Biospectroscopy insights into the multi-stage process of cervical cancer development: probing for spectral biomarkers in cytology to distinguish grades

Abstract

Cervical cancer screening programmes have greatly reduced the burden associated with this disease. However, conventional cervical cytology screening still lacks sensitivity and specificity. There is an urgent need for the development of a low-cost robust screening technique. By generating a spectral “biochemical-cell fingerprint”, Fourier-transform infrared (FTIR) spectroscopy has been touted as a tool capable of segregating grades of dysplasia. A total of 529 specimens were collected over a period of one year at two colposcopy centres in Dublin, Ireland. Of these, n = 128 were conventionally classed as high-grade, n = 186 as low-grade and n = 215 as normal. Following FTIR spectroscopy, derived spectra were examined for segregation between classes in scores plots generated with subsequent multivariate analysis. A degree of crossover between classes was noted and this could be associated with imperfect conventional screening resulting in an inaccurate diagnosis or an incomplete transition between classes. Maximal crossover associated with n = 102 of 390 specimens analyzed was found between normal and low-grade specimens. However, robust spectral differences (P ≤ 0.0001) were still observed at 1512 cm⁻¹, 1331 cm⁻¹ and 937 cm⁻¹. For high-grade vs. low-grade specimens, spectral differences (P ≤ 0.0001) were observed at Amide I (1624 cm⁻¹), Amide II (1551 cm⁻¹) and asymmetric phosphate stretching vibrations (ν_asPO₂⁻; 1215 cm⁻¹). Least crossover (n = 50 of 343 specimens analyzed) was seen when comparing high-grade vs. normal specimens; significant inter-class spectral differences (P ≤ 0.0001) were noted at Amide II (1547 cm⁻¹), 1400 cm⁻¹ and 995 cm⁻¹. Deeper understanding of the underlying changes in the transition between cervical cytology classes (normal vs. low-grade vs. high-grade) is required in order to develop biospectroscopy tools as a screening approach. This will then allow for the development of blind classification algorithms.