• Tania Jenkins,

    1. NERC Centre for Population Biology & Division of Biology, Department of Life Sciences, Imperial College London, Silwood Park, Ascot, Berkshire SL5 7PY, United Kingdom
    2. E-mail: tania.jenkins@gmail.com
    3. Computational and Molecular Population Genetics, Institute for Ecology and Evolution, University of Bern, Baltzerstrasse 6, CH-3012 Bern, Switzerland
    Search for more papers by this author
  • Gavin H. Thomas,

    1. NERC Centre for Population Biology & Division of Biology, Department of Life Sciences, Imperial College London, Silwood Park, Ascot, Berkshire SL5 7PY, United Kingdom
    2. School of Biological Sciences, University of Bristol, Woodland Road, Bristol BS8 IUG, United Kingdom
    Search for more papers by this author
  • Olof Hellgren,

    1. Department of Animal Ecology Animal Ecology Building Lund University, SE-22362 Lund, Sweden
    Search for more papers by this author
  • Ian P. F. Owens

    1. NERC Centre for Population Biology & Division of Biology, Department of Life Sciences, Imperial College London, Silwood Park, Ascot, Berkshire SL5 7PY, United Kingdom
    Search for more papers by this author


Host traits, such as migratory behavior, could facilitate the dispersal of disease-causing parasites, potentially leading to the transfer of infections both across geographic areas and between host species. There is, however, little quantitative information on whether variation in such host attributes does indeed affect the evolutionary outcome of host–parasite associations. Here, we employ Leucocytozoon blood parasites of birds, a group of parasites closely related to avian malaria, to study host–parasite coevolution in relation to host behavior using a phylogenetic comparative approach. We reconstruct the molecular phylogenies of both the hosts and parasites and use cophylogenetic tools to assess whether each host–parasite association contributes significantly to the overall congruence between the two phylogenies. We find evidence for a significant fit between host and parasite phylogenies in this system, but show that this is due only to associations between nonmigrant parasites and their hosts. We also show that migrant bird species harbor a greater genetic diversity of parasites compared with nonmigrant species. Taken together, these results suggest that the migratory habits of birds could influence their coevolutionary relationship with their parasites, and that consideration of host traits is important in predicting the outcome of coevolutionary interactions.

The emergence and transmission of diseases to novel hosts is thought to have led to the loss of a substantial number of species over the last century (van Riper 1986; Atkinson et al. 1995; Lanciotti et al. 1999; Daszak et al. 2000), with migratory birds in particular posing an interesting challenge with respect to the spread of disease (Olsen et al. 2006; Altizer et al. 2011). Understanding the host traits and ecological factors that can influence host–parasite interactions is an area of prime importance, particularly in the face of potential alteration of host and parasite ranges due to climate change (Brooks and Hoberg 2007). Although the role of dispersal on host–parasite coevolution has received much attention (Holt 1997; Lively 1999; Gandon 2002; Morgan et al. 2005), relatively little work has investigated the extent of coevolution in host species with different migratory behaviors, or looked at the effect of other host traits on coevolution.

Coevolution is the process of counteradaptation between two phylogenetically unrelated groups (Thompson 2005), and can lead to rapid diversification (Nunn et al. 2004; Thrall et al. 2007). In the case of host–parasite coevolution, the ability of the parasite to develop in the host (Combes 2001), the number of potential hosts present in the community that a parasite can infect (Fox 1988; Thrall et al. 2007), and the extent of dispersal in the system shapes the strength of coevolution (Lively 1999; Morgan et al. 2005). Strong coevolutionary associations between hosts and parasites can also result in cospeciation or codivergence, where a diversification event of the host is followed by a diversification event of the parasite—this is detected as a congruence between the phylogenies of the two groups (Hafner and Page 1995; Page 2003). Cospeciation is prominent in systems with high parasite to host specificity or when the hosts or parasites show limited dispersal. Cases where this process has been shown include the chewing lice and pocket gopher (Hafner et al. 1994; Hafner and Page 1995) and lice and bird systems (Page et al. 1998; Hughes et al. 2007). On the other hand, parasite generalism, defined as the ability to infect multiple host species, is usually associated with a weakening of cospeciation (Desdevises et al. 2002; Weckstein 2004, but see Banks et al. 2006). Community attributes such as the diversity of the available host community and their range distribution could play key roles in predicting spillovers and host shifts between sympatric species (Antonovics et al. 2002; Keesing et al. 2006). This could result in a weaker coevolutionary association and reduction in the signal of cospeciation (Weckstein 2004; Brooks and Ferrao 2005). It is likely therefore that different host ecological variables can alter the nature of coevolutionary interactions, although this has not yet been thoroughly investigated.

Host dispersal in particular is expected to affect the extent of coevolutionary interactions in many ways. Limited dispersal has been shown in experimental and theoretical situations to enhance coevolution locally by promoting local adaptation (Lively 1999; Gandon 2002; Morgan et al. 2005). On a macroevolutionary scale, dispersal during periods of range expansion following climatic change has been suggested to result in parasite diversification (Hoberg 1995) and host switching (reviewed in Hoberg and Brooks 2008). In monogean parasites of fish, dispersal of the mobile larval stages of the parasite—that could result in host switching—has been implicated in the lack of observed cospeciation (Desdevises et al. 2002). Despite this phylogenetic incongruence, the high specificity in the fish-monogean system is indicative of a strong ecological association, and could be due to the presence of functionally similar traits in phylogenetically distant hosts (Brooks 1979; Desdevises et al. 2002). Host specificity is an ecological index that describes the distribution of a parasite on the hosts in a community (Clayton et al. 2004) and, as in the fish-monogean example, is not necessarily indicative of phylogenetic congruence because specific parasites may descend from a generalist ancestor (Clayton et al. 2004). Dispersal events, resulting in a weakening of cospeciation, can also be aided by vectors or carriers: for example, in the case of Bruelia spp of lice and their Columbid bird hosts, dispersal of parasites is enabled by hippoboscid flies (Johnson et al. 2003).

Migratory birds are particularly interesting in the context of host–parasite coevolution because they are likely to encounter diverse avian communities at their wintering, breeding, and stop-over sites, and have been shown to carry a high diversity of blood parasites, often at a high prevalence (Figuerola and Green 2000; Hubalek 2004; Perez-Tris and Bensch 2005). Consequently, migrant birds are likely to act as carriers of parasites into new geographic regions, enabling greater opportunities for host switching (Garamszegi and Møller 2007). They could also increase parasite population size by allowing allopatric populations to be connected. On a longer time scale, this could result in a type of “diffuse coevolution,” whereby a parasite can coevolve with more than one host species (Futuyma and Slatkin 1983; Iwao and Rausher 1997), for example, by means of alternation between hosts, depending on which host has most reduced defenses at a particular time (Thompson 2005).

Recent studies of malaria parasites have found strong evidence for coevolution with both avian (Ricklefs et al. 2004) and primate (Garamszegi 2009) hosts, marked by a high frequency of host switching events, presumably mediated by the more generalist feeding preferences of the dipteran vectors. Leucocytozoon spp. are a diverse group of blood parasites, closely related to the avian malaria, and transmitted by Simulium blackfly species. Leucocytozoa are found in a wide range of hosts on all continents, apart from Antarctica, often at high prevalence (Valkiunas 2005), making them an ideal system for the study of the biogeography of avian blood parasites (Jenkins and Owens, 2011) and the role of migration on the extent of coevolution.

A range of cophylogenetic methods have been proposed to study cospeciation, originally based on assigning costs to coevolutionary events such as duplication, host switching, and sorting (for definitions see Page and Charleston 1998; Page 2003). A lack of congruence between host–parasite trees can be interpreted as host switching or as a combination of the above events (Page and Charleston 1998; Paterson and Banks 2001). Such techniques were originally designed for the “one host-one parasite” scenario, but when faced with multihost parasites the problem becomes too computationally intensive to provide an optimal solution (Legendre et al. 2002). An alternative approach, ParaFit (Legendre et al. 2002), uses matrices of phylogenetic distances and host–parasite associations and can deal with multihost parasitism, unbalanced numbers of hosts and parasites, large phylogenies, and phylogenetic uncertainty or polytomies (Hughes et al. 2007). ParaFit provides tests for overall fit, interpreted as congruence, between two phylogenies and assesses the contribution of individual host–parasite associations to the fit. This method has corroborated results from other approaches that have demonstrated cospeciation between pocket gophers and their chewing lice (Legendre et al. 2002), and Pelecaniform birds and their lice (Hughes et al. 2007). ParaFit has also produced results consistent with those other programs showing a lack of cospeciation between malaria parasites and their avian hosts (Ricklefs et al. 2004) and monogean parasites and their fish hosts (Desdevises et al. 2002).

In this study, we first test for evidence of a significant fit between the phylogenies of Leucocytozoon parasite species and their avian hosts, using the ParaFit method. Second, we assess whether host traits have an impact on coevolutionary associations. Finally, we focus on the effects related to host migratory behavior and test the prediction that migrant species harbor a more genetically diverse parasite fauna. We therefore use a quantitative framework to determine the effect of host traits on the coevolutionary relationships between hosts and parasites and suggest that the host traits affecting coevolutionary associations—at a macroevolutionary scale—have so far been overlooked.

Materials and Methods


The majority of Leucocytozoon spp. sequence data, with bird host species affiliation was collated by OH (Hellgren et al. 2008; Hellgren et al. 2009). TJ conducted a full search of GenBank in June 2008 to find all other deposited Leucocytozoon Cytb lineages that had information on which host they were isolated from, and for sources where there was published material, they were checked for details of other possible hosts (Merino 2008). To this dataset, we added our own sequence data of 14 lineages found in Parus major and/or Cyanistes caeruleus (Jenkins and Owens, 2011). Parasites were included in the dataset if we had sequence information from the hosts, in at least one of four protein coding mtDNA genes: Cytb, CO1, ND2, ND3 (obtained from GenBank by GT). The final dataset comprised 97 parasite lineages found in 53 host species, 34 Old World passerines, 11 New World passerines, 7 Old World nonpasserines, and 1 New World nonpasserine (see Table S1). A host–parasite association matrix was constructed to describe which parasites occurred in which hosts.

For the bird hosts, each gene was aligned by eye by GT in Se-Al version 2.0 (available at http://tree.bio.ed.ac.uk/software/seal/) and the aligned sequences were concatenated into a single alignment with a total length of 4064 base pairs. Each species had data for at least one of the four genes (coverage Cytb: 94% of taxa; CO1: 77%; ND1: 55%; ND2: 91%). The Cytb parasite sequences were also aligned by eye in Se-Al version 2.0.


We constructed both host and parasite phylogenies using BEAST version1.6.1 (Drummond and Rambaut 2007) with an uncorrelated lognormal relaxed molecular clock and Yule speciation priors. We used a General Time Reversible with gamma distributed among site rate variation (GTR +Γ) model with data partitioned by gene (COI, Cytb, ND2, ND3) and by codon position ([1 + 2], 3) for each gene. We explicitly allowed among partition rate variation. We defined the Galloanserae, which comprised five species in our dataset, as the outgroup. As the trees were not calibrated, the mean substitution rate was set to 1.0, thus fixing the rate of substitution of internal nodes at substitutions/site.

Four independent chains of 50 million generations (for the hosts) and 30 million generations (for the parasites) were run, with trees being sampled every 10K generations. The four separate runs were inspected in Tracer version1.4 (available at http://beast.bio.ed.ac.uk/tracer) to assess convergence and estimate the burn-in. Convergence was determined by visual examination of the plots and estimates of effective sample size (ESS), with ESS over 200 for each parameter indicating that the run had converged. After removing the burn-in, the log and tree files were combined using LogCombiner version 1.4.7 (Drummond and Rambaut 2007).

For the hosts, we subsequently repeated the phylogenetic reconstructions, as described above, for each of the four gene partitions separately.


Analyses of fit between the two trees were conducted using the ParaFit method (Legendre et al. 2002), implemented in the R package “ape” (Paradis et al. 2004). We preferred this to other methods of testing for congruence, due to the fact that ParaFit can cope with unbalanced numbers of hosts and parasites, large phylogenies that are not completely resolved and multihost parasitism. This is important because host switching is known to be common between avian blood parasites and their hosts (Ricklefs et al. 2004) rendering conventional cophylogenetic methods inappropriate (discussed in Johnson et al. 2003).

The ParaFit method requires matrices that describe the associations between hosts and parasites (A), the parasite phylogeny converted into a distance matrix (B), and the host phylogeny, converted into a distance matrix (C). Principal coordinates of the two matrices are used to calculate a fourth matrix (the fourth corner matrix, D) describing the associations between host and parasite phylogenies (following notation in Legendre et al. 2002):


The overall congruence between the host and parasite phylogenies, with respect to the presence–absence matrix of associations, is summarized by the sum of squares of the elements di,j of matrix D:


Statistical significance of the ParaFitGlobal statistic is assessed by comparison with a null distribution created by permuting the row entries of the host–parasite association matrix A while holding the phylogenetic matrices (B and C) fixed. This is analogous to random permutations of the hosts associated with each parasite. The ParaFitGlobal statistic for the observed associations is then compared to a uniform random distribution of ParaFitGlobal statistics derived from the permutation procedure (from 999 iterations). If the observed ParaFitGlobal is larger than the randomized statistic in ≥95% of cases then the null hypothesis of independent evolution can be rejected.

The significance of individual host–parasite phylogenetic associations can be assessed by testing their contribution to the ParaFitGlobal statistic. This is achieved by sequential removal of specific host–parasite associations from the matrix A and recalculation of the ParaFitGlobal statistic. If an important association is removed, then the ParaFitGlobal statistic should decrease significantly compared to that obtained from the true matrix A. Statistical significance is again assessed using a permutation process similar to that described above (see Legendre et al. 2002 for full details).


Analyses were repeated with 100 different host and parasite tree combinations, to account for potential phylogenetic error. We also performed analyses on the maximum clade credibility trees with branches collapsed where node support was less than 0.95. To account for biases due to uneven sampling of the regions and bird phylogeny, four separate analyses were conducted at different phylogenetic and geographic scales: (1) global analysis: all host–parasite associations, irrespective of geography; (2) global passerine analysis: all passerine hosts and their associated parasites; (3) Old World analysis: all host–parasite associations in Europe, Middle East, and Africa; and (4) Old World passerine analysis: all passerine hosts and their associated parasites in Europe, Middle East, and Africa. Overall, this resulted in 400 ParaFit analyses being conducted.

Host specificity is an ecological index that describes the contemporary distribution of parasites on a host (Clayton et al. 2004). To ensure that the ParaFit results were not simply a result of host specificity and that they were due to deeper coevolutionary history, we repeated the ParaFit analyses, maintaining the same associations but randomizing the tips on the parasite phylogeny. This maintained the same number of associations per parasite while breaking up the evolutionary structure among them. If our results were due to host specificity, we would expect the fit between host and parasite phylogenies to remain significant even after this randomization procedure and for distantly related parasites to contribute significantly to the ParaFitGlobal statistic. In contrast, if the results were at least partially due to coevolution, the randomization procedure would lead to the fit between the host and parasite phylogenies being no longer significant. This is because the coevolutionary hypothesis is dependent upon a particular pattern of phylogenetic divergence among the parasites, whereas host specificity is an ecological metric not based on a particular phylogenetic pattern. To conduct the randomization, we extracted the tip labels from the parasite trees, randomly sampled them, and replaced them onto the phylogeny, therefore maintaining the tree topography and the number of host–parasite associations intact. We randomized the parasite tips only, as parasites infect hosts and not vice versa, so as to not interfere with the permutation procedure of the hosts that occurs during the ParaFit procedure. This was repeated on all 100 host–parasite combinations for the full dataset.

To test whether our results were affected by uneven sampling: that is, some hosts had many parasites, whereas others had only one, we randomly sampled parasites occurring in each host and repeated the ParaFit analyses with only one parasite per host.


Our analyses of which host traits affected host–parasite associations was conducted in two stages. First, we determined whether each host–parasite association contributed significantly to the global ParaFitGlobal statistic. An association was considered to be significant if the median P value across 100 runs was less than 0.05, and if it was significant in more than 80% of analyses. For each host, we then counted the number of significant associations of parasites with the host versus the number of nonsignificant associations. Then, we analyzed the proportion of significant associations as a binomial response variable, fitted against the following predictor variables: migratory behavior of the host (migrant, partial migrant, or resident), breeding, wintering and total range size, body size, modal clutch size, habitat type, water use, diet type, colony nesting, group living, nesting habit and location, whether the species displays parental care, mating system and whether the species spends a portion of its time in the tropics (for explanations see legend Table 2). The binomial responses were analyzed using generalized linear mixed models with binomial errors and a logit link, fitted using the lme4 package in R (Bates and Sarkar 2007), with each variable in turn as the fixed effect. To account for phylogenetic effects, we also included bird family and order as random effects. Note that standard phylogenetic comparative methods such as independent contrasts and phylogenetic generalized least squares are not suitable for non-Gaussian responses and could not be used for these analyses. There were also not sufficient sister pairs to conduct a sister-pair comparison.

Table 2.  AIC of generalized linear mixed effects models of the following predictors on measure of overall association strength of each host with its parasites: body size (body weight of female, where available, or otherwise of adult of either sex), clutch size (modal number of eggs in a clutch), habitat score (1 = scrub/open habitat, 2 = mixture of forests/woodlands and scrub/open habitat, 3 = forests and woodlands only), water score (1 = habitat associated with freshwater, 0 = habitat not associated with water), diet score (0 = vegetable matter or nectar; 1 = omnivore; 2 = invertebrates and/or vertebrates), colony nesting (0 = no, 1 = yes), group living, for example, flocking (0 = no, 1 = yes), nesting habit (open; cup; dome; closed), nest location (ground, vegetation, hole), parental care (f = female only; mf = male and female; m = male only), mating system (m = monogamous; p = polygamous), tropical, that is, if the species inhabits the tropical region at any time in its life history, including winter range, (0 = no, 1 = yes). Data derived from appendices 1 and 2 of Bennett and Owens (2002) and from a global dataset of geographic distributions of birds (see Orme et al. 2005).
EffectAIC P
Null model77.4 
Sampling intensity 78.9    0.51
Log body size73.12   0.012
Clutch 79.4    0.94
Habitat78.66   0.39
Proximity to water 77.82    0.2
Diet type68.6<0.002
Colony living 78.1    0.253
Group living76.43   0.085
Nesting habit 73.92    0.024
Nesting location80.69   0.698
Parental care 76.14    0.07
Mating78.63   0.25
Tropical 79.4    0.97
Migration73.9   0.024
Geography 79.37    0.86
Breeding range79.4   0.93
Wintering range 91.6    0.99
Total range79.25   0.69

If we identified a significant association between a host trait and the proportion of significant host–parasite associations, this trait was tested in combination with the other significant variable(s) to determine whether the effect was confounded by the other variable(s). All analyses were corrected for sampling intensity, that is, the total number of each host species in each study found to be infected with Leucocytozoon spp., by including this as a covariate in the models. Significance was assessed by means of a likelihood ratio test (LRT Crawley, 2007) and model fit by calculation of Akaike weights (Wagenmakers and Farrell 2004). For the latter, models were fitted with all combinations of variables that were significant in the LRT and the Akaike Information Criterion (AIC) was extracted from each mixed model output and used to calculate Akaike weights The final model was the one with Akaike weights closest to 1, and included the combination of variables in that model contributed the most to model fit.


To understand whether hosts of a particular migratory class differed in their overall signal of cospeciation, we repeated the ParaFit analyses for residents and their parasites, and migrants and their parasites separately, across the 100 host–parasite trees. We further randomized the parasite trees to test that our results were not due to host specificity alone, and repeated the ParaFit analyses as described in the previous section. We also repeated ParaFit analyses having subsampled the tanglegram to include only one parasite per host.

To investigate whether migration was significantly over/underdispersed in the host phylogeny, and whether this was significantly different to a uniform random distribution, we calculated a median value of mean pairwise distance (MPD, Webb et al. 2002) between all migrant hosts on the tree, across 10% of trees in the posterior distribution. MPD was calculated using the appropriate function in the R package “picante” (Kembel et al. 2009). We then compared the observed median MPD value to the upper and lower 2.5% quantiles of a distribution based on a uniform random assignment of migration onto the phylogeny, to assess the extent of phylogenetic dispersion of migration. We also used MPD, calculated across 10% of trees from the posterior distribution, as a measure of the phylogenetic diversity of parasites found in each host: we chose this measure as it is not correlated with species richness. We used the median MPD to test whether migrants had a higher phylogenetic diversity of parasites than nonmigrants and partial migrants. Median parasite MPD was analyzed in a linear mixed effects model, with all predictor variables separately and bird order and family as random effects. This analysis was subsequently repeated with three outliers removed (Lanius collurio and Ficedula hypoleuca with exceptionally low MPD and Turdus viscivorus with exceptionally high MPD).

To correct for phylogenetic nonindependence among host species, where more closely related host species could have more similar values for MPD of their parasites due to shared ancestry, we repeated our analyses using a phylogenetic generalized linear model, incorporating the maximum likelihood value of the phylogenetic scaling parameter, λ (Freckleton 2002). As before, we repeated these analyses, first with only passerine hosts and second, with only passerine hosts from Europe and Africa.

All analyses were conducted in the statistical programming environment R version 2.12 (R Development Core Team, 2011).



The phylogenetic trees for both hosts and parasites converged at a global optimum, having reached the same likelihood space in four runs (hosts) and three out of four runs (parasites). The ESS values of the combined parameters for the four separate runs of the host and parasite trees were greater than 200, indicating that the phylogenies had converged (for maximum clade credibility trees see Figs. S1 and S6). The phylogenetic reconstructions for the separate gene trees of the hosts gave comparable results (Fig S2–S5).

Overall, we found a significant fit between the host and parasite phylogenies with the observed ParaFitGlobal statistic being significantly higher than expected by chance (P < 0.001 from 999 permutations). This was consistent across the 100 combinations of host and parasite trees drawn from the posterior distribution, repeated across the four phylogenetic and geographic scales (Table 1A). It was also consistent when analyses were repeated on the maximum clade credibility trees where nodes with less than 0.95 posterior support were collapsed (P < 0.001 from 999 permutations). However, when we looked at individual host–parasite associations, we found that only 40% of host–parasite associations on average significantly contributed to the ParaFitGlobal statistic (range 38–43% across the 100 analyses of the four data subsets; Fig. 1, Table 1A). There was a tendency for some associations to become more significant when phylogenetic resolution was improved (see Table S1). Testing for host specificity by randomizing the parasite tips, although maintaining the same host trees and association matrix showed no evidence for an overall significant fit between the two trees (Table 1C). This result was robust to some hosts having a higher number of parasites. Further reducing the number of associations to one parasite per host gave a significant ParaFit statistic in most (88%) analyses.

Table 1.  Values of the ParaFitGlobal statistic and significance as tested with 999 permutations. Also included are the range of the number of links found to be positive across 100 analyses in (A) the four separate geographic and phylogenetic analyses; (B) analyses of migrant and resident hosts separately; and (C) analyses conducted with randomized parasite tips. Median and range of values calculated across 100 host–parasite tree combinations.
(A) Geographic   TotalMedianRange
+ taxonomicMedian ParafitMedian PRange Pnumber of linkslinks positivelinks positive
  1. All = full analysis; Old world = data subset not including the Americas; passerine = analysis of data subset only including passerine birds; Old passerine = data subset only including passerine birds not found in the Americas.

Passerine 18,862 <0.001 <0.001–0.008 124 53 25–67
Old world14,749<0.001<0.0011184939–56
Old passerine 66,370 <0.001 <0.001–0.008 105 40 26–50
(B) Resident migrant      
 Resident 56,800 <0.001 <0.001  62 41 24–49
 Migrant3328   0.61   0.36–0.90 51 1 0–2
(C) Randomized       
 Global17,744   0.22   0.006–0.97139 8 0–31
 Resident 8787    0.25 <0.001–0.92  62  3  0–24
Figure 1.

Phylogenetic associations between Leucocytozoon parasites and their avian hosts. Green lines indicate associations that contributed significantly to the ParaFitGlobal statistic. Red lines indicate associations that did not significantly change the ParaFitGlobal test statistic. Blue colored host branches are from hosts that were classed as long-distance migrants. Points to the left indicate the taxonomic bird families for which we had more than two species represented in our dataset.

Although some parasites were found in multiple host families, the associations contributing significantly to overall fit were due to links between the parasite and a single host family (Fig. 1). For example, associations among parasites ChL2 and ChL3 and three hosts belonging to the host family Fringillidae, were found significant according to the ParaFit test; however, the associations of the same parasites with hosts from the families Certhiidae and Tyrannidae were not. Similarly, the associations between parasite REB11 and six members of the host family Ploceidae were significant, but those of REB11 with members of the Cisticolidae or Zosteropidae were not. An exception to this trend was the association between LCAP1 that was significant according to the ParaFit test, both with the nonpasserine host Tetrao tetrix (Tetraonidae) and the Old World passerine Phylloscopus trochilus (Phylloscopidae).


Feeding habit, body size, nesting habit, and migration were the only variables to significantly affect coevolution (assessed as the number of significant vs. nonsignificant associations, see Table 2), yet in a model with all four variables, body size and nesting habit no longer remained significant. Migratory behavior remained significant (P = 0.001): compared to migrant and partial migrants, nonmigratory host species harbored six times more significant associations with their parasites (Fig 2A). There were significant differences between all three classes of migratory behavior, as a model fitted with three levels of migration had the lowest AIC (see Fig 2A). Diet type also remained significant, with insectivorous birds harboring fewer parasites that significantly contributed to the ParaFitGlobal statistic (P < 0.001). Finally, a model with migration and feeding habit as fixed effects had the lowest AIC and this model accounted for 99% of Akaike weights (Table 3). To ensure that our results were robust to taxonomic biases, we repeated analyses on two subsets of the data: passerines alone and Old World passerines. The results on the effect of migration were consistent in the passerine analysis (migration P < 0.002; diet type P = 0.02). In the Old world passerine analysis migration was still significant (P = 0.04), but the final model also contained group living (P < 0.01), rather than diet type.

Figure 2.

The effect of host migratory behavior on host–parasite coevolution. The effect of migratory behavior of host species on (A) the proportion of host–parasite associations that contribute significantly to the ParaFitGlobal statistic. (B) Mean pairwise distance of parasites in each host migration type. There was a significant effect of host migratory behavior (P < 0.001). Factor level reduction revealed that migratory species were not significantly different from partially migratory species (P = 0.25) and that migratory and resident species were not significantly different (P = 0.15), although the model with the two classes of migratory behavior combined had the lowest AIC. When the outliers were removed, migratory species were marginally significantly different from resident species (P = 0.06) but were not different from partially migratory species (P = 0.24). Values presented are parameter estimates (best linear unbiased predictors) and their associated standard errors, from an analysis with migration type as the only fixed effect.

Table 3.  Factors affecting the strength of hosts–parasite coevolution. Summary of AIC, weighted AIC (wAIC), and pseudo R2 (model deviance—deviance of model with random intercept)/model deviance, for the models with all combinations of significant predictors. AICs were extracted from the model output. A full model refers to one with all variables that were tested.
ModelAICδAICwAICPseudo R2
Migration+diet type55.45  0.990.42
Full model 65.06  9.61   0.0014 0.48
Diet type68.613.15  0.00810.18
Migration 73.9 18.45 <0.0001 0.11
Random intercept7721.55<0.0001


When we repeated ParaFit analyses with parasites of residents and their hosts only, we found that there was a significant overall fit between the two phylogenies (P < 0.001), due to the significant contribution of a median of 66% significant associations (Table 1B). Links that were significant in the previous analysis, that included migrants, were also significant in this one. To exclude the effects of host specificity, we further repeated the ParaFit analyses with parasites with randomized tips and resident host trees and found there to be no evidence for a significant fit between the two (Table 1C). The ParaFit analyses between parasites of migrants and their hosts revealed that there was not a significant overall fit between the two phylogenies (Table 1C). Repeating the analyses with one randomly selected parasite per resident host also gave a significant overall fit for the residents but not for the migrants (see Table S2). Taken together, these results indicate that the significant overall fit between the two phylogenies was only due to significant associations between parasites of residents and their hosts.

Analyses of MPD showed that migration was not significantly over- or underdispersed with respect to phylogeny (median MPD = 0.93; 95% CI (i.e., lower and upper 2.5% quantiles from sampling): [0.76, 1.10]. Similarly, resident and partial migrant behaviors were not clustered with respect to phylogeny (resident: median MPD = 1.00, 95% CI: [0.78, 1.06]; partial migrant: median MPD = 0.77, 95% CI: [0.71, 1.14]). There was a significant effect of migratory behavior on median parasite MPD found in each host (P < 0.001, see Fig 2B), with migrants harboring a more phylogenetically diverse parasite fauna. One of our indices of the social behavior of hosts, group living, was also found to be significant in this analysis, with species that displayed flocking behavior having a significantly lower MPD of parasites than those that did not (P < 0.001), suggesting that these species tend to harbor less diverse parasites. A model including both migratory behavior and group living as factors accounted for 88% of Akaike weights, indicating the importance of both these variables (Table 4). In this model, there was an indication that group-living birds had lower MPD, with the effect being more prominent in residents. Note, however, that no migratory species were also classed as group living, and due to small sample size this potential interaction is not easily interpretable with our current dataset. Repeating the analysis using a phylogenetic generalized linear model revealed that there was a strong phylogenetic signal in the analysis (λ= 1, P < 0.001). Despite this strong signal, the effect of migratory behavior was still evident (F = 5.33, P = 0.01), as was that of group living (F = 19.36, P < 0.0001).

Table 4.  Factors affecting parasite diversity (calculated as mean pairwise distance) within host species. Summary of AIC, weighted AIC, and pseudo R2 (model deviance—deviance of model with random intercept)/model deviance, for the models with all combinations of significant predictors. AICs were extracted from the model output. A full model refers to one with all variables that were tested.
Group living and migration−89.35-880.20
Migration −85 4.35 11 0.15
Group living−79.429.93<0.0010.08
Null −67 22.35 <0.001  

Our results relating to parasite MPD in migrants were not confounded by sampling intensity (F = 0.02, P = 0.60) and none of the other host traits were significant in this analysis. When the analysis was repeated using passerines only, the effects of migration (F = 5.59, P = 0.01) and group living (F = 23.91, P < 0.001) on MPD were still present. Further restricting the analyses to passerines found in the Old World, still resulted in a significant effect of migration (F = 4.36, P = 0.03), and of group living (F = 15.58, P = 0.001).


Our results showed an overall significant congruence between the phylogenies of Leucocytozoon blood parasites and their avian hosts, which we interpret as evidence for coevolution. We also found that the host traits such as feeding type and migratory behavior had an impact on the congruence between hosts and parasites. We found that the observed congruence is restricted to resident hosts and their parasites. In addition, migratory host species also harbored more phylogenetically diverse parasite communities than did resident species. Our results show that there is a signature of host migratory behavior on host–parasite coevolution at this scale, supporting other work highlighting the importance of dispersal in both microevolutionary interactions between parasites and their hosts (Lively 1999; Gandon 2002; Morgan et al. 2005) and on macroevolutionary processes (Johnson et al. 2003; Clayton et al. 2004).

We found that migrants harbored a higher diversity of parasites, consistent with previous work (Figuerola and Green 2000; Hubalek 2004; Perez-Tris and Bensch 2005). This may be because migrants stop over at a larger number of sites, some of which may have high numbers of circulating vectors able to transmit haematozoan parasites (Figuerola and Green 2000). In addition, as migration is an energetically costly activity, resources may be traded-off from immune defense, making it likely that migrants are more susceptible than residents. This would result in migrants harboring a higher diversity and prevalence of parasites (reviewed in Ricklefs et al. 2005). This suggests that exposure to diverse parasites can be a hidden cost of migration (Waldenstrom et al. 2002).

Why does host–parasite coevolution appear less pronounced in relationships involving migratory host species? It may be that migrants are not coevolving with a particular group of parasites because they are exposed to a wider range of parasites, allowing greater opportunities for host switching to occur (Desdevises et al. 2002; Hoberg and Brooks 2008) thus preventing strong coevolution (Thompson 2005). In contrast, a closely related host species may not have evolved migration, making it beneficial for the evolution of resistance alleles to a particular parasite lineage(s), leading to a more specific immune response (Hellgren et al. 2009) thereby resulting in closer coevolution (Thompson 2005). Migrant hosts, on the other hand, may be unable to afford to mount a specific immune response to a particular strain (Taylor et al. 1998), and could evolve a more generalist immune system (as suggested in Hellgren et al. 2009). This may result in an overall lack of a phylogenetic signal of coevolution between parasites of and their migratory hosts leading to the observed differences between migrant and resident birds.

Another explanation for the observed pattern is that migrant birds carry parasites from one geographic region to another, allowing gene flow between allopatric populations of the parasite to occur, consequently enlarging the gene pool, and so precluding coevolution with a host in a single area (Banks and Paterson, 2005). One way to test for this would be to look for evidence of parasites occurring in resident host species from opposite ends of the migration route as well as in migratory species. In our dataset, we found no cases of this; however, the fact that Hellgren et al. (2007) found two very distinct, but non monophyletic, parasite faunas of Leucocytozoon between Europe and Africa suggests that such parasite dispersal events may be rare in ecological time but have happened on a macroevolutionary time scale. Previous work has shown that even rare dispersal events can affect coevolutionary analyses, resulting in reduced cospeciation (Clayton et al. 2004).

Our results also indicate that relatively subtle differences in host migratory behavior may have an impact on the outcome of host–parasite interactions. For instance, we found more associations among parasites of partial migrants and their hosts that significantly contributed to the ParaFitGlobal statistic than among parasites of migrants and their hosts. This could be explained by the fact that partial migrants typically travel between areas that are likely to be within a single transmission area, such as within a single continent (Newton 2007) and so these host species encounter a single parasite fauna (Hellgren et al. 2007). In addition, for those host species wintering in northern temperate regions, there would be no winter transmission of Leucocytozoon due to a lack of circulating vectors (Crosskey 1990). With fewer transmission opportunities and a less diverse parasite fauna, there are likely to be more opportunities for host–parasite coevolution.

The ecological and coevolutionary relationships of insect vectors and the parasites they transmit are also likely to play an important role in determining whether coevolution between hosts and parasites occurs. Migratory bird species are likely to encounter a more diverse vector community, and therefore the establishment of a novel parasite translocated by a migrant host into a new area/host would depend on the feeding preferences of the local host vectors. If the vectors in the new area have broad feeding preferences, and the parasite is able to complete its life cycle in the vector, it is likely to be transmitted to the local hosts. Differences in the specificity and coevolutionary history of the host-vector and vector–parasite interaction could—in part at least—explain some of the differences between the patterns we have shown here among Leucocytozoa and birds and those previously reported by Ricklefs et al. (2004) among Haemoproteus/Plasmodium and birds. Both studies found evidence for some taxonomic conservatism in host usage to the family level, yet Ricklefs et al.'s study (2004) found a nonsignificant fit between the two phylogenies, which they suggest is indicative of host switching resulting in weaker coevolution. We suggest that the differences between the two studies in detecting a signal of coevolution could be due to differences in vector–host specificity. A recent study of mosquitoes in Vanuatu and New Caledonia found the same species of Haemoproteus and Plasmodium in more than one mosquito vector, indicating the potential for vector shifts in this system (Ishtiaq et al. 2008). In contrast, Leucocytozoon spp. are transmitted by black flies (Malmqvist et al. 2004; Valkiunas 2005; Hellgren et al. 2008) and recent work investigating the specificity of simuliid vectors of Leucocytozoon spp. has shown them to have relatively species specific feeding preferences (Hellgren et al. 2008), indicating the potential for a stronger association between the host and parasite. Further work to elucidate the specificity and coevolutionary relationships among hosts and vectors and parasites and vectors is needed.

Diet type also affected the number of significant associations between hosts and parasites, with insectivorous birds harboring fewer significant associations with their parasites than omnivorous and nectar-feeding birds. This finding was however, due to the fact that there was only one species that was migratory and a plant feeder, and so may have been linked to the effect of migratory behavior. Flocking behavior was also found to affect within host parasite diversity, with flocking birds harboring a significantly lower genetic diversity of Leucocytozoon spp. This may be because birds often flock in groups of closely related species (Sridhar et al. 2009), thus reducing encounters with more distantly related birds and their parasites. This would prevent host switching between parasites of more distantly hosts.

We conducted a number of additional analyses to confirm the robustness of our key conclusions. These included analyses across multiple host and parasite trees, subsets of our trees at different geographic and phylogenetic scales, and subsets of the associations to include one parasite per host. Our results remained qualitatively unchanged in these additional analyses, although there was a suggestion that the results of the ParaFit analyses were sensitive to phylogenetic resolution.

We also conducted a series of further analyses to investigate whether our results were likely to be due to host specificity rather than coevolution. If this had been the case, then we would expect to find distantly related parasites contributing significantly to the ParaFit statistic. We did not however find this pattern (Fig. 1). We also tested for host specificity by randomizing the tips of the parasite trees and investigating the impact of this on the strength of host–parasite associations. If host specificity was responsible for the patterns found in this study, then such randomization should not have had a substantial impact on host–parasite associations. We found, however, that this resulted in a nonsignificant fit between host and parasite associations, consistent with coevolution. Further subsampling of our trees to include only one host–parasite association gave similar results, indicating that the effect was not driven by uneven sampling in our dataset. These tests therefore indicate that host specificity alone was not a sufficient explanation for our findings. However, we acknowledge that host specificity is also necessary for coevolution to occur, and that our results therefore imply evolutionary host specificity sometimes resulting in parasite lineages that radiate among closely related hosts: we detect this as a signature of coevolution.

In conclusion, this study has investigated the extent of phylogenetic congruence between the phylogenies of Leucocytozoon blood parasites and their avian hosts. We have shown that despite a strong overall fit between host and parasite phylogenies, this relationship is largely restricted to associations involving resident host species. We also found that migrants carried a higher phylogenetic diversity of parasites. The implication of these findings is that novel parasites may potentially be able to colonize new areas by “hitchhiking” on migrant hosts because such hosts have a less specific relationship with their parasite fauna. We exercise caution in this interpretation, however, as the successful establishment of the parasite in the new area is likely to depend on the capacity of invertebrates to vector these parasites and the ability of the parasite to infect a large enough number of hosts for there to be a viable number of infections allowing transmission to occur. Finally, we suggest that analyzing host traits, such as migration ability with respect to coevolution is highly informative and adds a lot to our understanding of the factors that affect the evolutionary ecology of host–parasite interactions.

Associate Editor: A. Read


We would like to thank Eric Allan for comments and help with R code. We would also like to thank Albert Phillimore, Staffan Bensch, and Michael Tristem for comments on earlier drafts of this manuscript as well as three anonymous reviewers. This work was funded by a Ph.D. grant awarded to Tania Jenkins at the Natural Environment Research Council (UK) Centre for Population Biology.