1In recent years there has been much interest in physiological trade-offs involving host immune function and parasite defence, with the suggestion that they could play a pivotal role in mediating well-documented life-history trade-offs, such as the cost of reproduction.
2Among studies of birds, the hypothesized link between reproductive effort and parasite defence has received particular attention, yet support for a trade-off between these two traits remains equivocal.
3We used meta-regression analysis and an information-theoretic approach to investigate, among avian studies, how strong the effect of experimentally altered reproductive effort is on (i) infection with blood parasites from four common genera (Haemoproteus, Leucocytozoon, Trypanosoma and Plasmodium) and (ii) the ability of hosts to mount an immune response to novel antigenic challenge.
4Across studies, there was a relatively weak but well-supported positive effect of reproductive effort on blood parasite infection levels. Importantly, this effect was significantly influenced by the parasitological measure employed; where parasitaemia (proportion of parasitized cells within infected hosts) was used as the response variable, effect size was almost three times as large as where infection prevalence (presence vs. absence of infection among hosts) was measured.
5A moderate negative effect of reproductive effort on immune responsiveness was also found across studies. This effect was greater the longer the time that had elapsed between manipulation of reproductive effort and measurement of immune responsiveness, and was also related to the stage at which reproductive effort was manipulated, with manipulation during brood rearing producing stronger effects than manipulation during incubation.
6Overall, these results provide evidence that reproductive effort can have pronounced effects on both parasitism and immune responses, but that effect size is influenced by methodology – what is measured and when. Exactly how such effects arise and whether they are sufficient to provide a mechanistic explanation for the cost of reproduction remains to be fully explored.
Central to this model of parasite-mediated reproductive costs is the relationship between reproductive effort and either parasitism, or immune function, which is presumed to affect susceptibility to parasitism. In recent years, many studies on birds have focused on the relationship between reproductive effort and either parasitism or immune function, often adopting a similar experimental approach. This involves the use of brood size manipulation to experimentally alter reproductive effort, followed by measurement of common parasitic infections such as blood parasites (e.g. Haemoproteus and Trypanosoma) or immune function shortly afterwards. Such studies have, in some cases, provided convincing support for this relationship (e.g. Deerenberg et al. 1997; Nordling et al. 1998) and are often cited in the literature. However, the support is by no means universal. There are studies that find no evidence for an effect of reproductive effort on parasitism (e.g. Merino et al. 2006) or immune function (e.g. Ilmonen et al. 2003). Additionally, null results contained within otherwise confirmatory papers are sometimes overlooked and publication may be biased towards confirmatory results. Discrepancies also exist between the findings of apparently similar studies, such as those conducted on the same host species and parasite genus (e.g. Merilä & Andersson 1999; Stjernman, Råberg & Nilsson 2004). Consequently, it is unclear from a qualitative review of the literature whether the proposed effects of reproductive effort on parasitism and/or immune function are generally supported by empirical data, and what factors might explain variation among studies.
Here, we employ meta-regression analysis using an information-theoretic approach to quantitatively review the literature on this issue; meta-regression is a type of meta-analysis which incorporates multiple predictors (or moderators) which may account for heterogeneity in a data set (Thompson & Higgins 2002). We aim to address the following questions:
1How much statistical support is there across experimental studies employing brood or clutch size manipulation for: (i) a positive effect of reproductive effort on levels of infection with avian blood parasites; (ii) a negative effect of reproductive effort on measures of immune responsiveness to a novel antigenic challenge?
2What biological and/or methodological factors are responsible for variation in effect size among studies?
Materials and methods
data collection and inclusion criteria
We attempted to locate all published studies in which the relationship between reproductive effort and either levels of blood parasitism or immune responsiveness to novel antigen had been investigated in birds. We performed forward searches on ISI Web of Science for articles published between 1994 and the end of 2007 that cited one of four prominent early studies on parasite or immune-mediated reproductive costs (Norris et al. 1994; Richner et al. 1995; Deerenberg et al. 1997; Nordling et al. 1998). We also performed keyword searches using the terms ‘immunocompetence’, ‘haematozoa’ and ‘brood size manipulation’. Abstracts of all articles retrieved by this search were inspected and those with potential to fulfil our inclusion criteria were consulted, and their references checked for any further relevant studies. We also attempted to locate unpublished studies through contacting those known personally to have worked in this field of research.
Our specific criteria for inclusion of a study were:
1The study involved experimental assessment of the relationship between reproductive effort (as determined by brood or clutch size manipulation) and either (i) levels of infection by blood parasites (parasites of the genera Haemoproteus, Plasmodium, Leucocytozoon or Trypanosoma) or (ii) immune function as assessed by the ability of individuals to mount an immune response against novel antigen, such as phytohaemaglutinin (PHA) or sheep red blood cells (SRBC).
2Brood or clutch size manipulations were assigned randomly with respect to laying date and original clutch size.
3For those studies measuring parasitism, this had been recorded in terms of either prevalence or parasitaemia of a specified genus of blood parasites, where prevalence is defined as the proportion of hosts infected within a given group and parasitaemia is defined as the percentage of infected cells within an infected host.
4The mean time period between manipulation of reproductive effort and measurement of parasitism or immune responsiveness could be (at least approximately) determined from the methods.
5Studies contained complete reporting of statistics for both significant and non-significant results concerning the relationship of interest.
We restricted our data set in this way for several reasons. First, we consider manipulative studies to provide the most plausible test for a causal relationship between reproductive effort and parasitism or immune function; correlative studies cannot isolate the effect of interest from that of confounding factors such as parental quality. We wanted to examine effect size variation among experiments using two contrasting types of response variable: those that measured levels of parasitic infection, and those that used a measure of immune function. Among the latter, we chose only to consider immune challenge studies and not studies that simply measured levels of circulating immune cells or factors (e.g. leukocyte counts or plasma proteins), because the latter are thought to represent constitutive immune function and are difficult to interpret since they are strongly affected by current health status (Lee 2006). By limiting our analyses to a relatively well-defined basic methodology and set of host–parasite systems, some of the possible sources of variation between studies that could obscure general patterns should have been eliminated.
Where insufficient data were available for calculating effect sizes (the minimum information required was a test statistic or P-value, sample size and effect direction) or methods were unclear, authors were contacted to request additional information, which proved fruitful in many cases. Overall, 28 studies were located which reported the effect of a brood/clutch size manipulation on either blood parasitism, immune responsiveness or both. Of these, one study was excluded on the basis of criterion 2 (Norris et al. 1994) and another on the basis of criterion 5 (Ots & Hõrak 1996). The total number of studies which matched the inclusion criteria was thus 26. However, the majority of studies contained multiple results pertaining to the effect of interest, such that there were 66 results in total for which effect sizes could be derived (Table 1; see Table S1 in Supporting Information for a full list of effect sizes). Multiple effect sizes within a paper arose in a number of ways. The most common instances involved reporting of the relationship between experimental reproductive effort and parasitism for several different blood parasite species or for more than one type of immune challenge (e.g. PHA response and SRBC antibody response) within the same host population. Where both prevalence and parasitaemia had been investigated as the response variable, an effect size was calculated for each of these measures separately. Similarly, where statistics on the relationship of interest were presented separately for males and females or different age cohorts of host, an effect size was calculated for each subgroup of the host population. The existence of multiple effect sizes per study introduces a source of non-independence: by adopting a meta-analytic method that uses linear mixed-effects model (Nakagawa et al. 2007) we were able to account for this non-independence, and thus maximise our use of the published data (see Meta-analytic techniques).
Table 1. Studies used in meta-analyses investigating the effect of experimental reproductive effort alteration on avian blood parasitism and immune responsiveness to novel antigen
Studies varied in the types of statistical tests they performed, the most notable difference in this respect being the use of either directional (e.g. isotonic regression, treatments represented as a continuous –1,0,1 variable) or non-directional (e.g. χ2, F and t) statistical tests. They therefore tested subtly different hypotheses, that is, either that parasitism or immune responsiveness differed significantly across treatment groups, or alternatively that they varied in a particular direction across treatment groups. It is important that all results from which an effect size is calculated are testing the same hypothesis, so they are comparable. There are obvious directional hypotheses for these experiments; if the manipulation is effective in altering reproductive effort, the expectation is for parasitism to increase across treatment groups in the order Reduced < Control < Enlarged and immune responsiveness to decrease across the treatment groups, in the order Reduced > Control > Enlarged. We used the ordered heterogeneity (OH) test (Rice & Gaines 1994) to convert all non-directional statistics into directional (one-tailed) ones that tested these two hypotheses. Positive effect sizes indicated the effect was in the direction predicted by the hypotheses above.
Standard methods (Rosenthal 1991) were used to convert all test statistics into a common measure of effect size, the Pearson product-moment correlation coefficient (r) and then into its normalising Fisher's z (Zr); the corresponding sample size was used to calculate the variance associated with each effect size estimate.
All meta-analyses were conducted in the R environment (version 2·6·1; R development Core Team 2006) using weighted linear mixed effects models (Pinheiro & Bates 2000). In all meta-analyses, we accounted for the hierarchical structure of our data set (e.g. with multiple effect sizes from the same population) by including host species and study population as nested random effects (Nakagawa et al. 2007). This allowed us to use all data from the papers we located and include multiple effect sizes from a study or host population in the same analysis, without violating the fundamental assumption of independence.
First, we conducted an all-inclusive meta-analysis, containing all effect sizes (i.e. both parasitism and immunity results). Subsequently, we split this into two meta-analyses according to the response variable involved, that is, we conducted a parasitism meta-analysis and an immunity meta-analysis. In all three meta-analyses, we used weighted linear mixed effects models and restricted maximum likelihood (REML) to determine overall mean effect size and 95% confidence intervals. Where the 95% confidence intervals for an effect size did not span zero, this effect could be considered statistically significant at the 5% level. We calculated the proportion of variation in effect size that could be explained by the random factors of study population and host species. We also calculated two measures of heterogeneity for each meta-analysis: QT (total heterogeneity) and QREML (the residual heterogeneity in random-effects models), which is a more appropriate measure of heterogeneity when meta-analyses contain random effects (Nakagawa et al. 2007). One of the limitations in using mixed-models is the difficulty of estimating accurate degrees of freedom for statistics when multiple random effects are used; for example, the estimate of statistical significance for QREML may be inaccurate due to inappropriate degrees of freedom used. Therefore, we used meta-regression analyses using an information theoretic approach based on AIC corrected for small sample size (AICc; Burnham & Anderson 2002; Anderson 2008), to avoid model selection based on heterogeneity statistics, which has been used traditionally (Cooper & Hedges 1994). In this way, we were able to compare null models (i.e. normal meta-analysis) and other models (i.e. meta-regression models incorporating predictors) to evaluate the importance of each relevant predictor. Model selection for the parasitism and immunity meta-regression analyses were conducted separately. In each case, we created a set of candidate models using maximum likelihood parameter estimation, which included, in addition to the random effects of host species and population, all possible combinations of our fixed factors (i.e. predictors) of interest. For parasitism, these fixed factors were: measure of parasitism (prevalence or parasitaemia), mean duration of reproductive effort manipulation (in days), host sex and parasite genus (Plasmodium, Haemoproteus, Leucocytozoon or Trypanosoma). For immunity, the fixed factors were: assay type (PHA, SRBC, diptheria–tetanus vaccine), mean duration of reproductive effort manipulation (in days), manipulation stage (during incubation only vs. during brood rearing only) and host sex. Mean duration of manipulation was unrelated to both manipulation stage and assay type (manipulation stage: F1,21 = 0·108, P = 0·745, assay type: F2,20 = 0·452, P = 0·642). During the process of model selection for immunity studies, three effect sizes were excluded since they involved an experimental procedure that was different from the majority and therefore could not be assigned a meaningful factor level for one of more of our fixed factors of interest. These were two effect sizes in which the assay involved use of an NDV vaccine (Nordling et al. 1998) and one where both incubation and brood rearing had been manipulated (Moreno et al. 1999). For each model an Akaike weight was calculated, which indicates its level of support (since Akaike weights sum to 1, models with Akaike weights approaching 1 receive the most support relative to other models). We subsequently used model averaging to determine the relative importance of each fixed factor, as expressed by the sum of Akaike weights from all models in which that factor was included.
A much discussed problem in meta-analysis is the possibility that published studies (i.e. those that are sampled for most meta-analyses) form a biased subset of all studies on a subject (Gurevitch et al. 2001; Møller & Jennions 2001) because non-significant results may be more likely to be filed away without submission, or rejected for publication. This bias has the potential to increase the risk of a type I error (Cooper & Hedges 1994). We explored the possibility of publication bias via inspection of funnel plots, as well as rank correlation between effect and sample sizes (Begg & Mazumdar 1994). Funnel plots depict the relationship between effect sizes and their associated sample size. In the absence of publication bias, the plot should be funnel-shaped, converging on the true population mean effect size as sample size increases; unusual gaps in such plots (for example an absence of data points relating to studies with small sample sizes or weak effects) can thus reveal the presence of publication bias (Møller & Jennions 2001).
A summary of results from the three meta-analyses are presented in Table 2, and Fig. 1a shows a visual comparison of their overall mean effect sizes and 95% confidence intervals. Among parasitism studies, manipulated reproductive effort had a relatively weak but well-supported effect on levels of host blood parasitism in the predicted direction (r = 0·090, 95% CI: 0·021 to 0·158, P = 0·016). Across immunity studies, manipulated reproductive effort had a moderate statistically significant effect on immune responsiveness, again in the expected direction (r = 0·297, 95% CI: 0·169 to 0·416, P < 0·001). In the meta-analysis in which both parasitism and immunity effects were included, this difference in magnitude of effect between immunity and parasitism studies was shown to be significant (t45 = 2·710, P = 0·010). In all three meta-analyses, the amount of variation in effect size explained by the random factors of host species and population was small, ranging from 0–2·3% for host species and 0–2·4% for host population (Table 2). This indicates that the effect of phylogenetic non-independence on our results (see Adams 2008) is likely to be negligible.
Table 2. Results summary for meta-analyses investigating effect size for the relationship between manipulated reproductive effort and (a) parasitism (b) immunity (c) both parasitism and immunity effects combined. For each meta-analysis, the level of heterogeneity (QREML and QT) detected and the percentage of variance explained by the random factors of host species (Vcsp) and population (Vcpop) are given. The number of effect sizes (k), number of host species (l), number of host populations (m) and the number of individuals (n) are also indicated. Statistically significant effect sizes are in bold
Effect size r (Zr)
t value (P, df)
95% CI for r (95% CI for Zr)
Heterogeneity QREML (P, df) [QT (P, df)]
0·021 to 0·158
38·800 (0·389, 37)
(0·021 to 0·160)
[51·058 (0·094, 39)]
–0·026 to 0·141
14·748 (0·543, 16)
(–0·026 to 0·142)
[22·973 (0·192, 18)]
0·079 to 0·216
19·722 (0·349, 18)
(0·079 to 0·220)
[19·722 (0·475, 20)]
0·169 to 0·416
24·594 (0·372, 23)
(0·170 to 0·442)
[56·126 (< 0·001, 25)]
0·084 to 0·323
74·096 (0·160, 63)
(0·084 to 0·335)
[137·885 (< 0·001, 65)]
The results of model selection exploring the relative importance of fixed factors (moderators) of interest are presented in Table 3 and the results of model averaging indicating the relative importance of each factor are shown in Table 4. Among parasitism studies, our AICc-based model selection showed that effect size was significantly dependent on the measure of parasitism employed, with effect size being almost three times as strong among studies measuring parasitaemia as among those with infection prevalence as the response variable (r = 0·155 and r = 0·058 respectively; Fig. 1b; Table 2). Indeed, mean effect size across studies measuring prevalence was not significantly different from zero, whereas for those where parasitaemia was the response variable, 95% confidence intervals were narrower and mean effect size was clearly positive (Fig. 1b). Host sex, mean duration of manipulation and parasite genus all explained little variation in effect size among parasitism studies.
Table 3. Model selection for parameters affecting the strength of relationship (effect size) between manipulation of reproductive effort and (i) parasitism (ii) immunity. Each candidate model is shown with its component variables, AICc value, AICc difference between the best model and that model (Δi), number of estimated parameters (K) and Akaike weight (ωi). Models are listed in order of decreasing Akaike weights, with the best model at the top. All models included Host Species and Population as random effects
Variable coding: (1) Intercept (2) Duration of manipulation (3) Host sex (4) Measure of parasitism (5) Parasite genus (6) Assay type (7) Manipulation stage.
(i) Parasitism model set
1 + 4
1 + 2 + 4
1 + 3 + 4
1 + 4 + 5
1 + 2
1 + 3
1 + 2 + 3 + 4
1 + 2 + 3
1 + 2 + 4 + 5
1 + 3 + 4 + 5
1 + 6
1 + 2 + 3 + 4 + 5
1 + 3 + 5
1 + 2 + 5
1 + 2 + 3 + 5
(ii) Immunity model set
1 + 2 + 7
1 + 2 + 3 + 7
1 + 2 + 6
1 + 7
1 + 2
1 + 2 + 6 + 7
1 + 6
1 + 3
1 + 3 + 7
1 + 2 + 3
1 + 2 + 3 + 6
1 + 6 + 7
1 + 3 + 6
1 + 2 + 3 + 6 + 7
1 + 3 + 6 + 7
Table 4. Relative importance of fixed factors tested in parasitism and immunity model sets. Indicated are the sum of Akaike weights for each fixed factor across all candidate models containing that factor (∑ ωi), which provides a measure of relative variable importance. Model averaged regression coefficients (b), their standard errors and 95% confidence intervals are also given. Factors are deemed to have a significant influence on effect size where 95% confidence intervals for b do not span zero; these cases are indicated in bold
b ± SE
95% CI for b
0·145 ± 0·054
0·040 to 0·250
Measure of parasitism (prevalence)
–0·109 ± 0·051
–0·208 to –0·010
Duration of manipulation
0·001 ± 0·004
–0·006 to 0·008
0·002 ± 0·052
–0·101 to 0·104
Parasite genus (Leucocytozoon)
0·100 ± 0·177
–0·247 to 0·446
Parasite genus (Plasmodium)
0·172 ± 0·094
–0·012 to 0·357
Parasite genus (Trypanosoma)
0·033 ± 0·046
–0·058 to 0·124
–0·057 ± 0·253
–0·554 to 0·440
Duration of manipulation
0·030 ± 0·013
0·004 to 0·056
Manipulation type (incubation)
–0·315 ± 0·140
–0·588 to –0·041
Assay type (PHA)
0·343 ± 0·138
0·073 to 0·613
Assay type (SRBC)
0·326 ± 0·134
0·063 to 0·588
0·151 ± 0·171
–0·185 to 0·487
Among immunity studies, several fixed factors explained variation in effect size. We found a positive correlation between the mean duration of manipulation (days between manipulation of effort and measurement of immune responsiveness) and effect size (Table 4; Fig. 2). We also found evidence that the stage at which reproductive effort had been manipulated influenced effect size, with manipulation during brood-rearing producing stronger effects than manipulation during incubation only (Table 4; Fig. 3). There was also weaker evidence that effect size varied according to the assay type used, with PHA and SRBC challenges generally producing stronger effect sizes than diptheria-tetanus vaccine (Table 4; Fig. 4). Funnel plots for the parasitism and immunity meta-analyses are presented in Fig. 5. Neither plot showed any obvious evidence that certain types of study (e.g. with small sample or effect sizes) were missing, although it is somewhat difficult to assess the shape of funnel plots with small sample sizes. Spearman's rank correlations between effect size (Zr) and sample size were also non-significant for both the parasitism (rs = –0·255, 95% CI: –0·525 to 0·061, P = 0·112) and immunity (rs = –0·132, 95% CI: –0·494 to 0·269, P = 0·521) analyses, again providing little evidence for publication bias.
overall results and predictors of effect size
Overall, our meta-analyses reveal that there is good support in the literature for causative relationships between reproductive effort and both parasitism and immune responsiveness in birds. Effect sizes were generally larger among immunity results than among parasitism results, which may be for a number of reasons. If immunity and parasitism responses are manifestations of the same general effect (e.g. immunosuppression caused by increased reproductive effort) then parasitism effects may be weaker since they constitute a downstream effect of changes in immune factors. Alternatively, parasitism effects may be weaker because parasitaemia levels prior to manipulation are rarely controlled for, yet they can explain a large amount of post-manipulation variation (Stjernman et al. 2004). This issue applies less to immunity results since, for example, no individuals possess novel antigen-specific antibodies before challenge (Deerenberg et al. 1997).
Importantly, the effect among parasitism results was influenced by the measure of parasitism employed: brood size manipulation had a significant effect on parasitaemia, yet there was no convincing evidence that it affected prevalence. This difference suggests the effect of manipulated reproductive effort is seen more strongly as changes in parasitaemia within existing haematozoan infections rather than as changes in infection status within individuals. However, the weaker effect size among prevalence results may have arisen for a number of non-mutually exclusive reasons. First, the time-scale of experiments will be critical in determining whether an effect of differential acquisition of infections across treatment groups, if present, could be detected. In the studies reviewed here, the average time period between manipulation of reproductive effort and blood sampling was just 12 days (range 6–26 days). However, the time it takes for newly acquired haematozoan infections to become detectable in the blood (the prepatent period) may range from as little as 4 days to as much as several weeks (Valkiūnas 2005). It is therefore unlikely that any effect of infection acquisition could be detected during the short period of these experiments; at best it would be expected to be weak. Second, variation in prevalence may be strongly influenced by environmental factors, since these can determine the degree of exposure an individual host experiences. Pronounced spatial and temporal variation in avian haematozoan prevalence is frequently reported in the literature, even within host populations (Bensch & Åkesson 2003; Wood et al. 2007; Cosgrove et al. 2008) and is often related to variation in the abundance of vectors (Sol et al. 2000). Environmentally driven variation in prevalence would, all else equal, reduce the size of effect detected, since it would add sampling error to any relationships. Third, prevalence not only reflects the presence or absence of infection, but also captures variation in parasitaemia. This arises because parasitaemia affects the likelihood that an infection will be detected, since heavy infections are easier to detect (Valkiūnas, 2005). Thus the weak effect found among prevalence results could simply constitute a coarse reflection of the stronger effect seen among parasitaemia results.
There has been much recent debate about how measures of immune responsiveness to novel antigenic challenge (e.g. PHA or sheep red blood cells) relate to natural host–parasite interactions (Adamo 2004; Kennedy & Nager 2006; Martin et al. 2006). Indeed, clear relationships between such assays and parasitism have yet to emerge (Saks et al. 2006; Owen & Clayton 2007). However, the fact that immune responsiveness and parasitaemia showed opposite responses to reproductive effort manipulation is consistent with the notion that a host's responsiveness to such immune challenges is negatively associated with its ability to control parasitic infections, as has been suggested by some avian field studies (Gonzalez et al. 1999; Navarro et al. 2003). Further work to examine how novel antigenic challenge assays relate to resistance against and control of parasitic infections present in the hosts’ natural environment would be of great value in this respect.
Among immunity studies, both the duration of reproductive effort manipulation and the stage of reproduction at which manipulation occurred influenced effect size. Since a relatively small number of studies were involved in this analysis, these effects should be viewed with a degree of caution; nevertheless their implications are intriguing. The positive relationship between effect size and duration of manipulated effort suggests that the longer parents care for extra chicks, the greater the immunosuppression they experience. If immunosuppression has fitness consequences for parents, this would imply that reproductive costs could be paid not only in relation to how many offspring are raised but also how long it takes to raise them. Greater declines in immune responsiveness occurred with manipulation during brood-rearing than manipulation during incubation. This may suggest that increased physical exertion has a greater impact on immunosuppression than increased energy-expenditure per se, since the latter appears to show little variation between incubation and brood-rearing stages in females (Bryant 1997).
possible explanations for the effect of reproductive effort on parasitism & immune function
There is a large body of literature from both human and animal studies on the negative effects of physical exertion on immune function (reviewed in Pederson & Hoffman-Goetz 2000). In birds, reduced responsiveness to novel antigenic challenge has been shown to occur in response to several non-reproductive types of physiological manipulation, such as cold stress and increased physical workload (Saino & Møller 1996; Deerenberg et al. 1997; Svensson et al. 1998). It seems plausible that common mechanisms may underpin these effects as well as those associated with reproductive effort revealed by our meta-analyses. However, the question remains as to exactly how and why, from an evolutionary perspective, such effects arise.
The traditional resource allocation-based hypothesis suggests that during stressful periods, immune function may be suppressed in order to reallocate resources to other costly functions such as provisioning offspring (Gustafsson et al. 1994; Sheldon & Verhulst 1996). Given that raising a brood is certainly costly in terms of resources, and that immune function has been shown to be condition-dependent (Saino et al. 1997; Gonzalez et al. 1999; Møller & Petrie 2002) it is possible that resource deficits created by reproductive effort are paid for in terms of reduced immune responsiveness. However, whether immune responses are sufficiently energetically expensive for this explanation to hold has been called into question (Svensson et al. 1998) and several other explanations warrant consideration and further investigation. For example, it has been suggested that hosts may adaptively suppress the immune system during periods of stress, to protect against harmful autoimmune responses (Råberg et al. 1998). Another, as yet unconsidered, possibility for the effect of reproductive effort on parasitism, is that it could reflect an active response of parasites to the treatment; for Haemoproteus, Leucocytozoon and Trypanosoma (those parasites amongst which an effect on parasitaemia was detected), parasite stages detected in the blood are those capable of transmitting to a vector. Therefore it is possible that parasitaemia increases detected in brood size manipulation experiments represent parasites shifting their reproductive strategy towards transmission in response to a changed (e.g. somehow worsened) host environment. Non-adaptive processes, such as damage to immune cells caused by increased production of free radicals during periods of high metabolic activity (Leeuwenburgh & Heinecke 2001) may also play a role in the effects seen here.
relevance of detected effects for the cost of reproduction
However the effects detected here arise, the question remains as to whether such changes in haematozoan parasitism or immune responsiveness could contribute significantly to a reduction in survival or future reproductive success. There is general support in the literature for a positive correlation between immune responsiveness and survival (Møller & Saino 2004). This suggests the effect on immune responsiveness detected here could have important fitness consequences, though whether and how this involves parasite defence is unclear. The link between haematozoan parasitaemia and survival is somewhat uncertain. The vast majority of haematozoan-infected birds caught in the wild (and therefore in the studies included here) have already passed the acute phase and harbour chronic infections that are rarely associated with signs of ill health or survival effects (Davidar & Morton 1993; Bensch et al. 2007; Stjernman et al. 2008). Indeed, even in cases where experimental primary infections are known to cause marked mortality during the acute phase, such as when Hawaiian Amikihi are infected with Plasmodium relictum, no fitness effects of chronic infection can be detected (Atkinson et al. 2000; Kilpatrick et al. 2006). On the other hand, several medication experiments have provided evidence that chronic haematozoan infections can have detrimental effects on reproductive success, which had not been appreciated from correlational studies (Merino et al. 2000; Marzal et al. 2005). One study included in our meta-analysis used a path analytical approach to investigate whether the increase in parasitaemia they detected could account for the reduced survival observed in parents of enlarged broods. Stjernman et al. (2004) concluded that the increase in Haemoproteus parasitaemia of female blue tits rearing enlarged broods could not account for their reduced survival. The significance of the effects detected here for reproductive costs thus remains questionable, and will require a greater understanding of how haematozoan parasites and measures of responsiveness to novel antigenic challenge influence host survival.
One question that studies on parasite-mediated costs of reproduction have yet to test adequately is whether changes in reproductive effort can affect the likelihood of acquiring novel infections. Such an effect could arise, if for example a higher workload meant more time spent foraging in parasite-infested or vector-inhabited microhabitats, or immunological changes that increased the likelihood of an infection establishing following inoculation. Since it is the initial acute phase of haematozoan infection that is most likely to have pronounced effects on fitness (i.e. a risk of mortality) (Valkiūnas 2005) we consider this an important avenue for future investigations. Studies of other endoparasites, such as bacteria and viruses which no doubt represent a pervasive selective pressure in avian populations, as well as non-avian host taxa would certainly be useful in assessing how general the relationships detected here are across a wider range of host–parasite systems.
Authors are grateful to P. Siikamäki, L. Gustafsson, D. Ardia, S. Merino and M.J. Wood for providing data. M. Szulkin, S. Bouwhuis, M. Liedvogel, M.J. Wood, S. Verhulst, C.M Perrins, T. Uller and two anonymous referees provided helpful comments on the manuscript. S.C.L.K. was funded by a studentship from the Natural Environment Research Council.