Fluorescence spectroscopy of an in vitro model of human cervical precancer identifies neoplastic phenotype



The clinical diagnosis of cervical neoplasia by spectroscopic methods is potentially a reliable, fast and cost-effective alternative to the conventional smear test. However, it is currently limited by significant inter-patient variation in the spectroscopic properties of the cervix. Characterisation of suitable in vitro models of the spectroscopic changes that take place during neoplastic progression may prove to be a significant step towards the successful development of reliable in vivo systems. In this study, we used organotypic epithelial raft culture as an in vitro model of cervical tissue to analyse changes in the fluorescence properties of surface squamous epithelium that are associated with the development of neoplastic disease. Collagen plugs lined by primary human keratinocytes (PHKs) were used to model the normal cervical epithelium, and plugs lined by cells of the SiHa line were used as a model of neoplastic cervical tissue. Fluorescence emission spectra of these rafts were recorded at excitation wavelengths in the 250–330 nm range, complementing previous work published at longer wavelengths. Normalised, truncated emission spectra were analysed using multivariate principal component analysis. We successfully distinguished between in vitro models of normal and neoplastic cervical tissue and demonstrated a differential effect of acetic acid, which enhances the discrimination of normal from neoplastic tissue. Identification of these differences between in vitro organotypic epithelial rafts may ultimately aid the discrimination of cervical lesions in vivo. © 2007 Wiley-Liss, Inc.

Worldwide, cervical cancer is the second most common malignant tumour in women after breast cancer, with an estimated 500,000 new cases and 200,000 deaths each year.1 Approximately 80% of cervical cancers are squamous cell carcinomas.1, 2 These are invariably preceded by dysplastic precancerous changes, termed cervical intraepithelial neoplasia (CIN), in which the histological changes associated with malignancy are confined to the epithelial layer.3

The potential of fluorescence spectroscopy to diagnose CIN in vitro was first demonstrated by Mahadevan et al., who showed that, at 330 nm excitation, dysplastic biopsy samples are significantly less fluorescent, especially at 385 nm emission, than histologically normal tissues.4 Further expansion of in vivo techniques led to analysis algorithms with high sensitivity, specificity and positive predictive value.5

Multiple excitation wavelengths were first investigated by comparing 337, 380 and 460 nm for the distinction of CIN, normal cervical stratified squamous epithelium and normal cervical simple columnar epithelium.6 These authors determined 337 nm to be the best excitation wavelength for distinguishing CIN from normal stratified epithelium. Further work investigating optimal excitation wavelengths identified specific regions between 300 and 470 nm that allowed discrimination between normal and dysplastic cervical tissue in vivo.7, 8

However, the clinical applicability of in vivo systems is limited by the significant variation in cervical fluorescence properties that occurs between patients.9 The combination of reflectance and fluorescence approaches has some benefits in improving the discrimination between normal and neoplastic tissues,8 and the incorporation of biomarkers into optical strategies may help to improve the detection of cervical precancers.10 Nevertheless, a better understanding of the changes that occur in the physical and biochemical properties of the cervix during the development of neoplastic disease is necessary to develop better diagnostic algorithms.11, 12, 13 For excitation wavelengths above 340 nm, changes in the concentrations of the fluorophores flavin adenine dinucleotide (FAD), the reduced form of nicotinamide adenine dinucleotide (NADH), and type 1 collagen may account for the observed alterations in fluorescence associated with neoplasia.14 Some of the most recent work in this field has shown that both age and follicle stimulating hormone (FSH) levels, which vary with menstrual status, affect the intensity of fluorescence of cervical tissue.15

In vitro methods provide robust systems free of many of the potential sources of variation and allow for the focussed study of simulated parameters. Organotypic epithelial raft culture is a well established in vitro system for investigating the molecular biology of both normal16, 17 and neoplastic18, 19, 20 stratified squamous epithelial tissue. It is known to reproduce many of the biochemical characteristics of stratified squamous epithelium and hence, provides a highly controlled system for recreating the three-dimensional architecture of cervical epithelium, which is grown on a type I collagen matrix that contains fibroblasts and therefore mimics cervical tissue.21, 22, 23

In this study, we aimed to assess the applicability of organotypic epithelial raft culture as an in vitro system for investigating the fluorescence properties of the cervix and the changes that occur in these properties during neoplastic progression. We also assessed the effect of acetic acid application, which is used clinically to assess cervical lesions,24 may be a valuable screening tool25 and has been shown to alter the optical properties of cervical neoplasia.26, 27, 28 We used organotypic epithelial rafts produced using normal, primary human keratinocytes (PHKs) and the cancer-derived cell line SiHa as models of normal and neoplastic cervical tissue respectively and concentrated on excitation between 250 and 330 nm as relatively little work has been done at these wavelengths.

We successfully distinguish between in vitro models of normal and neoplastic cervical tissue and demonstrate a differential effect of acetic acid, which enhances the discrimination of these tissues.

Material and methods


The following cell types were used: (i) J2 fibroblast cells (a gift from Simon Broad, Cancer Research UK), a derivative of the NIH-3T3 murine embryonic fibroblast cell line (ATCC CRL-1658)21; (ii) primary human keratinocytes (PHKs), normal human epidermal keratinocytes isolated from foreskin (PromoCell GmbH, Heidelberg, Germany); (iii) SiHa cells (ATCC HTB35), which were derived from a grade 2 poorly differentiated peritriploid (modal chromosome number 69; range 51–72) squamous cell carcinoma29 and are tumourigenic in nude mice.30

Cell culture

All cell culture and rafting was carried out under aseptic conditions using a class II laminar flow cabinet. All reagents used were supplied by Sigma-Aldrich, St. Louis, and Gibco, Invitrogen, Paisley, UK, unless otherwise stated. Cell culture was performed as described by Harrison and Rae.31 SiHa and J2-fibroblast cells were cultured at 37°C in an atmosphere of 5% CO2 in separate T75 (75 cm2) cell culture flasks containing 10 ml Dulbecco's Modified Eagle Medium (DMEM) supplemented with Fetal Calf Serum (10 vol %), penicillin (100 mg/ml), streptomycin (100 U/ml) and glutamine (2 mmol/l). PHKs were cultured in the same way except that they were grown in serum-free Keratinocyte Growth Medium 2 (KGM-2) (PromoCell, Heidelberg, Germany) rather than DMEM. KGM-2 is composed of Keratinocyte Basal Medium (KBM) together with SupplementMix (PromoCell). The medium in all flasks was changed 3 times a week and cells passaged when confluent.

Preparation of collagen plugs (dermal equivalents)

The collagen plugs (dermal equivalents) were prepared from rat tail type I collagen (Upstate Serologicals Cooperation, Norcross), 10x reconstitution buffer (a solution of 2.2% sodium bicarbonate and 4.8% HEPES buffer in milliQ water) and 10x DMEM solution. 2 ml collagen, 0.25 ml reconstitution buffer and 0.25 ml 10x DMEM solution were used per plug. For most of the plugs, 1 × 106 J2-cells per plug were added to the mix to produce rafts with J2 cells suspended within the collagen plug, although some control rafts containing no J2 cells were also made. Two milliliters of this mixture was then added to each of the wells of a 6 well plate (Iwaki, Japan) (well diameter 35 mm, depth 10 mm) and allowed to solidify by incubation overnight at 37°C.

Organotypic raft culture

The technique used to raft SiHa and PHK (passage 1) cells onto the collagen plugs is based on that described by Southern et al.18 Rafting was carried out using FAD Medium [3 parts DMEM (supplemented with penicillin, streptomycin and glutamine), 1 part Ham's F12 medium], supplemented with fetal calf serum (5 vol %), adenine (180 μmol/l), insulin (5 μg/ml), hydrocortisone (400 ng/ml), apo-transferase (5 μg/ml) tri-iodothyronine (0.2 mmol/l), cholera toxin (0.1 nmol/l) and epidermal growth factor (5 ng/ml).

Twenty-four hours after formation of the collagen plugs, 2 ml of FAD medium containing 1 × 106 keratinocytes (SiHa or PHKs as required), were added to each of the plugs in 6 well plates. Forty-eight hours later, the medium was removed from the plugs and each plug transferred to a circular stainless steel wire mesh support using a sterile spatula, and placed in a 90 mm diameter Petri dish. The mesh supports were made from circular stainless steel mesh discs that were specially made to order (Costacurta S.P.A., Milan, Italy). FAD medium was then added to the Petri dish until it just made contact with the bottom of the plug. Thus, the purpose of the mesh support is to suspend the raft just above the surface of the medium in order to induce epithelial differentiation. The medium in each Petri dish was changed every 72 hr and measurements were made on the rafts twelve days after transfer to the Petri dishes.

Autofluorescence measurements

All fluorescence measurements were made using a Cary Eclipse Fluorescence Spectrophotometer (Varian Optical Spectroscopy Instruments, Mulgrave, Victoria, Australia). A schematic diagram of the setup is shown in Figure 1. The spectrometer runs an auto-calibration protocol of both excitation and emission monochromators upon starting up. A further validation was performed using the Raman spectrum of water and the Varian Validation software. Data were corrected for the spectral response of the instrument using the correction file supplied by Varian, Inc. for this instrument. As a further validation, we compared the fluorescence spectra of dilute Fluorescein 27 and Rhodamine 6G solutions in quartz cuvettes recorded on the Cary Eclipse and a further spectrally corrected fluorimeter (Fluoromax).

Figure 1.

Schematic diagram of the apparatus showing excitation and emission geometries in the Cary Eclipse fluorimeter sample chamber from above.

Excitation wavelengths ranging from 250 to 330 nm (in 10 nm intervals) were used in ‘3D mode’ to illuminate a 2 × 10 mm area on the rafts and record emission spectra (260–660 nm, 1 nm resolution, 1 sec averaging time, 5 nm slit width). The intensity of the excitation beam was measured at all excitation wavelengths using a powermeter with UV head (Coherence Fieldmaster) and found to be of the order of 10 μW over the beam area. Measurements were repeated at all excitation wavelengths on the same raft and no evidence of photobleaching was observed. The averaging time was chosen to decrease the noise level while keeping the measurement time below 20 min for a set of spectra from each raft. Dehydration of the rafts was negligible over this time scale, and rafts were resuspended in medium immediately before being fixed for histology. For this reason, the rafts remained attached to the wire mesh on which they were grown throughout measurements, also limiting physical damage and necrosis. Samples were suspended vertically within the sample compartment of the fluorimeter. The wire mesh was found to be highly scattering at all angles of excitation incidence due to the three dimensional weaved geometry of its cylindrical wires, but was nevertheless used because of its suitability for raft culture.

Control measurements were made of DMEM medium both with and without phenol red content in quartz cuvettes (Hellma Scientific 10 mm/3.5 ml), and the wire mesh (at several angles of excitation incidence). All further measurements were made at 45° to both the incident excitation and the detector.

Four sets each of J2 (background), PHK (normal stratified epithelium) and SiHa (neoplastic epithelium) rafts were cultured separately and measured on different dates under the same conditions. Rafts were not washed of medium prior to measurement. The 4 sets of measurements were analysed together. Experiments were performed to determine the intra- and inter-raft variations in emission spectra.

Application of acetic acid

In a second set of measurements spectral data were recorded as described above, following which acetic acid (3% solution by volume) was applied to rafts outside the fluorimeter sample chamber and left to act for 2 min before repeating measurements. The sample was then placed back in the fluorimeter in the same position and measurements performed over the following 20 min as described above. The concentration of acetic acid used is consistent with local clinical practice and is within the range used in visual inspection after acetic acid (VIA) strategies.25, 32, 33 Clinical practice shows that the most prominent changes in optical properties following the application of acetic acid are rapid.24

Raft histology

To confirm the morphological features, rafts were removed from the mesh after measurements had been made and were then fixed in 10% neutral buffered formalin and embedded in paraffin wax. 5 μm sections were cut, stained with haematoxylin and eosin and viewed by light microscopy.

Data analysis

Spectral data were truncated below λ Ex + 30 nm and above 2λEx × 50 nm (where λEx is the excitation wavelength) to eliminate contributions from the source light and its second harmonic scattered off the wire mesh. The data were then normalised and analysed using principal component analysis (PCA), followed by a peak ratio analysis based on the first principal component.

For the normalisation, the total mean intensity of all the spectra at the same excitation wavelength was calculated. Then, each individual fluorescence spectrum was normalised by dividing it by this mean intensity. This procedure retains the correct global spectral intensity ratio between the different excitation wavelengths. At the same time, it corrects for intensity variations between spectra at the same excitation wavelength.

We used principal component analysis in conjunction with the covariance matrix as described by Chatfield and Collins.34 In the first instance, we looked at the first 2 principal components for a given excitation wavelength. These are represented in the so called ‘principal co-ordinates analysis’ to map the variations of the spectra and observe the clustering between the different cell lines. Using this method, it is possible to determine the optimum excitation wavelength to achieve best discrimination between the different cell lines.

The absolute minimum and maximum of the first principal component spectrum give the wavelengths of optimal discrimination between the spectra for each specific excitation wavelength. There is a possible instrumental contribution at 425 nm; so this emission wavelength was not used in the peak ratio analysis. Care was also taken to establish that peaks did not move with excitation wavelength, as would be expected from scattered light. The intensities corresponding to the 2 wavelengths selected by PCA were plotted as 2D scatter plots.


Figure 2a shows fluorescence emission spectra recorded at 290 nm excitation from rafting medium (with and without phenol red), the mesh upon which the rafts were grown, and from a collagen plug containing no J2 cells nor any surface keratinocytes and sitting on the mesh. Raw emission spectra recorded at 290 nm excitation from the bare mesh and from collagen, J2, PHK and SiHa rafts, are presented in Figure 2b. The excitation peak at 290 nm is shown. The intensity of fluorescence from the rafts lined with cells is much greater than that from the collagen or J2 rafts indicating that a significant fluorescence signal from the surface epithelial cells is present. This is consistent with the finding that appropriate cells were present on the surface of the PHK and SiHa rafts although, as the histology was performed after measurement, the features were mechanically disrupted by removal from the wire mesh (data not shown).

Figure 2.

(a) Emission spectra of the metal mesh, the DMEM medium with and without phenol red content and a collagen plug, excited at 290 nm. The medium was measured in cuvettes and spectra were scaled to fit the graph. (b) Raw emission spectra of the metal mesh, a collagen plug on the metal mesh, a J2 raft (background), a PHK raft (normal stratified epithelium) and a SiHa raft (neoplastic epithelium) excited at 290 nm.

Figure 3 shows normalized fluorescence spectra recorded from each of the 4 sets of J2 rafts (red), 4 PHK rafts (blue) and 4 SiHa rafts (green). Each graph displays spectra at a different excitation wavelength and displays the inter-raft consistency between the fluorescence measurements. A broad feature peaking at 334–340 nm dominates the spectra at excitation wavelengths between 250 and 300 nm inclusive. Spectra recorded from different areas of the same raft showed very good intra-raft consistency (data not shown). The variability in these spectra is smaller than that of the inter-raft data presented in Figure 3.

Figure 3.

Normalised fluorescence spectra and principal coordinates spectra from the 4 sets of J2 rafts (background, red squares), 4 PHK rafts (normal stratified epithelium, blue squares) and 4 SiHa cell rafts (neoplastic epithelium, green squares) excited at (a) 250, (b) 260, (c) 270, (d) 280, (e) 290 and (f) 300 nm.

Spectral shape differences between the PHK, SiHa and J2 rafts are readily identified by visual inspection of normalised emission spectra at all excitation wavelengths between 250 and 300 nm excitation inclusive (Fig. 3). Above 300 nm excitation, however, spectra become dominated by noise at background level (data not shown). Differences between spectra can be quantified using PCA, and principal co-ordinates plots and spectra (Fig. 3). Distinction between the cell types and background is achieved for excitation wavelengths between 250 and 300 nm.

Computation of the first principal component gives the emission wavelengths at which best discrimination between all input spectra is achieved. This is best visualised in scatter plots plotting the normalised emission intensities at the 2 wavelengths corresponding to the maximum and minimum values of the first principal component spectrum. A further advantage of this method of presentation is that peaks that are instrumental contributions, such as the feature at 425 nm emission, are readily avoided and hence do not affect the results.

The two systematically chosen emission wavelengths that were identified as best points of discrimination are shown as wavelengths x and y in Table 1. The entries 280, 290, 300, 510, 530 and 550 nm correspond to the edges of the truncated spectrum, and do not represent a physically significant wavelength. However the 334–340 nm peak picked up in this analysis is a distinctive feature. Normalised emission intensities from all rafts at wavelength y are plotted against those at wavelengths x (Fig. 4). Best clustering of rafts and separation between different raft types is found at 280 and 290 nm excitation.

Figure 4.

Emission intensity scatter plots at (a) 250, (b) 260, (c) 270, (d) 280, (e) 290 and (f) 300 nm excitation. Axes display arbitrary units of emission intensity at the labelled wavelengths. Wavelengths were chosen by PCA.

Table I. The Two Wavelengths at Which Distinction Between Raft Types is Maximised for Each Excitation Wavelength1
Excitation (nm)Wavelength x (nm)Wavelength y (nm)
  • 1

    These are derived from the absolute minima and maxima of the first principal component spectrum. The normalized intensities at wavelengths x and y are plotted against each other to give the scatter plots in Figures 4 and 5.


Following application of acetic acid to the PHK rafts, an increase in relative emission intensity at 300 nm excitation/340 nm emission was seen whilst for the SiHa rafts a corresponding decrease in fluorescence intensity was seen (Fig. 5). This trend was also present at shorter excitation wavelengths, but was less marked than at 300 nm.

Figure 5.

Emission intensity scatterplots demonstrating the effect of acetic acid on the fluorescence properties of (a) PHK and (b) SiHa rafts at 300nm excitation. Filled datapoints denote preapplication and outlined datapoints post-application measurements. Wavelengths were chosen by PCA.


The large emission peak at 334–340 nm is due to autofluorescence of tryptophan. Tyrosine emission at 300 nm features as a shoulder on this main peak. These peaks dominate the spectra recorded from all the rafts and from the medium. The presence of phenol red does not change the emission spectrum of the medium significantly at the wavelengths examined. Although the medium does contribute to the fluorescence observed, rafts lined with cells show a much higher intensity of fluorescence emission than do pure collagen or J2 rafts, indicating significant autofluorescence of the surface epithelial rafts over and above that from the medium, collagen and J2 cells. This allows clear identification of the presence of cells on top of the collagen and fibroblast background and is an essential step towards the further analysis, which focuses on discriminating between cell types grown on rafts.

Discrimination between raft types was achieved using emission intensity scatterplots to provide a simple, visual method of displaying the distinctive features of spectra. These features were determined quantitatively by PCA. Principal co-ordinate plots, based on differences across the whole of the spectra, show clear distinction between PHK and SiHa rafts for excitation wavelengths from 250 to 300 nm inclusive. The scatter plots however are based on analysis at just 2 emission wavelengths, which were systematically chosen to include only fluorescence contributions. Best discrimination between PHK and SiHa rafts was identified at 280 and 290 nm excitation. The identification of 2 significant emission wavelengths could ultimately allow detection of neoplastic tissue with simple apparatus using two-point sampling. The improved discrimination at 300 nm following the acetic acid treatment arises from spectral shifts, which increase the spectral differences between the PHK and SiHa rafts, a finding consistent with previously published work.26, 35

In conclusion, we have investigated the fluorescence properties of organotypic epithelial rafts produced using primary human keratinocytes and SiHa cells, as models of normal and neoplastic stratified squamous epithelium. We have shown that there is significant fluorescence signal from the surface keratinocyte layer over and above fluorescence associated with the medium, collagen and J2 cells. We also demonstrate discrimination between raft types at excitation wavelengths between 250 and 300 nm. These differences suggest that the development of neoplastic changes in the cervix is associated with alterations in fluorescence in this wavelength range. The application of acetic acid was found to enhance discrimination between normal and neoplastic models at 300 nm excitation. These results are a useful step towards understanding the spectral properties of tissue and identifying specific changes in biopsy samples. Identification of these differences may aid the discrimination of cervical lesions in vivo.


We thank Professor Malcolm White, Biomolecular Sciences, University of St. Andrews, for access to the Cary Eclipse Fluorescence Spectrophotometer. A.D.W. was funded by a Health Foundation Research Fellowship.