Patent Application
20160069809
The present application relates generally to spectroscopy and more particularly to the analysis of biological samples by spectroscopic techniques.
Cervical cancer kills approximately 300,000 women each year worldwide with 80% of the deaths occurring in developing countries. Epithelial abnormalities that are potentially capable of progression into an invasive neoplasm have traditionally been categorised either as dysplasia or as carcinoma in situ. Dysplastic changes within the epithelium are graded as being of a mild, moderate, or severe degree. Those cervical epithelial abnormalities associated with an increased risk of invasive carcinoma have now been classified into a single diagnostic category of cervical intraepitheilial neoplasia (CIN). Three grades of abnormality are recognised: CIN I which corresponds to mild dysplasia; CIN II which is equivalent to moderate dysplasia; and CIN III which encompasses both severe dysplasia and carcinoma in situ. There are two types of cervical cancer: squamous cell cancer and adenocarcinoma. They are named after the type of cell that becomes cancerous. Squamous cell cancer is the most common type of cervical cancer, which occurs in the ectocervix. Adenocarcinoma is cancer of gland cells that are located in the endocervical canal, which is less common than squamous cell cancer. A screening test is used to detect premalignant and maligant processes in the ectocervix. Significant changes can be treated, thus preventing the development of cervical cancer. The main method for cervical cancer screening is the Papanicolaou test (Pap test).
Vibrational spectroscopic techniques are potential tools for the non-invasive, label free investigation of biological samples at a molecular level. Over the course of the last few years, the range of applications of Infrared and Raman spectroscopy has extended from intact tissue to single cell analysis, elucidating different mechanisms involved in malignancy and cancer progression as well as probing the efficiency of new drugs and the toxicity of nanoparticles at cellular level. Vibrational spectroscopy is a fast developing discipline showing a strong potential in the field of cervical cancer screening. Since the first time vibrational spectroscopy was suggested as a tool for mass screening of cervical cancer in the 1990s by Wong et al. (1991), there has been significant progress. Numerous studies have shown the potential of vibrational spectroscopic techniques in the detection of cervical cancer and pre-cancerous lesions (Lyng et al. 2007, Diem et al. 2008). Moreover, spectroscopic data handling and multivariate analysis procedures have significantly been developed over the last two decades.
The challenge for vibrational spectroscopy remains for it to be recognized as a medical screening tool that could potentially be used to complement cytology screening techniques. The very high potential to replace the current cytology screening technologies has been proven many times and the high degree of objectivity of the assessment based on molecular composition make them really sensitive and specific tools. Moreover, the identification of specific markers could lead to the capability of detecting early abnormal changes that are not morphologically apparent and thus undetectable using standard methods. Ideally the process employed should be suitable for use for mass screening and accordingly a high degree of automation is desirable.
Extensive studies have been conducted on the study of single cells using Raman spectroscopy. The weak scattering efficiency of water and the high spatial resolution of the visible or near -IR wavelengths employed for Raman microspectroscopy place this technique as the favored tool for the study of living cells, whereby live cells in a physiological solution can be mapped with an immersion lens giving sub cellular resolution of the order of 1 μm. Equipped with high magnification objectives such as ×60 or ×100, the lateral resolution offered by this technique make it a perfectly suitable tool for single cells analysis.
Nevertheless the quality of the information which may be gleaned requires some improvement. The application of vibrational spectroscopy to biological samples is a complex task, mainly due to the numerous parameters that can possibly affect the interpretation of results. Each patient is unique with a specific medical history, lifestyle or genetic pool that could affect any of the organs or cells of the body at different levels with at present uncorrelated and uninvestigated impact on the spectra collected. Therefore, when studying human samples a source of variability related to each individual has to be tolerated while interpreting the data. However in most studies additional sources of variability, mostly instrumentation related, are added to the problem decreasing the specificity and reproducibility of the analysis. To discriminate between two populations often considered as normal and abnormal it is essential to identify markers specific to a pathological state but the importance of the control is often disregarded. In order to have a degree of relevancy the dataset representing the normal state of the sample is expected to exhibit a minimum level of variability. In other words, if the within group variability is greater than the between groups variability, the specificity of any discrimination reached has a limited degree of relevancy.
Unfortunately, results obtained previously have found a high degree of variability among cytology negative samples (normal samples). The present inventors have identified that contamination of the samples with blood residue is found to be present in a high percentage (close to 50%) of samples screened resulting in the variability seen. Accordingly, there is a need for a method of preparing biological samples which reduces the variability of the results, in particular when the samples are analysed by spectroscopy.
Accordingly, the invention provides a method of performing a spectroscopy process on a fixed biological sample, the method comprising the steps of:
(i) treating the fixed biological material to cause oxidation of haemoglobin present in the fixed biological material, and
(ii) performing spectroscopy on the treated fixed biological sample.
Suitably, the spectroscopy process is Raman spectroscopy.
In the present invention Raman spectroscopy was used for the analysis of 50 normal cervical cell samples. The samples were prepared using the Thin prep method of sample preparation which is routinely used for screening of cervical cancer. The present invention has been described with reference to the use of the Thin prep method for preparing samples, however it will be appreciated by the person skilled in the art that the method described herein may also be used in the SurePath method of sample preparation and for treating samples from direct smears.
The inventors of the method according to the invention have identified that contamination of the samples with blood residue is found to be present in a high percentage (close to 50%) of samples screened. This contamination is not visible to the human eye or with a microscope but introduces significant variability in the spectra obtained. The variability generated within normal samples is such that it is too high to be able to use the data for a discrimination purpose between normal and abnormal samples.
Accordingly, to reduce the variability, a washing step is employed to clean the fixed samples. More particularly, the method according to the invention employs the use of a washing step with a solution comprising H2O2 (hydrogen peroxide).
Therefore, the step of treating the fixed biological material suitably comprises treating the fixed biological material with a solution comprising hydrogen peroxide. The fixed biological material suitably comprises cervical cells fixed on a microscope slide.
Preferably, the microscopic slide comprises a glass slide.
Whilst the spectra produced from samples is modified very slightly by the use of hydrogen peroxide, it has been found that the variability between “normal” samples is dramatically reduced. This provides for the realistic use of multivariate analysis of spectral results obtained using Raman spectroscopy in performing automatic detection between normal and abnormal cells.
Suitably the treatment employs a solution with a concentration of hydrogen peroxide in the range of 3% to 30%, preferably 15% to 30%, most preferably 30%.
The fixed biological material may be treated with the hydrogen peroxide for a period of between 2 and 5 minutes.
Preferably, the fixed biological material is treated with the hydrogen peroxide for a period of 5 minutes.
Hemoglobin is particularly sensitive to oxidation which irreversibly changes its structure and thus its spectral signature. Accordingly, the present invention comprises the immersion of individual samples on the slides into a solution of 30% H2O2 for approximately 5 minutes. The sample may be washed afterwards using alcohol.
Preferably, the method according to the invention further comprises the step of washing the slide with alcohol after treatment to remove hydrogen peroxide. The slide may be washed with any suitable alcohol such as methanol, ethanol or industrial methylated spirits, for example.
The method according to the present invention demonstrates the feasibility to record data from a large set of samples with good reproducibility. It has been shown that by washing the slides using H2O2 before recording the spectra, in addition to using adapted acquisition parameters (such as accumulation time for example) and optimising substrate removal methods (such as the removal of the spectral signal obtained from the glass slide, for example) it has been possible to greatly reduce the variability present in the data set. Adapted acquisition parameters and optimised substrate removal methods suitable for use in the method according to the invention are described herein.
For example, the method according to the invention suitably comprises the step of pre-exposing the fixed biological sample to a laser source for an accumulation time of 30 seconds during the spectroscopy process.
It will be appreciated by the person skilled in the art that the spectral features of glass are usually broad and can overlap with the region of interest in a spectrum obtained from a biological sample. The spectral signal generated from a glass substrate is therefore another source of variability when conducting spectroscopy on a sample. The invention further provides a method of correcting spectra in order to remove the contribution of the glass to the spectra.
Suitably, the method according to the invention may further comprise the steps of
(i) obtaining a spectrum and
(ii) compensating the spectrum to remove a spectral signal caused by the glass of the slide, the step of compensating comprising the step of reducing the intensity of said signal by a factor of 10 and repeating this step until a reference calculation is met.
The reference calculation suitably comprises comparing the ratio between two peaks in the spectrum.
Preferably the two peaks are at 1007 cm−1 and 1098 cm−1.
It has been demonstrated that the flexibility of the technique according to the invention will ease the integration in the medical environment as the protocols currently used routinely in cytology can be used for Raman spectroscopy.
The invention will now be described in detail by way of example and/or with reference to the accompanying drawings in which:
The method according to the present invention is suitable for use with biological samples that have been fixed on a slide using liquid based cytology such as Thin prep and SurePath methods of sample preparation. It may also be used for direct smear samples.
The invention is described with reference to the following example in which the Thin prep method of preparing biological samples was used.
The process begins with the patient's gynaecologic sample being collected by the clinician using either a broom-like device or a brush/plastic spatula combination. The device is then rinsed in the specimen vial. The vial suitably contains a solution for preserving the cells. One exemplary solution is the PreservCyt™ solution provided by Hologic of Massachusetts, USA. The sample is then homogenized by spinning. This spinning process may be performed by a specific device for the purpose, e.g. The T2000™ or T3000™ ThinPrep processor from Hologic. The spinning process is selected so as to create shear forces in the fluid that are strong enough to separate randomly joined material, break up blood, mucus and non-diagnostic debris while keeping true cell clusters intact. The cells may then be collected onto the membrane of the filter by applying a gentle vacuum across the filter membrane to aspirate fluid through the membrane. The resulting cells trapped in the filter may then be transferred to provide a monolayer of cells on a glass slide in a 20 mm diameter circle. Once cell transfer is completed, the slide may be processed by immersion in a fixative bath containing 95% ethanol.
Human hemoglobin, hemin and proto-porphyrin (lyophilized powder) were purchased from Sigma Aldrich (Ireland). Each of the chemical compounds was deposited on a CaF2 window in the lyophilized form before recording.
An aliquot of blood comprising 200 microlitres of blood in 20 ml Preservcyte solution was mixed in a ThinPrep vial containing no cervical cells, transferred onto a CaF2 substrate and air dried before recording.
Raman studies were performed using a HORIBA Jobin Yvon XploRA™ system (Villeneuve d'Ascq, France), which incorporates an Olympus microscope BX41 equipped with a ×100 objective (MPlanN, Olympus). A 532 nm diode laser source was used throughout this work. In order to avoid any photo damage to the sample, the power of the laser was set at 50%, resulting in ˜8 mW at the objective. The confocal hole was set at 100 μm for all measurements, the specified setting for confocal operation. The system was pre-calibrated to the 520.7 cm−1 spectral line of Silicon. The XploRA™ system is a confocal spectrometer that contains 4 interchangeable gratings (600, 1200, 1800 and 2400 lines/mm) In the present study the 1200 lines/mm grating was used, which gave a spectral dispersion of ˜3 cm−1 per pixel. Although a higher resolution grating such as 2400 gr/mm may give a spectral resolution of <1 cm−1, the 1200 lines/mm grating was selected for this work to reduce the time of acquisition to 10s without decreasing the signal to noise ratio in the spectra collected. The backscattered light was measured using an air-cooled CCD detector (Andor, 1024×256 pixels). The spectrometer was controlled by Labspec V5.0 software. For each cell, 3 spectra were recorded from the nucleus, each of them corresponding to the average of 3 accumulations of 10s in the spectral range 400-1800 cm−1. For each slide, 10 cells were selected randomly for recording. Also, for each slide, a spectrum of the substrate was recorded for correction of the data. Concerning the recording of the spectra from pure chemicals, the laser power was set to 1% in order to avoid photo-damage of the samples. The acquisition time was increased to provide a final signal to noise ratio close to the one seen on the cervical cells. Thus, the spectra presented are the results of 3 accumulations of 100s.
The average Raman spectra obtained from 50 patient samples with negative cytology are presented in the
The application of Raman microspectroscopy for single cells in vitro has become very popular in the last decades, mainly due to the high resolution of the technique allowing access to sub-cellular information. Based on the numerous documents available in the literature, it is possible to determine what is a typical spectrum recorded from a single cell. Although Raman microspectroscopy can differentiate between different cell lines or cells in different phases of the cell cycle, the variations present in the spectra are usually minimal and can rarely be discerned by eye and the use of advanced multivariate analysis methods is required. Thus, in the present study, the black spectra in
Observation of the cells using an optical microscope does not give any indication of abnormal morphology of the cells.
3. Dealing with the Pattern 2: Reversion to Pattern 1
The origin of pattern 2 type spectra having been identified, the challenge is to remove the variability existing in the spectra to generate an acceptable control dataset. In order to solve the contamination of the samples with blood residues, three different approaches can be considered; (i) modify the protocol before the preparation of the samples while the cells are still in suspension, (ii) treat the slides and try to wash the cells after deposition and drying or (iii) manipulate the data and digitally remove the signal of the blood.
Signal decomposition and correction with approaches such as Independent Component Analysis (ICA) coupled with Non-negatively Constrained Least Squares Analysis (NCLSA) has been documented and proven promising for the application to Raman data. The first step consists of identifying the independent components of pattern 2 by analyzing the co-variance existing in the data set. In the second step, the unwanted hemoglobin component is subtracted from the data set using NCLSA. This method has been successfully used to remove the contribution of wax in embedded skin sample for Raman and Infrared analysis (Tfayli et al. 2009, Ly et al. 2008). However, the samples exhibiting a pattern 2 signature have been also found to be more susceptible to photodegradation, causing some difficulties to collect the data from ThinPrep slides heavily contaminated with blood residues which could represent up to 5% of the samples screened. Thus, although the digital removal of the hemoglobin would be a valid approach, the aim of this invention was to develop an alternative approach allowing the collection of Raman data from all the different samples, regardless of the quantity of hemoglobin initially present on the cell surface. For this reason the present invention is focused on sample preparation and handling before the collection of the datasets. Cell suspensions prepared for ThinPrep or smear samples are annotated according to a blood scale, whereby 0 indicates a clear solution and 3 indicates a bloody sample. In cytology laboratories, samples presenting a grade 2 or 3 on the blood scale are either rejected as being unsuitable or treated using Cytolyt or various solutions to wash the cells before preparing the samples. The Cytolyt solution is expensive and is usually used for bloody samples. It has been observed in this study that some samples with a grade 1 or 0 on the blood scale that would routinely be used for cytology present a pattern 2 signature. Therefore, for conformity with clinical procedures, all the samples would have to be washed with Cytolyt, which is unrealistic in terms of time and expense.
The only approach that can possibly be realistically applicable in a medical environment and for a large number of samples is the treatment of the slides after cell deposition and drying. To be present at this stage, the blood residues must have interacted with the cellular membrane, thus by destabilizing this interaction the residues may be washed away. Different solutions such as Cytolyt, eposti solution and acetic acid have been tested, but none delivered a reproducible result and features relating to the blood residue could still be seen in the spectra recorded (data not shown). Previous work published by Romeo et al. (2003) describes the feasibility to use red cell lysis buffer (RCLB) to reduce the presence of erythrocytes in cervical ThinPrep samples for IR spectroscopy. However, in our hands, this solution has been found to be inefficient after the cells have been deposited and dried on glass slides.
The action of H2O2 on the structure and conformation of hemoglobin is known (Vallelian et al. 2008, Zhou et al. 2007). Hemoglobin is particularly sensitive to oxidation, which irreversibly changes its structure. A solution of 30% H2O2 was therefore employed to treat the samples for 5 mins, after which the samples were washed in alcohol. The first concern was to preserve the cell morphology after such treatment.
FIG. 4III presents the mean spectra obtained from the 50 samples using the H2O2 treatment and the 30s (seconds) pre-exposure to the laser. The variability existing between the data sets corresponding to the samples originally displaying pattern 1 (black) and pattern 2 (grey) has been greatly reduced. Although the intensity is slightly different, the background remains similar. Thus, all the spectra have a profile corresponding to those commonly observed on cells grown in vitro.
In addition to the chemical contamination of the samples, other parameters can induce variability in the data sets recorded. For instance, as the cellular material tends to be relatively thin, the choice of substrate is critical. The spectral features of glass are usually broad and can overlap significantly with the region of interest of the cellular spectra and can therefore be another source of variability (Bonnier et al. 2010, 2011). Quartz or CaF2 are commonly used as an alternative, as, although these substrates can contribute to the final spectra, they do not contain any specific sharp peaks which interfere with the cellular spectrum and their contribution can be at least partially removed by a subtraction (Draux et al. 2009). However, the cost of Quartz or CaF2 slides is simply unrealistic for screening large numbers of samples. Moreover, as the present invention demonstrates the potential to apply Raman microspectroscopy for the screening of cervical ThinPrep samples, it is therefore important to develop a technique compatible with the protocols used in the medical field. Thus, glass slides have been used as substrate throughout this work, and all the samples are prepared on ThinPrep glass slides for cytology. Notably, although it may be argued that the contributions of residual hemoglobin would be greatly reduced by moving to longer, nonresonant wavelengths such as 785 nm, commonly employed for Raman analysis of biological samples, it was found that the glass slides exhibited a substantial background at this wavelength. Furthermore, using the 532 nm wavelength as a source, the confocality of the Raman can be greatly improved and the contribution from the substrate significantly decreased. The Raman microscope is equipped with a 100× objective and the different structures of the cells are easily discernable.
For comparison, samples have been prepared on CaF2 using the Thinprep protocol. The weak contribution of the CaF2 in the spectra collected (FIG. 5III B) allows visualisation of the cellular spectrum without the contribution of the glass (FIG. 5III A)
5. Substrate Removal from the Data Set
FIG. 6IA presents an example of a spectrum recorded from the nucleus of a cervical cell with a strong contribution from the glass substrate as presented in FIG. 5IIIA. This corresponds to a focus on the cytoplasm rather than the nucleus. Although during recording of the data the operator was careful to select an appropriate focus to have an optimal signal to background ratio with minimum contribution from the substrate, this example was selected to demonstrate the feasibility to remove the signal of the glass from the data recorded. The main difficulty is to estimate the contribution from the glass in the spectrum recorded. The laser being focused on the cell, it will be influenced by the density of the sample and how far the laser is from the surface of the glass. Because each cell has a different morphology and a different thickness, it is considered that every spectrum will have a different contribution from the glass. It is unlikely to have a perfect mix corresponding to 50% cellular features, 50% glass features. Based on the focus used to record the cellular spectrum, it is possible to move the stage beside the cell and record a spectrum of the glass with the same Z position; however the absence of biological material in the path of the laser makes the intensity of the signal collected from the substrate not suitable for correction. Therefore, the best approach remains to collect a spectrum from the glass, with the laser focused on the glass and try to match the intensity according to the intensity of the glass features. As shown in FIG. 6IB, the intensity of the glass spectrum has been matched with the intensity of the glass feature visible between 550-620 cm−1. Due to the different band ratios, the feature around 550 cm−1 seemed to be the most suitable in order to not have a greater intensity in the spectrum of the glass than the cellular spectrum which will result in negative values. However, after subtraction, the result demonstrates the lack of accuracy of this method. The spectrum seems over corrected compared to a spectrum recorded from the CaF2 substrate (FIG. 5IIIA). The baseline appears distorted, as shown by the dotted line (FIG. 6IIA). Similar attempts can be made using the band around 1100 cm−1, but the resulting spectrum appears even more over corrected (data not shown).
Rather than trying to match the intensity of the glass signal with the intensity of the cellular spectrum, the second approach consists of reducing its intensity by a factor of 10. Thus, the intensity of the glass signal becomes small compared to the spectrum of the nucleus (FIG. 6IC). Because it is difficult to estimate how much the glass contributes in the spectra collected from the cells, it has been considered that repeated subtraction of the signal of the glass divided by 10 will allow to fine tune the correction and after a number of subtractions, x, the glass will be completely removed without any visual effect of over-correction. However, in order to estimate when the glass has been completely removed a control was needed. For this reason, three fresh samples were prepared on both glass and CaF2 windows and recorded in similar conditions to the ThinPrep samples. To evaluate the presence of glass in the spectrum the band ratio 1007/1098 cm−1 has been used. The ratio has been calculated for the entire data set recorded from the CaF2 substrates and the average value has been used as the target for the glass correction. In order to compensate for the glass signal and to ensure the glass signal is completely removed, a reference calculation may be used. For example, using an algorithm in Matlab, the signal of the glass divided by 10 has been successively subtracted from the data until the ratio of the peaks at 1007/1098 cm−1 matches the ratio obtained on the CaF2 (FIG. 6IIA). The resulting mean spectrum after glass correction and pre-processing (baseline correction and normalization) is displayed in FIG. 6III in black, compared to the mean spectrum obtained from the CaF2 substrate. After correction, the two spectra have perfectly comparable profiles and no contribution from the glass can be seen, which indicates that the removal has been successful. This method has been used for the correction of the substrate contribution in the following steps of the study.
6. Evaluation of the Sample Variability after Pre-Processing
H2O2 is a strong reactive species and has an oxidizing effect on proteins. Thus, it has been necessary to first make sure the cells were not affected by the treatment. Although the cell morphology is intact after treatment and no visual alteration of the composition can be seen (
More importantly, the variability existing between the samples originally exhibiting a pattern 1 signature and a pattern 2 signature has to be investigated after treatment with H2O2.
The variability present in patient samples can always be an issue for the potential discrimination between normal samples and pathological or abnormal samples. Although it is often admitted that more specific analytical methods are needed to extract the relevant information, it is often forgotten that the sample preparation can have a crucial place in the success of such studies. Concerning the study of Thinprep samples using Raman spectroscopy, the presence of blood can be the limiting parameter as its spectral signatures overlap the cellular features therefore making the use of the raw data impossible for diagnostic purposes. However, the present invention clearly demonstrates that rather than try to develop advanced algorithms of multivariate analytical methods to correct the data, treatment of the samples can wash off the blood residue present on the cell membrane and, together with adapted background reduction and substrate removal methods, the collection of highly reproducible data can be achieved.
The words comprises/comprising when used herein are to specify the presence of stated features, integers, steps or components but do not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof.
| Number | Date | Country | Kind |
|---|---|---|---|
| 1306494.4 | Apr 2013 | GB | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2014/055932 | 3/25/2014 | WO | 00 |
