Although considerable progress has been made in reducing concentrations of persistent organic compounds in the environment, these contaminants are still found in many taxa. Here, we investigate the relationship between environmental contamination and avian egg coloration to examine whether egg coloration could be used as a bioindicator of contaminant load.
The Herring Gull Monitoring Programme has documented changes in contaminant levels found in herring gull Larus argentatus Pontoppidan eggs across the Great Lakes for nearly 40 years. We measured the coloration of these eggs using reflectance spectrophotometry and evaluated the influence of a suite of contaminants on egg colour using general linear mixed models.
Herring gull egg coloration was related to the levels of environmental contaminants found within the eggs. Our findings reveal a negative association between blue-green chroma and a principal component explaining contaminant load. We also found that ultraviolet chroma varied positively with contaminant load, whereas brown chroma was not significantly related to contaminant load. In addition, a cross-validated discriminant function analysis was able to correctly classify 84% of eggs to either high or low contaminant load. This is an important first step in assessing the utility of using egg coloration as a proxy for contaminant load in a colonially breeding waterbird.
Synthesis and applications. Our study utilized a large, multi-year data set to provide the first evidence that a suite of environmental contaminants appear to influence avian eggshell coloration. We also found that objective spectrophotometric measurements provide a reliable tool for assessment of egg contaminant load, and we provide a discriminant function for contaminant classification directly in the field. Our findings should be broadly relevant, because the pigments responsible for avian egg coloration are shared across all birds. The application of eggshell colour as a bio-monitoring tool has important conservation and management applications; measuring egg coloration may provide a rapid, inexpensive and nondestructive means of estimating contaminant levels in the environment. This may provide an essential tool for monitoring areas or species of concern, as well as evaluating potential human health risks, by identifying populations supported by an environment that requires more attention and potentially environmental remediation.
All organisms currently face significant challenges imposed by global climate change (Root et al. 2003) and other anthropogenic changes (Rosenzweig et al. 2008; Johnston & Roberts 2009); not even organisms living in the deep ocean can escape exposure to environmental contaminants (de Boer et al. 1998). Therefore, it is becoming increasingly important to develop new monitoring tools to assess habitat quality and manage sensitive areas and wildlife. Bird eggs may provide an efficient means of assessing the environmental quality of avian breeding habitats, particularly with respect to industrial processes and agricultural pesticides. Many persistent organic compounds associated with these activities are known to bio-accumulate in animal tissues as they are transferred from low trophic levels to higher ones. The influence of the bio-accumulation of these contaminants on avian reproduction became readily apparent during the late 1960s (Ratcliff 1967; Hickey & Anderson 1968), particularly through eggshell thinning induced by exposure to dichlorodiphenyldichloroethylene (DDE), a metabolite of the persistent insecticide dichlorodiphenyltrichloroethane (DDT) (Gilbertson 1974). Within the Great Lakes, the levels of persistent organic particulates have decreased dramatically over the last half-century (Hebert, Norstrom & Weseloh 1999); however, significant quantities of organic compounds are still prevalent in colonially nesting waterbirds (Lavoie et al. 2010). Recent models also suggest that cycles of contamination may be linked to oscillating currents and global patterns of climate change (Bustnes, Gabrelsen & Verreault 2010). Such studies reveal the importance of continued monitoring to track long-term patterns and evaluate potential risks to plants, animals and humans from environmental contamination.
Despite dramatic variation in avian egg coloration within and among species (Collias 1993; Kilner 2006), only two related pigments are primarily responsible for this variation: biliverdin (blue-green in colour) and porphyrin (brown in colour) (Kennedy & Vevers 1976), both of which are derived from haem biosynthesis (McGraw 2006). A variety of porphyrins are created endogenously through enzymatic interactions and these nonmetallic porphyrins create haem through the addition of an Fe+ ion (McGraw 2006), which can be oxidized to create biliverdin (McDonagh 2001). The concentration of one of these pigments, porphyrin, has been recommended as a bioindicator in faecal samples (Akins et al. 1993; Casini et al. 2003). However, as both porphyrin and biliverdin occur along the same biochemical pathway, arguments for the use of porphyrin may also be relevant for biliverdin (Mateo et al. 2004; Jagannath et al. 2008).
A number of factors support the possible utility of avian pigments, and particularly avian egg pigments, as bioindicators of environmental stress. Proximity to urbanization (Horak et al. 2000) and exposure to polychlorinated biphenyls (PCBs) (McCarthy & Secord 2000; Bortolotti, Fernie & Smits 2003; Bortolotti, Smits & Bird 2003) are known to influence avian plumage and soft part coloration. Egg coloration in birds may be similarly influenced by environmental quality as it has been linked with female body condition (Morales, Sanz & Moreno 2006; Soler et al. 2008), rainfall (Avilés et al. 2007), soil calcium availability (Gosler, Higham & Reynolds 2005) and health (Moreno et al. 2005; Martínez-de la Puente et al. 2007), yet this possibility has received surprisingly limited attention.
The relationship between environmental contaminants and eggshell pigmentation was recently examined in the Eurasian sparrowhawk Accipiter nisus Linnaeus (Jagannath et al. 2008). In a sample of eggs collected across the United Kingdom in a single year, Jagannath et al. (2008) found that blue hue was positively correlated with DDE concentration, while chroma decreased with DDE concentration. An experimental study has also shown that the contamination by lead caused a 53-fold increase in protoporphyrin and a 66-fold increase in biliverdin in faecal samples of mallards Anas platyrhynchos Linnaeus (Mateo et al. 2004). These findings are consistent with the observation that organochlorines, halogenated hydrocarbons and heavy metals influence the haem biosynthesis pathway (Kennedy et al. 1998; Casini et al. 2003; Mateo et al. 2003b, 2004).
Our objective was to determine whether eggshell coloration could serve as a nondestructive bioindicator of environmental stress, using herring gulls Larus argentatus Pontoppidan as an indicator species. Herring gulls can provide insight about changes in the health or quality of their ecosystem, and early research on Great Lakes herring gulls documented that DDT decreased egg hatchability (Keith 1966). Herring gulls have therefore been the focus of the long-term Great Lakes Herring Gull Monitoring Programme, spanning multiple colonies along the shores of the Great Lakes in Canada and the United States, with the objective of examining the concentrations and effects of environmental contaminants in herring gulls and their eggs. This programme has documented the levels of various organochlorines and metal contaminants in this species for 40 years (Hebert, Norstrom & Weseloh 1999). Fortunately, the levels of most legacy contaminants in Great Lakes herring gull eggs have declined significantly because the use of DDT was banned in 1972 and 1974 in the United States and Canada, respectively (Pekarik & Weseloh 1998; Jermyn-Gee et al. 2005). Herring gull eggshells contain both biliverdin and porphyrin (Kennedy & Vevers 1976), which allows for substantial variation in colour. This variability in eggshell coloration and dramatic temporal variation in contaminant load, as well as the herring gull's susceptibility to organochlorines (Neimi et al. 1986; Breton, Fox & Chardine 2008), makes this system ideal for examining the possible influence of contaminants on egg coloration. The eggs used in this long-term monitoring programme have been stored in a national archive and are available for continued research projects.
In this study, we investigated the relationship between egg coloration and environmental contaminants measured through the Great Lakes Herring Gull Monitoring Programme. Our objective was to examine whether environmental contaminants influence eggshell coloration, and to determine whether eggshell coloration could serve as a useful proxy for the degree of environmental contamination. First, we examined the possible influence of a suite of contaminants on eggshell coloration. On the basis of previous research on the effects of DDE concentration on egg colour (Jagannath et al. 2008), we predicted that contaminant load would be negatively related to blue-green chroma. We then used a cross-validated partial least squares linear discriminant function to determine whether raw spectral data obtained from field-portable spectrometers could be used to classify the contaminant load in the field.
Materials and methods
Long-term data set
The National Wildlife Research Centre Specimen Bank in Ottawa, Ontario, Canada, houses eggshells from eggs sampled through the Great Lakes Herring Gull Monitoring Programme (1971–2011). To our knowledge, this is the longest-running annual contaminants programme on an indicator species and has been conducted in a region that has experienced a dramatic change in the form and level of environmental contaminants (Hebert, Norstrom & Weseloh 1999; Jermyn-Gee et al. 2005). The data collection protocol for this project has been relatively consistent across all sampling years (Fox, Grasman & Campbell 2007). Briefly, 15 colonies (Fig. 1) were visited once per year, during early incubation, and 13 eggs, one per completed clutch, were collected from each location and stored at 4 °C (Fox, Grasman & Campbell 2007).
The protocol for organochlorine extraction has been described in detail (Pekarik & Weseloh 1998). Briefly, within 2 weeks of collection, egg contents were placed in hexane-rinsed jars, mixed with anhydrous sodium sulphate and stored at −20 °C. Gas chromatography was used to assess the level of contaminants in these aliquots. First, the lipid content was eluted from the column and assessed with a gravimetric analysis. Lipid soluble organochlorines were then separated and fractionated on Florisil. Another portion of the homogenized aliquot was analysed to determine organochlorine concentration from the lipids. The first fraction contained DDE, mirex, photo-mirex and a range of PCB Arochlors, whereas the second contained DDT, dichlorodiphenyldichloroethane, alpha-hexachlorocyclohexane, oxy-chlordane and beta-hexachlorocyclohexane. More details on the extraction methods, minor alterations to protocol and extraction of other contaminants can be found in published technical accounts (Bishop et al. 1992; Pekarik & Weseloh 1998; Pekarik et al. 1998; Jermyn-Gee et al. 2005). The levels of contaminants used in this study were from egg samples that were pooled by colony, where each pool was used to estimate average colony-level contaminant load (Pekarik & Weseloh 1998).
Egg colour assessment
We measured the coloration of 696 herring gull eggs from the National Wildlife Resource Specimen Bank at the National Wildlife Research Centre that had corresponding contaminant information for all 15 colonies from 6 different years: 1977, 1981, 1985, 1989, 1993 and 1997. We measured egg coloration using a reflectance spectrophotometer (USB4000; Ocean Optics, Dunedin, FL, USA) with a portable, full-spectrum light source (PX-2 pulsed xenon; Ocean Optics). All reflectance measurements were calculated relative to a Spectralon reflectance standard (WS-1-SL; Ocean Optics). We measured each egg twice, once on the blunt end and once on the pointed end, avoiding pigmented spots and averaged these measurements to obtain one spectrum (300–700 nm) per egg (Fig. 2). To approximate biliverdin content, we calculated blue-green chroma as a proportion of reflectance in the blue-green region (450–550 nm) relative to that of the entire avian visible spectrum (300–700nm). To approximate porphyrin content, we calculated brown chroma as a proportion of reflectance in the brown region (600–700 nm) relative to that of the avian visible spectrum, which is based on the absorbance spectrum of porphyrin (Scalise & Durantini 2004) and has been used in the related ring-billed gull Larus delawarensis Ord (Hanley & Doucet 2009). As with brown and blue-green chroma, we calculated ultraviolet chroma as the proportion of reflectance in ultraviolet region (300–400 nm) relative to that of the avian visible spectrum. We restricted our estimates to these regions of the spectrum to avoid areas where either biliverdin or porphyrin could be responsible for variation in reflectance.
Two studies have confirmed that blue-green chroma provides an adequate estimate for biliverdin content (Moreno et al. 2006; Lopez-Rull, Miksik & Gil 2008) in species that contain both biliverdin and porphyrin pigments (Kennedy & Vevers 1976; Miksik, Holant & Deyl 1996), while a recent study found that avian eggshell coloration was related to both porphyrin and biliverdin concentration across 49 species (Cassey et al. 2012). Our measures of blue-green and brown chroma were strongly negatively correlated (r = −0·53, N = 696, P <0·0001), suggesting that these chroma values distinguish between eggs differing in relative pigment composition, corresponding with large variation in blue-green to brown coloration found in herring gull eggs. Nevertheless, as we were unable to quantify the pigment concentration, and gulls have relatively high concentrations of both biliverdin and porphyrin (Cassey et al. 2012), our colour estimates should be viewed as a suggestive proxy for pigmentation only. As an applied tool, we are interested in how eggshell reflectance, not necessarily pigment concentration, responds to the levels of environmental contaminant. However, using colorimetric variables that have been shown to relate to pigment composition may allow us to elucidate possible mechanisms for the influence of contaminants on eggshell colour.
Although some studies have found no evidence of egg fading in long-term data sets (Soler et al. 2005; Jagannath et al. 2008), a more recent study found an influence of storage duration on the coloration of eggshells (Cassey et al. 2010). Eggshell fading should result in older eggs having lower chroma values than recently collected eggs because old eggs have been subject to more oxidation and degradation. For both blue-green and brown chroma, we found the opposite pattern, with chroma decreasing in more recent years (blue-green chroma: r = −0·07, n = 696, P =0·07; brown chroma: r = −0·21, n = 696, P <0·0001). This lack of substantial fading may not be surprising, because herring gull eggs are exposed infrequently after laying because of long attentive periods by the parents (Drent 1970; Pierotti & Good 1994), and these eggshells were stored in sealed containers away from light shortly after being collected. Nevertheless, we realize that some short-term fading may occur (Cassey et al. 2010; Moreno, Lobato & Morales 2011). To account for possible fading, we included year of collection as a covariate within all of our models.
All data exhibited normal kurtosis and skewness values (all < 2). To account for the fact that many of these contaminants are correlated and co-occur in herring gull diets, we summarized variation in contaminant concentration by running a principal components analysis (PCA) to simplify five contaminants that were consistently detected within eggs (i.e. not at trace or undetectable levels): DDE, PCB-1260, 2,3,7,8,-tetrachlorodienzo-p-dioxin (dioxin), hexachlorobenzene (HCB), and trans-non-Achlor; however, we realize that some of the contaminants we did not include might also influence egg coloration. Our PCA resulted in one principal component that accounted for 71·68% of the variation in contaminant load and received positive loadings from all five contaminants (PC1, factor loadings: DDE = 0·46, PCB-1260 = 0·35, dioxin = 0·44, HCB = 0·49, trans-non-Achlor = 0·48). We used this PC score as a general proxy for contaminant load (hereafter ‘contaminant load’). Contaminant load scores were calculated for years after 1986, since prior to this, values of some contaminants were lacking and the sampling protocol differed (Turle, Norstrom & Won 1986).
We used general linear mixed models to determine the relationship between individual egg coloration and our principal component score of colony-level contaminant load. This approach allowed us to account for within-colony variation in eggshell coloration that would be ignored if we used mean egg colour at the colony level. However, analyses using mean colony-level egg colour yielded results similar to those from the more inclusive general linear mixed models we present here (D. Hanley & S.M. Doucet, unpublished data). In each model, we included a colorimetric variable as the dependent variable, and year of collection, contaminant load, colony (random effect), as well as an interaction between colony and year as predictor variables. We included an interaction because contaminant levels have not changed equally across these colonies.
While the objective of our study was to investigate the influence of contaminant load on egg coloration, previous work has shown that egg colour can be influenced by climate (Avilés et al. 2007). Unfortunately, we could not control for local climate variation at our colonies because there were no meteorological stations located at the colonies. Nevertheless, to consider the possibility that directional patterns between egg colour and contaminant load could be explained by long-term climatic changes across the study area, we gathered temperature (°C) and total rainfall (mm) from the nearest available weather stations (30·4 ± 41·1 km) at the National Climate Data and Information Archive (Environment Canada 2012). We ran a general linear mixed model to determine whether changes in contaminant load were related to the changes in these climatic variables over the sampling period. Both colony and year were significant predictors of variation in contaminant load, while neither climatic variable was (whole model: r2 = 0·91, F15,10 = 6·87, P =0·002; colony: F12,25 = 3·22, P =0·04; year: β = −0·71, F1,25 = 38·51, P =0·0001; mean temperature: β = −0·57, F1,25 = 1·81, P =0·21; mean rainfall: β = 0·50, F1,25 = 4·61, P =0·06), suggesting that any patterns we detect between egg colour and contaminant load should be statistically independent of change in climatic conditions.
Is it possible to classify contaminant load based on eggshell reflectance?
We classified contaminant using MacQueen's K-means clustering (MacQueen 1967), which identifies natural groupings in data, resulting in two distinct groups representing low and high contaminant load (two-group cluster: F1,453 = 1371·40, P <0·0001). We then used partial least squares linear discriminant analysis (PLS-LDA) to classify eggs into these low and high contaminant categories based on their spectral reflectance (binned every 10 nm). We used a cross-validation approach where the initial analysis was performed on only 80% of the data set and the discriminant function was then tested on the final 20% of the data. We chose to use the PLS approach, which is robust against violations in multivariate normality and multicollinearity assumptions (Barker & Rayens 2003). As the spectral data are inherently correlated (e.g. reflectance at 300 nm is probably highly correlated with reflectance at 310 nm), the PLS method is appropriate while the standard LDA is not. The PLS analysis reduces the dimensionality of the raw spectra (here using 10 PLS axes), reducing potential overfit of the model which may occur in a LDA when there are many parameters but few cases (Delalieux et al. 2005). This technique has been used to classify spectral data in various fields (Chevallier et al. 2006; Latreille et al. 2007; Ariana & Lu 2008), and shares the familiar LDA classification function that can be easily used in the field. Although a feed forward neural network model also provides a method for mapping nonlinear relationships between a set of input and output variables (Moshou et al. 2004), our preliminary analyses showed that the PLS-LDA outperformed this analysis and has the additional benefit of producing a familiar classification function for use in the field.
Does contaminant load explain variation in egg coloration?
Contaminant load predicted variation in blue-green chroma (whole model: r2 = 0·22, F28,425 = 4·40, P <0·0001; colony: F13,28 = 2·76, P =0·001; year: β = −0·63, F1,453 = 20·14, P <0·0001; colony*year: F13,28 = 3·35, P <0·0001; PC1: β = −0·48, F1, 453 = 8·06, P =0·005) and ultraviolet chroma (whole model: r2 = 0·52, F28,425 = 16·45, P <0·0001; colony: F13,28 = 10·43, P <0·0001; year: β = 0·86, F1,453 = 59·22, P <0·0001; colony*year: F13,28 = 3·56, P <0·0001; PC1: β = 0·39, F1, 453 = 8·36, P =0·004). However, contaminant load did not significantly predict brown chroma (whole model: r2 = 0·16, F28,425 = 2·84, P <0·0001; colony: F13,28 = 2·78, P =0·001; year: β = −0·23, F1,453 = 2·50, P =0·11; colony*year: F13,28 = 0·78, P =0·68; PC1: β = 0·06, F1,453 = 0·12, P =0·73).
Can raw spectral reflectance data be used to predict contaminant load in the field?
The discriminant function trained on 80% of the original data (N = 363), correctly classified 81% of the eggs to the correct contaminant cluster of low or high contaminant load. The function performed similarly well with cross-validation when classifying the remaining 20% (N = 91) of eggs, correctly assigning 84% of the validation group to the proper contaminant cluster. Overall, a single linear discriminant performed reasonably well (Fig. 3) at predicting contaminant load, considering that this analysis does not control for colony or year effects that also predict colour. To use this discriminant function, two classification functions, D(low, x) and D(high,x) are compared, such that if D(low, x) > D(high, x), the egg is classified as having a low contaminant load and if D(high, x) > D(low, x), the egg is classified as having a high contaminant load.
This type of function would be easy to implement in the field through the use of computer templates or code (see Appendices S1 and S2 in Supporting Information for examples).
On the basis of the analyses of a long-term data set, our findings reveal significant associations between a principal component summarizing contaminant load, representing a suite of persistent organic contaminants and egg coloration in herring gulls. We found that as contaminant load increased, blue-green chroma decreased while ultraviolet chroma increased, when controlling for colony and time since collection. Our findings support the only other study to examine the relationship between egg colour and contaminant load (Jagannath et al. 2008) and provide the first evidence that this pattern holds across a suite of contaminants within a large, multi-year data set. We also provide a discriminant function that can effectively classify herring gull eggs into high and low contaminant classes in the field using egg reflectance spectra. Our findings suggest that egg coloration could be used as a bioindicator of contaminant load and provide further support for the use of colonial waterbirds for monitoring environmental quality (Kushlan 1993).
The only other study to examine the relationship between environmental contaminants and egg coloration found a negative association between DDE levels and chroma of Eurasian sparrowhawk eggs, and a positive association between DDE levels and blue hue (Jagannath et al. 2008). Although we did not consider the impact of contaminant load on hue, we did find that chroma, in our case blue-green chroma, was negatively associated with overall contaminant load, suggesting the potential for broader applicability of both studies, particularly as the eggs of many species contain the blue-green pigment biliverdin (Kennedy & Vevers 1976). Our findings build on those of Jagannath et al. (2008) by showing that this pattern was conserved across a long-term data set. Our findings also reveal that blue-green chroma was negatively related to a suite of contaminants rather than a single contaminant, which is important because wild birds will experience multiple environmental contaminants.
This relationship between environmental contaminants and eggshell coloration may relate to the physiological mechanisms of pigment production. Previous work has shown that the contaminants can influence the haem biosynthesis pathway by either enhancing or inhibiting the production of porphyrin or biliverdin (Casini et al. 2003). For instance, upstream degradation of haem through induced haem oxygenase activity has been proposed as a mechanism to explain increases in biliverdin associated with dioxin toxicity (Niittynen et al. 2002). In addition, some species exposed to lead poisoning produce excess biliverdin (Mateo et al. 2003a, 2004), whereas others produce excess haemoglobin (Styles & Phalen 1998; Pollock 2006). Based on this previous research, the negative relationship between contaminant load and blue-green chroma in herring gull eggs could be caused by the combined influence of those contaminants on the haem biosynthesis pathway. However, the herring gulls in this study were exposed to a suite of contaminants. Because the concentrations of these contaminants are correlated (Peakall & Fox 1987; Wiemeyer, Bunck & Stafford 1993), this type of data set cannot be used to assess the individual effects of contaminants on egg coloration.
Another nonmutually exclusive possibility is that demographical factors may mediate a relationship between egg coloration and exposure to contaminants. For example, egg coloration varies with age in some species (Moreno et al. 2005; but see, Hargitai, Herenyi & Torok 2008). If environmental contaminants have a differential influence on survival between age classes, variation in egg coloration may be mediated through indirect age-specific effects because of altered population level demographics, rather than a direct effect of contaminants on pigment production. Long-term population studies across a gradient of contaminant exposure would be required to assess such a hypothesis. Future studies involving the experimental manipulation of individual contaminants have the potential to greatly enhance our understanding of the direct and indirect effects of contaminants on the haem biosythesis pathway and, consequently, on avian eggshell pigment deposition.
By tracking decades of decline in contaminants on the Great Lakes, the Herring Gull Monitoring Programme provided a natural experiment to examine the long-term effects of environmental contaminants on eggshell coloration. In addition to demonstrating the capacity for contaminant load to predict blue-green and ultraviolet chroma, we also developed a discriminant function that allows raw spectral data to classify eggs into high and low contaminant levels with greater than 80% success. One might assume that the function would work best on the freshest eggs; however, the function performed similarly well across years (1989 = 87%, 1993 = 79%, 1997 = 82%). We chose to use raw spectral data rather than chroma values in our discriminant analyses to develop an approach that would minimize data processing so that eggs could quickly be classified in the field, and to provide consistency across studies, as researchers often use different chroma metrics (reviewed in, Cherry & Gosler 2010). Because most environmental data, including spectral data, violate the multivariate normality and heterogeneity of variance assumptions (e.g. Alden, Dauer & Rule 1982), we used a PLS-LDA which is not constrained by the same assumptions as a normal linear discriminant analysis (Barker & Rayens 2003). However, previous research using LDA for environmental assessment (Christman & Dauer 2003) has rightly pointed out that discriminant analysis is robust against violations of their assumptions (Lachenbruch 1975; Everitt & Dunn 1991; Sharma 1996) and when prediction is the goal, the percentage of correctly assigned values from an independent validation data set, which in our case was 84%, is the best means for assessing the influence of these violations. Our findings therefore suggest that egg coloration has the potential to be used to approximate contaminant load in the field, and the availability of numerous handheld, battery-operated spectrophotometers make the application of such a bioindicator very feasible.
While large variation in herring gull egg coloration was useful for an initial test of the potential utility of using egg colour as a bioindicator, large inter-clutch variation in egg colour may pose a challenge to assessing contaminant loads. Moreover, many other factors are known to influence egg coloration in birds including genetic determination (e.g., Punnett 1933; Hardiman, Collins & Urban 1975) physiological condition (Soler et al. 2008; Morales, Velando & Torres 2011), life-history traits such as age (Moreno et al. 2005; but see, Hargitai, Herenyi & Torok 2008) and environmental factors (Gosler, Higham & Reynolds 2005; Avilés et al. 2007). We therefore encourage future examinations of other potential ecological correlates of egg colour, such as age, temperature, rainfall and calcium availability, using geographically and temporally broad data sets.
As egg colour is likely to vary between herring gull colonies for reasons other than the influence of contaminants, it will be critical to determine the baseline variation in colour and examine the influence of contaminant load on this variation within target colonies. Similarly, such baseline data would need to be determined in any new species or population before reflectance data are used to predict contaminant load. We used eggs that were emptied and stored for several years, so future research should calibrate our discriminant function on fresh egg measurements, which will account for fading and potentially different eggshell transmission properties. The application of these findings as a monitoring tool would be most beneficial if future investigations target broadly distributed species with pigmented eggs, particularly those species that are already subject to widespread monitoring, such as herring gulls, bluebirds Sialia sialis Linnaeus, European starlings Sturnus vulgaris Linnaeus, or great tits Parus major Linnaeus.
The evaluation of contaminant levels in biota is important for the conservation of natural resources and for monitoring long-term health risks to humans. Long-term monitoring programmes provide a means to examine the progress of environmental remediation and for forecasting potential health risks. Recent research on great tits and blue tits Cyanistes caeruleus Linnaeus suggests that avian eggs may make valuable bioindicators in general, because these birds appear to be exposed to contaminants in the same way and exhibit similar mechanisms of accumulation and maternal transfer to their eggs (Van den Steen et al. 2010). Together with previous research, our findings suggest that the colour of avian eggs may serve as valuable bioindicator of contaminant load. Moreover, there are only two pigment classes controlling egg coloration in birds (Gorchein, Lim & Cassey 2009), and it is therefore possible that these patterns are conserved across all birds. Egg colour may provide a simple, inexpensive and nondestructive indicator of contaminant concentration, while numerous long-term monitoring programmes of colonial and semi-colonial birds worldwide should facilitate the global application of using avian egg coloration as a bioindicator of environmental contamination.
We thank the National Wildlife Research Centre Specimen Bank for providing access to the eggs, D.V. Chip Weseloh for his help and comments, and to R. McNeil, G. Savard, L. O'Keefe and D. Moore for their logistical support. We are grateful for funding from the Explorer's Club, the Mitacs Elevate programme and the Frank M. Chapman Fund of the American Museum of Natural History to DH and from the Natural Sciences and Engineering Research Council of Canada to SMD. We also thank J. Cuthbert, K. Drouillard, D. Edelstein, W. Green, V. La, D. Mennill, A. Mistakidis, D. Moore, T. Pitcher and L. Shutt for their contributions to this project.