371
13 Vibrational Spectroscopy: Disease Diagnostics and Beyond
nucleus) in cervical squamous cells [89]. Additionally, it is known that binding of
hPv dNA to host dNA disrupts the normal function of the cellular proteins and as
a consequence, the host cell accumulates more and more damaged dNA that cannot
be repaired [90].
CaSki cells are known to be the most malignant and in the spectra of the cytoplasm and nucleus representing this cell line the highest level of nucleic acids
was observed. A similar increase of the nucleic acid related peaks in cancerous
cervical cells was noticed in previous studies [91, 92] and is confirmed by these
observations. Changes in lipid levels are possibly associated with the disruption of
the membrane functionality caused by the virus which influences lipid rafts [93]. A
similar behaviour was noted and reported in the previous study on cervical cancer
cell lines, wherein both Raman and FtIR (Fig. 13.4) techniques were utilised [63].
moreover, it was reported previously by Wong et al. that the degree of disorder of
methylene chains in membrane lipids increases in cervical neoplastic cells [94]. the
absence of any clear relationship between protein levels and hPv copy number is
expected, as the presence of hPv in the cell results in functional over- and underexpression of only select biomarker proteins [73], while the level of the other proteins would not be affected. the variation of the spectroscopic features associated
with the nuclear lipids, proteins and nucleic acids as a function of the hPv copy
number was found to be sublinear and to be well fitted with a similar function to that
employed to describe the variation of the p16
INK4A
levels with hPv copy number in
Fig. 13.7. Notably, the dependences of the cytoplasmic lipids, proteins and nucleic
acids on the p16
INK4A
levels, as identified by fluorescence were found to be linear.
In order to further elucidate the multivariate spectral signatures which are specifically related to hPv infection, multivariate regression models were constructed using Partial Least Squares Regression (PLSR) and PLS Jack-knifing was employed
as a multivariate feature selection technique [94–96]. the PLSR algorithm seeks to
develop a model that relates the spectral data ( X-matrix) to a series of targets ( Ymatrix, e.g. concentration of reaction product or analyte) according to the equation
Y = XB + E, where B is a matrix of regression coefficients and E is the regression
residual. the PLS Jack-knifing method developed by Westad and martens [96] was
then used to determine the spectral features that were statistically significant at a
particular level of confidence using t-testing of the regression coefficients, B.
Figure 13.8 demonstrates that p16
INK4A
levels as measured by fluorescence intensity are approximately linearly correlated with the univariate spectral features
and thus, as PLS is a linear model, these values are employed as targets for the
PLS analysis. the model can therefore be applied to the FtIR data to elucidate
multivariate signatures which are correlated to p16
INK4A
expression level, and therefore, in accordance with Fig. 13.7, to hPv infection levels. once established, these
variation patterns can then be applied to unknown samples to screen for the biomarker levels. As p16
INK4A
was found to be predominantly expressed in cytoplasmic regions of the cells, signals recorded from the cytoplasm were utilised in the
analysis. Calibration (based on 40 % of the data) and test set performance (based
on 60 % of the data) are presented in Fig. 13.9a and demonstrate a very good fit to
the model. As shown in Fig. 13.9b, the PLS loading exhibits variations originating
13 Vibrational Spectroscopy: Disease Diagnostics and Beyond
nucleus) in cervical squamous cells [89]. Additionally, it is known that binding of
hPv dNA to host dNA disrupts the normal function of the cellular proteins and as
a consequence, the host cell accumulates more and more damaged dNA that cannot
be repaired [90].
CaSki cells are known to be the most malignant and in the spectra of the cytoplasm and nucleus representing this cell line the highest level of nucleic acids
was observed. A similar increase of the nucleic acid related peaks in cancerous
cervical cells was noticed in previous studies [91, 92] and is confirmed by these
observations. Changes in lipid levels are possibly associated with the disruption of
the membrane functionality caused by the virus which influences lipid rafts [93]. A
similar behaviour was noted and reported in the previous study on cervical cancer
cell lines, wherein both Raman and FtIR (Fig. 13.4) techniques were utilised [63].
moreover, it was reported previously by Wong et al. that the degree of disorder of
methylene chains in membrane lipids increases in cervical neoplastic cells [94]. the
absence of any clear relationship between protein levels and hPv copy number is
expected, as the presence of hPv in the cell results in functional over- and underexpression of only select biomarker proteins [73], while the level of the other proteins would not be affected. the variation of the spectroscopic features associated
with the nuclear lipids, proteins and nucleic acids as a function of the hPv copy
number was found to be sublinear and to be well fitted with a similar function to that
employed to describe the variation of the p16
INK4A
levels with hPv copy number in
Fig. 13.7. Notably, the dependences of the cytoplasmic lipids, proteins and nucleic
acids on the p16
INK4A
levels, as identified by fluorescence were found to be linear.
In order to further elucidate the multivariate spectral signatures which are specifically related to hPv infection, multivariate regression models were constructed using Partial Least Squares Regression (PLSR) and PLS Jack-knifing was employed
as a multivariate feature selection technique [94–96]. the PLSR algorithm seeks to
develop a model that relates the spectral data ( X-matrix) to a series of targets ( Ymatrix, e.g. concentration of reaction product or analyte) according to the equation
Y = XB + E, where B is a matrix of regression coefficients and E is the regression
residual. the PLS Jack-knifing method developed by Westad and martens [96] was
then used to determine the spectral features that were statistically significant at a
particular level of confidence using t-testing of the regression coefficients, B.
Figure 13.8 demonstrates that p16
INK4A
levels as measured by fluorescence intensity are approximately linearly correlated with the univariate spectral features
and thus, as PLS is a linear model, these values are employed as targets for the
PLS analysis. the model can therefore be applied to the FtIR data to elucidate
multivariate signatures which are correlated to p16
INK4A
expression level, and therefore, in accordance with Fig. 13.7, to hPv infection levels. once established, these
variation patterns can then be applied to unknown samples to screen for the biomarker levels. As p16
INK4A
was found to be predominantly expressed in cytoplasmic regions of the cells, signals recorded from the cytoplasm were utilised in the
analysis. Calibration (based on 40 % of the data) and test set performance (based
on 60 % of the data) are presented in Fig. 13.9a and demonstrate a very good fit to
the model. As shown in Fig. 13.9b, the PLS loading exhibits variations originating