Time trends of cadmium in kidney of moose (Alces alces) from south-central Sweden, 1980-1998.
Swedish monitoring programme in terrestrial biota
Contaminant Research Group
Swedish Museum of Natural History P O Box 50007
SE-104 05 Stockholm
Department of Chemistry
National Veterinary Institute SE-750 07 Uppsala
2000-03-31
TIME TRENDS OF CADMIUM IN KIDNEY OF MOOSE (Alces alces) FROM SOUTH-CENTRAL SWEDEN, 1980-1998
Compiled by Tjelvar Odsjö
Contaminant Research Group Swedish Museum of Natural History SE-104 05 Stockholm
Sweden
INTRODUCTION
The long-term monitoring of persistent and bio-accumulating chemicals in the Swedish environment is part of the Swedish National Environmental Monitoring Programme. It is based on chemical analyses of tissues and organs from species collected in selected reference areas of the Swedish mainland, lakes and coastal areas (Odsjö & Olsson 1979a,b, Bernes 1985). As part of the terrestrial
contaminant monitoring programme, specimens of muscle, liver and kidney of moose (Alces alces) have been collected since 1980 from Grimsö, a reference area of the monitoring programme and a coherent hunting district in the province of Västmanland in south-central Sweden.
Moose, with a diet dominated by twigs and leaves of trees and shrubs (Cederlund et al. 1980), was chosen in the monitoring programme as a representative matrix for biota in the Swedish forest areas.
By future extended collection of samples from further sites in the nation, moose was considered as an ideal matrix also for studies of spatial distribution of environmental pollution and bio-accumulation since the species is distributed almost all over the country.
This paper reports on levels of Cd in cortex of kidney of moose from the period 1980-98. It visualises the time series and summarises results from statistical treatments. The report also gives short accounts for reasons and arguments for selection of kidney of male calves of moose as an appropriate matrix for long-term trend monitoring of Cd.
MATERIAL AND METHODS
Since the start of the collection in 1980, samples of the tissues and organs mentioned above have been collected from approximately 45-50 individuals annually during the hunting season in the autumn and, with special permit also in the winter and spring. Samples were taken from all individuals shot in the area during hunting despite age and sex. This was done from the start of the programme to make it possible to select the most appropriate and homogeneous material for contaminant monitoring according to influence of biological variables (e.g. age, sex, etc.) on the concentrations. The samples were extracted at the slaughter, prepared in laboratory and stored in a temperature of -20 °C until analysis. Calves and, certain seasons also males were initially well represented in the material.
However, the age structure of the material has changed considerably during the period, which may have consequences for the choice of material from a smaller and spatial concentrated population like that in the Grimsö area.
According to the extended hunting period and date of collection, selection of individual calves for analyses was restricted to the period October 1 - April 30 each hunting season. The selection of specimens started with the earliest shot animals each season. (No significant variation in levels of Cd according to date of collection during the hunting season was revealed).
During the course of the period 1980-1996, cadmium was analysed annually by the Swedish Environmental Research Institute (IVL), Analytical Laboratory, Stockholm. The laboratory is accredited by SWEDAC and satisfies the demand according to SS-EN 45 001. From 1997
simultaneous analysis of Cd and 13 other elements was performed by use of a direct-current plasma- atomic emission spectrometer, DCP-AES (SpectraSpan IIIA, Applied Research Laboratories Inc., Valencia, CA, USA). For pre-treatment of the samples, see: Frank and Petersson 1983, 1985.
Those analyses were carried out by the Department of Chemistry, National Veterinary Institute, Uppsala.
For the chemical analysis a sample of approximately 5 g from the cortex of the kidney was prepared from each individual. The samples were analysed by use of AAS with flame according to a Swedish Standard method described by SIS. In order to achieve information on the within-year variation in concentration of the studied population, 10 individual analyses per year were normally carried out.
However, in some years it has not been able to achieve the required number of requested individuals.
From the individual analyses a geometric mean value was calculated and used as a value of the year in the time trend study. In order to study any differences in concentrations of Cd between specimens according to age and sex, analysis of kidney of males and females, one to four years old were carried out initially.
RESULT
Since males, females and different ages were represented in the collected material and a minimum of 10 individual analyses annually were planned for long-term trend monitoring, a decision on what type of material to be utilised was made initially. It was based on tests of the homogeneity of the material in relation to number of samples available in different age classes of the two sexes. In the beginning of the actual period the main part of the shot animals was calves of both sexes. During the course of the period the structure of the material shot has changed; the number of and proportion of calves has decreased. Hence, from 1986 and in some of the following years, less than 10 male calves were shot annually in the population. In many districts of Sweden, low numbers of calves were observed in 1993 compared to 1988-92 (Sand et al. 1994). That made it necessary either to reduce the number of analysed individuals per year or, if possible from a statistical point of view, to supplement the material with female calves or older males. The advantages or drawbacks of that type of addition are discussed below.
Differences in Cd concentrations between female and male calves
Paired t-test based on annual mean concentrations of Cd in kidney of female and male calves (<1 year), respectively from seven separate years during 1982-94 showed no significant difference between concentrations due to sex (p>0.05), (Table 1).
Differences in concentrations in relation to age
From the material collected in 1981, individual samples of male calves (<1 year) as well as of males one to four years old were analysed. Regression analysis showed a significant increase in
concentrations by age (r2=0.878, p<0.05, n=26). The greatest difference in concentration according to age was found between calves and one-year-old individuals, although levels tended to increase also for ages above one year.
Variation in concentration in females and males
To minimise the variation of the calculated mean concentration of Cd in the population, a minimum of 10 individuals per year should be analysed. In order to study the variation in concentrations in male and female calves respectively, the coefficients of variation (CV) and the interquartile range (IQR) of the mean concentrations of both sexes (Table 1) were compared. The IQR is defined as the 75th percentile minus the 25th percentile. It is the most commonly used resistant measure of spread (Helsel and Hirsch, 1995). The variation tended to be greater for females in the first part of the period and somewhat greater for males in the latest years. The difference in spread (both CV and IQR) between females and males were tested by paired t-test. None of the tests, however, showed any significant difference between the sexes. The variation in spread in the female population seems to be somewhat greater compared to the male population, especially in the first phase of the period.
Following that difference, it should be recommended to use a matrix of males prior to females. If samples from both sexes have to be utilised to satisfy the required number of individual analyses, the same share of males/females is recommended from year to year.
Variation in concentrations over time
The time trend study visualises that significant between-year variations could easily be found during the investigated period, despite there is no significant trend. This shows the need of annual sampling and analysis over a considerable time to detect significant trends.
Conclusions for a sampling and analysis strategy
For long-term monitoring of Cd by use of a matrix of kidney of moose, an annual selection of a minimum of 10 male calves for analyses is proposed on the following basis:
* Concentrations of Cd in kidney cortex tend to increase by age. Thus, samples of one age class only should be represented in the long-term trend study. If more than one age class must be utilised, the shares of different classes must be kept alike throughout the series.
* Males and calves tended to be over-represented in the material, at least during the first years of the actual period. However, the population structure changed and the individuals shot later on tended to belong to older age classes. That means that another choice of matrix for long-term trend studies could be necessary.
* Use of tissues from female calves may increase the variation of the calculated mean concentration in the population and should therefore, if possible, be restricted.
* In case of hyponumerary male calves, samples of female calves may be used as a supplement prior to use of older individuals.
* Time trend studies of cadmium should be carried out over a longer period in order to detect a significant trend due to significant normal between-year variation.
Long-term trend of cadmium in kidney of moose
The analytical results are displayed in the Figure, which visualises the trend in cadmium concentration in kidney of moose for the period 1980-1998 and the statistics. The time series is continuously updated once a year when new material is collected and analysed.
Cadmium concentrations in kidney of moose show no significant log-linear or linear change during the period (p<0.640).
The number of years required to detect an annual change of 5% is 15 years.
The ANOVA test shows that the smoothed line for concentration of cadmium in kidney indicates a significant non-linear trend component (p<0.034).
The overall geometric mean value of cadmium in kidney of moose from Grimsö is 910 ng/g (fresh weight) for the period 1980-1998.
DISCUSSION
The moose has elsewhere been shown to be a suitable animal for monitoring of bio-available and bio-accumulating elements occurring in the natural environment (Frank and Petersson 1984). In the present time trend study, the slope of Cd concentrations for the whole period indicates an annual increase of only 0.6% is indicated. However, the slope is not statistically significant. The smoothed line based on a 3-point running mean smoother indicates a significant non-linear trend component by ANOVA (p<0.034). Consequently, the number of years required to detect a log-linear trend will be longer (15 years to detect an annual change (log-linear) of 5% with a power of 80%, one-sided test). These facts favour long-term trend studies in regional and national monitoring before short-term studies. It is essential to know the normal between-year variation in order to distinguish true incidents from natural background variation. Frank et al. (1994) showed great differences in concentrations of Cr, Cu, Fe and Mo in liver of moose from south-western Sweden when comparing data from 1982 with corresponding data from 1992. Being aware of great natural differences between years, conclusions based on material from few years must take these statistical circumstances into account.
These facts must always be taken into consideration in trend monitoring of environmental contaminants from diffuse sources.
Statistical treatment and graphical presentation (According to A. Bignert, 1998)
Trend detection
One of the main purposes of the monitoring programme is to detect trends. The trend detection is carried out in three steps.
Log-linear regression analyses
Log-linear regression analyses is performed both for the entire investigated time period and for time series longer than ten years, also for the recent ten years. The slope of the line describes the yearly percentage change. A slope of 5% implies that the concentration is halved in 14 years whereas 10%
corresponds to a similar reduction in 7 years and 2% in 35 years. See table 1 below.
Table 1. The approximate number of years required to double or half the initial concentration assuming a continuos annual change of 1, 2, 3, 4, 5, 7, 10, 15 or 20% a year.
1% 2% 3% 4% 5% 7% 10% 12% 15% 20%
Increase 70 35 24 18 14 10 7 6 5 4
Decrease 69 35 23 17 14 10 7 6 4 3
Non-parametric trend test
The regression analyses presupposes, among other thing, that the regression line gives a good description of the trend. The leverage effect of points in the end of the line is also a well-known fact.
An exaggerated slope, caused 'by chance' by a single or a few points in the end of the line, increases the risk of a false significant result when no real trend exist. A non-parametric alternative to the regression analysis is the Mann-Kendall trend test (Gilbert, 1987, Helsel & Hirsch, 1995, Swertz, 1995). This test has generally lower power than the regression analysis and does not take differences in magnitude of the concentrations into account, it only counts the number of consecutive years where the concentration increases or decreases compared with the year before. If the regression analysis yields a significant result but not the Mann-Kendall test, the explanation could be either that the latter test has lower power or that the influence of endpoints in the time series has become unwarrantable great on the slope. Hence, the eighth line reports Kendall's 'ττ', and the corresponding p-value. The Kendall's 'ττ' ranges from 0 to 1 like the traditional correlation coefficient ‘r’ but will generally be lower. ‘Strong’ linear correlation of 0.9 or above corresponds to τ-values of about 0.7 or above (Helsel and Hirsch, 1995, p. 212). EPA recommended this test for use in water quality monitoring programmes with annual samples, in an evaluation comparing several other trend tests (Loftis et al. 1989).
Non-linear trend components
An alternative to the regression line in order to describe the development over time would be some kind of smoothed line. The smoother applied here is a simple 3-point running mean smoother fitted to the annual geometric mean values. In cases where the regression line is badly fitted the smoothed line may be more appropriate. The significance of this line is tested by means of an Analysis of Variance where the variance explained by the smoother and by the regression line is compared with the total variance. This procedure is used at assessments at ICES and is described by Nicholson et al.
(1995).
Outliers and values below the detection limit
Observations too far from the regression line considering from what could be expected from the residual variance around the line is subjected to special concern. These deviations may be caused by an atypical occurrence of something in the physical environment, a changed pollution load or errors in the sampling or analytical procedure. The procedure to detect suspected outliers in this presentation
is described by Hoaglin and Welsch (1978). It makes use of the leverage coefficients and the standardised residuals. The standardised residuals are tested against a t.05 distribution with n-2 degrees of freedom. When calculating the ith standardised residual the current observation is left out implying that the ith observation does not influence the slope nor the variance around the regression line. The suspected outliers are merely indicated in the figures and are included the statistical calculations except in a few cases, pointed out in the figures.
Values reported below the detection limit is substituted using the ‘robust’ method suggested by Helsel & Hirsch (1995) p 362, assuming a lognormal distribution within a year.
Legend to the plots
The analytical results from each of the investigated elements are displayed in figures. A separate plot except for time series shorter than 4 years represents each site/species.
The plot displays the geometric mean concentration of each year (circles) together with the individual analyses (small dots) and the 95% confidence intervals of the geometric means.
The overall geometric mean value for the time series is depicted as a horizontal, thin, dashed line.
The trend is presented by one or two regression lines (plotted if p < 0.10, two-sided regression analysis); one for the whole time period and one for the last ten years (if the time series is longer than ten years). Ten years is often too short a period to statistically detect a trend unless it is of
considerable magnitude. Nevertheless, the ten-year regression line will indicate a possible change in the direction of a trend. Furthermore, the residual variance around the line compared to the residual variance for the entire period will indicate if the sensitivity have increased as a result of e.g. an improved sampling technique or that problems in the chemical analysis have disappeared.
A smoother is applied to test for non-linear trend components. The smoothed line is plotted if p <
0.10. A broken line or a dashed line segment indicates a gap in the time series with a missing year.
The log-linear regression lines fitted through the geometric mean concentrations follow smooth exponential functions.
A cross inside a circle, indicates a suspected outlier, see above. The suspected outliers are merely indicated in the figures and are included the statistical calculations except in a few cases, pointed out in the figures.
Each plot has a header with element, species name, tissue and sampling locality. Below the header of each plot the results from several statistical calculations are reported:
n(tot) = The first line reports the total number of analyses included together with the number of years ( n(yrs) = ).
m = The overall geometric mean value together with its 95% confidence interval is reported on the second line of the plot (N.B. d.f .= n of years - 1).
slope = reports the slope, expressed as the yearly percentage change together with its 95%
confidence interval.
SD(lr) = reports the square root of the residual variance around the regression line, as a measure of between-year variation, together with the lowest detectable change in the current time series with a power of 80%, one-sided test, α=0.05. The last figure on this line is the estimated number of years required to detect an annual change of 5% with a power of 80%, one-sided test, α=0.05.
power = reports the power to detect a log-linear trend in the time series (Nicholson & Fryer, 1991). The first figure represents the power to detect an annual change of 5% with the number of years in the current time series. The second figure is the power estimated as if the slope where 5% a year and the number of years were ten. The third figure is the lowest detectable change for a ten- year period with the current between year variation at a power of 80%.
r2 = reports the coefficient of determination (r2) together with a p-value for a two-sided test (H0: slope = 0) i.e. a significant value is interpreted as a true change, provided that the assumptions of the regression analysis is fulfilled.
y(98) = reports the concentration estimated from the regression line for the last year together with a 95% confidence interval, e.g. y(98)=2.55(2.17,3.01) is the estimated concentration of year 1997 where the residual variance around the regression line is used to calculate the confidence interval.
Provided that the regression line is relevant to describe the trend, the residual variance might be more appropriate than the within-year variance in this respect.
tao = reports Kendall's 'ττ', and the corresponding p-value.
SD(sm) = reports the square root of the residual variance around the smoothed line. The
significance of this line could be tested by means of an Analysis of Variance. The p-value is reported for this test. A significant result will indicate a non-linear trend component.
Below these nine lines are additional lines with information concerning the regression of the last ten years.
Table 1. Mean levels (ng/g, fresh wt) of cadmium in kidney of female and male calves of moose, coefficient of variation (CV), interquartile range (IQR) and differences (diff) in per cent (%). Year equals hunting season, Oct. 1 - April 30.
females males
______________________ ______________________
Year n mean CV IQR n mean CV IQR diff. %
1982 17 1224 71 550 10 793 44 267 -431 -43
1986 4 1328 59 1468 10 1710 27 725 +382 +25
1987 5 1390 72 1475 9 1364 30 750 -26 -2
1989 3 680 (32) (440) 7 734 43 490 +54 +8
1990 5 928 61 945 3 500 (20) (190) -428 -60
1993 5 1328 17 280 4 1335 42 1055 +7 +1
1994 5 796 28 410 9 853 44 575 +57 +7
REFERENCES
Bernes, C. 1985. Monitor 1985. The National Swedish Environmental Monitoring Programme (PMK).
National Swedish Environmental Protection Board, INFORMS. Liber.
Bignert, A. 1998. Comments Concerning the National Swedish Contaminant Monitoring Programme in Marine Biota. Report to the Swedish Environmental Protection Agency.
Cederlund, G., Ljungqvist, H., Markgren, G. and Stålfelt, F. 1980. Food of Moose and Roe-deer at Grimsö in Central Sweden - Results of Rumen Content Analyses. Viltrevy 11:169-247.
Frank, A., Galgan, V. and Petersson, L. 1994. Secondary Copper Deficiency, Chromium Deficiency and trace Element Imbalance in the Moose (Alces alces L.): Effect of Anthropogenic Activity.
Ambio Vol.23, 4-5:315-317.
Frank, A. and Petersson, L.R. 1983. Selection of operating conditions and analytical procedure in multi- element analysis of animal tissue by d.c. plasma-atomic emission spectroscopy. Spectrochim.
Acta 38 B, 207-220.
Frank, A. and Petersson, L. 1984. Assessment of bioavailability of cadmium in the Swedish
environment using the Moose (Alces alces) as indicator. Fresenius Z.Anal.Chem., 317:652- 653.
Frank, A. and Petersson, L.R. 1985. Direct current plasma-atomic emission spectrometer as a simultaneous multi-element tool for analysis of biological materials. Kemia -Kemi 12, 426-430.
Gilbert, R.O. 1987. Statistical Methods for Environmental Pollution Monitoring. Van Nostrand Reinhold, New York.
Helsel, D.R. & Hirsch, R.M. 1995 .Statistical Methods in Water Resources, Studies in Environmental Sciences 49. Elsevier, Amsterdam.
Hoaglin, D.C. and Welsch, R.E. 1978. The hatmatrix in regression and ANOVA. Amer. Stat., 32:17- 22.
Loftis, J.C., Ward, R.C. and Phillips, R.D. 1989. An Evaluation of Trend Detection Techniques for Use in Water Quality Monitoring Programs. EPA/600/3-89/037. 139p.
Nicholson, M.D. & Fryer, R. 1991. The Power of the ICES Cooperative Monitoring Programme to Detect Linear Trends and Incidents. In: Anon. Report of the Working Group on Statistical Aspects of Trend Monitoring. ICES Doc CM 1991.
Nicholson, M.D., Fryer R. and Larsen, J.R. 1995. A Robust Method for Analysing Contaminant Trend Monitoring Data. Techniques in Marine Environmental Sciences. ICES.
Odsjö, T. and Olsson, M. 1979a. Program och arbetsbeskrivning vid miljöprovsbanken. Rapport till Statens naturvårdsverk, 1979-03-15. (In Swedish).
Odsjö, T. and Olsson, M. 1979b. Förslag till miljögiftsprogram inom Program för övervakning av miljökvalitet, PMK. 1979-06-14. (In Swedish).
Sand, H., Cederlund, G. and Pehrson, Å. 1994. Färre kalvar under älgjakten 1993. In: Skogsvilt II.
Sveriges Lantbruksuniversitet. Grimsö forskningsstation. (In Swedish).
SIS. Vattenundersökningar - Atomabsorbtionsspektrometri i flamma.
Swertz O. 1995. Trend assessment using the Mann-Kendall test. Report of the Working Group on Statistical Aspects of Trend Monitoring. ICES CM 1995/D:2.
Cd, in kidney of moose calves (Alces alces) ng/g fresh w. from Grimso
0 500 1000 1500 2000
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
n(tot)=218,n(yrs)=17 m=964 (822 , ) slope=.34%(-2.8,3.5) SD(lr)=.32,4.2%,16 yr power=.92/.35/9.9%
y(98)= 997 ( 705, 1411) r2=.00, p<.802 tao=.04, p<.837 SD(sm)=.29, n.s.
pia - 00.03.22 13:47, cdka
