This study was carried out in strict accordance with current laws of all the countries where the study was performed and following the recommendations of the ‘Guidelines to the Use of Wild Birds in Research’ (J. Fair, E. Paul and J. Jones, eds. 2010. Ornithological Council, Washington, D. C.). All efforts were made to ameliorate suffering of animals and minimize handling time. No birds were injured or killed during the study.

Blood sampling was carried out at locations in Belgium, Bulgaria, Denmark, Finland, Hungary, Norway, and Spain. The sampling consisted of collection of a tiny blood sample from the brachial vein. This procedure is a standard procedure performed by numerous bird banders throughout the world for the study of avian malaria. No birds showed signs of negative consequences of blood sampling. We have recorded as high survival rates from sampled birds as from other birds. Collection of blood was specifically authorized in each location under the permits issued for this work.

In January 2010 we requested 58 European scientists, who had published on bird-parasite interactions to participate in the project. Participants were mainly selected based on a recent review of the effects of bird parasites on the fitness of their hosts [8]. Three requests were made and eventually 37 scientists participated in the study. Some scientists did not participate because they were not working on parasites any more, or they were committed to other projects during 2010. The geographical distribution of the study sites is shown in Fig. 1. We deliberately collected all recent samples during a single breeding season (2010) to reduce potential bias related to variation in variables among years. Although this approach may reduce variation in estimated differences due to the second year being the same for all populations, geographic variation in climatic conditions during 2010 was sufficiently large among study populations to ensure conclusions that are independent of the particular climatic conditions of 2010. Geographic variation in temperature in 2010 did not differ from that of the first study year (Levene's test, F = 0.45, d.f. = 1, 88, P = 0.50), although the first study year differed among populations. We asked all participants to use exactly the same methods in 2010 as during the first year of their study, but also that the same person conduct or at least supervise the study, thus ensuring that all studies were consistent in methodology over time to avoid inter-observer variability.

We distinguished between four functional parasite groups based on their taxonomic and transmission status: protozoans including blood parasites, feather parasites, dipteran parasites and non-dipteran parasites. Protozoans are generally vector-transmitted. Feather parasites such as chewing lice and feather mites live in the plumage of birds. Dipteran parasites include louseflies and blowflies. Non-dipteran parasites included blood-sucking fleas, ticks and mites. When more than one taxon of parasite within one of these groups were investigated in a target host species and population, we estimated mean values, which were used for subsequent analyses. Variance explained by the four functional groups of parasites was significantly larger than the residual variance (that was explained by different kinds of parasites within the four functional groups) for parasite abundance (F = 4.31, d.f. = 3,44, P = 0.01) and prevalence (F = 4.14, d.f. = 3,82, P = 0.009), which further supports the use of mean values.

We requested information on prevalence (the fraction of adults or nests harboring a given parasite) and the mean abundance of parasites exactly as done in the first year of the study and again in 2010. We requested that participants record information on demographic traits and population density of hosts as described below.

Information on the methods used to quantify parasites can be found in the references in Møller et al. [8] and Merino et al. [43]. In brief, we quantified prevalence and intensity of infection (mean abundance of parasites per host/host nest). Most studies either used two independent methods to quantify parasites or quantified parasites repeatedly for the same individuals/nests to allow estimation of repeatability. All studies used statistically significant and highly repeatable estimators of parasite prevalence and abundance.

We used the E-OBS gridded dataset (version 5.0) maintained by the European Climate Assessment and Dataset (ECAandD) (http://eca.knmi.nl/) to estimate temperatures at the study sites [44]. We calculated mean monthly temperature for each 0.25×0.25 degree squares, covering the study sites from the daily mean temperature of the E-OBS gridded dataset. Change in temperature during the month best covering the laying, incubation and nestling period of each bird host species was estimated as the temperature in 2010 minus the temperature in the first study year divided by the interval between the two years. This change in temperature over time (°C/year) is referred to as change in temperature throughout the remainder of this paper.

We requested that all participants record for each host the date of laying of the first clutch, clutch size, reproductive success (no. fledglings) and body condition (estimated as body mass at the start of the breeding season, if possible). We also requested a local estimate of population density of hosts (such as the proportion of nest boxes occupied for nest box studies, the total number of occupied nest boxes, the number of individual birds captured, colony size, or in a few cases population density for open nesting species). The entire dataset is reported in File S1 in Tables S1–S2, while Table S3 in file S1 presents correlations between prevalence and response variables in the supplementary material (ESM).

All variables were log₁₀-transformed to achieve approximately normal distributions. The effects of time between sampling of host populations and parasite intensity and prevalence were explored by Repeated Measures ANOVAs with laying date, clutch size, brood size, body condition of adults, population density, parasite loads and parasite prevalence estimated for the two study years as within subject, and population identity, host (or parasite) identity, latitude, and temperature change per year (°C/years) as between subject factors. All analyses were performed individually in order to avoid inclusion of more than one interaction term of between and within factors (i.e. repeated measure).

The effects of parasitism on characteristics of host populations were also explored using Repeated Measures ANOVAs with laying date, clutch size, brood size, body condition of adults and population density estimated for the two study years as within subject factors and change in parasite abundance and prevalence as between subject factors. Parasite functional group, latitude and temperature change (°C/years) were also included in the models as additional between factors to statistically control for the effect of these factors on changes in host population characteristics. A significant interaction between repeated measure and between factors would be consistent with an effect of the latter on change in host population characteristics.

Statistical analyses assume that each observation provides equally precise information about the deterministic part of the process variance [38]. Sample sizes differed among species due to differences in abundance of hosts, and such variation may affect the precision of estimates and hence the outcome of the statistical analyses [45], [46]. Therefore, we weighted all analyses by sample size to ensure that individual observations contributed relative to their precision.

The findings might depend on duration of the interval between first and second study year. Therefore, we conducted a second series of analyses based on datasets with an interval of 5–15 years, and these results are reported in Tables S4–6 in File S1. All analyses were performed with Statistica 10 [47]. Values reported are means (SE).

The mean change in temperature per year across the 89 data sets was + 0.085°C (SE = 0.029), with a range from −0.65°C to +1.07°C, differing significantly from zero which would be expected if temperature on average was the same in the first and the last year of study (t-test, t = 2.90, P = 0.0047). A similar increase in temperature over time was found for studies based on intervals of 5–15 years (+0.076°C (SE = 0.020), with a range from −0.55°C to +0.50°C). Temperature in the first year was lower than in the second year (t = 2.52, P = 0.013), and that was even more so for studies with an interval of 5–15 years (t = 4.30, P<0.0001). The number of years between the first study year and 2010 was on average 10 years (SE = 0.63), range 1 to 39 years with 78% within the interval 5–15 years. We also detected a large range in change in temperature across studies from −2.94°C to +4.04°C, on average +0.40°C, for the mean period of study. Because change in temperature between the first year of study and 2010 decreased slightly with the number of years of study (F = 4.02, d.f. = 1,87, r² = 0.04, P = 0.048, estimate (SE) = −0.053 (0.027)), implying that temperatures decreased more in studies of long duration, we used temperature change among study years divided by number of years elapsed between the first year and 2010 in the subsequent analyses. There were no additional effects of latitude or longitude on change in temperature (latitude: F = 0.05, d.f. = 1,85, P = 0.83; longitude: F = 0.39, d.f. = 1,85, P = 0.54), suggesting that these variables were not confounding effects on change in temperature.

Total sample size was 9935 hosts (or host nests depending on study) in the first year and 7956 in 2010, or in total 17,891 hosts/nests. A total of 37.2% of the 89 parasites studied were ectoparasites, and 76.4% of the 89 parasites lived permanently on the host, with the remaining species living apart from the host at ambient temperature at least part of the annual cycle. Prevalence and abundance of parasites in the first study year was independent of the duration of the study period (GLM: prevalence: log-likelihood χ² = 2.12, d.f. = 1, P = 0.14; abundance: log-likelihood χ² = 0.42, d.f. = 1, P = 0.51).

Host populations on average started to reproduce later during 2010 than during the first study year (Fig. 2). Moreover, temporal change in laying date differed significantly among localities (Table 1), which may be related to differential effects of climatic change at different latitudes (Table 1, Fig. 3). Change in clutch size decreased with increasing change in temperature (Table 1; Fig. 4A). Change in brood size differed among localities (Table 1), but not when considering studies with intervals between 5–15 years (Table S4 in File S1). Change in body condition tended to decrease with increasing temperature (Table 1; Fig. 4B). Host populations were less abundant in 2010 and this did not depend on host identity when studies with intervals between 5–15 years were considered (Table 1, Table S4 in File S1).

We detected 18 gains and 8 losses of parasites among studies, and for studies with 5–15 years intervals there were 10 gains and 10 losses. Abundance of parasites was larger in 2010 than in the first study year. However, none of the variables explained change in abundance of parasites (Table 2; Fig. 5A). Variables that explained change in parasite prevalence depended on interval between study years. Locality was associated with parasite prevalence when considering studies with 5–15 years intervals (Table S5 in File S1), while parasite identity did so when all studies were considered (Table 2). Non-dipteran parasites were the only group that tended to decrease in prevalence between sampling years although this category did not differ significantly from tendencies for other categories (Table 2; Fig. 5B).

When considering the entire dataset, change in parasite abundance, but not in prevalence was associated with delayed laying (beta (SE) = 0.70 (0.21), P = 0.004) and reduced body condition (beta (SE) = −0.49 (0.16), P = 0.038) by their avian hosts (Table 3). Moreover, the rate of change in body condition and laying date of hosts depended on parasite group (Table 3). These associations were detected after controlling for the effects of latitude and temperature change (Table 3).

When only considering studies with an interval of 5–15 years, the association between laying date and parasite abundance did not reach statistical significance, but change in laying date differed significantly among parasite categories when either parasite abundance or parasite prevalence were included in the model (Table S6 in File S1). Interestingly, these restricted analyses showed not only the detected negative association between change in parasite abundance and change in body condition of hosts (beta (SE) = −0.49 (0.16), P = 0.038), but also significant effects of parasite abundance on clutch size (beta (SE) = −0.61 (0.23), P = 0.026) and brood size (beta (SE) = −0.58 (0.21), P = 0.021) (Table S6 in File S1). Finally, when considering this restricted sample of studies, the results indicate that parasite categories explained changes in population density of hosts when parasite abundance was included in the model (Table S6 in File S1). All these associations were detected after controlling for the effects of latitude and temperature change (Table S6 in File S1).