382
H. J. Byrne et al.
membrane. however, in analysing the spectral response of the cells, it is observed
that the spectra recorded from the edge of the cells contain a high variability due to
the transition from the membrane to the substrate, resulting in quite noisy spectra.
When the K-means analysis is performed, the variability between these spectra can
interfere with the analysis, resulting in the creation of new clusters. thus, in order
to best visualise the different structures existing inside the cells the numbers, the
number of clusters has to be increased to 10 groups, which produced the best reproducibility. In this way, the variability in the spectra obtained at the edge of the
cells are contained in different clusters and do not interfere with the identifications
of the different structures present within the cells. the initiation step was selected
randomly for each iteration and it was seen to have no effect on the final results of
analysis. After analysis, the spectra were gathered in the 10 different groups according to their similarities and a colour (arbitrarily) attributed to each cluster. False colour maps were then constructed representing the partition of the different clusters
in the cells. the resulting images are presented in Fig. 13.13b, d. A correspondence
between the organisation of the clusters and the structure present in the cells can be
seen. In Fig. 13.13b, specific clusters relating to the cytoplasm, nucleus and nucleoli
appear. the specificity of the information contained in the spectra recorded allows
discrimination of the different cellular compartments using K-means analysis. the
sensitivity of the analysis furthermore allows visualization of the two different compartments present in the cytoplasm of the A549 cell and identifies a specific cluster
related to the nucleoli in the haCat cell, although they are not clearly seen in the
visible image.
K-means clustering analysis appears to be a powerful tool to identify the different structures present in the cells. the false colour maps give a representation of
the partition of the different clusters but do not provide any information about the
chemical composition of the samples. this biochemical information and the variations between the different clusters can be visualised through the mean spectra of
the clusters. Figure 13.14 presents the average spectra obtained from selected clusters of the K-means analysis performed on an A549 cell (Fig. 13.13b). (A: nucleus
from the cluster number 4, B: nucleolus from the cluster number 6, C: cytoplasm
from the cluster number 5 and d: cytoplasm from the cluster number 7). the other
clusters correspond to the spectra recorded from the collagen gel and those located
at the edge between the cell and the substrate (data not shown). variations can be
observed in different spectral windows according to the different clusters (highlighted in grey in Fig. 13.14).
the characterisation of live cells on 2d substrates is well documented [123, 124,
147–149] and, based on literature; assignments of the different peaks can be performed. For example, the variations in the spectral range 718–725 cm
−1
are related
to the lipids and nucleic acids, between 815 and 830 cm
−1
to RNA and proteins, from
1,088 to 1,102 cm
−1
to the nucleic acids, the lipids and carbohydrates, from 1,243–
1,266 cm
−1
to the nucleic acids, the proteins and lipids, from 1,296–1,322 cm
−1
to
the lipids and from 1,315–1,348 cm
−1
to the nucleic acids and proteins. In analysing
the cluster average spectra, it is important to note that, with a ~ 1.5 µm spot size,
while the nucleus can be specifically targeted, the overlying cellular membrane and
H. J. Byrne et al.
membrane. however, in analysing the spectral response of the cells, it is observed
that the spectra recorded from the edge of the cells contain a high variability due to
the transition from the membrane to the substrate, resulting in quite noisy spectra.
When the K-means analysis is performed, the variability between these spectra can
interfere with the analysis, resulting in the creation of new clusters. thus, in order
to best visualise the different structures existing inside the cells the numbers, the
number of clusters has to be increased to 10 groups, which produced the best reproducibility. In this way, the variability in the spectra obtained at the edge of the
cells are contained in different clusters and do not interfere with the identifications
of the different structures present within the cells. the initiation step was selected
randomly for each iteration and it was seen to have no effect on the final results of
analysis. After analysis, the spectra were gathered in the 10 different groups according to their similarities and a colour (arbitrarily) attributed to each cluster. False colour maps were then constructed representing the partition of the different clusters
in the cells. the resulting images are presented in Fig. 13.13b, d. A correspondence
between the organisation of the clusters and the structure present in the cells can be
seen. In Fig. 13.13b, specific clusters relating to the cytoplasm, nucleus and nucleoli
appear. the specificity of the information contained in the spectra recorded allows
discrimination of the different cellular compartments using K-means analysis. the
sensitivity of the analysis furthermore allows visualization of the two different compartments present in the cytoplasm of the A549 cell and identifies a specific cluster
related to the nucleoli in the haCat cell, although they are not clearly seen in the
visible image.
K-means clustering analysis appears to be a powerful tool to identify the different structures present in the cells. the false colour maps give a representation of
the partition of the different clusters but do not provide any information about the
chemical composition of the samples. this biochemical information and the variations between the different clusters can be visualised through the mean spectra of
the clusters. Figure 13.14 presents the average spectra obtained from selected clusters of the K-means analysis performed on an A549 cell (Fig. 13.13b). (A: nucleus
from the cluster number 4, B: nucleolus from the cluster number 6, C: cytoplasm
from the cluster number 5 and d: cytoplasm from the cluster number 7). the other
clusters correspond to the spectra recorded from the collagen gel and those located
at the edge between the cell and the substrate (data not shown). variations can be
observed in different spectral windows according to the different clusters (highlighted in grey in Fig. 13.14).
the characterisation of live cells on 2d substrates is well documented [123, 124,
147–149] and, based on literature; assignments of the different peaks can be performed. For example, the variations in the spectral range 718–725 cm
−1
are related
to the lipids and nucleic acids, between 815 and 830 cm
−1
to RNA and proteins, from
1,088 to 1,102 cm
−1
to the nucleic acids, the lipids and carbohydrates, from 1,243–
1,266 cm
−1
to the nucleic acids, the proteins and lipids, from 1,296–1,322 cm
−1
to
the lipids and from 1,315–1,348 cm
−1
to the nucleic acids and proteins. In analysing
the cluster average spectra, it is important to note that, with a ~ 1.5 µm spot size,
while the nucleus can be specifically targeted, the overlying cellular membrane and