1.

Introduction

Cervical cancer is the fourth most common cancer in women worldwide with an estimated 528,000 cases and 266,000 deaths in 2012.¹ The link between cervical cancer and high risk human papilloma virus (HPV) infection and the molecular control of the cell cycle is firmly established.² The presence of an HPV infection does not automatically mean the development of cervical cancer or precancer [cervical intraepithelial neoplasia (CIN)]. However, the HPV virus can integrate its viral DNA into the host cell’s DNA located in the cell’s nucleus. This integration disrupts the host cell’s control of cell growth and replication and can lead to malignant transformation.³ The most common method for cervical screening is based on the papanicolaou test (pap test), which is designed to screen for signs of precancer and cancer of the cervix. The pap test requires cells to be scraped from the cervix, fixed to a glass slide, stained, and reviewed by a trained cytologist. Cellular abnormalities are identified based on cellular morphology and staining characteristics and classified according to the degree of dysplasia. The ThinPrep liquid-based cytology (LBC) method involves the patient’s cells being transferred into a specimen vial containing PreservCyt transport medium. PreservCyt is an alcohol-based solution that preserves cell morphology and breaks down biological components, such as mucin and blood. The normal smear pattern is made up of epithelial and nonepithelial components [Fig. 1(a)]. The epithelial cell components include parabasal cells, intermediate cells, superficial cells, endocervical and endometrial cells. Nonepithelial cell components include red blood cells, lymphocytes, histiocytes, polymorphs, mucus, and bacteria. Red blood cells are 7 to $8\mu \mathrm{m}$ in size and stain red or pink with the pap stain. They are often found in cervical samples, either due to menstruation or inflammation and their presence can also be used as an indicator of disease (precancer).⁴ A patient’s cervix with a high-grade (HG) lesion will often bleed on contact with the cervical brush and, as a result, the smear becomes contaminated with blood. The presence of blood in the sample will often dilute diagnostic cells, cover and surround cell groups, increase screening time, and make microscopic interpretation difficult [Fig. 1(b)]. PreservCyt contains a lysis agent for mucus and red blood cells but excessive blood in the sample vial will prevent a satisfactory smear being produced. High volumes of red blood cells will prevent the collection of epithelial cells onto the polycarbonate filter membrane during processing resulting in a ringed halo effect with red blood cells deposited mostly around the edges of the smear with very few epithelial cells present [Fig. 1(c)].

Fig. 1

(a) Pap smear after staining. Note the presence of intermediate (blue), superficial (pink) cells, and lymphocytes (indicated by arrows), (b) bloody smear pattern with red blood cells, obscuring diagnostic cells. $\text{Bar}=35\mu \mathrm{m}$. (c) Pap smear contaminated with blood showing ringed halo effect.

Cytology laboratories often annotate pap smears according to a blood scale index, where 0 indicates no visible evidence of blood present and 3 indicates an extremely bloody sample (Fig. 2). Samples presenting as a 2 or 3 on the blood scale are often either rejected as being unsatisfactory or treated with a wash solution to lyse more of the red blood cells to increase the number of epithelial cells present on the slide. This blood scale grading is subjective and is only based on a visual appearance of a sample. The advantage of the pap test is that it is a noninvasive and widely accepted screening-based test. However, despite a high specificity of 96.5%, a lower sensitivity of 85% can be due to sampling, technical and interobserver errors associated with the subjectivity of the test.⁴ Over the past 20 years, Raman spectroscopy has shown potential for the diagnosis of cervical cancer both in vivo and ex vivo.^5–12 Raman spectroscopy is based on inelastic light scattering in which a sample is illuminated by monochromatic laser light and interactions between the incident photons and molecules in the sample result in the scattering of the light. The coupling of the light generates vibrations within the sample, which are characteristic of the chemical structure. The energy of the scattered light is reduced by an amount equal to the vibrational energy, which is called the Raman shift. This means that the position, peaks, and shape of the Raman bands carry information about the molecular makeup of the sample. The Raman spectrum of cells and tissues is made up of contributions from many biochemical components, including DNA, RNA, proteins, lipids, and carbohydrates.¹² To date, most likely due to confounding factors, such as blood contamination, there have been relatively few studies on Raman spectroscopy of cervical smear samples.^13–19 Rubina et al.¹⁶ described the influence of blood on the Raman spectra of cervical cell pellets and how this might lead to false interpretations. Classification efficiency of 86% and 84% was reported for normal and abnormal samples, respectively, and this decreased to 78% and 79% after samples were treated with red blood cell lysis buffer to reduce blood contamination. A previous study by our group¹⁷ addressed many of the issues involved in recording Raman spectra from LBC (ThinPrep) samples and described a method to remove blood contamination before Raman spectroscopy based on pretreatment of the slides with hydrogen peroxide. This was shown to minimize variability and to result in the collection of highly reproducible data with excellent discrimination between negative and HG samples. However, only blood scale 0 and blood scale 1 samples were used for this study as the treatment was not found to be effective for excessively contaminated blood scale 2 and 3 samples. Kerr et al.²⁰ investigated the effect of blood contamination on Raman spectra of urine cytology samples. They determined that spectra recorded from samples with a low level of blood appeared free from contamination due to the effective removal of many of the red blood cells using the lysing agent CytoLyt. However, for excessively bloody samples, several additional Raman peaks associated with blood were observed leading to the conclusion that excessively bloody samples are not suitable for diagnostics with Raman microspectroscopy. The aim of the present study was to extend our previous study¹⁷ to samples annotated as blood scale 2 and 3 to investigate if these excessively bloody samples could be treated to render them suitable for Raman spectroscopy.

Fig. 2

ThinPrep vials graded according to the blood scale index, where 0 indicates no visible evidence of blood and 3 indicates an extremely bloody sample.

2.

Materials and Methods

3.

Results

Figure 3(a) shows an untreated Pap smear, blood scale 3. The slide was coated in blood and cellular debris. Figure 3(b) shows Raman spectra recorded from cells present on the slide. Very few cells were present and often burned on contact with the laser. A high background was observed together with Raman bands associated with blood at 754, 1311, 1374, 1398, 1588, $1640{\mathrm{cm}}^{-1}$. The high background can be attributed to the cellular debris or a subunit of the oxidized hemoglobin on the surface of the cell nuclei, which has a weak fluorescence.¹¹ Figure 3(c) shows the same sample from Fig. 3(a) after undergoing the blood removal method. Any blood residue has been oxidized with the ${\mathrm{H}}_2{\mathrm{O}}_2$ treatment and cellular debris has been reduced allowing for high quality spectra to be recorded as demonstrated in Fig. 3(d).

Fig. 3

(a) Untreated pap smear, blood scale 3, stars indicate recorded nuclei in field-of-view. $\text{Bar}=35\mu \mathrm{m}$. (b) Raman spectra recorded from untreated pap smear. (c) Treated pap smear, stars indicate recorded nuclei in field-of-view. $\text{Bar}=35\mu \mathrm{m}$. (d) Raman spectra recorded from treated pap smear.

In order to determine if the Raman specific method of blood removal using ${\mathrm{H}}_2{\mathrm{O}}_2$ had any effect on the spectra recorded without any external confounding factors, such as inflammation, HPV infection, etc., the protocol was performed on HeLa cervical cancer cells. HeLa cells were seeded with blood to mimic cytology samples with blood scale index 2 and 3 and HeLa cells with no added blood acted as controls. Both sample types were treated with ${\mathrm{H}}_2{\mathrm{O}}_2$ and slides prepared as for the ThinPrep cytology samples. Figure 4(a) shows the mean Raman spectra recorded from 30 HeLa cells with and without blood contamination (blood scale 2 and 3) and treated with ${\mathrm{H}}_2{\mathrm{O}}_2$. The mean spectra display strong similarities with only minor changes at $1242{\mathrm{cm}}^{-1}$ (Amide III), $1350{\mathrm{cm}}^{-1}$ (DNA/RNA, CH def. in proteins and carbohydrates), $1430{\mathrm{cm}}^{-1}$ (protein, lipids), and $1669{\mathrm{cm}}^{-1}$ (Amide I). Figure 4(b) shows the mean Raman spectra of negative cytology samples (TN) with and without blood contamination (blood scale 2 and 3) and treated with ${\mathrm{H}}_2{\mathrm{O}}_2$. The mean spectra display strong similarities with small changes at 1242, 1350, 1430, and $1669{\mathrm{cm}}^{-1}$ as observed for the HeLa cells. These small changes in the spectra were consistent throughout all samples tested including the HG samples [Fig. 4(c)].

Fig. 4

(a) Mean Raman spectra of HeLa cervical cancer cells treated with blood (blood scale 2 and 3) and subsequently treated with ${\mathrm{H}}_2{\mathrm{O}}_2$ (red) and HeLa cervical cancer cells not treated with blood but treated with ${\mathrm{H}}_2{\mathrm{O}}_2$ (blue), (b) mean Raman spectra of true negative samples (TN), blood scale 2 and 3, treated with ${\mathrm{H}}_2{\mathrm{O}}_2$ (red) and blood scale 0, treated with ${\mathrm{H}}_2{\mathrm{O}}_2$, (c) mean Raman spectra of HG samples, blood scale 2 and 3 treated with ${\mathrm{H}}_2{\mathrm{O}}_2$ (red) and blood scale 0, treated with ${\mathrm{H}}_2{\mathrm{O}}_2$.

PLS-DA classification of negative cytology samples and HG cytology samples, blood scale index 0, is shown in Fig. 5. Figure 5(a) is a latent variables (LV) scores scatter plot of LV1 and LV2, which shows good discrimination along LV1. The loadings, as shown in Fig. 5(b), show that the discrimination is based around Raman peaks at 482, 851, 936, 1128, 1338, 1449, 1572, 1655, and $1669{\mathrm{cm}}^{-1}$, which are mainly related to DNA, proteins, and glycogen. Figure 5(c) is a PLS-DA prediction plot showing good classification of negative and HG samples with a sensitivity of 92% and a specificity of 93% (Table 1).

Fig. 5

(a) LV score scatter plot of Raman spectra acquired from 15 TN samples and 15 HG samples (blood scale index 0), (b) LV 1 loadings, and (c) PLS-DA prediction plot showing good discrimination between TN (blue) and HG (red) ThinPrep cytology samples.

Table 1

Sensitivity and specificity from PLS-DA classification of TN and HG cervical cytology samples with and without blood contamination and combined.

PLS-DA classification of TN and HG cytology samples, blood scale index 2 and 3, is shown in Fig. 6. Both negative and HG samples have undergone the Raman-specific blood removal method. Figure 6(a) is a LV scores scatter plot of treated TN and HG samples, which show overlap between the samples. The LV1 loadings, as shown in Fig. 6(b), show that the discrimination is based around Raman peaks at 482, 549, 851, 936, 1145, 1182, 1237, 1338, 1449, 1576, and $1669{\mathrm{cm}}^{-1}$ and are similar to the loadings shown in Fig. 5(b). The PLS-DA prediction plot is shown in Fig. 6(c) and has a sensitivity of 89% and a specificity of 88% (Table 1).

Fig. 6

(a) LV score scatter plot of Raman spectra acquired from 15 TN and 15 HG samples (blood scale index 2 and 3), (b) LV1 loadings, (c) PLS-DA prediction plot showing some discrimination between TN (blue) and HG (red) ThinPrep cytology samples, blood scale 2 and 3 and treated with ${\mathrm{H}}_2{\mathrm{O}}_2$.

In order to determine if excessively bloody samples could be treated the same way as nonbloody samples in terms of data analysis, the Raman spectral data from the two groups (blood scale 0 and blood scale 2 and 3) were combined. Figure 7(a) shows the LV scores scatter plot, which highlights decreased discrimination between TN and HG cytology. However, the loadings shown in Fig. 7(b) remain similar to the loadings shown in Figs. 5(b) and 6(b) with Raman peaks at 482, 851, 1129, 1184, 1251, 1338, 1449, and $1669{\mathrm{cm}}^{-1}$. The PLS-DA prediction plot is shown in Fig. 7(c) and has a sensitivity of 82% and a specificity of 87% (Table 1).

Fig. 7

(a) LV score scatter plot of Raman spectra from TN and HG samples (combined blood scale 0 and blood scale 2 and 3), (b) LV1 loadings, and (c) PLS-DA prediction plot showing some discrimination between TN (blue) and HG (red) ThinPrep cytology samples, combined blood scale 0 and blood scale 2 and 3 and treated with ${\mathrm{H}}_2{\mathrm{O}}_2$.

Interestingly, the loadings (482, 851, 936, 1149, 1338 and $1669{\mathrm{cm}}^{-1}$) remain almost the same between all three datasets, blood scale 0, blood scale 2 and 3 and combined, confirming that glycogen, DNA, and proteins are the main discriminating features between negative cytology and HG cytology samples regardless of blood contamination. Similar findings have been reported previously.^17,19 However, as mentioned earlier, some spectral features undergo minor changes following the blood removal method. As these features overlap with the discriminating features, this results in a reduction in classification efficiency when excessively bloody (blood scale index 2 and 3) samples and nonbloody samples (blood scale index 0) samples are combined for analysis. Comparing sensitivity and specificity of the test groups shows that blood scale 0 samples with no blood contamination performed best with a sensitivity of 92% and specificity of 93% when compared to excessively bloody blood scale 2 and 3 samples, which had a sensitivity of 89% and specificity of 88%. Combining the blood scale 0 and blood scale 2 and 3 samples for analysis resulted in a sensitivity of 82% and specificity of 87%. The high specificity indicates a good ability to accurately exclude negative samples (few false positives), however, the reduction in sensitivity indicates a reduced ability to accurately detect HG samples and an increased likelihood of false negatives.

Despite the reduction in classification efficiency, the values are higher than previously reported for samples with blood contamination. Rubina et al.¹⁶ reported a classification efficiency of 78% and 79% for normal and abnormal bloody samples treated with red blood cell lysis buffer. In addition, Kerr et al.²⁰ concluded that excessively bloody samples were not suitable for diagnostics with Raman spectroscopy.

4.

Conclusion

This study demonstrates for the first time that excessively bloody (blood scale 2 and 3) ThinPrep cervical smear samples can be used for diagnostics with Raman spectroscopy following a hydrogen peroxide treatment in the collection vial. However, the presence of excessive blood contamination (blood scale index 2 and 3) did result in reduced sensitivity and specificity for classification of negative and HG samples. A higher sensitivity and specificity could be achieved by keeping excessively bloody samples separate from nonbloody samples for analysis. Although this study has focused only on cervical cytology samples, thyroid, urine, and serous fluid cytology samples may also benefit from such an approach for Raman spectroscopic analysis. Serous fluids for cytological examination are often contaminated with a high level of blood (pericardium, pleura, and peritoneum fluids) and often require additional treatment methods, such a density gradient technique to remove red blood cells, which can result in a loss of diagnostic cellular material. The application of our blood treatment method would prevent this loss of material and allow Raman spectroscopy to be performed.

While these results are promising for the development of Raman spectroscopy for cervical cancer screening, Raman spectroscopy in its current form does not have the ability to replace cytology for cervical cancer screening and its associated benefits (e.g., detection of Herpes Simplex virus or actinomyces like organisms). However, it could be used to supplement the pap test or it could have potential as a triage test after HPV testing.