FTIR Microspectroscopy Coupled with Two-Class Discrimination Segregates Markers Responsible for Inter- and Intra-Category Variance in Exfoliative Cervical Cytology

Abstract

Infrared (IR) absorbance of cellular biomolecules generates a vibrational spectrum, which can be exploited as a "biochemical fingerprint" of a particular cell type. Biomolecules absorb in the mid-IR (2-20 mum) and Fourier-transform infrared (FTIR) microspectroscopy applied to discriminate different cell types (exfoliative cervical cytology collected into buffered fixative solution) was evaluated. This consisted of cervical cytology free of atypia (i.e. normal; n = 60), specimens categorised as containing low-grade changes (i.e. CIN1 or LSIL; n = 60) and a further cohort designated as high-grade (CIN2/3 or HSIL; n = 60). IR spectral analysis was coupled with principal component analysis (PCA), with or without subsequent linear discriminant analysis (LDA), to determine if normal versus low-grade versus high-grade exfoliative cytology could be segregated. With increasing severity of atypia, decreases in absorbance intensity were observable throughout the 1,500 cm(-1) to 1,100 cm(-1) spectral region; this included proteins (1,460 cm(-1)), glycoproteins (1,380 cm(-1)), amide III (1,260 cm(-1)), asymmetric (nu(as)) PO(2) (-) (1,225 cm(-1)) and carbohydrates (1,155 cm(-1)). In contrast, symmetric (nu(s)) PO(2) (-) (1,080 cm(-1)) appeared to have an elevated intensity in high-grade cytology. Inter-category variance was associated with protein and DNA conformational changes whereas glycogen status strongly influenced intra-category. Multivariate data reduction of IR spectra using PCA with LDA maximises inter-category variance whilst reducing the influence of intra-class variation towards an objective approach to class cervical cytology based on a biochemical profile.

Figure 1

(A) Average absorbance spectra of the biochemical-cell fingerprint region (1,800 cm⁻¹ to 900 cm⁻¹) comparing normal (black), low-grade (green) and high-grade (blue) categories of exfoliative cervical cytology. (B) Chronology of the process from the collection of exfoliative cytology smears and identification of histological subtype to the multivariate data analysis.

Figure 2

(A) PCA scores plot comparing normal (n = 60, black squares) and low-grade (n = 60, green diamonds) categories of exfoliative cervical cytology derived using the biochemical-cell fingerprint region (1,800 cm⁻¹ to 900 cm⁻¹). The region of overlap between scores for the two categories of exfoliative cervical cytology is identified between the dotted red lines. The corresponding PCA loadings plot which indicates the spectral regions responsible for segregation were derived for PC1 (black), PC2 (red) and PC5 (purple). (B) A second PCA scores plot derived from normal (black squares) and low-grade (green diamonds) categories of exfoliative cervical cytology, which were located in the region of overlap or were outliers from the initial PCA run. Loadings plot of the second PCA for PC1 (black), PC2 (red) and PC5 (purple). (C) A final scores plot of normal (black squares) and low-grade (green diamonds) categories of exfoliative cervical cytology located in the overlap region or were outliers from the second PCA run. Loadings plot of the final PCA were obtained for PC2 (black), PC3 (red) and PC4 (purple).

Figure 3

(A) PCA scores plot comparing normal (n = 60, black squares) and high-grade (n = 60, blue stars) categories of exfoliative cervical cytology derived using the biochemical-cell fingerprint region (1,800 cm⁻¹ to 900 cm⁻¹). The region of overlap between the scores for the two cell categories is identified between the dotted red lines. Corresponding PCA loadings plot which indicates the spectral regions responsible for segregation were derived for PC1 (black), PC2 (red) and PC3 (purple). (B) A second PCA scores plot derived from specimens that were located in the region of overlap or were outliers from the initial PCA run. Loadings plot of the second PCA for PC1 (black), PC3 (red) and PC4 (purple). (C) A final scores plot of specimens located in the overlap region or were outliers from the second PCA. Loadings plots of the final PCA were obtained for PC2 (black), PC3 (red) and PC5 (purple).

Figure 4

(A) PCA scores plot comparing low-grade (n = 60, green diamonds) and high-grade (n = 60, blue stars) categories of exfoliative cervical cytology derived using the biochemical-cell fingerprint region (1,800 cm⁻¹ to 900 cm⁻¹). The region of overlap between the two cell categories is identified between the dotted red lines. Corresponding PCA loadings plot which indicates the spectral regions responsible for segregation were derived for PC2 (black), PC3 (red) and PC4 (purple). (B) A second PCA scores plot derived from specimens that were located in the region of overlap or were outliers from the initial PCA. Loadings plot of the second PCA for PC1 (black), PC4 (red) and PC5 (purple). (C) A final scores plot of specimens located in the overlap region or were outliers from the second PCA. Loadings plots of the final PCA were obtained for PC1 (black), PC4 (red) and PC5 (purple).

Figure 5

PCA-LDA scores plot using the (A) biochemical-cell fingerprint region (1,800 cm⁻¹ to 900 cm⁻¹) or (B) the DNA/RNA region (1,425 cm⁻¹ to 900 cm⁻¹) with classes assigned as normal (n = 60, black squares), low-grade (n = 60, green diamonds) or high-grade (n = 60, blue stars) categories of exfoliative cervical cytology. PCA-LDA loadings plots were derived for normal (black), low-grade (green) and high-grade (blue), to demonstrate the spectral differences between the average spectrum of each category compared to the average spectrum of all categories.

Figure 6

(A) PCA scores plot using the biochemical-cell fingerprint region (1,800 cm⁻¹ to 900 cm⁻¹) of low-grade specimens (n = 60, green diamonds) to identify apparent CIN1 subgroups. (B) A corresponding PCA loadings plot derived for PCs 1 (black), 2 (red) and 3 (purple).