An approach to identify causal components of complex air pollution mixtures was explored. Rats and mice were exposed by inhalation 6 h daily for 1 week or 6 months to dilutions of simulated downwind coal emissions, diesel and gasoline exhausts and wood smoke. Organ weights, hematology, serum chemistry, bronchoalveolar lavage, central vascular and respiratory allergic responses were measured. Multiple additive regression tree (MART) analysis of the combined database ranked 45 exposure (predictor) variables for importance to models best fitting 47 significant responses. Single-predictor concentration-response data were examined for evidence of single response functions across all exposure groups. Replication of the responses by the combined influences of the two most important predictors was tested. Statistical power was limited by inclusion of only four mixtures, albeit in multiple concentrations each and with particles removed for some groups. Results gave suggestive or strong evidence of causation of 19 of the 47 responses. The top two predictors of the 19 responses included only 12 organic and 6 inorganic species or classes. An increase in red blood cell count of rats by ammonia and pro-atherosclerotic vascular responses of mice by inorganic gases yielded the strongest evidence for causation and the best opportunity for confirmation. The former was a novel finding; the latter was consistent with other results. The results demonstrated the plausibility of identifying putative causal components of highly complex mixtures, given a database in which the ratios of the components are varied sufficiently and exposures and response measurements are conducted using a consistent protocol.
Keywords: Air pollution; coal emissions; diesel exhaust; gasoline exhaust; mixtures; regression tree; wood smoke.