4.5. FTIR Data Acquisition

The FTIR samples were prepared as previously reported [56] with some modifications. The 2D and 3D SK-MEL-2 cells were trypsinized (final concentration of 0.25%) and centrifuged at 540 g for 5 min. These cell pellets were twice washed using 0.9% of NaCl (w/v) and re-suspended in 50 µL of 0.9% of NaCl (w/v). This step was gently performed to avoid the abrupt change of the osmolality between the culture media and the physiological saline solution. The drop of re-suspended cells was transferred onto a barium fluoride window (BaF₂) and vacuum-dried for 30 min in a desiccator [17,56]. The cells on the window were rinsed with distilled water then vacuum-dried. This step was repeated to completely remove the salt. The washed and dried cell monolayer was kept in a desiccator prior to use. The spotted cells on the BaF₂ window were analyzed in the transmission mode by a FTIR spectrometer (Bruker Vertex 70) connected to a Bruker Hyperion 2000 microscope (Bruker optic Inc, Ettlingen, Germany), using synchrotron radiation as the light source (Synchrotron Light Research Institute, Thailand) [57,58]. The FTIR spectrometer was coupled with a potassium bromide beam splitter and a MCT (HgCdTe) detector under liquid nitrogen. The scanning range was between 800 and 4000 cm⁻¹ at a spectral resolution of 6 cm⁻¹. Each cell drop was scanned 30–40 times to obtain 30–40 spectra. For each spectrum, 64 scans, 10 × 10 µm² aperture size, were acquired using OPUS 6.5 software (Bruker optic Inc, Ettlingen, Germany). Using OPUS 6.5 software, we determined the water-compensation—spectra cut between 900 and 3000 cm⁻¹—baseline correction and integration areas: between 2813 and 2992 cm⁻¹ for the lipid region, 1480 and 1700 cm⁻¹ for the amide I and II region, 1180 and 1280 cm⁻¹ for the DNA region, 1040 and 1140 cm⁻¹ for the RNA region, and 2813 and 2992 cm⁻¹, with 937 and 1772 cm⁻¹ for the whole cell region. After water compensation and baseline correction, the spectra with a peak height at the amide I region (1660 cm⁻¹) lower than 0.4 a.u. was discarded, such that 20–26 spectra were obtained for each cell drop. There were three replications (3 drops from 3 spheroids) from each cell seeding condition (cell group), so 60–76 spectra were obtained per cell group for further analysis.

Multivariate analysis was performed using Unscrambler^® X software (version 10.2, CAMO Software AS, Oslo, Norway). Unsupervised methods were completed using principal component analysis (PCA) and hierarchical cluster analysis using Ward’s algorithm. The primary spectra were smoothed using the Savitzky–Golay method (polynomial order 3 with 13 smoothing points). Then, the spectra (between 2813 and 2992 cm⁻¹, 937 and 1772 cm⁻¹) were baseline corrected and normalized using Extended Multiplicative Signal Correction (EMSC), to correct for differences in sample thickness and light scattering artifacts before performing the multivariate analysis. The top two principal components (PCs) were chosen for analysis. Score plots (2D) and loading plots were used to display the clustering of the spectra and variations (wavenumber) from each range of spectra in the dataset.

Unsupervised hierarchical cluster analysis of the FTIR spectral data sets was performed using Ward’s algorithm, utilizing a matrix defining the inter-spectral distances to identify and confirm the discrimination of the FTIR spectra from the PCA analysis. The spectral regions used in the cluster analysis were obtained from the loading plot, which mainly represents the biological components inside the cells. RStudio version 1.2.1355 was used to perform the hierarchical cluster analysis.

Curve fitting was used to analyze the average absorbance spectra of the amide I and amide II regions (1480–1720 cm⁻¹) so as to differentiate the secondary structure of the protein. A second derivative (Savitzky–Golay method with 13 points of smoothing) was used to estimate the protein secondary structure. The primary spectra were baseline corrected and deconvoluted into the specific areas for the amide I (1593–1723 cm⁻¹) and amide II (1483–1590 cm⁻¹) regions. The 50% Gaussian and Lorentzian function of the OPUS 6.5 software was used to perform the curve fitting. The peak position, integral area, and width were obtained from the curve fitting. The goodness of fit was determined by assessing the residual RMS error. The secondary structures of protein were assigned according to the previous reports [22,24,25,26,59].