At the individual level, larger birds were more likely to be infected with haemosporidians (Fig. 1). None of the population-level factors significantly predicted infection risk, which suggests that characteristics of malaria infection prevalence in Kenyan house sparrows are complex. Capture month, average annual precipitation, and population age (distance from Mombasa) all remained in the best model for infection prevalence, but no factor alone strongly predicted population prevalence (Table 3). Below, we interpret these unexpected patterns and discuss their implications for range expansion in HOSP and related species.
Individual level factors: body mass
We expected body mass would either be lower in infected individuals, consistent with experimental Plasmodium (SGS1) infections of HOSP (Palinauskas et al. 2008), or be unrelated to haemosporidian infection as was the case in North and South American populations of HOSP (Martin et al. 2007). However, our finding that larger birds were more likely to be infected has precedence: in a massive study of malaria infections of European passerines (14 812 individuals belonging to 74 species in 39 genera in 17 families), Scheuerlein and Ricklefs (2004) found that besides host species identity, body mass was one of the most important predictors of haemosporidian infection status; as in our study, larger birds were more likely to be infected. A similar pattern was also found in a smaller study of southern Missouri songbirds for Plasmodium and Haemoproteus parasites (Ricklefs et al. 2005). Why this pattern occurs is still unresolved. It may be that heavier birds are more attractive to vectors (Suom et al. 2010) due to increased chemical attractants (Takken and Kline 1989), such as carbon dioxide (Brady et al. 1997). Alternatively or additionally, heavier birds might have higher encounter rates with mosquitoes due to particular behaviors (e.g. increased foraging at mosquito-dense sites) or by chance because of greater surface area (Atkinson and Van Riper III 1991), which in turn increases the risk of exposure.
Another possibility is that mass and immunity are related in HOSP. If this is the case, our positive correlation between mass and probability of infection may be explained in one of two ways. First, larger sparrows may invest more resources in growth and less in immunity, making them more susceptible to haemosporidian infections (Sheldon and Verhulst 1996, Lochmiller and Deerenberg 2000, Ricklefs and Wikelski 2002). Or, because haemosporidian infections have negative fitness consequences in many passerines (Atkinson et al. 2000, Merino et al. 2000, Valkiūnas 2005, Palinauskas et al. 2008, Knowles et al. 2009, 2010), it is feasible that the pattern is a product of differential survival rates among infected individuals: only birds with sufficient resources to maintain large body masses also have the resources to survive chronic haemosporidian infections. A study of Spanish HOSP supports this explanation; Navarro et al. (2003) found that larger birds mounted stronger immune responses than smaller birds. Future studies comparing these hypotheses could reveal why larger birds are more likely to be infected.
We predicted that population age would be an important risk factor at the population level. Though there are assumptions regarding the effect of change in haemosporidian prevalence on a invasion success (Colautti et al. 2004), we expected that the mechanisms that often permit enemy release (e.g. reduced host population density, preference of vectors for native species) and novel weapons (e.g. differential response to parasite infection) could still operate and benefit introduced HOSP along the invasion gradient. Lower prevalence in more recently colonized locations would result if establishment was promoted by enemy release. High or similar prevalence along the invasion gradient (from old to young populations) could result if novel weapons promoted establishment. In this latter scenario, the cost of parasitism would be outweighed by the benefits gained from competitive advantage over native passerines. As predicted, population age was included in the best model and the coefficient suggests a slightly positive relationship between distance from Mombasa and prevalence, a pattern which supports novel weapons. However, this pattern was neither obvious (Fig. 4) nor statistically significant (Table 3).
One possible explanation for the lack of evidence for both enemy release and novel weapons is that we missed the important period of invasion. For example, in the enemy release described in another successful, range- expanding vertebrate, the Australian cane toad Bufo marinus, reduced prevalence of their native lungworm Rhabdias pseudophaerocephala lasted only 2.5 yr on average before the lungworm recolonized cane toad populations (Phillips et al. 2010). Our youngest population likely arrived between 5 and 10 yr ago. As such, our model may be picking up a residual importance of population age that is no longer statistically or biologically important. Alternatively, because we know that Kenyan HOSP are exposed to and being infected by generalist haemosporidians (Bennett and Herman 1976, Bensch et al. 2009, Marzal et al. 2011) and that native congeners are present in this system (P. griseus and P. motitensis), Kenyan HOSP may be at a similar risk of infection at all parts of their introduced range.
Capture month had a much stronger effect on infection risk than either distance from Mombasa or precipitation, but even this factor was non-significant. Interestingly, variation in prevalence with capture month is more likely attributable to seasonal fluctuations in host physiology (Martin et al. 2008) than to vector abundance or activity. Specifically, in our study, prevalence was highest in populations sampled in July (as compared to October/November), yet vector abundance is likely to be highest in March/April/May and October/November, coincident with the rainy seasons and appropriate weather conditions for vectors. A peak in prevalence in July also does not match up with previous findings by Bennett et al. (1974) who reported a major peak in haemosporidian prevalence in October and a smaller peak in April in native Ugandan birds. Some procedural disparities could explain why we found different patterns than Bennett and co-workers. First, Bennett et al. used blood smears to diagnose infections, which underestimate prevalence relative to PCR (Richard et al. 2002). Second, the Bennett et al. study did not include any species from Passeridae, the family to which HOSP belong. Yet, the most likely explanation for the differences in timing of peak prevalence involves specific differences in host–, parasite– or vector– species interactions. Previous studies in our laboratory suggest that HOSP may have physiological, genetic and epigenetic traits that impact range expansion success (Martin et al. 2010, Liebl and Martin 2012, 2013, Schrey et al. 2012). Such adaptations may affect other behavioral and physiological processes including parasite exposure and response.
Similar to distance from Mombasa's pattern of statistical non-significance but inclusion in the best model, average annual precipitation at capture location was not significantly correlated with haemosporidian prevalence in HOSP despite its inclusion in the best model (Fig. 3a). Typically, probability of haemosporidian infection is positively correlated with precipitation and proximity to water bodies (Foley et al. 2003, Lachish et al. 2011a). These results are normally attributed to the fact that haemosporidians are vectored by arthropods whose reproduction and population density is precipitation- and water-body dependent (Bennett et al. 1974, Chandler et al. 1977, Minakawa et al. 2002). Perhaps because the majority of our samples were collected between the two major rainy seasons (Table 1, Methods), when birds were still breeding (pers. obs. LBM, CACC), infected birds in our study were a mix of both new infections acquired in the previous 2–4 months, and relapsed, previously latent infections acquired during a previous rainy season or before (Applegate and Beaudoin 1970, Applegate 1971). Such a mixture would obscure a strong signal of precipitation on prevalence.
Our interest in patterns of prevalence led us to choose a methodology that did not permit identification of specific parasite species or lineage. Consequently, there are limitations on the conclusions we can draw. For example, though both this study and that by Marzal et al. found evidence of only Plasmodium relictum infections in Kenyan HOSP, there could be other haemosporidian species infecting HOSP which were not detected. Because different lineages of haemosporidians can have effects on passerines that are species- and/or individual-specific, we cannot presume the level of parasite virulence (Valkiūnas 2005). In fact, variation in haemosporidian species or lineages between populations and/or parasite–environment interactions (Chasar et al. 2009) may explain the variation we found between populations (Fig. 4). Also, without knowing the identity of the parasite strain (or the exact source population of HOSP) we cannot know the parasite's origin and without this information we cannot make conclusive statements about the loss of parasite diversity, or the possibility of parasite spillback (Kelly et al. 2009), parasite spillover (Callaway and Ridenour 2004), and/or co-introduction of one or more species of haemosporidians with HOSP (Ewen et al. 2012).
Regardless of the roles of the risk factors or identity of haemosporidians infecting HOSP, it is interesting that despite infection, in some cases at high levels of prevalence (Fig. 4), HOSP are still able to establish themselves in new areas in Kenya. Indeed, an alternative hypothesis to enemy release and tangential to the idea of novel weapons involves parasite tolerance: house sparrows may be successful because they can maintain fitness even when infected by haemosporidians or other parasites. Indeed, when (native) European HOSP are challenged with the same P. relictum lineage found in Kenya (SGS1) (Marzal et al. 2011), they do not lose mass, engage fever, or change hematocrit levels (Palinauskas et al. 2008). This lack of response is not abnormal in HOSP. In North America, HOSP did not exhibit strong sickness behaviors in response to simulated infections with diverse antigens (Coon et al. 2011) and they also maintained reproductive output after simulated infections better than a less successful introduced congener, the Eurasian tree sparrow (P. montanus) (Lee et al. 2005). Moreover, Kenyan HOSP gain body mass when their immune systems are stimulated whereas a native congener (P. motitensis) loses body mass (Martin et al. 2010). If HOSP are as unaffected by haemosporidians as they are with simulated infections, then haemosporidian prevalence in Kenyan populations should be primarily predicted by patterns of exposure rather than population age, as is the case here.
We found little evidence of enemy release or use of novel weapons by range-expanding HOSP in Kenya. Instead, haemosporidian prevalence was best predicted by characteristics that likely affect vectors at both the individual and population levels (i.e. rainfall and host size). It may be that host qualities, such as HOSP predisposition towards minimal behavioral and physiological response to parasites, may be more important to HOSP range expansion success than parasitological phenomena such as enemy release or novel weapons. On the other hand, more specialized, virulent and/or directly transmitted parasites might be less affected by environmental input and thus better fit a pattern of enemy release or novel weapons. A comparison of the roles of a variety of parasite guilds among diverse range expanding host taxa would be worthy of investigation, especially given that many hosts and parasites will undergo climate change induced range shifts in the coming century.