A score plot of PCA of the spectra of cells with and without virus 12 h after the addition of virus is shown in Fig. 5(a). Up to seven PCs were obtained and the PCA score plot was composed of two characteristic PCs with major differences between these two groups. The datasets of cells with and without virus showed significant differences. At the border area, some of the data from the virus-treated group overlapped with that of cells not treated with virus. These cells seemed not to sense the virus invasion yet. As described earlier, this data were collected from three independent experiments. When PCA was carried out for each experimental dataset, these three datasets highlighted consistent differences with high reproducibility. Loading plots for PC 1 are shown in Fig. 5(b). As the spectra were normalized with the band due to phenylalanine at , the bands in the loading plot indicate relative changes to the concentration of phenylalanine. Bands at 1445 and were observed in the positive direction, which may be attributed to changes in the composition of lipids. Bands at 1257, 1080, and were attributed to DNA. The PCA score plot for the spectra recorded 24 h after the virus addition is shown in Fig. 6(A). PC 1 (not shown) was unrelated to viral infection and was probably due to variations in culture conditions. PC 2 contributed to the discrimination of two groups. Very few spectra of the cell-with-virus group were classified into the cell-without-virus group, suggesting that the virus infection progressed and most of the cells had already sensed the virus invasion. The control (cell-without-virus) data separated into two groups along the PC 3 axis, which was attributed to the fluctuation of culturing conditions. The loading plot of PC 2 is shown in Fig. 6(B). The noisy nature of the plot suggested that the differences between spectra without and with virus were considerably small. A weak and broadband near , which was almost buried in noise, was probably due to a CH bending mode. A positive band near was attributed to DNA. Bands at 985 and , observed in the positive direction in PC 3, were due to polysaccharides. The PCA score plot for spectra 48 h after the virus addition is shown in Fig. 7(A). The plot showed good discrimination between two groups within the PC 2 axis. There was some overlap between the two groups. The loading plot of PC 2 is shown in Fig. 7(B-b). A negative band near was due to the CH bending mode. A positive band at does not have strong accompanying bands. This feature resembles the Raman spectrum of the group and it may indicate that the normal uninfected cells had more ions than did the virus-infected cells. The features of the loading plots for the cells 12, 24, and 48 h after the virus addition were different, which appeared to reflect the dynamics of the reaction of the cells. During the early stages of infection (), several cells with virus were found to be classified into the group of cells without virus, because the virus had not yet succeeded in invading these cells. At 24 h, the overlap between the groups of cells with and without the virus was much smaller. It appears that the spectral changes reflected the self-defense reaction of the cells to the virus infection. As evident from the fluorescence images, in the cells, the virus started replicating 48 h after its addition. The noise level in PC 2 of the data collected 48 h after virus addition was much lower than those corresponding to data collected after 12 and 24 h, indicating a much larger spectral alteration. The analysis of fluorescence emission showed that 5% to 10% of the cells emitted intense fluorescence, whereas the Raman analysis classified nearly 100% of the cells into the infected group. Judging from the MOI (), six virus particles were used to infect one cell, a ratio sufficiently high to ensure that all cells are infected. Therefore, the Raman analysis was more sensitive in detecting the viral infection than the conventional fluorescence imaging methods.