Sources of error and variation in QSAR approaches to modeling the population growth impairment of Tetrahymena pyriformis

Author

Julie Rebecca Seward

Date of Award

12-2000

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Ecology and Evolutionary Biology

Major Professor

T. Wayne Schultz

Committee Members

Dewey Bunting, John Kennedy, Alice Layton, Mark Cronin

Abstract

Within predictive toxicology, quantitative structure-activity relationships (QSARs) are powerful tools used to predict potency of a chemical based on molecular structure. Recently, two approaches to QSAR have been employed in derivation of such predictive models. These include the mechanism of action approach, which collectively models only chemicals eliciting toxicity via the same mode or mechanism of toxic action and the response-surface approach, which allows modeling of diverse chemicals across mechanism and mode of action. Models derived using the mechanism of action approach produces models of superior fit, as determined by the coefficient of determination, and thus are of higher predictive power. However, the use of such models to predict toxicity of untested chemicals is tempered by the difficulties in a priori assignment of mechanism of toxic action and therefore the use of the correct model. Assignment of mechanism has classically been done by rules, which assign mechanism based on the presence or absence of a particular toxicophore or substituent. The response-surface approach circumvents the necessity of knowledge of mechanism of action. However, response-surface models are of lesser predictive power than mechanistic QSARs. The experiments described here investigate the relative advantages and disadvantages of the two approaches to predicting toxic potency of substituted pyridines to Tetrahymena pyriformis. Previous work has shown that growth kinetic trends in exposed T. pyriformis populations are mode of action specific. Expanding this work, growth kinetic experiments were performed exposing T. pyriformis populations to two groups of pyridines representing different mechanisms of action assigned by the presence of classic substituents or toxicophores holding hydrophobicity constant. Growth kinetic trends of the two neutral narcotics (pyridine and 3-chloropyridine) exhibited hydrophobicity-dependent increases in lag phase as has been established in the literature. The hydroxylated pyridine exposed populations exhibited similar trends. Growth kinetic trends were hydrophobicity-independent and unique to neutral narcotic exposed populations confirming polar narcosis as a distinct non-covalent mechanism. The growth kinetic trends of the reactive nitropyridines demonstrated markedly different growth kinetic trends. Populations exposed to 2-chloro-3,5-dinitropyridine demonstrated similar trends observed for electrophiles acting via Michael-type acceptance: concentration dependent death in initial inoculum of cells at threshold concentrations. A unique trend was observed for populations exposed to 2- bromo-5-nitropyridine, which was a hybrid between the neutral narcotics and the electrophiles. These results suggest that different mechanisms do exist within broad modes of toxic action as exhibited by unique growth kinetic trends. However, different growth kinetic trends were established for covalent pyridines with similar substituents. Rules governing mechanism of action selection would place these chemicals in the same mechanism although growth kinetic trends imply different mechanisms exist. This points toward the difficulty in a priori assignment of mechanism of action to a pyridine, particularly those chemicals acting via mechanisms other than nonpolar narcosis. Therefore, the response-surface would be of greater use because of the risk of wrong QSAR selection. A robust response-surface has been derived for substituted benzenes. Because benzenes and pyridines, though structurally similar, have different reactivity and electronic configuration, a response-surface was derived for substituted pyridines to investigate the potential of extension of the benzene domain to nitrogen heterocyclic pyridines. The toxicity of over 100 pyridines was used in model formation. The resultant surface response was similar to the benzene surface as determined by the overlap of standard deviations of the intercept and slopes of the two equations. However, the increased reactivity of pyridines resulted in a lesser model fit as demonstrated by the coefficient of determination. Thus, the domain of the surface described for benzenes may be extended to encompass pyridines, but the amount of unexplained variation is increased. Although the response-surface QSAR models pyridines across mechanism of action, this inclusiveness is tempered by a reduction in model fit and thus predictive power. While the toxicity and hydrophobicity data used in both regression approaches are the same, one difference between these approaches is the use of a quantum chemical (QC) descriptor, particularly molecular orbital (MO) energy values such as the energy of the lowest unoccupied molecular orbital (E_LUMO). Thus, the reduced statistical fit exhibited by QSAR models, which include these QC-MO descriptors, could be a result of the variability inherent in the calculation of these descriptors. An investigation employing a structurally and mechanistically diverse set of pyridines revealed that variability is associated with the calculation of the MO descriptor E_LUMO both between selected Hamiltonians and selected software packages. However, this variability in no way affects the statistical significance of QSARs for toxicity using these values. Variability in biological toxicity data was also investigated as a potential source of increased uncertainty in response-surface QSARs due to the inclusiveness of reactive chemicals. The reproducibility of growth impairment of the freshwater ciliate T. pyriformis exposed to a structurally diverse group of chemicals of varying hydrophobicity across modes of toxic action, either non-covalent narcosis or covalent electro(nucleo)philicity was investigated. Toxicity values of 28 of the 50 re-tested chemicals conformed to the criterion set for reproducible values. The majority of the chemicals that did not have reproducible potency measurements were electro(nucleo)philic. Toxicophores largely represented in this group were quinone derivatives, electron releasing amino and hydroxyl moieties and withdrawing nitro substituents, often in tandem with strong leaving groups (i.e., halogens), and unsaturated alcohols. These results suggest that toxic potency values of chemicals acting via the covalent mode of toxic action could be more susceptible to non-reproducibility.

Recommended Citation

Seward, Julie Rebecca, "Sources of error and variation in QSAR approaches to modeling the population growth impairment of Tetrahymena pyriformis. " PhD diss., University of Tennessee, 2000.
https://trace.tennessee.edu/utk_graddiss/8421