377
13 Vibrational Spectroscopy: Disease Diagnostics and Beyond
Comparing the regression co-efficients obtained via spectral regression against
cisplatin concentration (which highlight the primary biochemical effect of the action of the agent) with those obtained via regression against cell viability (which
highlight biochemical markers of the change in the viability of the cell), very different spectral features are selected as being statistically significant and thus allow the differentiation of the chemotherapeutic response. the spectral features of
thymine (737–742 cm
−1
, 1,589–1,595 cm
−1
), guanine (1,548–1,553 cm
−1
) and adenine (1,578 cm
−1
, 1,604–1,611 cm
−1
) emerge as prominent as a result of regression against the concentration of the chemotherapeutic agent and are attributable to
direct binding of the cisplatin molecule with these dNA bases. In addition, changes
related to proteins feature in the regression against the concentration of the chemotherapeutic agent are significantly different than those changes of protein associated bands which appear as a result of regression against the cellular viability. the
changes in the features of the proteins due to the chemical effect are attributable to
structural alterations in nuclear histone proteins due to cisplatin interactions with
the cellular dNA. these features are not selected as being statistically significant
by the analysis for regression against the viability endpoint, suggesting that they are
markers of the biochemical interaction between cisplatin and dNA. Contrastingly,
the conformational changes to the dNA A-form (798–813 cm
−1
) and dNA B-form
(832–833 cm
−1
), among the changes associated with the protein features, especially
the amide-III random coil (1,225–1,228 cm
−1
), amide-III β-sheet (1,250–1,259),
amide-III α-helix, among others, are selected as being statistically significant in
the regression against cellular viability, suggesting that they are markers of this
physiological endpoint in the chemotherapeutic response. overall, Fig. 13.11 demonstrates that the Raman spectrum of the cell may be resolved into distinct groups
of statistically significant spectral markers for distinct types of effect. A similar
analysis of the spectral changes as a result of the action of the chemotherapeutic
agent in the cytoplasm and cell membrane can be performed [99]. Raman microspectroscopy can therefore clearly fingerprint the effects of the interaction of the
chemotherapeutic agent with dNA, and identify that fingerprint within the sub cellular spectra of exposed cells.
PLSR models of A549 nuclear spectra versus concentration and reference viability measurements from the mtt assay can then be constructed to determine
the ability of the spectral data to predict the level of exposure to cisplatin and the
associated level of physiological effect from Raman spectra. All spectra were compiled into a matrix, and were randomly sorted. A total of 60 % of the spectra were
used to train the PLSR model and 40 % of the total was retained as an independent
test set to assess the performance of the model in predicting the level of exposure,
and the cellular viability, with unseen data. Leave-one out cross validation with
the calibration set was used to determine the optimal model complexity for use in
testing [115]. this process was performed on fifty separate occasions, with randomization of the data matrix and splitting of the data on each occasion to prevent
data bias [116]. the results of prediction of levels of cellular viability is shown in
Fig. 13.12, for the example of the nuclear region, where the values of the root mean
square error of calibration (RmSEC) and the root mean square error of prediction
13 Vibrational Spectroscopy: Disease Diagnostics and Beyond
Comparing the regression co-efficients obtained via spectral regression against
cisplatin concentration (which highlight the primary biochemical effect of the action of the agent) with those obtained via regression against cell viability (which
highlight biochemical markers of the change in the viability of the cell), very different spectral features are selected as being statistically significant and thus allow the differentiation of the chemotherapeutic response. the spectral features of
thymine (737–742 cm
−1
, 1,589–1,595 cm
−1
), guanine (1,548–1,553 cm
−1
) and adenine (1,578 cm
−1
, 1,604–1,611 cm
−1
) emerge as prominent as a result of regression against the concentration of the chemotherapeutic agent and are attributable to
direct binding of the cisplatin molecule with these dNA bases. In addition, changes
related to proteins feature in the regression against the concentration of the chemotherapeutic agent are significantly different than those changes of protein associated bands which appear as a result of regression against the cellular viability. the
changes in the features of the proteins due to the chemical effect are attributable to
structural alterations in nuclear histone proteins due to cisplatin interactions with
the cellular dNA. these features are not selected as being statistically significant
by the analysis for regression against the viability endpoint, suggesting that they are
markers of the biochemical interaction between cisplatin and dNA. Contrastingly,
the conformational changes to the dNA A-form (798–813 cm
−1
) and dNA B-form
(832–833 cm
−1
), among the changes associated with the protein features, especially
the amide-III random coil (1,225–1,228 cm
−1
), amide-III β-sheet (1,250–1,259),
amide-III α-helix, among others, are selected as being statistically significant in
the regression against cellular viability, suggesting that they are markers of this
physiological endpoint in the chemotherapeutic response. overall, Fig. 13.11 demonstrates that the Raman spectrum of the cell may be resolved into distinct groups
of statistically significant spectral markers for distinct types of effect. A similar
analysis of the spectral changes as a result of the action of the chemotherapeutic
agent in the cytoplasm and cell membrane can be performed [99]. Raman microspectroscopy can therefore clearly fingerprint the effects of the interaction of the
chemotherapeutic agent with dNA, and identify that fingerprint within the sub cellular spectra of exposed cells.
PLSR models of A549 nuclear spectra versus concentration and reference viability measurements from the mtt assay can then be constructed to determine
the ability of the spectral data to predict the level of exposure to cisplatin and the
associated level of physiological effect from Raman spectra. All spectra were compiled into a matrix, and were randomly sorted. A total of 60 % of the spectra were
used to train the PLSR model and 40 % of the total was retained as an independent
test set to assess the performance of the model in predicting the level of exposure,
and the cellular viability, with unseen data. Leave-one out cross validation with
the calibration set was used to determine the optimal model complexity for use in
testing [115]. this process was performed on fifty separate occasions, with randomization of the data matrix and splitting of the data on each occasion to prevent
data bias [116]. the results of prediction of levels of cellular viability is shown in
Fig. 13.12, for the example of the nuclear region, where the values of the root mean
square error of calibration (RmSEC) and the root mean square error of prediction