application of pattern recognition techniques to ... - Science Direct

Chemosphere, Voi.25, Nos.l-2, Printed in Great Britain

pp 129-134, 1992

0045-6535/92 $5.00 + 0.00 Pergamon Press Ltd.

APPLICATION OF PATTERN RECOGNITION TECHNIQUES TO ASSESSMENT OF BIOMAGNIFICATION AND SOURCES OF POLYCHLORINATED MULTICOMPONENT POLLUTANTS, SUCH AS PCBs, PCDDs AND PCDFs. C.R. Macdonald, R.J. Norstrom and R. Turle. National Wildlife Research Centre, Canadian Wildlife Service, Environment Canada, Hull, Quebec, Canada. ABSTRACT Principal component analysis of PCB congeners and organochlorines in herring gull eggs from four of the Great Lakes reveals that, with the exception of Lake Erie, the two colonies within each lake contain distinctive patterns which did not change significantly between 1983 and 1990. Lake Ontario colonies were modelled by high percentages of mirex and photomirex, while high proportions of dieldrin, heptachlor epoxide and oxychlordane grouped the colonies in lakes Huron and Superior. The Middle Is. colony in the western end of Lake Erie showed high relative amounts of the higher chlorinated PCB congeners, due to continuing contamination from the Detroit R., or resuspension of contaminated sediments. PCDD/F patterns in the same colonies were dominated by 1,2,3,6,7,8 HxCDD and 1,2,3,7,8 PeCDD in lakes Ede, Huron and Superior while Lake Ontario patterns were dominated by 2,3,7,8 TCDD. Lake trout and walleye in the same lakes showed high levels of 2,3,7,8 TCDF which does not accumulate in herring gulls. It is concluded that the application of PCA techniques to Great Lakes data sets provides a useful screening procedure which can generate hypotheses on lake loading patterns and species residue differences by grouping samples with similar residue patterns. INTRODUCTION Pattern recognition techniques such as principal components analysis (PCA), SIMCA modelling and PLS regression have been used increasingly to reveal spatial and temporal patterns in complex data sets, such as PCB and PCDD/F congener patterns in environmental samples. Pre-processing techniques such as transformations, normalisation and standardisation (equal weighting for each variable) 1 permit considerable flexibility as to the type of information which can be extracted from the data. Because of this versatility and its ability to simplify complex data sets, PCA has been used to compare residue patterns between different species of biota2, populations in contaminated and control sites3, and between samples of a single species collected from several sites contaminated from a single source 4. The purpose of the present work is to apply unsupervised PCA techniques (i.e. no comparison to standard mixtures or to a heavily contaminated site) to PCB congener, organochlorine pesticides and PCDD/Fs data in herring gull eggs collected from two colonies from each of lakes Ontario, Erie, Huron and Superior. Our objective is to determine 1) if there are extensive differences in the residue patterns within and

129

]gO

between the four lakes and 2) if there are substantial changes in the dominant residues between 1983 and 1990. Two data sets were used: 1) 32 PCB congeners and 14 organochlorines analysed between 1983 and 1990 and 2) 9 PCDD and PCDF congeners analysed between 1984 and 1990 in the same colonies. METHODS Herring gull eggs were collected and archived as part of the routine herring gull monitoring program of the Canadian Wildlife Service s. Collection sites were: Muggs Is./Leslie St. Spit (O) and Snake Is. (S) in Lake Ontario; Middle Is. (M) and Port Colborne (P) in Lake Erie; Double Is. (D) and Chantry Is. (C) in Lake Huron; Agawa Rocks (A) and Granite Is. (G) in Lake Superior. All contaminants were analysed either annually as part of the herring gull program or from archived samples6. Detailed methods for the collection and storage of specimens, and analytical methods for PCBs, organochlorine pesticides, and PCDD/Fs are given elsewhere6'7'8. Egg samples were analysed for 32 PCB congeners and 11 organochlorine pesticides, 5 PCDD congeners (2,3,7,8-TCDD, 1,2,3,7,8-PCDD, 1,2,3,6,7,8-HxCDD, 1,2,3,4,6,7,8-HpCDD and OCDD) and 4 PCDF congeners (2,3,7,8-TCDF, 2,3,4,7,8-PCDF, 1,2,3,4,7,8-HxCDF, and 1,2,3,6,7,8-HxCDF). Principal components analysis was conducted using SIRIUS Ver 2.2 (Pattern Recognition Systems Ltd., Bergen, Norway). PCDD/F data for lake trout (Salvelinus namaycush) from lakes Ontario, Huron and Superior and walleye (Stizostedionvitreum) from Lake Erie~'1° were added to the herring gull egg data for comparison. The data were analysed after normalising to the total of the respective data sets. Residue levels below the detection limit were replaced either by a value equal to half of the detection limit (PCDD/Fs) or by a random number less than the detection limit (PCB/OCs) 1. Statistical significance of the individual components was determined by cross-validation.

RESULTS AND DISCUSSION In 1983, the total concentration of 32 PCB congeners ranged from a maximum of 23 mg/kg at Snake Is. in Lake Ontario to approximately 7 mg/kg in the Lake Huron and Lake Superior colonies. By 1990, the concentration had decreased to approximately 5 mg/kg in all colonies, except for Middle Island in western Lake Erie which remained at approximately 20 mg/kg. DDE concentrations in 1983 ranged from a maximum of 9.6 and 5.2 mg/kg in the Lake Ontario colonies (Snake Is. and Muggs Is., respectively) to approximately 2 mg/kg in all other colonies. By 1990, DDE in all colonies had declined to 2 to 3 mg/kg. Principal components analysis of the PCB congener/OC normalised data indicated large differences in the patterns between lakes, and in the case of Lake Erie, between colonies. The first two statistically significant components explained 92.9% of the total variation and were based on a relatively small number of contaminants (Fig. 1). The first component (83.7%) grouped most of the samples from the two Lake Superior colonies on the basis of relatively high proportions of DDE, and all of the Middle Is., western Lake

131

PCB180,153, 138,182

DDE

A ¢~1

L Ontario tO

mirex photomirex

L Ede

O.

E

8 "6 L,.

11.

[

//

,o

dieldrin,PCB66 heptachlorepoxide oxychlordane

1

Principal Component 1 (83.7%) Figure 1 - PC analysis of 32 PCB congeners and organochlorine pesticides in herring gull eggs in the 8 colonies. See text for symbol legend. Contaminant names on axes refer to major modelling compounds. Erie, samples on the basis of the higher chlorinated PCBs. Port Colborne (eastern Lake Erie) samples were less strongly modelled by PC 1 indicating that they contain levels of the higher chlorinated PCB congeners intermediate to that of Middle Is. and the other colonies. PC 2 (9.2%) grouped all of the Lake Ontario samples on the basis of high relative amounts of mirex and photomirex, while the Lake Huron and Lake Superior samples were grouped by dieldrin, PCB 66, heptachlor epoxide and oxychlordane. The comparison of residue patterns in the herring gull colonies using PCA supports several intedake comparative studies which have shown similar patterns in sediments" and fish TM. The concentration of PCBs in the western end of Lake Erie have been shown to be among the highest in the Great Lakes 1~, and to be comprised predominantly of highly chlorinated congeners, due to input from the Detroit Riveris, consistent with the pattem observed in the gulls. The close grouping of the Middle Island samples collected from 1983 to 1990 in Figure 1 suggests that the character of the source of the PCBs has not changed markedly during that time, as a result of continuing input or resuspension of contaminated sediment. Lower levels of the PCBs in the Port Colborne samples suggest that the level of contamination decreases substantially from west to east in the lake, although enough of the odginal signal is present to differentiate the Port Colborne samples from the other colonies in the Great Lakes examined here. The high mirex and photomirex residues in Lake Ontario biota have been reported elsewhere 14'15and have been attributed to manufacturing sources in the Niagara and

132

Oswego Rivers. The clear separation of the lake Ontario samples in the present analysis and close grouping of the 1983 to 1990 samples indicates that mirex/photomirex remain a unique feature of the residues in Lake Ontario. The source patterns of contaminants, particularly the organochlorine pesticides, to Lake Huron and Superior are less clearly defined. Mass balance estimates TM have suggested that atmospheric deposition was responsible for approximately 90% of the PCB and DDT residues entering lakes Superior and Huron, but only 13% of the PCB and 22% of the DDT entering Lake Erie. PCA of the PCB congeners without the organochlorines (Macdonald, Norstrom and Turle, unpubl, data) shows that the Huron and Superior colonies contain primarily moderately and less chlorinated congeners, consistent with an atmospheric source to the lakes lz. Assuming that the residues present in the gulls are transferred through the food web at similar rates, then the higher percentage of organochlorine residues and metabolites in the gull eggs from lake Superior and lake Huron colonies probably indicates that organochlorine pesticides make up a larger fraction of the source signal in these lakes. In 1984, total PCDDs ranged from relatively constant values of 45 to 55 p.g/kg (w.w.) for the colonies in lakes Superior and Huron to maxima of 114 and 160 i~g/kg in Muggs Island and Snake Is., respectively, in Lake Ontario. In contrast, total PCDFs showed little spatial variability and remained relatively constant at approximately 10 to 20 i~g/kg in all colonies between 1984 and 1990. There was a significant decline in total PCDDs in Chantry Is., Port Colborne, Muggs Is. and Snake Is. colonies between 1984 and 1990, however total PCDFs declined significantly only in the two lake Erie colonies (Middle and Port Colborne) and Chantry Is. in Lake Huron. Spatial variability of total concentration of the same PCDD congeners in lake trout 9 was greater than in the herring gull, and ranged from a maximum of 55.6 p.g/kg in Lake Ontario to only 6.72 p.g/kg in Lake Superior. Total PCDFs in fish followed a similar pattern, reaching 55.4 in lake Ontario lake bout and only 19.4 p.g/kg in Lake Superior trout. Comparison of the PCDD/F patterns in the gulls and fish in the four lakes showed extensive spatial and species differences (Fig 2). The first principal component accounted for 71.6% of the variance in the data set and separated three of the fish groups (Huron, Erie and Superior) from the herring gull and Lake Ontario bout samples on the basis of very high proportions (30-40%) of TCDF of the total PCDD/F residues in trout. PC 2 (18.4%) grouped together the gull samples in a gradient ranging from high percentages of TCDD (Lake Ontario) to relatively high percentages of HxCDD, PeCDD and PeCDF, consistent with other interlake comparisons of gull residues ~'18. The lake Ontario lake bout were intermediate to the fish from the other lakes and the herring gull eggs in that they contain relatively high proportions of both TCDF and TCDD. The first conclusion from this analysis is that there are significant differences in the ability of the fish and gulls to accumulate TCDF, which comprised a large proportion of the total PCDD/F residues in the top fish predators throughout the Great Lakes. Gull eggs contain low levels of TCDF which are often below the detection limit of 1 p.g/kg. For example, gull egg/fish ratios range from less than 0.10 for TCDF to 1.5 and 2.7 for TCDD in lakes Ontario and Huron, respectively.

4"TCDD H,xCDD TCDF'

ce 1

,/,,~

I~J

l~etm=

II

-=2 ........

H!,

IPecDD'

Principal Component I (71.6%)

Figure 2 - PC analysis of PCDD/F residues in herring gulls and fish in lakes Ontario, Erie, Huron and Superior. This analysis also indicates that, of the congeners examined here, the pattern of PCDD/F residues in the biota of lakes Erie, Huron and Superior are similar, and are comprised primarily of TCDF, HxCDD, PeCDD and PeCDF. Sediment cores in Lake Huron suggest that the source pattern to the lakes is combustion related and is dominated by the OCDD, HpCDD and HpCDF congeners, although most homologue groups (except TCDD) were present TM. Large differences in bioavaJlability~° and biomagnification of the congeners, however, results in high proportions of TCDF, PeCDD and HxCDD accumulating in fish, particularly at the top of the food web. The use of factors to account for the differences in bioavailability =° and biomagnlfication 7 of PCDD/F congeners have been used previously to reconstruct source patterns from residue data in eggs from herring gulls and other birds2~. In lake trout and walleyeg, TCDF and PeCDF comprise the largest fraction of the PCDD/F residues in lakes Erie, Huron and Superior. in contrast, the dominant residue in both fish and gulls in Lake Ontario is TCDD, which comprises 60-70% of the total residues in gulls end approximately 36% of the residues in lake trout~. This analysis shows that the pattern of PCDD/F residues in the Great Lakes is dominated by the high levels of TCDD in Lake Ontario, while the other three lakes tested here show a general pattern which probably represents the relative distribution of contaminants which results from atmospheric deposition. The large differences in residue patterns in commonly used indicator organisms such as the herring gull and lake trout is due to the inability of the gull to accumulate TCDF which forms a large part of the PCDD/F residue in the lakes.

134

In summary, PCA, the simplest form of pattern recognition techniques, has been used to demonstrate the differences in residue potterns of herring gulls in four of the Great Lakes which can be attributed to differences in source patterns to the lakes. The analysis showed that the residue patterns did not change substantially dudng the 1980's. Patterns were also relatively consistent within each lake, with the exception of western Lake Erie which is influenced by local sources of PCBs. These data support the conclusion that the herring gull is a good regional indicator which reflects the amounts and source patterns of contaminants in an area, and, when used in conjunction with sediments and fish, can provide a useful profile on the characteristics of the residues in their home lake. REFERENCES 1. Sharaf, M.A., D.L. IIIman and B.R. Kowalski (1986). Chemometrics. In_.."P.J. Eiving and J.D. Winefordner. (ed). Chemical Analysis, Monographs on Analytical Chemistry and Its Applications Vol 82. Wiley-lnterscience, Toronto. 332 p. 2. Stalling, D.L., R.J. Norstrom, L.M. Smith and M.Simon (1985). Chemosphere 14: 627643. 3. Schwartz, T.R. and D.L. Stalling (1991). Arch. Environ. Contarn. Toxicol. 20:183-199. 4. Oehme, M., A. Bartonova and J. Knutzen (1990). Environ. Sci. Technol. 24:18361841. 5. Mineau, P., G.A. Fox, R.J. Norstrom, D.V. Weseloh, D.J. Hallett and J.A. Ellenton (1985). In J.O. Nriagu and M.S. Simmons. (ec~. Toxic Chemicals in the Great Lakes. Wiley-lnterscience, Toronto. 6. Turle, R., R.J. Norstrom and B. Collins (1991). Chemosphere 22:201-213. 7. Braune, B.M. and R.J. Norstrom (1989). Environ. Toxicol. Chem. 8:957-968. 8. Norstrom, R.J., M. Simon and D.C.G. Muir (1990). Environ. Pollut. 66:1-19. 9. De Vault, D., W. Dunn, P.-A. Bergqvist, K. Wiberg and C. Rappe (1989). Environ. Toxicol. Chem. 8:1013-1022. 10. Zacharewski, T., L. Safe, S. Safe, B. Chittim, D. De Vault, K. Wiberg, P.-A. Bergqvist and C. Rappe (1989). Environ. Sci. Technol. 23:730-735. 11. Thomas, R.L. and R. Frank. 1983. In: D. Mackay, S. Paterson, S.J. Eisenreich and M.S. Simmons. (eds.) Ph.ysical Behavior of PCBs in the Great Lakes. Ann Arbor, MI: Ann Arbor Science. p. 245-269. 12. Baumann, P.C., and D. M. Whittle. (1988). Aquat. Toxicol. 11:241-257. 13. Oliver, B.G. and R.A. Bourbonniere. (1985). J. Great Lakes Res. 11:366-372. 14. Kaiser, K.L.E. 1974. Science 185:523-525. 15. Norstrom, R.J., D.J. Hallett, F.I. Onuska and M.E. Comba (1980). Environ. Sci. Technol. 14: 860-866. 16. Strachan, W.J.M., and S.J. Eisenreich (1988)o Report to the International Joint Commission, Windsor, Ontario, Canada. 113 p. 17. Macdonald, C.R. and C.D. Metcalfe (1991). Can. J. Fish. Aquat. Sci. 48:371-381. 18. Stalling, D.L., P.H. Peterman, L.M. Smith, R.J. Norstrom and M. Simon (1986). Chemosphere 15:1435-1443. 19. Czuczwa, J.M. and R.A. Hites (1984). Environ. Sci. Technol. 18:444-450. 20. Kuehl, D.W., P.M. Cook, A.R. Batterman, D. Lochenbach and B.C. Butterworth (1987). Chemosphere 16: 667-679. 21. Norstrom, R.J., B.M. Braune, C.R. Macdonald, M. Simon and D.V. Weseloh (1989). Presented at DIOXIN 89, Toronto, Ont.