Spectroscopic and Chemometric Approaches to Radiobiological Analyses

Aidan Meade (Thesis), Dublin Institute of Technology

Document Type Theses, Ph.D

Successfully submitted for the award of Doctor of Philosophy (Ph.D) to the Dublin Institute of Technology, July, 2010.


Vibrational spectroscopy is an attractive modality for the analysis of biological samples, providing a complete non-invasive acquisition of the biochemical fingerprint of the sample. In this study, applications of such techniques in the interrogation of radiobiological effects in-vitro were examined using Fourier-transform infrared microspectroscopy (FTIRM) and confocal Raman microspectroscopy (CRM). Preliminary investigations were conducted for the purposes of evaluating the effect of cell culture with spectroscopic substrates and the effect of cell fixation on spectroscopic measurements in-vitro. It was observed that cell culture on commonly used spectroscopic substrates affects cell morphology and physiology in an adverse manner, and that coating of the substrates with a gelatin solution ameliorated this affect in a manner that produced similar FTIRM and CRM spectral correlations with cellular physiology. It was also observed in studies comparing CRM spectra of live cells to fixed cells, that fixation of various cell lines in 4% formalin, followed by storage of the sample in deionised water, provided an effective means of preservation of the sample in a form producing spectral signatures in the fixed cell close to that in the live cell. In a direct-irradiation study, normal human skin cells (HaCaT) were exposed to gamma radiation (at ten dose points from 0Gy to 5Gy) and assayed using FTIRM and CRM in addition to parallel measurements of biochemical function, at times ranging from 6 hours to 96 hours post-irradiation. Studies on the dose response in the cell line were conducted using linear and nonlinear multivariate partial least squares regression (PLSR) and generalized regression neural networks (GRNN). This investigation demonstrated that the variation in the biochemical fingerprint of the cell with dose and time after irradiation is non-linear by virtue of the higher modelling efficiency yielded from use of the non-linear GRNN algorithm. Dose prediction accuracies of approximately ±10mGy were achieved at 96 hours after irradiation, highlighting the potential applications of the methodology in radiobiological dosimetry. Further studies were conducted on the effect of spectral preprocessing of FTIRM and CRM spectra on the prediction of radiation dose, again using linear and nonlinear PLS algorithms in addition to support vector regression (SVR). The optimal preprocessing methodology (which comprised combinations of spectral filtering, baseline subtraction, scaling and normalization options) was selected using a genetic algorithm (GA) with the root mean squared error of prediction (RMSEP) used as the fitness criterion for selection of the preprocessing chromosome (where this was calculated on an independent set of test spectra randomly selected from the dataset on each pass of the algorithm). The results indicated that GA selection of the optimal preprocessing methodology substantially improved the predictive capacity of the algorithms over baseline methodologies, although the optimal preprocessing chromosomes were similar for various regression algorithms, suggesting an optimal preprocessing methodology for radiobiological analyses with biospectroscopy. Feature selection of both FTIRM and CRM spectra using genetic algorithms and multivariate regression provided further decreases in RMSEP, but only within non-linear multivariate regression algorithms. Detailed investigation of the origin of the non-linearity in the spectral variation with dose demonstrated that spectral signatures in all cellular components (nucleic acid, protein, lipid and carbohydrate) contain evidence of low-dose hyper-radiosensitivity, increased radioresistance and high dose radiation damage, and that the overall trend in variation of these components agrees well with the Induced Repair (IR) model of cell survival. Investigations of the correlation between spectral signatures and measures of various physiological characteristics of the cell after irradiation uncovered evidence of two photon ionization events causing quadratic relationships between spectral signatures of the cell and its physiological state, particularly at high doses. Further analysis demonstrated that distinct physiological responses in the cell could be predicted using both FTIRM and CRM with PLS regression and GA-based feature selection approaches. Finally, studies were conducted with Raman spectroscopy to examine its sensitivity in the categorization of bystander mediated radiation damage in HaCaT cells with inhibition of the bystander effect mediated via the mitogen-activated protein (MAP) kinase transduction pathway. Spectral variations were seen that were associated with the response of the cell to exposure to bystander inhibited medium, and supervised cluster analysis with either support vector machines (SVM) or principal components analysis combined with linear discriminant analysis (PCA-LDA) resulted in spectral classifications that correlated well with the operation of the MAP kinase transduction pathway. The study demonstrates that vibrational spectroscopy can provide the means to assay the total biochemical composition of the cell after exposure to ionizing radiation, and various aspects of the effect of ionizing radiation on individual biochemical species, in addition to assaying cellular physiology, in-situ without extensive sample preparation, where suitable analytical methodologies are employed. Further challenges and opportunities in the development of vibrational spectroscopy for radiobiological applications are explored in the thesis conclusions.