Strategies for toxicological evaluation of mixtures

Pergamon

Food and Chemical Toxicology 34 (1996) 1147-1149

Strategies

for Toxicological Mixtures

Evaluation

of

1. EIDE Statoil

Research

Centre,

N-7005

Trondheim,

Norway

Abstract-Different strategies for the toxicological evaluation of mixtures are presented. The purpose is to determine the effects of each component (variable) in the mixture, and possible interactions between variables. The examples presented have 3-5 predictor variables and 1-3 responses, and are based on statistical experimental design, multivariate data analysis and modelling. The following examples are presented: (1) inhalation experiments with synthetic vapour mixtures of hydrocarbons formulated by means of mixture design at different vapour concentrations (the experimental region covers both partial and complete evaporation of the liquid mixtures); (2) combination of refinery streams for fuel blending by means of mixture design with constraints, followed by engine tests and determination of exhaust particles; (3) fractionation of organic extracts of diesel exhaust particles, and recombination of the extracts by means of mixture design, followed by mutagenicity testing of the recombined extracts in the Ames Salmonellu assay; (4) spiking complex mixtures with individual compounds using factorial design, and subsequent mutagenicity testing. The data obtained from these four examples were analysed by means of Projections to Latent Structures (PLS). The effects of each variable and possible interactions, were quantified by means of PLS regression coefficients. Furthermore, the empirical models obtained were evaluated by means of correlation coefficients, cross validation and predictions. Copyright 0 1997 Elsevier Science Ltd

Introduction Different strategies have been described for the evaluation of toxicological properties of complex mixtures: these are integrative (studying the mixture as a whole), dissective (dissecting or fractionating a mixture to determine causative constituents), and synthetic (studying interactions between agents in simple combinations) (Mauderly, 1993). The two latter give information about the effects of the individual compounda,; only the synthetic approach gives information about interactions. In order to evaluate interactions, it is necessary to vary the composition of the mixture and the relative amounts of the different compounds. Synthetic mixtures have to some extent been formulated by means of factorial design (Groten et al., 1991; Krewski and Thomas, 1992; Svendsgaard and Hertzberg, 1994). The present study presents alternative strategies for evaluation of mixtures, illustrated by four examples. The purpose is to dletermine the effects of each component (variable) in the mixture, and possible interactions between them. Two of the examples represent a further development of the synthetic category. The two other examples deal with strategies for evaluating existmg mixtures by means of recombination of fractions, and spiking of mixtures with individual compounds. In this short review, Abbreviations:

DCM = dichloromethane; DMSO = dimethyl sulfoxide; PAH = polycyclic aromatic hydrocarbon; PLS = Projections to Latent Structures.

0278-6915/97/$17.00 + 0 00 0 PII 0278-6915(97)00086- 5

emphasis is placed details are presented

on design and strategies; in the original papers.

Materials and methods: experimental multivariate data analysis and modelling

the

design,

The examples are based on factorial (Box et al., 1978) or mixture (Cornell, 1990) design. Factorial design requires orthogonality (independent variables), which is not always the case in liquid mixtures. Factorial and mixture design may support empirical models with linear, interaction and square terms in a Taylor polynome (Box et al., 1978), giving possibilities for response surface modelling (RSM), which is useful for visualizing the results. PLS (Projections to Latent Structures) have been found practica1 for the analysis of mixture data as it overcomes many of the problems inherent in inter-correlated (dependent) predictor variables (Kettaneh-Wold, 1992; Kvalheim, 1989). Furthermore, with PLS many responses may be analysed and considered simultaneously. The models are evaluated by means of scaled and centred PLS regression coefficients, 95% confidence intervals, pure error versus model error, correlation coefficients (r’) and prediction coefficients Q’. The latter is obtained after cross validation (1-CV), and is important to avoid overfit. The models may also be verified by comparing data from a few new experiments with values predicted from the model. It is emphasized that interaction terms are only indications of possible synergism or antagonism and

1997 Elsevier Science Ltd. Al1 rights

reserved.

Printed

in Great

Britain

1148

1. Eide

Fig. 1. Mixture design with three components at three vapour levels. The dotted lines represent factorial design. From Eide and Zahlsen (1996).

distribution of one hydrocarbon is affected by simultaneous exposure to one or both of the others. The hydrocarbon mixtures were formulated by means of mixture design (Cornell, 1990) at different vapour levels. The experimental region is visualized as a pyramid in Fig. 1. With three variables, three concentration levels with a total of 15 different mixtures (plus replicates in the centrepoint at leve1 2) are sufficient to support an empirical model with linear, interaction and square terms. The study could, however, have been carried out with factorial design as the predictor variables are vaporized hydrocarbons delivered independently. With more complex mixtures this is difficult, and the design should incorporate the different compounds in a liquid mixture as variables, using mixture design. lf a liquid mixture is evaporated completely to give different vapour concentrations, the design wil1 correspond to the design used in the present work (Fig. 1). If the liquid mixture is partially evaporated, and the hydrocarbons have slightly different vapour the experimental region wil1 be conpressures, strained, with a more uneven distribution of the points representing the different combinations. Consequently, factorial design represents an experimental space (the dotted tube in Fig. 1), that is of less relevante for vapours from liquid mixtures. E‘cample 2: combination of rejinery streams in jïuel blending

x4

Fig. 2. Mixture design with constraints. The experimental domain is represented by the three-dimensional polyhedron within the four-component tetrahedron. From Eide and Johansson (1994).

have to be evaluated with respect and response-additivity.

to dose-additivity

Results and Discussion Example 1: inhalation experiments with vapour from liquid hydrocarbons Synthetic mixtures of three C9 n-paraffinic, naphthenic and aromatic hydrocarbons were studied in the rat after inhalation for 12 hr (Eide and Zahlsen, 1996). In this work, the concentration of each hydrocarbon in different organs was measured-not a toxic effect or a response. Hence, possible interactions are present if the uptake and

Petroleum products are blended from different refinery streams to give a variety of products. The present example is related to diesel fuel, but the principles may be applied to any toxicological evaluation of petroleum products blended from refinery streams. A drawback with automotive diesel is the emission of exhaust particles that are composed of a carbon core and adsorbed organic substances such as polycyclic aromatic hydrocarbons (PAHs). Organic extracts of diesel exhaust particles are mutagenic, and diesel exhaust particles are classified as being probably carcinogenic (IARC, 1989). The present example demonstrates how the exhaust emissions may be related to the composition of a four-component diesel fuel (Eide and Johansson, 1994). First, the ‘permitted’ combinations of four different refinery streams were determined according to the technical specifications for commercial autodiesel, covering a wide range of physical/chemical parameters. These specifications impose mixture constraints (Cornell, 1990; Crosier, 1984; Snee, 1981). The four-component mixture with constraints is visualized as a three-dimensional polyhedron with 11 corners within a tetrahedron (three-dimensional simplex) as shown in Fig. 2. The blending matrix consisted of 19 diesel blendings; three of these were replicates representing the centroid of the polyhedron.

Strategies for evaluation of mixtures The different fuels were then tested in engine tests, diesel exhaust particles were sampled and the amount (mass) of particles were quantified. PLS were used to relate the particulate emissions to the different refinery streams, and to identify possible interactions. A further evaluation of the emissions may include characterization and mutagenicity testing of the organic extracts of the particles. E.xample

3: fractionation

and recombination

The present example shows a strategy for genotoxic evaluation of an exisl:ing mixture, in this case an organic extract of diesel exhaust particles (nstby et al., 1996). After extraction with dichloromethane (DCM), the crude extracts were fractionated according to polarity into five fractions: (1) aliphatic hydrocarbons; (2) PAHs; (3) nitro-PAHs; (4) dinitro-PAHs; (5) polar compounds. Prior to mutagenicity testing in the Ames Salmonella assay, DCM was replaced by dimethyl sulfoxide (DMSO), and the fractions con!.aining the primary mutagens were recombined to create new extracts in order to determine the mutagenicity of each fraction and to identify new

possible

extracts

interactions.

was

design at different inp to the desian

The composition

determined

by means

concentration shown

in Fig.

of the

of mixture

levels, correspond1.

Mutagenicity testing of the new extracts was carried out in the Ames test using different strains of Salmonella typhimurium. The incorporation of dose in the design reducmzd the number of samr>les (recombined-extracts) significantly, compared with the determination of ‘close-response curves on each sample (i.e. recombined extracts in different dilutions).

Furthermore,

instead

of

running

two

independent experiments, as required in the standard procedure for the .4mes test, predictions and verifications of a few new samples were used. Example

4: spiking

Spiking may be used to evaluate the mutagenicity or toxicity of individual compounds in a mixture. In the present study, mutagenicities of individual PAHs were evaluated in an organic extract of diesel exhaust particles (E. Bostrom, S. Engen and 1. Eide, unpublished data). The particles were extracted with DCM. After the change to DMSO, the extract was spiked with four individual PAHs: these were benzo[a]pyrene, benzo[a]anthracene, pyrene and fluoroanthene. The PAHs were added separately and in various combinations to the extract to determine the effects of each val-iable and to identify possible interactions between the individual PAHs and between the PAHs and the extract. The study was designed

as a fractional

factorial

experiment

with

the five variables (the extract and the four PAHs), giving 16 (instead of 32) mixtures plus a triplicate centrepoint (i.e. a total of 19). The mixtures were

1149

tested for mutagenicity in the Ames test using four strains of S. typhimurium. PLS was used to quantify the mutagenicity of each compound and possible interactions. The fractionated factorial design used in the present work, supports a model with linear and interaction terms (not quadratic). The centrepoint then becomes important to verify linearity. The replicates are used to evaluate repeatability (pure error). Fractionated design with five variables is very useful, giving resolution V which does not confound main effects and two-factor interactions (Box et al., 1978).

REFERENCES

Box G. E. P., Hunter W. G. and Hunter J. S. (1978) Statistics for Experimenters: An Inrroduction io Design, Data Analysis, and Model Building. John Wiley & Sons, New York. Cornell J. A. (1990) E,xperiments with Mixfures: Designs, Models, and the Analysis of Mixture Data. 2nd Ed. John Wiley & Sons, New York: Crosier R. B. (1984) Mixture experiments: geometry and pseudocomponents. Technometrics 26, 209-216. Eide 1. and Johansson E. (1994) Statistical experimental design and projections to latent structures analysis in the evaluation of fuel blends with respect to particulate emissions. Chemometrics and Intelligent Laboratory Systems 22, 77-85. Eide I. and Zahlsen K. (1996) Inhalation experiments with mixtures of hvdrocarbons: Exuerimental design. ., statistics

and internretation

of kinetics and oossible interactions.

Archives 01‘Toxicology 70, 397404:

Groten J. P., Sinkeldam E. J., Muys T., Luten J. B. and van Bladeren P. J. (1991) Interaction of dietary Ca, P, Mg, Mn, Cu, Fe, Zn, and Se with the accumulation and oral toxicity of cadmium in rats. Food and Chemical Toxicology 29, 249-258. IARC (1989) IARC Monographs on the Evaluarion of Carcinogenic Risks to Humans. Vol. 46. Diesel and Gasoline Enpine Exhausts and Some Nitroarenes. InternationalAgency for Research on Cancer, Lyon. Kettaneh-Wold N. (1992) Analysis of mixture data with partial least squares. Chemometrics and Intelligent Laboratory Systems 14, 57-69. Krewski D. and Thomas R. D. (1992) Carcinogenic mixtures. Risk Analysis 12, 105-113. Kvalheim 0. M. (1989) Model-building in chemistry, a unified approach. Analytica Chimica Acra 223, 53-73. Mauderly J. L. (1993) Toxicological approaches to complex mixtures. Environmenral Health Perspectives 101 (Suppl. 4), 155-165. Snee R. D. (198 1) Developing blending models for gasoline and other mixtures. Technomelrics 23, 119~130. Svendsgaard D. J. and Hertzberg R. C. (1994) Statistical methods for the toxicological evaluation of the additivity assumption as used in the Environmental Protection Agency chemical mixture risk assessment guidelines. In Toxicologv of Chemical Mixrures. Case Studies. Mechanisms, and‘N;cel Approaches. Edited by R. S. H. Yang. pp. 599-642. Academie Press, San Diego, CA. Ostby L., Engen S., Melbye A. and Eide 1. (1996) Mutagenicity testing of organic extracts of diesel exhaust particles after fractionation and recombination. Archiues of Toxicology. In press.