Syst. Biol. 54(5):710–718, 2005 Copyright c Society of Systematic Biologists ISSN: 1063-5157 print / 1076-836X online DOI: 10.1080/10635150500221036 Comparing Host and Parasite Phylogenies: Gyrodactylus Jumping from Goby to Goby

TINE HUYSE AND FILIP A. M. VOLCKAERT Katholieke Universiteit Leuven, Laboratory of Aquatic Ecology, Ch. De Beriotstraat´ 32, B-3000 Leuven, Belgium; E-mail: [email protected] (T.H.)

Abstract.—The combination of exceptionally high species diversity, high host specificity, and a complex reproduction sys- tem raises many questions about the underlying mechanisms triggering speciation in the flatworm genus Gyrodactylus. The coevolutionary history with their goby hosts was investigated using both topology- and distance-based approaches; phylogenies were constructed of the V4 region of the 18S rRNA and the complete ITS rDNA region for the parasites, and 12S and 16S mtDNA fragments for the hosts. The overall fit between both trees was significant according to the topol- ogy-based programs (TreeMap 1.0, 2.0β and TreeFitter), but not according to the timed analysis in TreeMap 2.0β and the distance-based method (ParaFit). An absolute timing of speciation events in host and parasite ruled out the possibility of synchronous speciation for the gill parasites, favouring the distance-based result. Based on this information together with the biological background of host and parasite, the following TreeMap solution was selected. The group of gill parasites evolved from a host switch from G. arcuatus, parasitizing the three-spined stickleback onto the gobies, followed by several host-switching events among the respective goby hosts. The timing of these events is estimated to date back to the Late Pleistocene, suggesting a role for refugia-mediated mixing of parasite species. In contrast, it is suggested that cospeciation in the fin-parasites resulted in several host-associated species complexes. This illustrates that phylogenetically conserved host-switching mimics the phylogenetic signature of cospeciation, confounding topology-based programs. [Coevolution; Gobies; Gyrodactylus; host-parasite interactions; host-switching; speciation.]

One of the most significant radiations of platyhelminth scribed as well (Bakke et al., 2002). As such, all speciation fish parasites is documented in the monogenean “super- modes appear plausible in this group. genus” Gyrodactylus (Cribb et al., 2002). The 409 Gyro- In spite of these interesting characteristics, together dactylus species named so far (Harris et al., 2004) are with the fact that Gyrodactylus may pose serious threats to expected to represent only 2% to 10% of the estimated aquaculture (e.g., the highly pathogenic salmon parasite diversity (Bakke et al., 2002). Gyrodactylus displays the G. salaris), quantitative host-parasite cospeciation stud- widest host range of any monogenean family (19 orders ies are lacking (Bakke et al., 2002). According to Page of bony fish), encompassing both highly specific and gen- et al. (1996), the prerequisites for coevolutionary stud- eralist species. These remarkable facts can be linked with ies are (1) the availability of a sound alpha of their even more remarkable reproduction mode. Gyro- both host and parasite, (2) robust phylogenies of hosts dactylus spp. contain a fully grown daughter in utero, and parasites, (3) although not essential, molecular phy- which in turn encloses a developing embryo, boxed in- logenies are preferably based on homologous characters, side one another like Russian dolls; asexual, partheno- (4) wide taxon sampling, and (5) quantitative compari- genetic and sexual reproduction alternate (Cable and son of host and parasite trees by means of explicit sta- Harris, 2002). In optimal conditions, a first-born daugh- tistical tests. The first four prerequisites are fulfilled in ter can be produced within 24 hours of her parent’s birth our model consisting of the Gyrodactylus parasites living (Cable and Harris, 2002). This can result in explosive pop- on gobies of the genus Pomatoschistus. (1) The various ulation growths, and the direct life cycle (only one host Gyrodactylus taxa have been described by morphologi- involved) allows them to invade new habitats with their cal, morphometric, and molecular methods (Huyse and hosts. All these factors may accelerate population differ- Malmberg, 2004; Huyse et al., 2004a), whereas the host entiation and thus speciation. species have been extensively studied by Miller (1986). On the one hand, the direct life cycle and the high host- They occur in the Atlantic and Mediterranean coasts of specificity enforce a tight relationship of a Gyrodacty- Europe; some goby species are sympatric and hybridiza- lus species and its host, promoting coevolution (Poulin, tion has been reported (Webb,1980). (2) The phylogenetic 1992; Kearn, 1994). On the other hand, the ability to relationships of Gyrodactylus have been studied using the produce an entire population from a single individual complete ITS rDNA region and the V4 region of the 18S increases the chance for sympatric speciation and speci- rRNA (Huyse et al., 2003); the phylogeny of the host has ation by host-switching (Brooks and McLennan, 1993). been reconstructed from 12S and 16S mtDNA fragments, They may also survive for some time independently and in order to compare homologous characters. (3) The of their host and a “swimming behavior” has been ob- ITS 1 locus has been sequenced as well (Huyse et al., served (Cable et al., 2002). These features add to the 2004b). (4) In order to minimize sampling bias, sampling colonizing capability of Gyrodactylus, illustrated by sev- has been extended over time and space throughout the eral switches between fish families (Harris, 1993; Zietara natural geographical distribution of the fish hosts (see and Lumme, 2002) and even fish orders (Huyse et al., below). 2003). However, phylogenetic radiations (diversification All respective Gyrodactylus species display phylo- constrained/triggered by host phylogeny) have been de- genetic host specificity towards gobies of the genus

710 2005 HUYSE AND VOLCKAERT—COEVOLUTION: DISTANCE- VERSUS TOPOLOGY-BASED PROGRAMS 711

Pomatoschistus (Huyse et al., 2003). They can be split into position for all sequences was compared usinga5%χ 2 two groups by their genetic and ecological parameters. test on the average composition (TREE-PUZZLE v. 5.0, Group A represents a monophyletic group of closely re- Schmidt et al., 2002). Plotting transitions and transver- lated and host-specific species, mainly infecting gills, sions against divergence of the complete data set using all belonging to the subgenus G.(Mesonephrotus). Host DAMBE v. 4.0.75 (Xia and Xie, 2001) did not show sat- specificity is lower in group B, which comprises species uration. The molecular-clock hypothesis was tested as- that are dominantly found on fin and skin and belong to suming the HKY+ model with the likelihood-ratio test G. (Paranephrotus). In this article, we fulfill the last prereq- (LRT) for the clock hypothesis implemented in TREE- uisite for coevolutionary studies by comparing host and PUZZLE. ModelTest v. 3.06 (Posada and Crandall, 1998) parasite phylogenies using a combination of statistical was used to estimate the optimal model of molecular approaches. In addition, we test whether the difference evolution in a likelihood-testing framework. These pa- in niche is reflected in the evolution and distribution on rameters were used in the maximum likelihood (ML) the host species: Are exposed fin parasites more subject method using PAUP∗;trees were statistically tested by to host-switching than the more “sheltered” gill para- calculating P values for the individual branches. This sites? Did the monophyletic group of gill parasites evolve likelihood-ratio test assesses whether branch-lengths are through cospeciation with the host, as generally expected significantly different from zero by optimizing all branch for host-specific parasites (Poulin, 1992; Kearn, 1994)? lengths under the constraint that one of the branches is Alternatively, did the group of the more generalist fin zero, for each branch tested. For maximum parsimony parasites evolve through repeated host-switching events (MP) the exhaustive search method was performed us- instead of cospeciation? ing the branch and bound algorithm. In these analyses gaps were treated both as missing data and fifth character because indels might be phylogenetically informative; all MATERIAL AND METHODS sites were equally weighted. The neighbor-joining (NJ) Host and Parasite Data search was conducted (1000 replicates of tree-bissection During 1999 to 2002, 15 Gyrodactylus taxa were reconnection branch swapping) from a matrix of ML ge- collected on Gobiusculus flavescens and several Po- netic distances calculated under the optimized model. matoschistus species (Gobiidae, Teleostei) along the ML branch lengths have been generated under the spec- northeastern Atlantic continental shelf, the Baltic, Adri- ified model and topology, with and without a molec- atic, and Mediterranean Seas (see Huyse et al., 2003). ular clock assumption. Bayesian inference (BI) using Two Gyrodactylus parasites of the three-spined stickle- MrBayes v. 3.0b4 (Huelsenbeck and Ronquist, 2001) was back, Gasterosteus aculeatus, were collected as outgroup performed, using the model obtained by ModelTest. Pos- species (G. arcuatus and G. branchicus). Phylogenetic re- terior probabilities were estimated over 2.0 × 106 gen- lationships of the Gyrodactylus parasites were inferred erations, sampling the Markov chains at an interval of from the V4 region of the 18S rRNA and the complete 100 generations. In addition, four incrementally heated ITS rDNA region obtained in a previous study (Huyse Markov chains were used. Sample points generated be- et al., 2002, 2003). The phylogenetic relationships of the fore reaching stationary were discarded as “burn in” goby hosts were derived from ITS 1 rDNA and 12S and samples. Shimodaira-Hasegawa (SH) tests were used to 16S mtDNA (Huyse et al., 2004b). compare alternative MP and ML topologies obtained during the analyses (Shimodaira and Hasegawa, 1999), ∗ Phylogeny Reconstruction as implemented in PAUP .Ifnosignificant differences were found, only the ML topology was used; otherwise The V4 and ITS region of the parasites and the 12S all topologies were investigated. and 16S sequences of the hosts were treated as one data set because the incongruence-length difference test (Farris et al., 1995) implemented in PAUP∗ (Swofford, Testing for Cospeciation 2001) provided no evidence for significant difference Several methods for testing cospeciation are avail- in the phylogenetic signal of both regions (P = 0.10, able; most of them have been reviewed by Paterson and P = 0.65, respectively). Analyses have been conducted Banks (2001). Four of these methods were used to ana- on three data sets consisting of groups A and B sepa- lyze the host-parasite interactions in the present system. rately and all parasite species together. Within the re- The first method, TreeMap 1.0 (Page, 1994), reconciles spective groups the 5.8S sequence was identical, so only the host and the parasite tree by introducing four types the ITS 1, ITS 2, and V4 sequences were aligned us- of events: cospeciation (C), host-switching (H), dupli- ing Clustal X v. 1.81 (Thompson et al., 1997). When cation or intrahost speciation of the parasite (D), and pooling groups, the highly variable ITS 1 region was sorting or extinction of the parasite lineage (S). Using omitted and only the V4, 5.8S, and ITS 2 sequences a parsimony argument, the program attempts to explain were aligned using the program SOAP (L¨oytynoja and the differences between both phylogenies by postulat- Milinkovitch, 2002). This program identifies unstable ing the fewest possible number of events, and maximiz- sites, which are easily excluded and reincluded during ing the number of cospeciation events. A randomization subsequent analyses. All data matrices are available from test was performed to assess if both phylogenies are more TreeBASE (http://www.treebase.org/). The base com- similar to each other than expected by chance alone. 712 SYSTEMATIC BIOLOGY VOL. 54

Completely resolved trees are necessary, but alternative nificantly different (P > 0.05), the ML tree was used as tree topologies can be imported and evaluated. Recently, input file. The relationships between the closely related the beta version of TreeMap 2.0 was released (Charleston Pomatoschistus minutes, P. lozanoi, and P. norvegicus were and Page, 2002 available at http://taxonomy.zoology. not resolved, but the three possible branching patterns gla.ac.uk/∼mac/treemap/), which computes all opti- imported in TreeMap did not affect the number of cospe- mal solutions by exhaustive search, represented by Jun- ciating nodes. Exclusion of the unstable characters in the gles (Charleston, 1998). In the timed analysis, branch V4 and complete ITS region of the parasites resulted length information can be taken into account. It also in a 675-bp fragment. An alternative evaluation of the allows an assignment of different costs to each of the alignment by means of dot plots implemented in the Ge- four cophylogenetic events, a feature also available in neWorks software (Intelligenetics, Oxford, UK) resulted the program TreeFitter v. 1.1 (Ronquist, 2001, available in a similar fragment of 690 bp. The resulting V4-5.8S-ITS at http://www.ebc.uu.se/systzoo/research/treefitter/ 2 data set had no deviating base composition (P = 0.70– treefitter.html). The optimal reconstruction is that which 0.92) and yielded 118 parsimony informative sites. When minimizes the global cost. TreeFitter uses a permuta- the unstable alignment positions were reincluded in the tional procedure to statistically test the overall cost and analysis or when gaps were treated as a fifth character, contribution of each type of event. the number of parsimony informative sites increased to Whereas these methods are topology based, the pro- 136 and 126, respectively. This did not affect the topol- gram ParaFit (Legendre et al., 2002) makes use of raw ogy; the bootstrap values varied only slightly. The com- or patristic distances, thus overcoming the need for plete parasite tree, including groups A (gill parasites) well-resolved topologies. It statistically assesses the fit and B (fin parasites), was not completely resolved. Sepa- between host and parasite phylogenetic distance ma- rate analyses (A and B alone) included the complete data trices mediated by the matrix of host-parasite links. set (a total of 1125 bp) of group A; because the 5 ITS 1 Moreover, ParaFit can estimate and test the contribu- region of group B was still too variable, only the 3 end of tion of each individual link to this global fit. These ITS 1 was used, resulting in a fragment of 1125 bp. Both distance matrices can be computed from any data in- data sets displayed homogenous base composition and cluding sequence data, DNA/DNA hybridization data, resulted in fully resolved phylogenies; GTR+ was used or morphological characters. In the present study both with gamma shape parameter estimated at 0.3. These the ML genetic distances and the patristic distances phylogenies were used as backbone constraints in the inferred from the host and parasite phylogenies were analysis on the complete data set. Both host and parasite used. sequences evolved in a clock-like fashion as tested by Host-switching and cospeciation can also be detected TREE-PUZZLE. by analyzing genetic distances (Paterson and Banks, 2001), on the condition that hosts and parasites evolve at Comparison of ITS 1 Variation between Host and Parasite a constant rate (which may differ in both groups). There- The ITS 1 sequences of the gobies and the Gyrodactylus fore, the genetic distances of a homologous rDNA frag- spp. of group A behaved clock-like according to the LRT ments, i.e., the ITS 1 region, from both the host and the in TREE-PUZZLE, in contrast to group B. The genetic dis- parasite were compared (see Table 1). tances constructed from the ITS 1 region of the hosts Po- matoschistus minutus and P. microps, and their respective RESULTS parasites, are presented in Table 1. Each time, the para- Phylogenetic Analyses: Are the Phylogenies Resolved? site sister species were compared (i.e., comparison within group A and group B, respectively). With the exception of A total of about 800 bp of 12S and 16S mtDNA were G. rugiensoides–G. micropsi, genetic distances were much used for the construction of the host phylogeny; the base + + higher between the host-pair than the respective parasite composition was homogenous and the GTR I model species-pairs. was selected with the gamma shape parameter and the proportion of invariable sites estimated at 0.6. Because Is There Evidence for Coevolution? the Bayesian, ML, MP, and NJ topologies were not sig- TreeFitter.—Using the default settings (C = 0, D = 0, S = 1, H = 2), the fit between the host and parasite TABLE 1. The uncorrected pairwise distances (excluding gaps) con- phylogenies showed that the overall cost is significantly structed from the ITS 1 sequences of Pomatoschistus minutus and P. lower than expected by chance (P = 0.01; 1000 permu- microps amounts to 11.6%. The distances between the respective Gyro- tations). However, it could not be determined which co- dactylus parasites found on these hosts is summarized below (based on ITS 1 rDNA). phylogenetic event contributed to this fit because none of the individual P-values were significant. Lowering Gyrodactylus species pair p-Distance (%) the cost of host-switching from 2 to 1.5 (considering G. gondae–G. ostendicus 4.7 high parasite dispersal capabilities and sympatry of host G. gondae–G. branchialis 1.3 species), resulted in significant values for the number G. micropsi–G. cf micropsi2 3.6 of codivergence events (six to eight events; P = 0.034) G. rugiensoides–G. rugiensis 2.1 and the number of host-switching events (six to eight G. rugiensoides–G. micropsi 13.0 events, P = 0.008). Applying maximum codivergence 2005 HUYSE AND VOLCKAERT—COEVOLUTION: DISTANCE- VERSUS TOPOLOGY-BASED PROGRAMS 713 settings (C =−1, D = 0, S = 0, and H = 0), the global parasite phylograms and the respective host-parasite fit between the two trees was not significant anymore associations. (P = 0.10). By assigning a cost of 1 to sorting and TreeMap 2.0β.—In the nontimed analysis (using only host-switching events, the fit was significant (P = 0.017) topology and no branch length information), 22 optimal with host-switching as the main factor contributing to reconstructions were found. The optimal solutions pos- this (four to six events; P = 0.023). If cospeciation and tulated 16 cospeciations, five losses, six switches, and 16 sorting were assigned a very high cost (Fitch optimi- duplication events (Fig. 2). The randomization test on the sation), the significant values disappeared, confirming complete data set suggested that the global fit between the signal of cospeciation in the present host-parasite the host and parasite trees was statistically significant system. (P = 0.02 ± 0.01). When the data set was split in groups A TreeMap 1.0.—Without invoking any host-switching and B, only the association between group A and the host event, TreeMap had to introduce seven cospeciation tree appeared to be significant (P = 0.01 ± 0.01) with 11 events, nine duplications, and 27 sorting events to optimal reconstructions; the level of congruence between reconcile the trees. Adding host-switching events (us- group B and the host tree was not higher than expected ing a heuristic search) lead to the postulation of by chance (P = 0.11 ± 0.03). In the timed analyses (us- seven cospeciation events, eight duplications, one host- ing branch length information but not having the host switching, and 23 sorting events. By randomizing the and parasite tree set on the same scale), the number of parasite tree with the proportional-to-distinguishable cospeciation events in the gill parasites was not signifi- option, a null frequency distribution was generated. cant anymore. The observed number of cospeciations appeared sig- ParaFit.—For each of the three data sets, both raw and nificantly higher (P = 0.01) than expected by chance. patristic distances were used as input matrix. The global The percentage of cospeciating nodes (i.e., the num- test of cospeciation on the complete data set of raw dis- ber of cospeciating nodes divided by the total num- tances did not reveal a global association between hosts ber of nodes in the parasite phylogeny, multiplied by and parasites (P = 0.095). Using patristic distances did 100) amounted to 44%. Figure 1 shows the host and not influence the results (P = 0.094). Considering the

FIGURE 1. Evolutionary patterns of host association in Gyrodactylus species. Comparison of the goby host (left) and Gyrodactylus (right) phylograms, constructed from 12S and 16S mtDNA and the V4 and ITS region, respectively. TreeMap 1.0 (Page, 1994) called upon seven cospeciation events (denoted as a black circle; P = 0.01) to reconcile both trees. Branch lengths are proportional to the amount of evolutionary change, except for the branches connecting the ingroup with the outgroup sequences. Bootstrap support from minimum evolution analysis = (n 1000). Evolutionary rate for host and parasite 1% and 5.5%, respectively, per million years (Huyse et al., 2004b; Zietara and Lumme, 2002). “B” and “V” following the species G. ostendicus and G. branchialis stand for the collection sites “Belgium” and “Venice” respectively. Groups A and Bare boxed with dotted lines. (Picture from Pomatoschistus microps by Van Kampen; scanning electron micrograph of Gyrodacylus ostendicuswith scale bar = 100 µm). 714 SYSTEMATIC BIOLOGY VOL. 54

FIGURE 2. Reconciliation by TreeMap 2.0β (Page and Charleston, 2002) of goby and Gyrodactylus phylogenies. A shows the optimization for the complete data set (P = 0.02 ± 0.01); B and C represent two alternative jungles for group A (P = 0.01 ± 0.01). Arrows indicate a host-switching event; black circles, cospeciation; grey circles, sorting; and white squares and circles, duplication events. individual host-parasite links, only the links between tus/P. lozanoi; G. rugiensoides and P. minutus/P. lozanoi/P. the outgoup host and parasites were significant. How- pictus. In contrast, when taking the gill parasites sepa- ever, when a separate matrix was constructed from the rately, only the link between G. gondae and P. minutus/P. complete data set of group B (about 1300 bp), the re- lozanoi turned out to be significant (P = 0.023), be- sult of the global test was highly significant (P = 0.001). sides the G. arcuatus and G aculeatus link (P = 0.029). Significant host-parasite links include: G. cf. micropsi However, the overall signal was also significant (P = and P. minutus/P. lozanoi; G. cf. micropsi 2 and P. minu- 0.029). 2005 HUYSE AND VOLCKAERT—COEVOLUTION: DISTANCE- VERSUS TOPOLOGY-BASED PROGRAMS 715

DISCUSSION tion of both host and parasite trees is necessary (see So far, phylogenetic congruence is imperfect or absent below). in most interactions (Poulin, 1998). As also illustrated in this study, most associations represent a combination of Is Host-Switching More Frequent in the Fin Parasites? cospeciation and host-switching (Page and Hafner, 1996; Roy et al., 2001; Weiblen and Bush, 2002; Ricklefs et al., In contrast to the previous case, the number of cospe- 2004). Examples of strict cospeciation are found in sys- ciation events was not significant according to TreeMap tems where host-switching is prevented by the asocial whereas the fit between host and parasite trees was lifestyle of the host and the low mobility of the parasite. highly significant according to ParaFit. Page (1993) Examples include the rodent-lice (Hafner and Nadler, pointed out that the presence of one or more parasite lin- 1988; Page and Hafner, 1996), and insect-symbiont asso- eages on the same host results in incongruent host and ciations where the bacteria, needed for host reproduc- parasite phylogenies. As seen on Figure 1, up to four Gy- tion, are transmitted maternally (Clark et al., 2000). The rodactylus species were present on a single host species. only other study on flatworm-fish associations has been However, TreeMap reduces the multihost–parasite as- conducted by Desdevises et al. (2002). No significant sig- sociations to a single association as can be seen in Fig- nal of cospeciation could be detected and they concluded ure 2. The low host specificity of some species further that specialization is mainly influenced by ecological complicates the comparison of the host and parasite factors. Solitary fish species harbored a single parasite phylogenies in TreeMap. This might partly explain the species, whereas the highest species richness was found incongruent results of both programs. on gregarious fish living in sympatry. As is shown here, host specificity is not necessarily a Because Gyrodactylus species can readily disperse to prerequisite for cospeciation. Although G. rugiensoides is new hosts (see introductory section), speciation by host- found on Pomatoschistus minutus, P. lozanoi, and P. pictus, switching is expected to be an important speciation mode it might have coevolved with one of these hosts and col- (Brooks and McLennan, 1993; Bakke et al., 2002). The im- onized the other hosts without speciating (Brooks and pressive colonizing capability of Gyrodactylus is demon- McLennan, 1993). Generalism, like specificity, is some- strated by several ecological radiations (diversification times a pure transitional state rather than a fixed trait triggered by ecological factors) on distant fish species (Nosil, 2002). A specialist colonizing a new host might first become a generalist (increases its host range), and (Harris, 1993; Zietara and Lumme, 2002) and the epi- demic spread of G. salaris on Atlantic salmon in Norway eventually speciates on its new host(s). Laboratory ex- (Johnsen and Jensen, 1991). As such, the significant sig- periments are required to test such hypothesis. For exam- nal of cospeciation produced by the topology-based pro- ple, P. minutus and P. lozanoi occur in sympatry and share grams was rather unexpected. On the other hand, due to Gyrodactylus gondae, but this parasite species is not found their direct life cycle and progenetic development, Gyro- on the closely related P. norvegicus.Byexperimentally dactylus species live in close association with their host. transferring G. gondae populations grown on P. lozanoi It is shown that Gyrodactylus-host interactions are very to P. minutus and P. norvegicus (and analogous for pop- subtle and host specificity among monogeneans is gov- ulations grown on P. minutus), the reproductive success erned by a number of dynamic interactions (Buchmann in the different cross-infections can be compared. If each and Lindenstrøm, 2002). population performs best on its original host, this might In contrast to the topology-based programs, the indicate local adaptation. This can be regarded as an on- distance-based approaches suggested that there was no set to specialisation. If each population performs equally significant fit between both data sets. These contrast- well on all hosts, G. gondae should be regarded as a gen- ing results reflect the differences between the underly- eralist species. Information on the mtDNA variation be- ing methods (topology versus distance) and possibly the tween G. gondae populations collected from the two host genes under study. Therefore, we used a complementary species might already give a clue on possible popula- approach by absolute and relative timing of speciation tion differentiation. According to Futuyma and Moreno in both host and parasite (see below). (1988), the ultimate cause for specificity is linked to be- havior. However, an experiment by Clayton et al. (2003) highlighted the importance of host defence in reinforc- Is Cospeciation More Frequent in the Host-Specific ing host specificity of feather lice infecting birds. To test and Monophyletic Gill Parasites? whether parasite behavior rather than host defence is the TreeMap 1.0 and nontimed 2.0β suggested a signif- underlying factor in this system, all host species can be icant fit between the host and parasite trees, in con- mixed in a subsequent experiment and monitored for in- trast to the timed analysis and ParaFit. According to fection levels. If the result is different from the previous the topology-based programs the number of cospeci- experiment, this is likely due to preferential host colo- ation events was higher than expected by chance, al- nization by the parasite. though host-switching events were suggested as well. The P. minutus complex consists of the closely related The 11 proposed solutions by TreeMap 2.0β could be gobies P. minutus, P. lozanoi, and P. norvegicus. Sympatry divided in either predominantly cospeciating or pre- of P. minutus and P. lozanoi might explain why they share dominantly host-switching scenarios. To discriminate so many Gyrodactylus species (e.g., G. gondae, G. rugien- between both scenarios, relative or absolute calibra- soides, G. cf. micropsi). However, the fact that P. lozanoi also 716 SYSTEMATIC BIOLOGY VOL. 54 harbors a unique parasite G. longidactylus (Geets et al., TABLE 2. Probabilities computed by ParaFit (999 permutations); 1998) proves that host-switching does not always occur the H0 hypothesis of the global test is that the Gyrodactylus fin parasites (group B) colonized the goby hosts at random (see bottom of the table); whenever possible. This might also be seen as cases of in- the H0 hypotheses in the tests of the individual host-parasite association ertia (Paterson and Banks, 2001), where a parasite species links is that the link under evaluation has no significant contribution to fails to speciate despite speciation of its host (Johnson the global fit (Legendre et al., 2002). Probabilities in bold are significant et al., 2003). at a level of 5%. Pomatoschistus norvegicus, which is ecologically more isolated by occupying the deeper sections of the con- Gyrodactylus species Host species Probability tinental shelf down to 200 m (Miller, 1986), is caught G. branchicus Gasterosteus aculeatus 0.001 G.cf. micropsi 1 Pomatoschistus lozanoi 0.037 infected with a similar species here referred to as G. G.cf. micropsi Pomatoschistus minutus 0.015 cf. longidactylus. Morphological analysis showed dis- G.cf. micropsi P. lozanoi 0.034 tinct differences between both species (personal data); G. micropsi Pomatoschistus microps 0.062 a molecular comparison is in progress. This indicates the G.cf. micropsi 2 P. minutus 0.015 . possibility of another host-associated species complex as G cf. micropsi 2 P. lozanoi 0.036 G.cf. longidactylus Pomatoschistus norvegicus 0.062 previously described for the species pairs G. rugiensis–G. G. rugiensis P. microps 0.069 rugiensoides and G. micropsi–G. cf. micropsi (Huyse and G. rugiensoides P. minutus 0.013 Volckaert, 2002). G. rugiensoides P. lozanoi 0.023 G. rugiensoides Pomatoschistus pictus 0.030 Are There Instances of Sympatric Speciation? Global test 0.001 The dominant mode of speciation in the present sys- tem appeared to be allopatric. However, one instance of recent intrahost speciation might have led to Gyrodacty- ciation events in groups A and B fall within the Late Pleis- lus cf. micropsi and G. cf. micropsi 1, who are each other’s tocene period. In case of the hosts, speciation is thought closest relative found on the same host. The programs to have occurred in the early Pliocene, with exception of TreeMap and TreeFitter also pointed to the importance of the P. minutus complex that originated in the Pleistocene historical duplication or intrahost speciation events. This (Huyse et al., 2004b). In this case, all recent speciation can be expected from the biology and population struc- events in Gyrodactylus should be the result of duplica- ture of Gyrodactylus characterized by high host specificity tion or host-switching. This is in agreement with ParaFit and autoinfection of their hosts. It has been stated that if and the comparison of genetic distances based on ho- sympatric speciation occurs, it is most likely in parasite mologous gene fragments in host and parasite (ITS 1). groups like monogeneans (Brooks and McLennan, 1993; The divergence in the hosts was much greater than the Gusev, 1995; Poulin, 1998, 2002). A recent molecular phy- distance between the respective parasites (Table 1). If ITS logenetic study showed that the main process of Dactylo- 1inGyrodactylus does not evolve at a slower rate than in gyrus () diversification on cyprinid hosts was fish, this would indicate a more recent speciation of the ˇ parasites then of the hosts. Due to the rapid generation intrahost speciation (Simkov´aetal., 2004). The authors µ found that sister taxa on the same host tend to occupy time (24 to 72 hours) and a mean body size of 350 m, niches differing at least in one niche parameter. Because we might actually expect a relatively faster rate for Gy- monogeneans infect long-lived hosts, they may generate rodactylus compared to the host (as illustrated for COI several generations on the same host individual, ensur- mtDNA, see below). ing a deme continuous in time and space and reinforced According to the clock calibration, the split between G. by inbreeding to such an extent that they generate incip- arcuatus and the Gyrodactylus species. found on the go- ient species. Also, gene flow might initially be severed bies occurred 1.82 to 2.05 Mya. Such timing would sup- by asexual or parthenogenetic reproduction (Brooks and port the scenario of a host-switching from three-spined McLennan, 1993) or niche differentiation and specializa- stickleback onto Pomatoschistus gobies (see Fig. 2A and tion (de Meeus ˆ et al., 1998). C), possibly when sharing the same refugium, e.g., the Bay of Biscay during the Pleistocene ice ages (Garcia- Marin, 2001). This event might have been followed by Tentative Timing of Events a combination of host-switching and cospeciation with It is always difficult to estimate the evolutionary rate the respective gobies. for a given gene fragment, especially when no fossil data The speciation between G. gondae and G. flavescensis is are at hand. In case of the tiny, soft bodied Gyrodacty- estimated at 0.18 Mya. However, because the speciation lus flatworms, only well-documented (recent) vicariance between their hosts Gobiusculus flavescens and P. minu- events can provide critical information. Recently such an tus is much older (Huyse et al., 2004b), cospeciation, as attempt has been made by Zietara and Lumme (2002). suggested by Figure 2B, is rejected. Rather, the scenario Their estimate was based on the divergence of G. aphyae described in Figure 2C, characterized by successive host- living on the minnow Phoxinus phoxinus on opposite switching events between the goby species, is favored. sides of the Baltic–White Sea watershed. Connecting this Such phylogenetically constrained host-switching pro- divergence with the well-documented divergence time duce nonrandom patterns of phylogenetic congruence, of the host, the authors obtained a rate of 5.5% per million which has also been described in plant-insect interactions years (My). If this rate is applied, most of the (sister) spe- (Percy et al., 2004). Information on the absolute timing 2005 HUYSE AND VOLCKAERT—COEVOLUTION: DISTANCE- VERSUS TOPOLOGY-BASED PROGRAMS 717 of host and parasite speciations may help to distinguish ACKNOWLEDEGMENTS this pattern from genuine cospeciation. T.H. has been supported by a scholarship of the Institute of Sci- The occurrence of very closely related taxa on the sis- entific and Technological research (IWT-Vlaanderen) and by a travel ter species P. microps and P. marmoratus might falsely ap- grant from FWO Vlaanderen. We would like to thank Y. Desdevises pear as two cospeciating nodes, heavily influencing the and J. Van Houdt for the much appreciated discussions and J. Cotton and M. Charleston for superb help with TreeMap 2.0β.Y.D. and M.C. topology-based programs. G. branchialis V, collected in provided helpful comments on an earlier draft of this manuscript. This Venice from P. marmoratus, differs only in 0.2% (complete study has been funded by the Belgian government (Federal Science ITS region) from G. branchialis B, collected in Belgium Policy, contract MN/DD/42 and EV/01/25C). from P. microps, whereas the hosts differ in 3% based on 16S mtDNA. In a study by Huyse et al. (2003), it is estimated that the two hosts speciated about 3 Mya. REFERENCES If the parasites speciated at the same time, the ITS re- Bakke, T. A., P. D. Harris, and J. Cable. 2002. Host specificity dynamics: Observations on gyrodactylid monogeneans. Int. J. Parasitol. 32:281– gion should evolve at a rate of 0.07% per My. Because 308. the internal transcribed spacers are generally thought to Brooks, D. R., and D. A. McLennan. 1993. Parascript. Parasites and the be rapidly evolving because they are noncoding (Hillis language of evolution, Smithsonian Institution Press, Washington, et al., 1996), and the rate in Gyrodactylus is estimated to be DC, and London. much higher (5.5% per My, see above), this association Buchmann K., and T. Lindenstrøm. 2002. Interactions between mono- genean parasites and their fish hosts. Int. J. Parasitol. 32:309–319. cannot be the result of a cospeciation event. Excluding Cable, J., and P. D. Harris. 2002. Gyrodactylid developmental biology: one taxon from a single association (G. branchialis V), the Historical review, current status and future trends. Int. J. Parasitol. amount of cospeciation events indicated by TreeMap 1.0 32:255–280. is not significant anymore. This suggests that the signal Cable, J., E. C. G. Scott, R. C. Tinsley, and P. D. Harris. 2002. Behavior favoring transmission in the viviparous monogenean Gyrodactylus of cospeciation was, at least partly, generated by phylo- turnbulli.J.Parasitol. 88:183–184. genetically constrained host-switching events. Charleston, M. A. 1998. Jungles: A new solution to the host/parasite phylogeny reconciliation problem. Math. Biosci. 149:191–223. Clark, M. A., N. A. Moran, P. Baumann, and J. J. Wernegreen. Future Perspectives 2000. Cospeciation between bacterial endosymbionts (Buchnera) and a recent radiation of aphids (Uroleucon). Evolution 54:517– Attempts have been made to find another homolo- 525. gous marker. Currently, the 16S mtDNA is being se- Clayton, D. H., S. E. Bush, B. M. Goates, and K. P. Johnson. 2003. Host quenced for Gyrodactylus species. Distances between a defense reinforces host–parasite cospeciation. Proc. Natl. Acad. Sci. 350-bp fragment sequenced in G. branchialis and G. ar- USA 100:15694–15699. Cribb T. H., L. A. Chisholm, and R. A. Bray.2002. Diversity in the Mono- cuatoides reached 3.7%, whereas their respective hosts, genea and Digenea: Does lifestyle matter? Int. J. Parasitol. 32:321–328. P. microps and P. pictus, differed 3.6% in a 550-bp frag- de Meeus, ˆ T., Y. Michalakis, and F. Renaud. 1998. Santa Rosalia revis- ment of 16 mtDNA (Kimura 2-parameter model). This ited: ‘Or why are there so many kinds of parasites in the garden of is according to expectations because the COI mtDNA early delights?’ Parasitol. Today 14:10–13. has been estimated to evolve at a rate of 28% per million Desdevises Y, S. Morand, O. Jousson, and P. Legendre. 2002. Co- evolution between Lamellodiscus (Monogenea; Diplectanidae) and years in G. salaris (Meinil¨aetal., 2004). Therefore, the COI Sparidae (Teleostei): The study of a complex host-parasite system. mtDNA gene in parasites might be too variable for re- Evolution 56:2459–2471. liable phylogeny reconstruction. Although homologous Farris J. S., M. Kallersjo, A. G. Kluge, and C. Bult. 1995. Constructing a genes are a requirement for meaningful comparisons of significance test for incongruence. Syst. Biol. 44:570–572. Futuyma, D. J., and G. Moreno. 1988. The evolution of ecological spe- the rate of evolution in hosts and parasites, it is not, by cialization. Annu. Rev. Ecol. Syst. 19:207–233. itself, enough. Any statement about overall difference in Geets, A., G. Malmberg, and F. Ollevier. 1998. Gyrodactylus longidactylus rate requires that the same gene is evolving in a corre- n. sp., a monogenean from Pomatoschistus lozanoi (de Buen, 1923) from lated fashion in both host and parasite, such that branch the North Sea. Syst. Parasitol. 41:63–70. lengths in the host are a constant proportion of those Gusev, A. V. 1995. Some pathways and factors of monogenean mi- croevolution. Can. J. Fish. Aquat. Sci. 52:52–56. in the parasite. The life history features of Gyrodactylus Hafner, M.S., and S.A. Nadler. 1988. Phylogenetic trees support the (mean body size of 350 µm and generation time in op- coevolution of parasites and their hosts. Nature 332:258–259. timal conditions of 24 hours) are likely to have a higher Harris, P. D. 1993. Interactions between reproduction and population impact on the evolutionary rate of a gene than its func- biology in gyrodactylid monogeneans—A review. Bull. Fr. Peche Pis- cic. 1:47–65. tion. Therefore we argue that the present data set, consist- Harris, P. D., A. P. Shinn, J. Cable, and T. A. Bakke. 2004. Nominal ing of nonhomologous genes but with a clock-like rate, species of the genus Gyrodactylus von Nordmann 1832 (Monogenea: should be suitable to address questions on cospeciation ), with a list of principal host species. Syst. Parasitol. between Gyrodactylus and its fish hosts. 59:1–27. In conclusion, the evolutionary history of Gyrodactylus Hillis, D. M., B. K. Mable, and C. Moritz. 1996. Applications of molec- ular systematics. Sinauer Associates, Sunderland, Massachusetts. species and their goby hosts has gone through periods of Huelsenbeck, J. P., and F.Ronquist. 2001. MRBAYES.Bayesian inference cospeciation, followed by (and intermingled with) peri- of phylogeny. Bioinformatics 17:754–755. ods of successive host-switching events. These switches Huyse, T., V., Audenaert, and F. A. M. Volckaert. 2003. Speciation have most likely been triggered by alternating Pleis- and host-parasite relationships in the parasite genus Gyrodactylus (Monogenea, Platyhelminthes) infecting gobies of the genus Po- tocene ice ages. Eventually a new period of coevolution matoschistus (Gobiidae, Teleostei). Int. J. Parasitol. 33:1679–1689. may appear, but considering the colonizing capacities of Huyse, T., and G. Malmberg. 2004. Molecular and morphological Gyrodactylus, future host-switches are bound to happen. comparisons between Gyrodactylus ostendicus n. sp. (Monogenea: 718 SYSTEMATIC BIOLOGY VOL. 54

Gyrodactylidae) on Pomatoschistus microps (Krøyer) and G. harengi Percy, D. M., R. D. M. Page, and G. C. B Cronk. 2004. Plant-insect inter- Malmberg on Clupea harengus membras L. Syst. Parasitol. 58:105–113. actions: Double-dating associated insect and plant lineages reveals Huyse, T., G. Malmberg, and F. A. M. Volckaert. 2004a. Discrimination asynchronous radiations. Syst. Biol. 53:120–127. of four new species of Gyrodactylus (Monogenea, Platyhelminthes) Posada, D., and K. A. Crandall. 1998. ModelTest: Testing the model of on gobiid fish: Combined DNA and morphological analyses. Syst. DNA substitution. Bioinformatics 14:817–818. Parasitol. 59:103–120. Poulin, R. 1992. Determinants of host-specificity in parasites of fresh- Huyse, T., J. Van Houdt, and F. A. M. Volckaert. 2004b. Paleoclimatic water fishes. Int. J. Parasitol. 22:753–758. history and vicariant speciation in the “sand goby” group (Gobiidae, Poulin, R. 1998. Evolutionary ecology of parasites: From individuals to Teleostei). Mol. Phylog. Evol. 32:324–336. communities. Chapman and Hall, London. Huyse, T., and F. A. M. Volckaert. 2002. Identification of a host- Poulin, R. 2002. The evolution of monogenean diversity.Int. J. Parasitol. associated species complex using molecular and morphometric 32:245–254. analyses, with the description of Gyrodactylus rugiensoides n. sp. Ricklefs, R. E., S. M. Fallon, and E. Bermingham. 2004. Evolutionary (Gyrodactylidae, Monogenea). Int. J. Parasitol. 32:907–919. relationships, co-speciation and host switching in avian malaria par- Johnsen, B. O., and A. J. Jensen. 1991. The Gyrodactylus story in Norway. asites. Syst. Biol. 53:111–119. Aquaculture 98:289–302. Roy, B. A. 2001. Patterns of association between crucifers and their Johnson, K. P., R. J. Adams, R. D. M. Page, and D. H. Clayton. 2003. flower-mimic pathogens: Host jumps are more common than coevo- When do parasites fail to speciate in response to host speciation? lution or cospeciation. Evolution 55:41–53. Syst. Biol. 52:37–47. Schmidt, H. A., K. Strimmer, M. Vingron, and A. von Haeseler. 2002. Kearn, G. C. 1994. Evolutionary expansion of the Monogenea. Int. J. TREE-PUZZLE: Maximum likelihood phylogenetic analysis using Parasitol. 24:1227–1271. quartets and parallel computing. Bioinformatics 18:502–504. Legendre, P., Y. Desdevises, and E. Bazin. 2002. A statistical test for Shimodaira, H., and M. Hasegawa. 1999. Multiple comparisons of log- host-parasite coevolution. Syst. Biol. 51:217–234. likelihoods with applications to phylogenetic inference. Mol. Biol. L¨oytynoja, A., and M. C. Milinkovitch. 2001. SOAP: Cleaning multiple Evol. 16:1114–1116. alignments from unstable blocks. Bioinformatics 17:573–574. Simkov´ˇ a, A., S. Morand, E. Jobet, M. Gelnar, and O. Verneau. 2004. Meinil¨a, M, J. Kuusela, M. S. Zietara, and J. Lumme. 2004. Initial steps Molecular phylogeny of congeneric monogenean parasites (Dactyl- of speciation by geographic isolation and host switch in salmonid ogyrus): A case of intrahost speciation. Evolution 58:1001–1018. ∗ pathogen (Monogenea: Gyrodactylidae). Int. J. Swofford, D. L. 2001. PAUP : Phylogenetic analysis using parsimony Parasitol. 34:515–526. (and other methods), version 4.0b8. Sinauer Associates, Sunderland, Miller, P. J. 1986. Gobiidae. Pages 1019–1085 in Fishes of the north- Massachussetts. eastern Atlantic and the Mediterranean. (P. J. P. Whitehead, M.-L. Thompson, J. D., T. J. Gibson, F. Plewniak, F. Jeanmougin, and D. G. Bauchot, J.-C. Hureau, J. Nielsen, and E. Tortonese, eds.). UNESCO, Higgins. 1997. The Clustal X windows interface: Flexible strategies Paris. for multiple sequence alignment aided by quality analysis tools. Nu- Nosil, P. 2002. Transition rates between specialization and generaliza- cleic Acids Res. 25:4876–4882. tion in phytophagous insects. Evolution 56:1701–1706. van Herwerden, L., D. Blair, and T. Agatsuma. 1999. Genetic diversity Page, R. D. M. 1993. Parasites, phylogeny and cospeciation. Int. J. Par- in parthenogenetic triploid paragonimus westermani. Int. J. Parasitol. asitol. 23:499–506. 29:1477–1482. Page, R. D. M. 1994. Parallel phylogenies: Reconstructing the history Webb, C. J. 1980. Systematics of the Pomatoschistus minutus complex of host-parasite assemblages. Cladistics 10:155–173. (Teleostei: Gobiidae). Philos. Trans. R. Soc. Lond. Ser. B 291:201–241. Page, R. D. M., D. H. Clayton, and A. M. Paterson. 1996. Lice Weiblen, G. D., and G. L. Bush. 2002. Speciation in fig pollinators and and cospeciation: A response to Barker. Int. J. Parasitol. 26:213– parasites. Mol. Ecol. 11:1573–1578. 218. Xia, X., and Z. Xie. 2001. DAMBE: Software package for data analysis Page, R. D. M., and M. S. Hafner. 1996. Molecular phylogenies and in molecular biology and evolution. J. Hered. 92:371–373. host-parasite cospeciation: Gophers and lice a model system. Pages Zietara, M. S., and J. Lumme. 2002. Speciation by host-switching and 255–270 in New uses for new phylogenies (P. H. Harvey, A. J. L. adaptive radiation in a fish parasite genus Gyrodactylus (Monogenea, Brown, J. M. Smith, and S. Nee, eds.). Oxford University Press, Gyrodactylidae). Evolution 56:2445–2458. New York. Paterson, A. M., and J. Banks. 2001. Analytical approaches to measuring First submitted 9 August 2004; reviews returned 8 November 2004; cospeciation of host and parasites: Through a glass, darkly. Int. J. final acceptance 21 January 2005 Parasitol. 31:1012–1022. Associate Editor: Adrian Paterson