In 1991 the Environmental Protection Agency (EPA) reported that General Electric (GE) plants at Fort Edward and Hudson Falls illegally discharged up to 1.3 million pounds of PCBs into the Hudson River between 1947 and 1977 [27]. Today, the Hudson River basin continues to contain high PCB levels that vary throughout the drainage.

From 2006 to 2009, 48 recording sites and 7 mist-netting sites were established in five regions in New York State (Figure 1). These regions were chosen based on the documented historic variability in the levels of PCBs found in the Hudson River basin [28]. Three regions were identified with no local point source PCB inputs (labelled as − Ithaca, − Adirondacks, − Hudson), and two regions were identified with varying environmental historic PCB inputs (labelled as + Hudson and ++ Hudson) [28]. − Ithaca was located in Ithaca, New York (N = 5 recording sites, N = 1 mist-netting site), − Adirondacks was located in Adirondack Park, New York (N = 14 recording sites, N = 1 mist-netting site), and − Hudson was located above the point source of PCB inputs on the Hudson River (N = 5 recording sites, N = 2 mist-netting sites) [28]. The region downstream from the point source of PCB inputs, with historically intermediate PCB levels, was labelled + Hudson (N = 12 recording sites, N = 1 mist-netting site) [28]. The region directly below the pollution point source, with the highest historical PCB level, was labelled ++ Hudson (N = 12 recording sites, N = 2 mist-netting sites) [28]. The sites at − Ithaca and − Adirondacks were within 40 m of a river or lakeshore, and the sites at − Hudson, + Hudson, and ++ Hudson were within 40 m of the Hudson River. All sites were in or on the edge of mixed deciduous/evergreen forests, with similar amounts of tree cover.

We used black-capped chickadees and song sparrows as study species because both have life history traits and song characteristics that make them particularly vulnerable to PCB exposure. Adults feed their young insects and also primarily eat insects themselves during the breeding season [29]–[32]. Both sexes have relatively short natal dispersal distances, with males dispersing shorter distances than females [29], [30], [33], [34]. Additonally, both species have territories within foraging distance of water and are either year-round residents or have high between-year site fidelity [29], [30], [35], [36]. Because insects with an aquatic larval stage have higher amounts of pollutants if collected near a pollution source [37], it is reasonable to infer that birds hatched in areas with high PCBs are likely to be ingesting them for their entire life. In addition, black-capped chickadees and song sparrows both have well-studied songs and researchers have identified specific song characteristics important in communication. In this study the glissando and interval ratios were measured in male black-capped chickadee songs (Figure S1, [38]), and the trill-rate frequency-bandwidth trade-off was measured in male song sparrows (Figure S2, [39]).

In May and June 2008–09, black-capped chickadees and song sparrows were captured using 3 m or 6 m 30 mm mist nests (Avinet, Inc. Dryden, NY 13053 USA; USFWS bird banding permit 22669: André A. Dhondt). To attract birds to the mist nets, birdfeeders with sunflower seeds were hung near net locations. Playbacks with recordings from the target species were played directly adjacent to the nets to attract territorial males. The playbacks were all songs from wild individuals recorded during the two previous field seasons (2006–07) at comparable sites. All sampling and field measurements were performed on birds within 20 minutes of capture, individuals were released directly following the procedure, and all efforts were made to minimize distress. All bird-handling procedures were approved under the Institutional Animal Care and Use Committee Protocol 2008-0022 of Cornell University.

Species were identified and sexed according to Pyle [40], and then banded with metal USFWS bands. Wing chord, tarsus length, weight, muscle score (score from 0–5 of the size of the pectoral muscle), fat score (score from 0–3 of the fat stores in the furcula) and mites (score from 0–3 of ectoparasite damage on the primary wing feathers) were recorded.

Blood samples were collected, when possible, from all individuals. Approximately 100 µL of whole blood was taken from the left brachial vein with a 27-gauge single-use needle (BD Medical, 1 Becton Dr., Franklin Lakes, NJ 07417 USA; length: 12.7 mm). The blood samples for PCB analysis were collected with a non-heparinized capillary tube and immediately transferred to 1.5 mL centrifuge tubes (EDTA washed to prevent coagulation) and stored on ice. PCB samples were separated into red blood cells and plasma with a Mini Centrifuge (Fisher Scientific, Inc., 2000 Park Lane Dr., Pittsburgh, PA 15275 USA) in the field, flash frozen on dry ice or liquid nitrogen, and stored at −80°C for later analysis. The blood samples for mercury analysis were collected with a heparinized capillary tube, stored briefly on ice, and then transferred to a −20°C freezer until analysis.

Blood serum samples from black-capped chickadees (22.1±2.2 mg) and song sparrows (50.6±2.2 mg) were fortified with 25 µL of ¹³C₁₂-labeled internal standard mixture (EC-5087; Cambridge Isotope Laboratories, Inc., Andover, MA 01810 USA). Samples were extracted twice with 100 µL of n-hexane as described by Rivera-Rodriguez et al. [10]. The hexane extracts were dried over 100 mg anhydrous sodium sulphate and the volume reduced to 50 µL under a gentle stream of nitrogen. PCB levels were determined by gas chromatography mass spectrometry (GCMS) on a Varian Saturn 2200 ion trap GCMS system (Varian, Inc., Walnut Creek, CA 94598 USA). One µL aliquots were injected into the GC at an injector temperature of 300°C. PCBs were separated on a VF-5MS capillary column (30 m×0.25 mm ID, 0.25 micron film thickness; Varian, Lake Forest, CA 92630 USA) with helium as carrier gas at a flow rate of 1 mL min⁻¹ and the following temperature gradient: 40°C for 1 min, at 30°C min⁻¹ to 160°C, at 4°C min⁻¹ to 220°C, at 12°C to 300°C min⁻¹, and 300°C for 5 min. Congener-specific selected reaction monitoring (SRM) conditions were optimized for the quantitation of 41 of the 48 most abundant congeners reported for the Hudson River Superfund Site with quantitation limits of 0.1 to 1 ppb (Table S1C, [41]). Calibration curves were generated using a dilution series made from pure PCB standards (AccuStandard, Inc., New Haven, CT 06513 USA).

A large enough blood sample to measure both mercury and PCB content could only be collected from song sparrows. Blood samples collected in 2008 from song sparrows were analyzed for total mercury content (THg) at the Centre for Environmental Sciences and Engineering (University of Connecticut, Storrs, CT 06269 USA) using Environmental Protection Agency Method 1631. Blood was first digested with sulphuric and nitric acids, oxidized using bromine monochloride, and trapped on a gold trap. The trap was purged into a cold vapour atomic fluorescence (CVAFS) unit for analysis. Detection limit for blood THg was 0.02 ng/g. Standard quality assurance methods were used: analysis of duplicate samples; method blanks; spiked samples; laboratory control samples; and standard reference materials (DOLT-3, DORM-2, 966). Instrument response was evaluated using calibration verification standard and blank before, during (every 20 samples), and after each analytic run.

All recordings were made with a Fostex FR-2 Field Memory Recorder (Fostex Company, 3-2-35 Musashino, Akishima, Tokyo, Japan 196-0021) or a Tascam HD-P2 Recorder (TEAC Corporation, Nakacho, Musashino-shi, Tokyo 180-8550, Japan), a Universal Telinga Pro 24-inch Parabola (Telinga Microphones, PI. 129 Botarbo, S-748 96 Tobo, Sweden), and a Sennheiser ME62 Omni Microphone (Sennheiser Electronic Corporation, 1 Enterprise Drive, Old Lyme, CT 06371, USA). Recordings were performed with a sampling width of 24 bits and a sampling rate of 44.1 kHz. All songs were recorded between 0400 and 1000 EST.

Recordings were made of spontaneous male black-capped chickadee and song sparrow song. To increase the likelihood of non-repeated sampling, individuals were distinguished by either visual or auditory cues. If there still was uncertainty of replication, the next recording was made at least one territory's distance away (∼160 m). In 2009, a subset of the song sparrows from which we obtained blood samples were also color banded and recorded (N = 29).

The familiar fee-bee song (Figure S1) is usually only sung by male black-capped chickadees to defend a territory and to attract and arouse females during the breeding season [42]. The glissando ratio (see Figure S1) is thought to indicate species identity and varies <2% over the species range [38], [43]. The interval ratio (see Figure S1) is thought to indicate male quality [43]. High quality males are able to reproduce both ratios consistently at all frequencies (see Figure S3), and the ratios vary <2% over their range [43].

Although black-capped chickadee songs were recorded from April–August (2006–07), the recordings used in the analysis were limited to recordings made in May through early July (2006–07), while adults in our populations were breeding and seasonal song variation was at a minimum. Only recordings of black-capped chickadees singing at least eight consecutive non pitch-shifted fee-bee songs were used in this analysis (glissando ratio: N = 210; interval ratio: N = 218). Spectrograms were generated in RavenPro 1.4 (Bioacoustics Research Program, Cornell University, Ithaca, NY 14850 USA) with a 512 sample size, a 512 DFT size, spectral overlap of 50% (Hann window), 3 dB filter bandwidth at 135 Hz, 256 sample hop size, grid spacing of 93.8 Hz, no clipping, and averaging 1 spectrum. The glissando ratio was calculated by dividing the 95% fee frequency by the 5% fee frequency. The 95% fee frequency is the frequency value for a selected fee note where 95% of the energy is below that value. Similarly the 5% fee frequency is the frequency value for a fee note where 5% of the energy is below that value. These two measurements calculated by RavenPro give a stereotyped frequency value of the beginning fee note frequency and the end fee note frequency. The interval ratio was calculated by dividing the center frequency of the end of the fee note by the center frequency at the beginning of the bee note. Because a stereotyped measurement was not possible to use for the interval ratio, interval ratio measurements were performed manually, while blind to the source of the song. To look at the variation of the song structure within an individual black-capped chickadee, the coefficient of variation (CV) of the glissando ratio and the interval ratio was calculated for each individual. We conservatively classified individuals as having inconsistent (hereafter ‘variable’) songs if they had a glissando ratio or interval ratio CV value greater than 0.03, and as having consistent (hereafter ‘stereotyped’) songs if the CV was less than 0.03. We choose 0.03 as the cut-off point because it resulted in the most conservative grouping; only individuals singing with extremely high variation were classified as variable singers. We statistically analyzed the proportion of variable singers in a population because, although we expected variation in the song of each individual, we were especially interested in the variation between regions of outlier singers with especially high CVs.

Song sparrows are known for their melodic and varied songs. Territorial males typically sing a repertoire of approximately 4–12 song types (Figure S2) during the breeding season [44]. Although large repertoire size has been linked to large territory size, longer territory retention, and higher lifetime reproductive success [45], accurate estimates of repertoire size are difficult to obtain (minimum of 200 recorded songs needed [44]). Therefore, in this study we classified song sparrow song performance as the trill-rate frequency-bandwidth trade-off [39]. The trill is a phrase present in most song sparrow song types, characterized by the rapid repetition of a syllable. The trill rate is the number of syllables produced per unit time (Hz), and the frequency bandwidth is the difference between the maximum and minimum frequencies of the syllables in the trill. The bird is constrained in how quickly it is able to transition between low frequency production (with a relatively closed beak and a fully inflated nasopharyngeal cavity) and high frequency production (with a relatively open beak and an only partially inflated nasopharyngeal cavity), resulting in the observed trill-rate frequency-bandwidth trade-off [39].

Song sparrow songs used in the analysis were recorded from April to August in 2006, 2007, and 2009. Song sparrows that sang song types containing a trill were included in the analysis if they sang a minimum of five repetitions of the trill, and if the trill contained a minimum five syllable repetitions (N = 155). Many male song sparrows were recorded singing multiple trill types. One trill type was randomly chosen for each male to avoid pseudoreplication. Spectrograms were generated in RavenPro 1.4 (Bioacoustics Research Program, Cornell University, Ithaca, NY 14850 USA) with a 557 sample size, a 1024 DFT size, spectral overlap of 80.1% (Hann window), 3 dB filter bandwidth at 124 Hz, 111 sample hop size, grid spacing of 46.9 Hz, no clipping, and averaging 1 spectrum. The trill rate of a trill of n syllables was calculated by dividing the count of n−1 syllables by the time span of n−1 syllables (to account for inter-syllable time). The bandwidth was measured by subtracting the 0.5% frequency from the 99.5% frequency and averaged for at least five syllables per trill. Trills were binned into 5-Hz trill rate increments and from each bin the trill with the maximum frequency bandwidth was selected and used to produce the linear regression as described in Podos 1997 [39]. We classified individuals singing trills with a trill rate and a bandwidth that fell less than 4 kHz from the regression line as singing high performance trills and individuals singing trills with a trill rate and a bandwidth that fell more than 4 kHz from the regression line as singing low performance trills. Four kHz was chosen as the dividing point because it resulted in the most conservative grouping; only individuals singing trills with extremely low performances were classified as low performance singers. We statistically analyzed trill performance as a dichotomous variable because, although we expected variation in trill performance of song in each individual, we were especially interested in the variation between regions of outlier singers with especially low trill performance. Trill types varied greatly within and between regions. However, no attempt was made to classify the trill types, and all trills were pooled for analysis. This greatly increased the noise of the data, decreasing the likelihood of finding any pattern, and therefore making this a conservative analytical approach.

Total PCB data were log-transformed and analyzed with a two-way ANOVA in which species identity and region were fixed effects, followed by a Tukey post-hoc test. The difference in total PCB concentration between sexes and age classes were analyzed with an ANOVA with sex and age class as fixed effects. Congener-specific PCB data were analyzed in the ++ Hudson region with a principal components analysis (PCA). The proportions of tetra- and penta-chloro (4-Cl+5-Cl), and hepta- and octa-chloro (7-Cl+8-Cl) PCBs in the ++ Hudson were compared separately between the two species with Student's t-tests.

The regional differences of black-capped chickadee glissando and interval ratios were analyzed with an ANOVA with region as a fixed effect, and significant differences were followed with a Tukey post-hoc test. The average glissando and interval ratios in each region were compared to ratios published in Christie et al. [43]. A Chi-square test was used to assess whether proportions of stereotypical and variable individual black-capped chickadee singers differed between regions.

In the song sparrow analysis, a Chi-square test was used to test whether proportions of individual song sparrows singing high performance and low performance trills differed between regions. In the 2009 subset of song sparrows in which individual males were sampled for song, location of the individual, PCB profile, and body condition, Generalized Linear Models (GLMs) were built and ranked using the Akaike Information Criterion corrected for small sample sizes (AICc).

Body condition variables and blood mercury concentrations were analyzed with an ANOVA with region as a fixed effect, and Tukey post-hoc tests.

Analysis of PCB profiles in the blood of black-capped chickadees and song sparrows along the contamination gradient in northeastern New York (Figure 1) revealed region- and species-dependent differences in PCB concentration and congener composition. The average concentration of the 41 congeners analyzed for black-capped chickadees (Table S1A) and song sparrows (Table S1B) varied greatly among the five regions and between species, but not between sexes or age classes (Black-capped chickadees: 74% male, ANOVA: F = 1.31, P = 0.29; Song sparrows: 90% male, ANOVA: F = 1.39, P = 0.25).

The black-capped chickadee total PCB concentration in the ++ Hudson region was higher, but not significantly so, than total PCB concentrations found in individuals from the + Hudson region (Figure 2A). Chickadees in the − Ithaca, − Adirondack, and − Hudson regions had significantly lower total PCB concentrations than individuals from the ++ Hudson regions (Figure 2A). The PCB congener profile in black-capped chickadees varied between regions, and chickadees in the + Hudson region appeared to have the largest proportion of higher chlorinated PCBs than in all other regions (Figure 2B).

Song sparrows had significantly higher total PCB concentrations in the ++ Hudson region (the area of highest historic PCB contamination [28]) when compared to all other regions (Figure 2A). While the total PCB concentrations in birds from the + Hudson and − Hudson regions did not differ significantly from each other, the total PCB concentration in song sparrows from the + Hudson region were significantly higher than the PCB concentrations found in song sparrows from the − Ithaca and − Adirondacks regions (Figure 2A). The PCB congener profile in song sparrows varied between regions, and song sparrows in the ++ Hudson region appear to have the largest proportion of lower chlorinated PCBs than in all other regions (Figure 2B).

In regions − Ithaca and − Hudson, the two bird species had a similar congener profile (Figure 2B). In region − Adirondacks, chickadees had higher penta- and hexa- chlorinated PCB levels, while song sparrows had more evenly distributed levels of PCB chlorination (Figure 2B). The greatest difference in PCB content of the two species was in the ++ Hudson region, directly below the point source pollution from the GE plant. To examine the different congener profiles between the chickadees and sparrows in the ++ Hudson region in more detail, a PCA was performed on the PCB congener data (Figure 2C). The difference in congener profiles for chickadees and sparrows partitioned clearly along PC2, which corresponds to the proportion of congeners with a lower degree of chlorination based on factor loading. In our sample from the ++ Hudson region, chickadees and song sparrows did not differ significantly in proportion of hepta- and octa- chlorinated congeners in their PCB load (Figure 2D; t = 1.30, P = 0.20), while song sparrows had a significantly higher proportion of lower chlorinated PCBs (4-Cl+5-Cl) than chickadees (Figure 2D; t = −3.36, P = 0.0016). Similarly, the proportion of lower chlorinated PCBs was higher in song sparrows in the + Hudson region, further downstream from the point source of pollution (Figure 2D; t = −2.66, P = 0.01).

Male black-capped chickadees are known to have a highly stereotyped song, with the glissando and interval ratios varying less than 2% across their entire North American range, according to published values (Figure S1, Table S2, [43]). Yet in our study regions, the glissando ratio varied more than 2% from published values in both the + Hudson and the ++ Hudson regions, and the interval ratio varied slightly more than 2% from the published value in the − Hudson region (Table S2, [43]).

Birds from + Hudson had a significantly higher glissando ratio than birds from all other regions (Figure 3A; ANOVA: F = 9.78, P<0.001). Birds from the − Hudson had a significantly higher interval ratio than birds from all other regions (Figure 3B; ANOVA: F = 6.53, P<0.001). Additionally, individual black-capped chickadees within a region varied in the stereotypy of their glissando ratios (see Figure S3 for examples), but not in the stereotypy of their interval ratios. The proportion of individual chickadees singing inconsistent glissando differed between regions (Figure 3C; Chi-square = 42.54, df = 4, P<0.001). The proportion of individual chickadees singing inconsistent interval ratios was not significantly different between regions (Figure 3D; Chi-square = 7.23, df = 4, P = 0.12).

All trills recorded from male song sparrows during 2006–07 were plotted as trill rate (Hz) versus frequency bandwidth (kHz), with the upper-bound regression line indicating the trill performance limit (Figure 4A, [39]). The proportion of song sparrows that sang high performance trills were significantly different between regions (Figure 4B; Chi Square = 12.80, df = 4, P = 0.01).

In 2009 a subset of song sparrows were sampled for blood PCB loads and body condition, and their songs were recorded. Table 1 shows the GLMs that were built and their corresponding AIC-based weight of support from the data. The variation in the average trill performance for an individual male was best described by the individuals' level of mono-ortho PCBs. The best-supported model contained only mono-ortho PCB concentrations as a predictor, and this model had approximately twice the support of the second strongest model that included both the level of mono-ortho PCBs and a body condition index (tarsus/mass). Although inclusion of body condition creates a model with half the support of the best model, body condition does not appear to be an independent or strong predictor of trill rate: the model with body condition index alone was the third-lowest ranked model and had almost no support from the data. Additionally, our analyses show that the effect of PCB concentrations on trill rate is not mediated by body condition. The model with an interaction between the effects of mono-ortho PCB level and body condition has little support from the data, with only 11% of the support of the best model. Latitude was also a poor predictor of trill performance, with only 3% of the support of the best model.

Black-capped chickadees captured from the five regions showed no differences in wing chord length, body weight, muscle score, fat score, or feather ectoparasite damage (ANOVA: P>0.05), but differed in tarsus length (ANOVA: F = 3.751, P = 0.0141), with ++ Hudson birds having significantly longer tarsi than individuals from − Adirondacks and − Hudson (Tukey's test: P<0.05). Song sparrows captured from the five regions showed no differences in wing chord length, tarsus length, body weight, or feather ectoparasite damage (ANOVA: P>0.05), but differed in muscle score (ANOVA: F = 3.63, P = 0.0079), with ++ Hudson birds having significantly higher muscle scores than + Hudson' birds (Tukey's post hoc test: P<0.05). Sparrows also differed in their fat score (ANOVA: F = 5.286, P = 0.0006), with + Hudson birds having higher fat scores than − Adirondacks and ++ Hudson birds (Tukey's post hoc: P<0.05). Song sparrows from − Adirondacks had higher blood mercury concentrations than song sparrows from all other regions except − Hudson (Figure S4; ANOVA: F = 6.58, P = 0.0025).