All four species feed predominantly on nuts and seeds in winter. However, blue- and great tits maintain a proportion of arthropods (e.g., overwintering insects and spiders) in their winter diet, whereas tree- and house sparrows are largely granivorous at this time of the year. To varying degrees, all species exploit human-provided food sources, such as sunflower seeds, peanuts and bread (Jokimäki and Suhonen, 1998). For practical reasons, it was not possible to monitor individual or species-specific diet composition in the present study.

The study was approved by the Malmö/Lund Ethical Committee for Animal Research under the Swedish Board of Agriculture (permit no. M454 12:1), and capture was performed under license from the Swedish Museum of Natural History (to CI, license no. 681). Fieldwork was performed in the province of Scania in southern Sweden from 15-Feb-2013 to 6-April-2013. Urban birds were caught at four locations within the city limits of Malmö, the largest city in southernmost Sweden with approximately 300,000 inhabitants (3,600 inhabitants/ km²; study areas centered at 55°35′ N, 12°59′ E). Samples were collected either in city parks and recreational areas (tree sparrows, blue tits, great tits; two sites), on ruderal land with rich deciduous secondary growth (tree sparrows, blue tits, great tits; one site), or along hedgerows adjacent to granaries (house sparrows; one site). Rural birds were caught at seven locations 15–45 km ENE to NNE of Malmö (with most activity centered on 55°40′ N, 13°31′ E). Sampling was undertaken in deciduous forest stands on grassland (blue tits, great tits, tree sparrows; two sites), along hedgerows on rural farms (all four species; five sites), and in granaries on the same farms (house sparrows, tree sparrows; four of five sites). The rural areas were sparsely inhabited by humans (< 5 inhabitants per km²). Sampling took place throughout the day, from 6 a.m. to 7 p.m., at both urban and rural sites.

All species were caught with mist nets (n_tot. = 243 birds), but not all individuals could be measured for all physiological assays due to limited plasma volumes. The assay- and model-specific sample sizes are provided in Table 1, and in figure legends when relevant. All birds were individually ringed and (when plumage characteristics permitted) sexed (great tits, blue tits, house sparrows) according to Svensson (1992). Tree sparrows, which are sexually monomorphic, were molecularly sexed using the primers P2 and P8 (Griffiths et al., 1998). A blood sample (100–125 μl) was collected from the jugular vein with a heparinized syringe, and kept on ice until centrifugation (1,800 rpm for 10 min) and separation of blood plasma 0–1 h later. All samples were stored at −80°C until biochemical analyses.

Fatty acids were extracted following the protocol described in Andersson et al. (2015). Briefly, a total lipid extraction of 5 μl plasma was performed for 1 h at room temperature using 50 μl chloroform:methanol (2:1 v/v). The solvent was then evaporated, and the lipid extracts were subjected to base methanolysis for conversion into corresponding fatty acid methyl esters (FAMEs), which were then extracted in n-hexane. Extracts were washed with H₂O and then dried over anhydrous sodium sulfate.

The FAME extracts were analyzed using an Agilent 5975 MS coupled to an Agilent 6890 GC equipped with an HP-88 capillary column [(88%-Cyanopropy)aryl-polysiloxane; 30 m, 0.25 mm id, d_f 0.20 μm; Agilent, CA, USA]. The oven temperature was set to 80°C for 1 min, then increased by 10°C/min to 230°C and held for 20 min. Helium was used as carrier gas at a constant flow of 1 ml/min. FAMEs present in plasma were identified by comparing their mass spectra and retention times with those of synthetic standards (Supelco 37-Component FAME Mix, Sigma-Aldrich, Stockholm, Sweden).

We quantified malondialdehyde (MDA), the most frequently used biomarker of overall lipid peroxidation, following the protocol described in Eikenaar et al. (2016). Briefly, 15 μl plasma was mixed with 50 μl O– (2, 3, 4, 5, 6-pentafluorbenzyl) solution (1 mM in sodium acetate buffer, pH 5.0) and the micro-reaction was conducted for 1 h at room temperature. The resulting MDA-bis-(PFB-oxime) derivatives were extracted in 300 μl n-hexane containing 1.65 pg/μl of 3-bromofluorobenzene as internal standard, and analyzed using an Agilent 5975 MS coupled to an Agilent 6890 GC. A non-polar capillary column, HP-5MS (30 m, 0.25 mm id, d_f 0.25 μm; J&W Scientific, USA), was used for GC/MS analysis, and the characteristic ion at m/z 250 was measured under selected ion monitoring mode to quantify the target derivatives of MDA. The GC oven was programmed to 60°C for 1 min, followed by 15°C/min to 150°C, and then 10°C/min to 270°C, which was held for 5 min. All chemicals were purchased from Sigma-Aldrich (Stockholm, Sweden).

In total, 19 FAs were identified and quantified in all species (Table 1). The proportion of each FA was calculated by dividing each peak area with the sum of all FA peak areas in each individual. All FA proportions were logit-transformed (log(y/[1-y])), prior to statistical analyses (Warton and Hui, 2011). Individual saturated fatty acids, SFAs (i.e., ∑tot[SFA] = 14:0 + 16:0 + 18:0 + 20:0) were pooled, because they are all mainly used as metabolic fuel, in contrast to mono- and polyunsaturated FAs (i.e., MUFAs and PUFAs) that typically have specific biological functions.

Firstly, general linear models (GLMs) were performed on nineteen variables; fifteen individual FAs, _totSFA, total ω-6/ω-3 PUFA ratio, peroxidation index (PI), and MDA. The PI was calculated as PI = [(% monoenoic × 0.025) + (% dienoic × 1) + (% trienoic × 2) + (% tetraenoic × 4) + (% pentaenoic × 6) + (% hexaenoic × 8)] (Pamplona et al., 1998; Kang et al., 2004). PI takes into account that the sensitivity to peroxidation increases as a power function of the number of double bonds in each FA molecule (Pamplona et al., 2000); in other words, the PI provides an estimate of oxidative stress susceptibility in terms of lipid peroxidation. Apart from the GLMs for MDA (see below), all models included two fixed factors: “species” (four levels) and “habitat” (two levels), along with their interaction (species × habitat). In addition, our previous work on great tits has shown that sampling hour affects the proportion of FAs (Isaksson et al., 2015). Therefore, “sampling time” was always included as a covariate in these GLMs. Furthermore, when the interaction between species and habitat was significant in the above model, we proceeded by fitting species-specific GLMs, including time as a covariate. When the main effect of species or habitat was significant, but the interaction between them non-significant, we proceeded by traditional post-hoc analyses (Student's t-test), and removed the interaction from the final model (Supplementary Material 1). There were no sex differences among any of the FAs (ANOVAs: F = 0.0002–2.237, p = 0.14–0.96). Accordingly, sex was excluded from all FA models.

Secondly, multivariate canonical discriminant analysis (CDA) was performed to complement the GLMs for the individual FAs. Here, the proportions of each unsaturated FA along with the proportions of _totSFA were included, and the analysis performed both for each habitat separately, and with habitats combined, with species as a categorical variable. These analyses allowed us to test whether the FA profiles showed distinct species- and family (exploiter vs. adapter) signals, and whether these signals were less pronounced in the urban environment due to expected dietary convergence (i.e., reliance on human-provided food sources). The CDA predicts an individual's classification in a category, i.e., species/family, based on observed values of several continuous variables (here, the FAs). A quadratic discriminant method was used, because the covariances between FAs were not expected to be equal across species.

Lastly, MDA was log₁₀-transformed to obtain a Gaussian distribution. The GLM with MDA included the three fixed factors “species,” “habitat,” and “sex,” and the three covariates “sampling hour,” “PI” and the “total ω-6/ω-3 PUFA ratio.” In addition, we included all possible two-way interactions. For MDA, we derived final models using backward-elimination of non-significant (P > 0.05) interactions and covariates, starting with the least significant interaction followed by the covariates.

All analyses were performed in JMP Pro 11.0 (2013 SAS Institute Inc.). For clarity, we only present significant results in the Results Section [see Supplementary Materials 1–3 for all GLMs and species-specific models for FA, FA-derived indices (PI and total ω-6/ω-3 PUFA ratio), and MDA].

The CDA analyses revealed that the first three discriminant functions significantly discriminated the four species in both urban and rural habitats [urban: P < 0.001, eigenvalues: Canonical 1 (C1) = 10.325, C2 = 1.091, C3 = 0.611; Figure 3A; rural: P < 0.001, eigenvalues: C1 = 11.808, C2 = 1.103, C3 = 0.743; Figure 3B]. In the urban habitat, C1 accounted for 85.85%, C2 for 9.07%, and C3 for 5.08% of the variation [Wilk's lambda: 0.026, F_{(48, 233)} = 11.65, P < 0.001]. In the rural habitat, C1 accounted for 86.47%, C2 for 8.083% and C3 for 5.45% of the variation [Wilk's lambda: 0.021, F_{(48, 325)} = 17.93, P < 0.0001]. 98.97 and 99.22% of the individuals were correctly assigned to the pre-determined groups (species) in the urban and rural habitat, respectively. When birds from the two habitats were analyzed together the correct species assignment was reduced to 93.34%, but the model was still highly significant [Wilk's lambda: 0.044, F_{(48, 613)} = 23.80, P < 0.001; Figure 3C]. Most misclassified individuals were still assigned to the correct family, with only one blue tit misclassified as a tree sparrow. Similarly, there was a clear distinction between the two families (adapters vs. exploiters), despite the significant species signals [Wilk's lambda: 0.118, F_{(16, 208)} = 97.02, P < 0.001].

In the full-species GLMs, there was a clear effect of the main factor “species” for two of the MUFAs: palmitoleic acid [F_{(3, 220)} = 54.11, P < 0.001] and eicosenoic acid [F_{(3, 220)} = 25.55, P < 0.001]. The post-hoc analyses revealed that specifically palmitoleic acid was significantly higher in house sparrows than in any of the other species, and eicosenoic acid was significantly higher in the two tit species than in the sparrows (P < 0.05 in all cases; Figures 1C,E). Finally, the ω-6 PUFA, arachidonic acid, differed significantly between species [F_{(3, 220)} = 35.90, P < 0.001], with the two tit species having higher proportions than the two sparrow species (P < 0.05 in all cases; Figure 2I).

The exploiter and adapter species often showed opposite differences in plasma FA composition between the urban and rural environments (Figures 1, 2). For instance, urban blue tits had higher proportions of the main MUFA oleic acid (great tits showed a similar tendency), whereas both urban house- and tree sparrows had lower oleic acid than their rural counterparts. For _totSFA, a habitat difference was only found in house sparrows, with urban individuals having lower _totSFA than the rural conspecifics. Since both SFAs and MUFAs can be biosynthesized by birds, their relative levels can be controlled to a larger extent than the levels of PUFAs. Thus, the results for _totSFA and MUFA are likely due to a combination of metabolic regulation, and habitat- and species-specific diets.

Moreover, urban exploiters had lower proportions of several ω-3 PUFAs, whereas urban adapters had higher proportions than their rural conspecifics. Differences between the habitats in plasma levels of the essential ω-3 PUFA α-LNA should be a direct reflection of dietary differences. This could be a result of either different food items being eaten/or available across the habitats, or habitat-specific differences in fatty acid composition of the same food items. Our previous diet experiments on great tits showed that the proportion of all ω-3 PUFA in plasma is highly affected by availability in diet, with plasma amounts rapidly declining (within a few days) after restricting ω-3 PUFAs in the diet (Andersson et al., unpublished). Thus, the results also for the other ω-3 PUFAs in the present study are likely to be at least partly affected by dietary differences across the habitats. In addition, it is also possible that there are species/family-specific effects on metabolic conversion across habitats (Jing et al., 2013; Watson et al., 2017). Regardless, the lower ω-3 PUFA of rural house- and tree sparrows compared to urban ones indicates a “poorer” nutritional status at the rural farms, rather than in the urban habitat. In support of this argument, mead acid—a suggested biomarker of malnutrition or deficiency of essential FAs (Mead and Slaton, 1956; Smit et al., 2004)—was higher in the rural house- and tree sparrows than in the urban conspecifics. Furthermore, the strictly dietary ω-6 PUFA LA was lower for urban great tits, but higher for urban house sparrows compared to their rural conspecifics. Such a contrasting effect of habitat was also evident for eicosadienoic acid, with both adapters having lower levels in the urban habitat, whereas urban house sparrows had higher levels than rural ones. Whether these interspecific differences have any adaptive values in the urban-rural context requires further investigation.

The present study is the first to investigate interspecific differences in FA profiles among terrestrial bird species. Previously, a similar approach for FA profiles was used on pelagic seabirds in relation to their feeding ecology (Iverson et al., 2007). In that study, a captive feeding experiment was conducted along with sampling of adipose tissue in four species of seabirds and their natural prey. Distinct species-specific discrimination of FA signatures was evident, which could be explained by interspecific differences in diet (Iverson et al., 2007). It was suggested that species FA signatures could be a complement or even an alternative to stable isotope analyses, because FA profiles can provide more information about the diet composition and not simply indicate trophic levels. Although the current study cannot link the plasma FA content with the FA profile of specific food items, it supports the previous conclusions given the clear species and family distinction in FA profiles.

More specifically, our CDA revealed that the two bird families (or the adapter vs. exploiter categorization) were separated by the first canonical function (C1) (Figure 3). The separation between adapters and exploiters could be a combined result of shared evolutionary history, as well as similar life-styles and diets among the two species within each group. Unfortunately, the relative contribution of each of these factors could not be determined. However, despite the resemblance between the species of the same taxonomic family, and the variation in FAs in birds across the habitats, the four species were also discriminated by the FA loadings on C2 and C3 both within each habitat and across habitats. Overall, 99% of the birds were successfully assigned into the correct species within each habitat, and although the correctness of species classification was reduced when birds from the two habitats were combined, it was still high (93%). Interestingly, both α-LNA and mead acid signified house sparrows with a decrease in α-LNA and an increase in mead acid likely indicating malnutrition or nutrient deficiency (Mead and Slaton, 1956; Smit et al., 2004). The effect of mead acid was only present in the rural habitat, which is in line with the suggestion that the rural house sparrows are more malnourished than urban conspecifics (see above).

Despite the interspecific differences across the urban-rural habitats in several ω-3 and ω-6 PUFAs, only the blue tit differed in the total ω-6/ω-3 ratio between the habitats. As predicted, the urban blue tits had a higher ratio than the rural blue tits, suggesting that urban birds have a diet-induced increased susceptibility to inflammation when exposed to antigens, such as traffic-generated particulate matter and pathogens (Ezenwa, 2004; Romieu et al., 2008; Isaksson, 2015). Since non-pathogenic inflammatory agents are more common in the urban environment, future studies should investigate whether urban blue tits have a higher inflammatory response than their rural conspecifics, and whether this could be explained by their higher ω-6/ω-3 PUFA ratio. However, at the species level, great tits had the highest ratio of all species across habitats (Figure 4A). Thus, also great tits are likely to suffer from increased, dietary induced, negative effects, if exposed to pro-inflammatory agents, such as urban air pollution.

MDA is a commonly used biomarker for lipid peroxidation (i.e., oxidative damage) that is generated by pro-oxidant exposure (e.g., Del Rio et al., 2005; Lykkesfeldt, 2007). The two exploiter species followed our prediction, with higher MDA in the urban habitat as compared to the rural habitat. Surprisingly, blue tits had lower MDA in the urban compared to the rural environment. Previous studies on lipid peroxidation in birds have not detected a difference between urban and rural populations (great tits in Gothenburg: Isaksson et al., 2009; house sparrows in Madrid: Herrera-Dueñas et al., 2014). Possibly, our more sensitive quantification of MDA using GC/MS, compared to the absorbance measurements used in the previous studies (thiobarbituric acid [TBARS] assays), is able to more accurately detect smaller differences in lipid peroxidation (Del Rio et al., 2005). Alternatively, lipid peroxidation may be more context-, nutrition-, and species-specific than previously appreciated, particularly since it exhibited the opposite patterns in the two bird families. In support of this argument, both urban exploiter species also showed higher susceptibility to lipid peroxidation, as indicated by the higher PI, than rural birds. This result might be a consequence of population differences in dietary intake or biosynthesis of the long-chain, peroxidation prone, PUFAs, especially the highly unsaturated DHA.

Moreover, due to high levels of traffic-generated air pollution (acting as pro-oxidants) in the urban habitat, we predicted that MDA levels in urban birds would be positively associated with the ω-6/ω-3 PUFA ratio (via increased ROS production during a pro-inflammatory response), and/or the PI (via increased susceptibility to FA peroxidation). However, in contrast to our predictions, there was no significant interaction between habitat and either of the two FA-derived estimates. Instead, there was an interaction between species and PI, revealing a negative association between PI and MDA in tree sparrows, but not in the other species. The direction of this association in tree sparrows is in contrast to our prediction, and also to the results at the population level (where both MDA and PI were higher in the urban compared to the rural environment). Possibly, individual tree sparrows with a high PI have a corresponding increase in dietary antioxidants (Eikenaar et al., 2016, 2017). This could prevent lipid peroxidation to some degree, but not enough to override the population/habitat effect. Alternatively, individuals with a high peroxidation rate, independent of habitat (e.g., due to other unknown factors such as age and life history), might reduce their plasma PUFA levels, thus reducing the PI.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer CH and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.