conditions other than cospeciation could generate congruent host and parasite phylogenies.
We modelled the arrival of a new parasite onto a speciose host clade whose phylogeny was known. This parasite then colonizes new hosts by switching between terminal lineages with a probability that depends on the relatedness between hosts and speciates either immediately or after a time lag. Once all hosts are parasitized, we construct the phylogenetic tree of the parasites and examine its shape and its congruence with the host tree. We performed simulations with three host trees having different topologies, from completely unbalanced to highly balanced. We examined the effect of different parameters concerning host and parasite behaviours and evolution on the congruence between host and parasite trees, such as host-switch probabilities, first host parasitized, shape of the host tree and time lag between a switch and the speciation following this switch. Note that here we did not simulate coevolution, the hosts neither evolving nor speciating during the adaptive radiation of their parasites. This situation can arise during biological invasions or shifts of a parasite onto a new clade of pre-existing hosts. Clearly, such a process does not preclude that the hosts can also speciate during the adaptive radiation of their parasites.
Obviously, some simplifying assumptions and rules had to be imposed: (1) we did not allow parasites to sequentially replace each other on a particular host species; (2) we did not allow more than one parasite species to parasitize the same host; (3) parasites speciated much more rapidly than did hosts (the phylogeny of the host remained unchanged during the adaptive radiation of the parasites); (4) no parasites went extinct; and (5) speciation always followed a host-switch, although more or less rapidly.
Our study thus differs from the one carried out by Charleston & Robertson (2002) by the wider range of hypotheses tested concerning host tree shape, first host parasitized, ‘preferential’ host-switches performed by parasites and time lag between switch and speciation, as well as by precluding the replacement of parasites on a given host.
Here, we address the following questions: (i) What conditions of host-switch probabilities, time lag before speciation, first host parasitized and topology of the host tree generate the most congruence? (ii) Can the comparison between host and parasite phylogenies give us any information about the evolutionary history of the two interacting species? (iii) Can the model we propose here contribute to a better understanding of the genetic basis of host–parasite interactions?
Congruence between host and parasite trees was assessed using cophylogenetic analysis software including the event-based parsimony method implemented in TREEFITTER (Ronquist, 1995). This software estimates, for two given phylogenies, the most parsimonious evolutionary history of the two lineages by assigning to cospeciation, duplication, sorting and switching events different costs, without requiring knowledge of branch lengths. We also used TREEMAP and COMPONENT in one particular case where our model generated high congruence to render our results comparable with previous experimental work. Using these analytical methods also allowed us to explore their limits because the only type of evolutionary event that was simulated here was host-switch, so all cospeciation, sorting and duplication events inferred by TREEFITTER and TREEMAP were artefacts.
We simulated the colonization of a set of hosts with known phylogenetic relationships by a new parasite, through the following steps: (1) arrival of the parasite on one host; (2) sequential switches from one tip of the host phylogeny to another (host range expansion followed by speciation after a certain time lag), each switch thus giving rise to two parasite species specialized on two different hosts; we considered that a given host could not be parasitized by more than one parasite species; (3) once all the hosts were parasitized, the phylogeny of the parasites was constructed based on the order of the switches and the timing of the speciations; and (4) the topological congruence between this tree and the host tree was estimated.
Simulations were performed under varying values for the following parameters: (1) the topology of the host tree; (2) the identity of the host that the parasite initially infects (e.g. its degree of isolation); (3) how the probability of switch between two hosts depended on their phylogenetic distance (e.g. switches more likely between closely related hosts or switches more likely between distantly related hosts); and (iv) the time lag before speciation after a host-switch, depending on the phylogenetic distance between the hosts.
Note that the trees in this study have no branch lengths sensu stricto. The term ‘phylogenetic distance’ will be used to describe the number of evolutionary units separating two given species in the trees. An evolutionary unit is the minimal time between two cladogenesis events, i.e. the shortest distance between two nodes (see Fig. 1a; for example, the phylogenetic distance between hosts 2 and 3 is four units).
Host trees used for the simulations
Three rooted host trees with 12 species each and particular topologies were chosen. The first one was completely unbalanced, the second one was intermediate and the last one was the most balanced possible (Fig. 1A–C respectively).
The first host parasitized
We assessed the effect of the degree of isolation in the phylogeny of the first host parasitized on the congruence between the host tree and the parasite tree resulting from simulations. The degree of isolation of a given host species in the phylogeny was estimated as its mean phylogenetic distance to the other hosts divided by the mean of the distances between all pairs of hosts. Simulations were run with each of the 12 hosts being parasitized first.
The probability for a switch from an host X (parasitized) to an host Y (not yet parasitized) was determined by: (1) the phylogenetic distance between hosts X and Y, dXY; and (2) by the isolation of host X in the tree, which depends on the distances between each nonparasitized host (Hj) with all other hosts (H) in the tree, dHHj. Three assumptions on host-switch probabilities were considered, depending on host phylogenetic distances.
(1) Higher probability for switches between phylogenetically close hosts. The probability p for a parasite to switch from the host X to the host Y was computed as:
Note that in this case, on host tree A, when host 12 is the first host parasitized, any host-switch is highly improbable because host 12 is highly isolated, but if a switch does occur each of the possible hosts is equally likely to be colonized.
(2) Lower probability for switches between phylogenetically close hosts. The probability P for a switch from host X to host Y was computed as:
The use of exponential functions as well as the fact that we took into account the isolation of the host in these two formulae allowed us to maximize the effect of the between-host phylogenetic distance on the behaviour of the parasites. The denominator allows one to have a complete system of events (PH→Hj).
(3) Equiprobability for all switches. The probability p for a switch from host X to host Y was computed as:
where NHi and NHj are the number of already parasitized hosts and not yet parasitized hosts respectively.
To illustrate the above equations, we represented the switch probabilities for a parasite that first colonized host 1 of host tree A to switch to each of the 11 other (not yet parasitized) hosts as a function of the pairwise phylogenetic distances between host 1 and each other host (Fig. 2). Note that these probabilities are perfectly symmetrical and very stringent.
Time lag between host range expansion and speciation
Initial simulations considered that speciation immediately followed host range expansion. We subsequently included the ability for the parasite to remain generalist for a certain time. Two types of time lag between host range expansion and speciation were considered.
(1) Shorter time lag for switches between phylogenetically closer hosts. The time lag between the switch of the parasite on the host X to the host Y and speciation between the parasites on hosts X and Y (tXY) was then computed as:
with k being the ‘generalism coefficient’, a dimension-free constant calibrating the global propensity of parasites to remain generalist, and dXY the phylogenetic distance between the hosts X and Y.
(2) Shorter time lag for switches between phylogenetically more distant hosts. The time lag between the switch of the parasite on the host X to the host Y and speciation between the parasites on hosts X and Y (tXY) was then:
We ran simulations with k ranging from 10 to 300 with steps of 10 units. We could find no studies that allowed us to estimate this parameter but our aim was to investigate whether a tendency to remain generalist longer would influence the degree of congruence between host and parasite phylogenies.
Note that host-switches continued to take place during the period where a particular parasite remains generalist and even for generalist parasites that occupy more than one host.
Generation, comparison and imbalance of the trees
The program was written using the software R, version 2.0.1 (Ihaka & Gentleman, 1996), with the package ‘ape’ (Analyses of Phylogenetics and Evolution, Paradis et al., 2004). The code of the program is available on request.
Random trees were generated to: (1) compare observed vs. expected congruence between host and parasite trees; and (2) explore the impact of the imbalance of the host trees on the expected degree of congruence among random trees to control for this parameter in our comparisons. These random trees were generated using COMPONENT 2.0 (Page, 1993) whose seed for the random number generator is taken from the system clock and whose algorithm for generating rooted labelled trees is the one described by Furnas (1984).
We ran 1000 simulations of adaptive radiations for each of the different combinations of host tree shape, initially parasitized host, assumption on host-switch probability and assumption on time lag between host range expansion and speciation. Using the software TREEFITTER version 1.0 (Ronquist, 1995, 1997), the host trees were compared with: (1) the 1000 parasite trees resulting from our simulation for each combination of parameters; and (2) 1000 randomly generated trees, to compare the congruence observed in (1) to the level of congruence expected by chance. Note that TREEFITTER can generate random trees for each comparison of one host tree to one parasite tree, but we chose not to use this function because the time needed to do the tests was too long given the large number of comparisons. We used the default values of the parameters, suggested in the software, giving costs for codivergence, duplication, sorting and switching events were set to 0, 0, 1 and 2, as respectively. TREEFITTER estimates the minimum and maximum numbers of each of these four types of events for each pair of host and parasite trees, representing the range over all equally optimal reconstructions (i.e. same cost). We chose the minimum number of cospeciation events estimated to give us a conservative estimate of congruence. Clearly, the maximum number of cospeciation events yielded an even higher degree of congruence. For each comparison of one host tree to the 1000 corresponding parasite trees generated by our model, as well as to the 1000 randomly generated ones, the mean proportion of estimated cospeciation events among the total number of events was calculated and taken as the degree of congruence between the host and parasite trees. Indeed, if the trees are highly congruent, TREEFITTER estimates a high number of cospeciation events (the maximum number of cospeciation events being the number of nodes in the phylogenies) and few, if any, duplication, switching or sorting events. In this case, the proportion of cospeciation events among the total number of events will be close to or equal to one. If the trees are not topologically more similar than expected by chance alone, the number of cospeciation events estimated by TREEFITTER will be low and the number of duplication, switching and sorting events will be high. The mean proportion of estimated cospeciation events among the total number of events will then be close to what is obtained when comparing one host tree with 1000 randomly generated trees.
TREEFITTER was preferred to other available similar software packages, such as TREEMAP (Page, 1994), because it allows fitting any number of parasite trees to a given host tree and can execute commands from external files, allowing us to perform a high number of comparisons. We used two additional software packages for the set of parameters that generated the most completely congruent host and parasite phylogenies: (1) COMPONENT 2.0 (Page, 1993) assessed the plain topological congruence between trees without inferring evolutionary events by counting the number of species that had to be removed in the host and the parasites phylogenies to obtain identical trees (Kubicka et al., 1995); the smaller this number, the more congruent the phylogenies. (2) TREEMAP v1.0a (Page, 1994) was used because it is one of the most frequently used software packages for cophylogenetic analyses.
Because our study was based on simulations, we did not perform statistical tests, as the significance of a test depends upon the number of runs. We therefore decided to analyse and present the results in a qualitative instead of a quantitative manner.
Congruence with random trees
Congruence with random trees, estimated as the mean proportion of estimated cospeciation events when comparing one host tree with 1000 randomly generated trees, differed for the three host trees used in this study: these values were 0.266, 0.298 and 0.307 for host trees A, B and C respectively. This discrepancy is probably due to the fact that the three host trees do not have the same probability of being obtained by chance. Indeed, over 10 000 randomly generated trees with 12 species, when the range of the imbalance values of these trees was divided into 20 classes, the imbalance values for the trees A, B and C fell within classes containing 6.86%, 4.08% and 0.05% of the total number of trees respectively (data not shown). Because two completely congruent trees necessarily have the same level of imbalance, different probabilities of generating trees with particular degrees of imbalance can explain the difference between the three host trees we used on their level of congruence with random trees.
Speciation immediately following host range expansion
Effects of the topology of the host tree and of the host-switch probabilities
We first investigated the effects of the topology of the host tree and of the three different host-switch probabilities on the congruence between the host tree and the parasite trees when parasites speciate immediately after host shift. For the host trees B and C, and for the host tree A with high probability of distant switches or equiprobability for all switches, each tree was compared with 12 000 parasite trees generated by our program (1000 simulations for each of the 12 hosts initially parasitized). For the host tree A and a high probability for close switches, the parasite had too low a probability to perform the first switch when initially on hosts 9, 10, 11 or 12, so the host tree A was compared with only 8000 parasite trees (hosts 1–8 and 1000 parasite trees generated for each host parasitized first).
Most combinations of host tree topology and relatedness-dependent probabilities of host-switches yielded parasite trees that were not different from random trees in terms of the mean proportion of estimated cospeciation events (Fig. 3). This was always true for host tree A. However, for host trees B and C, when switches were more likely towards phylogenetically close hosts, parasite adaptive radiation yielded trees that were more congruent to host trees than were random trees (Fig. 3, cases 2), leading TREEFITTER to estimate a higher mean proportion of estimated cospeciation events than for random trees. This effect was most pronounced for the highly balanced host tree C. However, congruence was never very high, the mean proportion of estimated cospeciation events never approaching 1. For host trees B and C, when the probability for shifts to phylogenetically distant hosts was high, the mean proportion of estimated cospeciation events was slightly lower than for random parasite trees (Fig. 3, cases 3).
Effect of the first host parasitized
The first host parasitized further influenced the congruence, but mainly when the switches were not equiprobable and when the host tree was not completely balanced. For the unbalanced host tree A, when the first host parasitized was not very isolated (hosts 1, 2, 3 and 4), the resulting parasite trees were less congruent with the host tree than were random trees. From hosts 5 to 9, as isolation increased, congruence became closer to the congruence obtained with random trees. For the very isolated hosts (hosts 10, 11 and 12) parasitized first, when testing was possible (i.e. when the probability for switches to distant hosts was high; see above), congruence slightly decreased. For the host tree C, no effect of the first host parasitized was detectable. Finally, for the host tree B, the congruence appeared to improve with increasing isolation of the first host parasitized (the mean proportion of estimated cospeciation events was higher for hosts 5–8 than for hosts 9–12 parasitized first, and higher for hosts 9–12 than for hosts 1–4 parasitized first; Fig. 3).
To summarize, the host tree topology, the host-switch probabilities and the degree of isolation of the first host parasitized influenced the congruence between the host and the parasite trees following a parasite adaptive radiation onto an existing set of hosts. In some particular cases, the congruence could be higher than with random trees, although it was never very high and never total.
Time lag between switch and speciation
A time lag between a host-switch and subsequent speciation that was a function of: (1) the phylogenetic distance between the hosts from which and to which the parasite was switching and (2) the coefficient k (see eqns 4 and 5) further influenced tree congruence. For clarity, for each host tree topology, we only represented congruence with parasite trees resulting from a first colonization event to a small subset of hosts. We chose first hosts colonized that had different levels of isolation within the host tree: hosts 1, 8 and 12 for the intermediate host tree B and only hosts 1 and 12 for the balanced host tree C. For the host tree A, the pectinate comb, the chosen hosts depended on the host-switch probability that was tested: hosts 1, 6 and 12 when the probability for switches to phylogenetically distant hosts was high or for equiprobability for all switches, and hosts 1, 5 and 8 when this probability was small. This was because, in the latter case, if the first host parasitized was very isolated, the first switch seldom occurred and adaptive radiation did not take place.
Equiprobability for all switches or preferential switches to distant hosts
Adding a time lag between host-switch and speciation did not influence the congruence between host and parasite trees when all switches were equiprobable or when switches were more likely between phylogenetically distant hosts.
Preferential switches to proximate hosts
In contrast to the above-described situation, when switches occurred preferentially between phylogenetically close hosts and the lag-time was short for such switches between phylogenetically close hosts, the congruence between the trees was closer to what was obtained in random trees than the cases where no time lag was allowed (k = 0). This was true for all values of k greater than 0.
However, when the time lag before speciation was shorter for switches between phylogenetically distant hosts (Fig. 4) the shapes of the curves were completely different. For all host trees, increasing the coefficient k increased the congruence, but the strength of this increase depended on the shape of the host tree. For the host tree A, the congruence was only slightly affected. For the host trees B and C, this increase was sharper. Some parasite trees were even completely congruent with their host tree, leading to a mean proportion of estimated cospeciation events estimated close to 1. For example, for the host tree C, for k equal to 300 and for an initial infection of host 12, 21.3% of the parasite trees were totally congruent with the host tree, leading COMPONENT to estimate that a very small number of species had to be pruned to obtain identical trees (Fig. 5) and TREEMAP to estimate a mean number of cospeciation events close to 10, the maximum possible being 11.
The effect of the first host parasitized was not consistent across the different cases of host tree topology. For the host trees A and C, it appeared that the more isolated the first host was, the more congruent were parasite phylogenies resulting from adaptive radiation of parasites: the mean proportion of estimated cospeciation events across all the values of k increased with increasing isolation of the first host parasitized. This was not true for host tree B, however, as the best congruence was obtained with host 12 initially infected, whereas host 8 was more isolated (Fig. 4)
The results obtained under the different hypotheses are recapitulated in Table 1. It appears clearly that substantial congruence is only obtainable when considering preferential switches between closely related hosts, and that a long time lag before speciation after this kind of switch enormously increases this congruence. The shape of the host tree also seems to have a strong impact, as the simulations performed with the host tree A never gave a noticeable congruence with the parasite trees, in contrast to the ones preformed with the host trees B and C. Finally, the influence of the first host parasitized was not clear (Figs 3 and 4) and was therefore not represented in Table 1.
Table 1. Recapitulation of the congruence between host and parasite trees under the nine hypotheses tested and the three host trees (A, B and C; see Fig. 1).
Equiprobability for all switches
−: no noticeable congruence between host and parasite trees; ±: congruence slightly higher than expected by chance, +: congruence clearly higher than expected by chance; ++: clear congruence with at least 1% of trees completely identical to the host tree.
No time lag
Shorter time lag before speciation for switches between phylogenetically close than distant hosts
Shorter time lag before speciation for switches between phylogenetically far than near hosts
Congruence between host and parasite trees following host shifts
Parasites that colonize and radiate on a new host phylum can have phylogenies highly congruent to that of their hosts even without cospeciation. Congruence was maximized: (1) when parasites were better at switching to hosts phylogenetically closely related to the host they were already parasitizing, as previously shown on a specific host–parasite case (Charleston & Robertson, 2002); and (2) when speciation occurred more slowly, with parasites remaining generalists for longer after switching between closely related hosts than when they switched between phylogenetically distantly related ones. Under these conditions, TREEFITTER and TREEMAP estimated that most, if not all, of the evolutionary events were cospeciations.
Assumptions on host and parasite behaviour and evolution
Several of our assumptions on the behaviour and the evolution of hosts and the parasites may influence our results. First, in contrast to Charleston & Robertson's (2002) model, we did not allow parasites to replace each other on a particular host species. It seems indeed realistic to consider that unoccupied hosts were more readily colonized than those already parasitized. Thus, in our model, we considered the extreme case where the probability for a parasite to infect an already parasitized host was null.
Second, we did not allow for colonization of an already parasitized host species by an additional parasite. Because stable coexistence of distinct species having exactly the same ecological niche is unlikely in theory, one can assume that highly related species of parasites are unlikely to be able to form distinct species parasitizing the same host species, although such cases have been reported (Desdevises et al., 2002; Percy et al., 2004; Le Gac et al., 2007). These existing cases of infections of a single host species by multiple related parasite species can be due to the fact that the host represents various ecological niches, because of its anatomy (e.g. different parasites on different organs), its own life history (e.g. existence of different classes of individuals, such as male and female, old and young, susceptible and resistant, etc.) or because of its habitat (Thomas et al., 2002). However, we do not know if these parasites specialized on the same organs, for example, are more closely related to each other between host species than parasites on different organs within a single host species.
Third, parasites had a higher rate of speciation than hosts. Indeed, in our model, hosts did not speciate and parasites radiated onto a pre-existing clade. Although host–parasite coevolutionary history is likely to be characterized by cospeciation events as well as parasite speciation on pre-existing host species, we here address an ecological context involving only the latter. In particular, we consider cases of parasites with rapid evolutionary rates compared with those of their hosts colonizing new environments previously unoccupied by such parasites (Ricklefs & Fallon, 2002).
Fourth, we did not consider parasite extinctions or parasite speciation in the absence of a host-switch, i.e. duplication. These events are generally considered as factors increasing the discordance between host and parasite phylogenies (for a review, see Page, 2003, chapter I) and we expect that parasite extinctions and speciation in situ would have similar effects in our study.
Probabilities of host-switches and time lag before speciation
Host-switches have been inferred from incongruent cophylogenies in many host–parasite systems. However, details, such as the relative probability of switching to phylogenetically more or less distant hosts or the probability of speciation following such a range expansion, remain unclear. Host shifts occur more readily to phylogenetically more similar hosts in several systems (Reed & Hafner, 1997; Janz & Nylin, 1998; Nishiguchi et al., 1998; Ricklefs & Fallon, 2002), but there is no correlation between the phylogenetic distance between host plants and the switches of their leaf beetle parasites (Futuyma et al. (1995). Both scenarios of host-switches are plausible: switches will occur preferentially between closely related hosts if the mechanisms employed by a parasite to attack its habitual host are more effective on a closely related host or if closely related hosts have more similar habitat requirements. This will be the case if there is a phylogenetic signal in anti-parasite defence or habitat use. On the other hand, switches will occur preferentially between distantly related hosts if phylogenetically similar hosts are geographically or ecologically separated as would be the case during incipient allopatric speciation or when there was strong character displacement between newly diverged species.
Speciation following host-switches has been observed in several parasites (Zietara & Lumme, 2002; Lopez-Villavicencio et al., 2005), but the rapidity of speciation as a function of phylogenetic distance between the hosts has not, as far as we know, been assessed. Again, the two hypotheses that we considered are plausible: when closely related hosts are ecologically and/or geographically separated by character displacement or allopatry, switches between these more closely related species will be rare and thus lead to rapid parasite speciation in isolation. On the other hand, when closely related hosts overlap more in ecology and defence against parasites, rare switches to distantly related hosts would lead to rapid speciation in isolation facilitated by strong trade-offs for infecting very different hosts (Timms & Read, 1999). Although we tested all combinations of preferential host-switches and time lag before speciation, the cases where the highest and even perfect congruence was obtained – preferential host-switch to closely related host species but with relatively long lag time until speciation after these switches – corresponded to the more probable scenario we test. Indeed, it seems intuitive: (1) that parasites can jump more easily between more closely related hosts, those being more likely to share general ecological, physiological and chemical properties; and (2) that they will be able to remain generalists longer (i.e. speciate slower) on such similar hosts where trade-offs in performance are less strong and where repeated exchange maintains selection for success on both hosts. Our model thus shows that interpreting congruence as evidence of cospeciation should be treated most cautiously for species groups where closely related species share ecological and life-history traits.
Indeed, Charleston & Robertson (2002) observed that the phylogenies of primates and their lentiviruses were more congruent than expected by chance but that the nodes in the viral phylogeny were much younger than those in the host phylogeny. Similarly, Hirose et al. (2005) found a high degree of congruence between the phylogenies of maple powdery mildews (Sawadaea, Erysiphaceae) and their maple tree hosts (Acer spp.) despite the divergence of the different species of Acer occurring many millions of years before that of Sawadaea. In both cases, as in our model, congruent phylogenies resulted from host shifts and not from cospeciation. Hence, even though the choices of parameter values, such as the generalism coefficient k or of the functions governing the probabilities of host shifts for our model, were not based on empirical studies, we generated theoretical predictions consistent with real-world observations.
Adaptive radiation and coevolution
For simplicity, we decided to consider in our model parasites undergoing an adaptive radiation on a set of pre-existing and phylogenetically stable hosts. This way, the model was simpler than if we incorporated coevolution with the hosts and the results were much easier to interpret, avoiding having to tease out cospeciation and pseudocospeciation for congruent nodes. Pseudocospeciation should also occur when the hosts and the parasites are coevolving, as was initially proposed (Hafner & Nadler, 1988). In the context of contemporaneous speciations by both hosts and parasites the opportunity for pseudocospeciation to occur and its relative importance compared with cospeciation will depend on the proportion of empty niches (hosts free of parasites) left by the cospeciation process. If many host speciations are followed by lineage sorting of the parasite (i.e. missing the boat) or if parasite extinctions are frequent, then hosts scattered across the host phylogeny will be, from time to time, free of parasites. This will provide the opportunity for host shifts between more or less distant hosts. Under such conditions, symmetrical trees, preferential host shifts to phylogenetically similar hosts and longer time lag to speciation for these compared with the rarer shifts to phylogenetically distant hosts should artificially inflate congruence even if cospeciations also occur contemporaneously.
Implications of this study
Our model confirmed that congruence between host and parasite phylogenies could arise without cospeciation but through a rapid adaptive radiation under some plausible hypotheses about host-switches, time lag before speciation and host tree shape. This ‘pseudocospeciation’ mechanism was proposed a long time ago, but the conditions favouring its appearance had never been explored in a general way. Further, we hope that our model will reinforce the idea that congruence does not necessarily imply cospeciation and that the two following points will now be systematically taken into consideration: (i) assessing temporal congruence between host and parasite phylogenies is absolutely necessary, in addition to topological congruence, for concluding that cospeciation occurred; it should not be considered only as optional additional evidence; and (ii) results obtained with softwares like TREEFITTER or TREEMAP, that infer a coevolutionary scenario from the topologies of two given phylogenies, have to be considered with great care, as a high number of cospeciation events can be estimated in some cases where none actually occurred.
Furthermore, applied to real phylogenies of recently interacting hosts and parasites, this simulation approach may bring insights into some aspects of the genetics of resistance and infectivity. Indeed, if some recently interacting hosts and parasites have phylogenies with significant congruence, one can conclude that these parasites more easily switch to hosts that are phylogenetically close to their native host than more distant ones, and that the rare distant switches are rapidly followed by speciation.
M. Blum and E. Paradis have been of invaluable help for the programming of the model. We also thank R. Page, A. Paterson, R. Cruickshank and two anonymous referees for their interesting comments and suggestions on the manuscript.