S. Bensch, Department of Animal Ecology, Ecology Building, Lund University, S-223 62 Lund, Sweden. Tel.: +46 46 2224292; Fax: +46 46 2224716; E-mail: email@example.com
1We have used molecular methods to unravel a remarkable diversity of parasite lineages in a long-term population study of great reed warblers Acrocephalus arundinaceus that was not foreseen from traditional microscopic examination of blood smears. This diversity includes eight Haemoproteus and 10 Plasmodium lineages of which most probably represent good biological species.
2Contrary to expectation, the relative frequency of parasite lineages seemed not to change over the 17-year study period and we found no effects of the parasites on a male secondary sexual ornament (song repertoire size) and two measures of fitness (adult survival and production of recruited offspring).
3We discuss whether the absence of fitness consequences of the parasites might relate to the fact that we have studied the host at the breeding sites in Europe, whereas the transmission seems to take place at the wintering sites in Africa, where the naïve birds encounter the parasites for the first time and the resulting primary infections likely make them sicker than during the chronic phase of the infection.
4The prevalence of the three most common lineages appeared to fluctuate in parallel with a periodicity of approximately 3–4 years. Theoretical models based on intrinsic interactions between parasite antigen and host immune genes cannot explain such dynamics, suggesting that knowledge of extrinsic parameters such as vector distribution and alternative hosts are required to understand these patterns.
The view prevailing 50 years ago that parasites and diseases were unimportant to understand population fluctuations (Lack 1954) is now radically different. Research over the past decades have made it increasingly clear that parasites play a decisive role for host population dynamics and evolution (Ricklefs 1992; Hudson, Dobson & Newborn 1998; Lively & Dybdahl 2000).
To understand mechanisms of parasite–host coevolution and population dynamics of the parasites and the host, research must address the temporal and geographical variation of parasite communities, patterns of interaction between parasite lineages and the presence and distribution of alternative hosts for these lineages (Bensch & Åkesson 2003; Fallon et al. 2005). Elucidating these patterns would be important also for sexual selection theory. In the influential hypothesis put forward by Hamilton & Zuk (1982), it is assumed that the expression of elaborate male traits indicates parasite resistance. In this model, temporal changes of the composition of a parasite community are assumed to drive frequency-dependent changes of host immune genes, and females choosing ornamented males as mates will produce disease resistant and ornamented sons. Some comparative between-species analyses have found support for a positive association between parasite prevalence and plumage brightness (Read 1987; Scheuerlein & Ricklefs 2004) but others have failed to find this relationship (Read & Harvey 1989; Ricklefs et al. 2005). Within species, however, there is now convincing empirical evidence of negative associations between male ornamentation and parasite loads, and a female preference for ornamented and parasite-free males (Milinski & Bakker 1990; Møller 1990). Such within-species relationships require genetic variation for fitness and it has been proposed that positive parent–offspring regression of fitness can be maintained if there is cyclic coevolution between hosts and parasites (Hamilton & Zuk 1982). This appears particularly likely if the host population is exposed to many different parasites that show nonsynchronized temporal cycles, each with moderate or minor fitness effects as governed by variation at different nonepistatic loci (Eshel & Hamilton 1984).
It is clear that the role of avian blood parasites in studies of host–parasite interactions, population dynamics and sexual selection would benefit from detailed studies of the dynamics and potential fitness consequences of specific parasite species, rather than treating them collectively as anonymous members of species-rich genera. Recent molecular analyses of avian malaria parasites indicate that different lineages indeed compete with one another for access to the available hosts (Fallon et al. 2004). Studies that we have been undertaken in great reed warblers Acrocephalus arundinaceus (Linnaeus), show that the species is infected by one common lineage of Haemoproteus (GRW1) and two common lineages of Plasmodium (GRW2 and GRW4) along with a handful of more rare lineages of both genera (Bensch et al. 2000; Waldenström et al. 2004). The three common lineages (GRW1, GRW2 and GRW4) are more divergent in the cytochrome b gene than are the human and chimpanzee parasites P. falciparum and P. reichinowi (Escalante et al. 1998) that recently have been demonstrated to exhibit species-specific interactions with the human immune system in cell cultures (Martin et al. 2005). Also, experiments with human malaria parasites have demonstrated only little cross-resistance between species, suggesting that their frequent co-occurrence in single hosts has led to antigenic divergence (Richie 1988).We therefore assume that parasites with well differentiated cytochrome b genes (such as GRW1, GRW2 and GRW4) are exhibiting antigenic differences that will influence the effectiveness of the host immune system. At least the three common malaria parasites (GRW1, GRW2 and GRW4) have active transmission in Africa as they have been found also in African resident songbirds coexisting with the great reed warbler in its wintering area (Waldenström et al. 2002). On the contrary, there is no evidence of transmission of these parasites in the European breeding quarters as they have not been detected in first year birds still on breeding grounds (in about 100 juveniles from Sweden in August and 30 from Bulgaria in September, only SGS1 has been found, S. Bensch et al. unpublished). In our Swedish study population, the Plasmodium lineage GRW2 has been shown to be positively associated both with MHC diversity and a specific MHC allele (Westerdahl et al. 2005), suggesting a link between the host's resistance genes and the likelihood and outcome of infections. Also worth noting is that the Plasmodium lineage GRW4 is genetically identical at the cytochrome b gene to Plasmodium relictum (Ricklefs & Fallon 2002; Beadell et al. 2006), the species of avian malaria on Hawaii, supposedly responsible for extinction and range contraction of many endemic bird species during the twentieth century (van Riper et al. 1986).
In the present study, we explore a long-term data set of great reed warblers collected at a breeding site in Sweden during 17 years to look at the diversity and temporal variation of avian malaria parasites. For the three most common parasite lineages we use maximum likelihood to estimate the probabilities of gain and loss of infections. We attempt to identify patterns of parasite diversity within and across generations, i.e. whether there is stable, directional or cyclic variation in the parasite community. This is to address whether there is apparent competition between the lineages (Gupta & Galvani 1999; de Roode et al. 2005) and to what extent the host is likely to keep its immune system tuned to the parasites (stable community) or if it is more likely to be lagging behind (cyclic or instable variation). For malaria parasites, fitness costs to the host are supposed to be most apparent during the primary infection when the naïve individual has its first encounter with the parasite, and in long-distance migratory birds as the great reed warbler, this often occurs in the winter quarter. This part of the life cycle is inherently difficult to study in migratory songbirds, so we have restricted this study to investigate the fitness consequences of these tropically transmitted parasites in the birds’ summer quarter, when they circulate in the blood as chronic infections with less (if any) fitness consequences for the infected individual. Thus, a failure to find fitness effects would only support the hypothesis that chronic infections have limited fitness consequences for the bearer, whereas the parasite still can have profound effects on naïve individuals.
To investigate the relationships between host phenotypic quality and parasite prevalence we test whether male song repertoire size, which is the main secondary sexual ornament in great reed warblers (Hasselquist, Bensch & von Schantz 1996; Nowicki et al. 2000), is higher in unparasitized than in parasitized males. We also investigate to what extent chronic infections of these parasites have negative effects on host fitness, by examining two components of fitness; number of offspring recruits and between-year survival.
fieldwork and study population
The study was carried out at lake Kvismaren (59°10′N, 15°25′E), south Central Sweden, between 1987 and 2003. Adults were captured in mist nets, aged into one of two age classes (1 year or at least 2 years old) based on a combination of leg colour, eye colour and presence of tongue spots (Bensch et al. 1998). A blood sample of c. 20–80 µL was collected from each captured bird, which was stored in SET-buffer until DNA extraction and molecular analyses of avian malaria parasites (Waldenström et al. 2004). We took several morphological measurements and before its release, each bird was given a unique combination of one aluminium and two to three colour rings that later could be used for identification in the field. Each year, we captured or identified (from colour rings) almost all (> 95%) breeding adults in the population. Between early May and mid July, we made daily visits to each male territory (range; 14 territories in 1987 and 33 in 1997) to score whether males were singing ‘short song’ or ‘long song’, a behaviour that accurately reflects their present mating status and serves as a guide to locate and identify newly settled females (Hasselquist & Bensch 1991; Bensch & Hasselquist 1992). We spent more time in territories where males recently had switched from long song to short song, until we could observe and identify a nest building females and locate the nest. At least 5 min of spontaneous long song was recorded from each male for sonographic analyses of song repertoire size as previously described (Hasselquist 1998). All nests were inspected regularly to obtain information on clutch size, hatching day, hatching success and fledging success (Bensch 1996). Chicks were given an aluminium ring at day 9 post-hatch when we also took blood sample for paternity analyses (Hasselquist, Bensch & von Schantz 1995; Arlt et al. 2004). About 15% of the locally fledged chicks return as breeders to Kvismaren, which is approximately 50% of the chicks that survive to breeding age (Hansson et al. 2002). The breeding population consisted on average of approximately 50% breeders returning from the previous year, 25% first-time breeders that were locally hatched and 25% first-time breeders that were immigrants. Measuring return rate to the study area is a good approximation for adult survival rate because we individually identify almost all birds every year and very few adults change breeding site between years once becoming a breeder (Hansson et al. 2002).
Genomic DNA was extracted and diluted to a polymerase chain reaction (PCR) working concentration of 25 ng µL−1 (Waldenström et al. 2004). A partial segment (479 bp) of the cytochrome b gene of avian malaria parasites of the genera Plasmodium and Haemoproteus was amplified in two steps by a nested PCR approach, exactly as described in Waldenström et al. (2004). For the first PCR we used the primers HAEMNF (5′-CATATATTAAGAGAATTATGGAG-3′) and HAEMNR2 (5′-AGAGGTGTAGCATATCTATCTAC-3′) of which final product 1 µL was taken as a template in a second PCR with the primers HAEMF (5′-ATGGTGCTTTCGATATATGCATG-3′) and HAEMR2 (5′-GCATTATCTGGATGTGATAATGGT-3′). We used multiple negative controls and a previous experiment has verified that this nested PCR protocol has a high (99·8%) repeatability (Waldenström et al. 2004). Positive amplifications, as evaluated by running a 2·5 µL aliquot of the final PCR on agarose gels, were sequenced directly with the primer HAEMF. Electropherograms were manually inspected and the proof-read sequences were compared with an alignment containing the accumulated number of unique cytochrome b sequences. Potentially new and unique sequences were checked by sequencing the fragment with the primer HAEMR2. We checked the electropherograms for double nucleotide peaks to infer possible cases of mixed infections of two different parasites. Such cases were resolved by TA cloning as described elsewhere (Pérez-Tris & Bensch 2005a). Dilution experiments of mixes between lineages (GRW1 and GRW2) have demonstrated that concomitant infections can be detected with this method when differences in the level of parasite intensities are within two orders of magnitude (Pérez-Tris & Bensch 2005a). The method will, however, score individuals as single infected when differences in intensities are larger and in cases where the underlying sequence may favour one lineage over the another (Valkiûnas et al. 2006).
Let Xi,t be the infectious state of bird i (i = 1:N) in year t (t = 1987–2003), where state is 1 if infected and 0 if not infected. We treated the parasite-lineage-specific probabilities of gain and loss of infection as independent and the following analyses were carried out for GRW1, GRW2 and GRW4 separately.
If the probability of a bird being in a given state (infected or not infected) at a given time is only dependent on its state at the previous time step, we have a two-state Markov model with transition matrix
( eqn 1)
where p01 and p10 are the annual probabilities of gain and loss of infection, respectively. As the lineages GRW1, GRW2 and GRW4 appear to be transmitted in Africa (Waldenström et al. 2002; Westerdahl et al. 2005) we assume that the birds are parasite free when they leave the breeding area the year they are born. Hence, we restrict the analyses to the actual transitions starting from state 0, and for each transition we estimate the probability of moving from state i to state j in r time steps, where r ≠ 1 means that there are missing observations and the transition matrix needs to be raised to the rth power. To test for a difference in likelihood of gaining parasites during the first winter and the following winters, we expressed p01 as
( eqn 2)
for the first transition and
( eqn 3)
for the following transitions for each individual. Equation 1 is a logistic regression model – a standard model when the response variable is binary (e.g. Quinn & Keough 2002) – without any explanatory variables. The parameter α01 to be estimated can take any values in [–∞, +∞], which translates into an estimated probability of transmission between 0 (α01→–∞) and 1 (α01→+∞). We tested for a difference between the first and later transitions by calculating a 95% confidence interval for the parameter d, which is a measure of the difference (if any), using maximum likelihood. The 95% confidence interval is the range of the parameters for which the log-likelihood is within 1·92 of the maximum values of the log-likelihood (Hilborn & Mangel 1997). If the confidence interval contains zero, the null hypothesis of no difference cannot be rejected at the 5% level. A negative (positive) d would indicate a lower (higher) probability of gaining an infection in older birds. For more details on two-state Markov models and their use in ecology (Clark & Rosenzweig 1994).
The Markov model was also contrasted with a binomial model where the transitions are independent of state (hence, we estimate a single parameter p in the binomial distribution corresponding to the proportion of the population being infected). However, the Akaike Information Criteria corrected for small sample size (AICc; Burnham and Anderson 1998) was in strong favour of the Markov model for all parasite lineages (GRW1: Δi = AICc(i) − min(AICc) = 66·7, GRW2: Δi = 32·7, GRW4: Δi = 75·6).
We used generalized linear models (GLMs) to investigate potential fitness costs of the parasites and their effects on song repertoire size, by examining the effect of the state of infection for each bird during its first breeding year at our study site (mainly when 1 or 2 years old). We analysed two fitness correlates, number of offspring recruits and between year survival. The distribution of number of recruits was positively skewed (many individuals counted zero and a few scored higher values) and the survival data was binomially distributed. Thus, we used a logarithmic link function and negative binomial errors in the model of number of recruits, and a logistic link function and binomial errors for the survival data, following McCullagh & Nelder (1989). For song repertoire size, which was close to normally distributed, we used an identity link function and normal errors in the model. To allow tests of effects of infection when accounting for covariates of potential importance, we included arrival date, sex, year (16 years) and age (1 years or ≥ 2 years old) in the models. We used a backward elimination strategy and dropped parameters and interaction terms at P > 0·1. Type 3 tests were used. Least square means were transformed to an appropriate unit (e.g. from the logarithmic value to the linear value in the case of number of recruits). The GLMs were carried out in the SAS 8·02 GENMOD module (SAS 1990).
overall parasite prevalence and diversity
Among the 782 analysed samples (involving 470 individuals), the overall prevalence of Haemoproteus and Plasmodium combined was 43·2%. With the aid of DNA sequencing we found that the 338 samples from infected individuals consisted of 18 different parasite lineages, eight being Haemoproteus and 10 Plasmodium (Table 1). Three parasite lineages were found at frequencies > 5%, three lineages between 1 and 5% and the remaining 11 lineages at frequencies < 1%. The genetic distances between the parasite lineages (Fig. 1) varied from just one substitution (between GRW1 and GRW8) but were in most cases well beyond distances corresponding to good species (Bensch et al. 2004). We found 14 birds (1·8%) that carried concomitant infections, i.e. simultaneous infections of two parasite lineages. Given the frequency of the parasites (Table 1), we calculated the number of birds expected to be infected by at least two parasites to be 6% (47 birds) and the observed incidence of concomitant infections was significantly lower than this ( = 25·3, P < 0·001).
Table 1. Parasite lineages encountered in 782 samples of great reed warblers at lake Kvismaren (Sweden)
GenBank accession no. (% prevalence)
New morphological species (G. Valkiûnas et al. unpubl.).
The data included 183 individuals tested for malaria parasites in more than one (2–7) year. For each of the three most common malaria lineages (GRW1, GRW2 and GRW4) we estimated the likelihood of gaining parasites while being in African winter quarters. This was highest and similar for GRW1 and GRW4, and somewhat lower for GRW2 (Fig. 2). For GRW2 (d = −0·69) and GRW4 (d = −0·47) the likelihood of gaining parasites was higher (i.e. d was negative) during the first than during successive winters in Africa, though not statistically significant at the 5% level (as 0 was not excluded from the 95% confidence interval; GRW2: −1·56–0·06; GRW4: −0·99–0·03). However, especially for GRW4, the pattern was very close to significance. No such tendency was found for GRW1 (d = 0·22; 95% CI: −0·22–0·66). The likelihood of annual loss of infection (i.e. that a bird infected with linage x should be scored free from lineage x in the following year) was about 50% for all three lineages (Fig. 2) and independent of age.
In order to avoid statistical problems associated with pseudo-replication when investigating annual variation, we included the individuals only once, the first year when they were recorded as breeders (n = 444). Before examining temporal variation in parasite transmission it is important to evaluate if it is accurate to use the annual prevalence at the time of sampling as a measure of transmission during the preceding winter. If birds frequently get infected during their first winter in Africa and then carry the parasite for several years, the prevalence at the year of sampling would not directly correspond to the time of transmission unless the bird is 1 year old. The majority of the first time breeders were either 1 (n = 231) or 2 years old (n = 193), and whereas the 1 year olds must have got their parasites during the preceding winter, the 2 year olds may either have been infected during the preceding or next-preceding winter. If the prevalence is mainly a result of infections during the preceding winter, and assuming that there is annual variation in infectivity, we would expect the prevalence between these groups to be correlated during the year of sampling (1-year-old and 2-year-old first-time breeders). However, if the 2-year-old birds mainly were infected during their first winter, we would expect a stronger positive correlation if we used cohort rather than year of sampling in the comparisons. The correlation was actually negative when comparing the age groups at the year of sampling (r = −0·51) indicating that the infections were not gained during the previous winter. In contrast, Fig. 3 illustrates that there is a significant positive correlation if the comparisons are made within cohorts (r = 0·48, P = 0·03, one-tailed). In the following we therefore base the temporal analyses on the prevalence in first-time breeders grouped by cohorts.
Prevalence of the three most common parasite lineages (GRW1, GRW2 and GRW4) showed large variation between cohorts (Fig. 4). However, analyses using logistic regression did not support that the variation in prevalence between cohorts were larger than expected from sampling variance (GRW1, = 21·7, P = 0·19; GRW2, = 19·0, P = 0·33; GRW4. = 17·4, P = 0·43). Note, however, that the statistical power of these tests are fairly low. Annual prevalence of the three most common parasites and pooled groups of rare Plasmodium and Haemoproteus, respectively, did not correlate (all P > 0·15). However, the prevalence of GRW1 and GRW4 showed a similar pattern of change, apart from a phase shift for two cohorts starting 1993, when GRW4 peaked and GRW1 lagged 1 year behind. Combining GRW1, GRW2 and GRW4, the temporal prevalence pattern based on cohorts showed seemingly regular fluctuations around a positive trend (Fig. 5a). After removing the linear trend, the fluctuations were dominated by a periodicity of 3–4 years as revealed by the autocorrelation function (Chatfield 1999) and the partial autocorrelation function (Fig. 5b,c) had a significant (P < 0·05) negative lag 2 typical of many cyclic natural populations (Stenseth 1999).
parasites and correlates to fitness
Among the first-time breeders, we investigated whether overall prevalence and any of the three most common parasite lineages (GRW1, GRW2, GRW4) were associated to song repertoire size (males only) and two measures of fitness. Song repertoire size differed between 1-year and older males but not between parasitized and nonparasitized males (Table 2). Number of recruits was significantly associated with arrival date, sex, breeding year and age of the birds, and there was a nonsignificant trend for parasitized birds to get more recruits than unparasitized ones (Table 2). This nonsignificant association was mainly driven by GRW1-infected birds (Table 2). Survival rates differed between years but was not explained by status of infection (Table 2). Hence, the presence of malaria parasites among breeding adult great reed warblers seems not to be associated with low fitness. This conclusion should be robust as the analyses are based on quite large sample sizes (n = 442 for recruits and survival) and that the nonsignificant results were in the direction of higher fitness for parasitized than unparasitized birds (Table 2).
Table 2. Comparison of song repertoire size, number of recruits and survival of nonparasitized and parasitized great reed warblers breeding at Lake Kvismaren in 1987–2002 (n = 104 males for song repertoire size, n = 442 males and females for recruits and survival). Data are from generalized linear models (see text for details). The parameter values were similar for the different models (A–D) and therefore we present the range of χ2- and P-values
Theoretical analyses of host–parasite systems have demonstrated that genetic heterogeneity of the parasite should lead to independent dynamics of the lineages, or if there is cross-immunity between lineages, to negatively coupled periodicity (Gupta, Swinton & Anderson 1994). Our finding of the common parasite lineages in great reed warblers tending to show positive covariation, which is contrary to the expectation from the model by Gupta et al. (1994), might be explained by violation of some of the assumptions of the model, as developed for strains rather than species of malaria parasites. Further theoretical work has shown that genetically diverse parasites tend to self-organize into lineages with exclusive and nonoverlapping antigenic variants, excised by selection from the host immune system (Gupta & Galvani 1999). Such genetically different lineages are more likely to coexist in a host population if individual hosts respond in a lineage-specific manner (Gupta & Galvani 1999). Although we are dealing with species of malaria and not genetic strains, this reasoning seems to hold also for great reed warblers where the probability of being infected by lineage GRW2 depends on the MHC genotype of the host (Westerdahl et al. 2005).
We found fewer concomitant infections than predicted based on the prevalence of each parasite lineage. At present we cannot determine whether this is due to: (1) the tendency of the lineage with the highest parasitaemia to take over the PCR; (2) increased mortality among birds infected by two different parasites; (3) that the parasites are competing in the hosts and thus exclude each other; or (4) that there is cross-immunity between lineages, i.e. that infection by one lineage induces the host defence to also block infection success of other lineages.
Cyclic variation in malaria prevalence has previously been observed in lizards (Schall & Marghoob 1995) and humans (Molineaux 1988) with a periodicity in both cases about 10 years. Climatic variation, affecting vector populations, has been proposed as explanation. If vectors are governing parallel cycles of GRW1 (Haemoproteus) and GRW4 (Plasmodium), it requires a mechanism that control vectors of both Ceratopogonidae and Culiciidae (Atkinson & van Riper 1991). In order to build a model accounting for the intriguing parallel periodicity of these parasite lineages, we need more information about the vector species and the alternative hosts of the parasites, as well as the level of competition between parasite lineages and their effects on host fitness.
parasites or commensals?
One explanation that could account for the observed lack of fitness consequences of lineages of Plasmodium (GRW2 and GRW4) and Haemoproteus (GRW1) is that they are harmless to the great reed warbler and therefore should be recognized as commensals rather than parasites. Before accepting this explanation, however, one should bear in mind that we have studied the malaria parasites during a time when there is no transmission. The primary infection of GRW1, GRW2 and GRW4 most likely takes place at winter quarters in Africa (Waldenström et al. 2002), and the birds we are studying may be those that have recovered from the primary infection and managed to migrate back to Sweden (Westerdahl et al. 2005). The infection intensities of the two Plasmodium lineages (GRW2 and GRW4) in our study area are mainly below one infected erythrocyte per 10 000 (Waldenström et al. 2004). These levels of parasitaemia characterize chronic infections (Jarvi et al. 2003) when few if any fitness effects are expected (Valkiûnas 2005). Even species that suffer from high mortality at primary infections of malaria parasites, such as the Hawaii amakihi Haemignathus virens infected with Plasmodium relictum (i.e. GRW4), show no disease symptoms when chronically infected individuals are challenged with superinfections (Atkinson, Dusek & Lease 2001). For species of Haemoproetus (e.g. GRW1), the majority of published correlative studies (such as ours) have not found any effects on host fitness (Weatherhead 1990; Atkinson & van Riper 1991; Edler, Klump & Friedl 2004; Stjernman, Råberg & Nilsson 2004). However, experimental infections (Valkiûnas 1993) and medication experiments (Merino et al. 2000; Marzal et al. 2005) demonstrate that Haemoproetus spp. can have substantial fitness consequences for the host. Similarly, although human malaria parasites are responsible for millions of deaths annually, correlative studies in endemic areas characteristically do not reveal any costs of carrying malaria infections (Bruce et al. 2000). Hence, although we did not find any fitness costs for the lineages GRW1, GRW2 and GRW4, it is premature to conclude that they are benign to great reed warblers.
infections by rare lineages
Most of the parasite lineages were observed in less than a handful of individuals, a pattern detected in several recent studies of birds (Pérez-Tris & Bensch 2005b; Ishtiaq et al. 2006; Reullier et al. 2006). One could argue that because parasites are rare, they should have relatively little effect on the host population even if associated with stronger fitness consequences than the common lineages. This reasoning ignores that an unknown number of birds might have died from parasite infections before reaching the European breeding grounds. It is, however, quite unlikely that all these rare parasites can maintain their populations solely within the great reed warbler. A more likely scenario is that they mainly depend on other hosts and that the observed infections are sporadic ‘spillovers’ from such main hosts (Woolhouse, Taylor & Haydon 2001). Indeed, eight of the rare lineages in great reed warblers have been encountered in other host species (S. Bensch et al. unpublished). The summed spillover from the 14 rare lineages contributed to a prevalence of 6·5%. Hence, it seems as great reed warblers are exposed to a high risk of contracting malaria parasites that mainly reside in other host species. The fact that these parasites are rare might, however, indicate that the great reed warbler immune system can handle them effectively. This might be true for most or all of these rare lineages but it is well-known that ‘species jumps’ can result in changed virulence and be the starting point of emerging diseases (Woolhouse, Haydon & Antia 2005). Thus, although difficult to study statistically, rare parasites spilling over from other hosts may constitute important selective agents on the host's genetic immune system. Migratory birds, dividing their year in vastly different areas with different parasite communities (Møller & Erritzoe 1998) may be particularly exposed to such attacks of sporadic parasites, a potentially important agent of ‘cost of migration’ (Waldenström et al. 2002).
avian malaria parasites in hamilton–zuk models
Any good-genes model of sexual selection depends on a mechanism that maintains the genetic variation in the host population, which is linked to variation in the phenotypic traits in focus of mate selection (Andersson 1994). Although we found a remarkable diversity of avian malaria parasites (18 lineages) in the studied great reed warbler population, we failed to find support for several of the assumptions underlying the model proposed by Hamilton & Zuk (1982). First, the composition of parasite lineages did not seem to change over the study period. A similar conclusion was reached for island parasite communities in the West Indies where the composition of lineages largely remained the same over a 10-year period (Fallon et al. 2004). Second, the parasites did not seem to inflict fitness costs because the parasitized individuals produced as many surviving young as nonparasitized individuals. Third, song repertoire size, a sexually selected trait in this species (Catchpole 1986; Hasselquist et al. 1996), did not differ between parasitized and parasite-free males.
By studying chronic infections only, it is clear that we cannot properly evaluate the second and third point, as it is still possible that primary infections in Africa affect both survival chances and development of song performance. That song complexity is a particularly sensitive trait to developmental stress has recently been demonstrated in canaries Serinus canaria experimentally infected with Plasmodium relictum (Spencer et al. 2005). The assumption from the Hamilton and Zuk model of a negative cycling between parasites was not upheld by our data, but as we have only studied a house-keeping gene (cytochrome b) it is possible that such cycles are present but at a finer level within the different lineages. This possibility may be examined by looking at variation at fast evolving genes used by the parasites to evade the host immune system (Topolska et al. 2004), but the inherent rapid molecular evolution of such genes makes it difficult to design primers that can be of broad use.
The diversity of malaria parasites seen in the present study is not unique for great reed warblers (Hellgren 2005; Pérez-Tris & Bensch 2005b; Bonneaud et al. 2006; Ishtiaq et al. 2006). It is not clear whether the large parasite diversity as such will result in diversifying selection on the host immune genes. Such a scenario depends on the parasites having fitness consequences, a pattern we have not established for chronic infections of these parasites in great reed warblers. However this has conclusively been shown for several Plasmodium and Haemoproteus species in experimental studies (Valkiûnas 1993; McCutchan et al. 2004; Marzal et al. 2005) and we therefore stress that absence of fitness effects in correlative studies never should be taken as evidence of parasites being harmless. To examine whether the avian malaria community, with its high lineage diversity and the propensity of lineages to switch hosts, is driving parasite-mediated sexual selection as envisioned in good-genes models, we propose that successful strategies should combine long-term field studies with infection experiments on molecularly typed lineages.
Lots of people have help out with the field work over the years and in particular we wish to thank Bo Nielsen, Anna-Karin Olsson, Martin Stervander, Debora Arlt, Mikael Åkesson and Örjan Östman. We thank Isabella Cattadori and Robert E. Ricklefs for valuable improvements of the manuscript. The study was supported by the Swedish Research Council (VR), the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (Formas), Lunds Djurskyddsfond, Crafoordska Stiftelsen, Carl Tryggers Stiftelse and Stiftelsen Olle Engkvist Byggmästare. This is report 144 from Kvismare Bird Observatory.