Assessment of paraffin removal from prostate FFPE sections using transmission mode FTIR-FPA imaging†

Manchester Institute of Biotechnology, Univ Manchester, M1 7DN, UK Genito Urinary Cancer Research Group, Institute for Cancer Research, The Univers Health Science Centre, The Christie NHS Fou School of Chemical Engineering and Analyti Oxford Road, Manchester, M13 9PL, UK Department of Urology, The Christie NHS F UK Department of Urology, Salford Royal Hos Salford, M6 8HD, UK † Electronic supplementary informa 10.1039/c3ay41308j Cite this: DOI: 10.1039/c3ay41308j


Introduction
In standard histopathology, a wide variety of stains are utilised to aid disease-based detection during inspection of tissue morphology at a microscopic level. Spectral pathology, however, is a complementary, stain-free approach. Infrared spectroscopy is just one application of spectral pathology that can be used to probe the global biochemistry of the sample. When an infrared spectrometer is coupled to a microscope equipped with an appropriate detector, it is also possible to achieve spatiallyresolved biochemical information.
Histology sample preparation for infrared spectroscopy is minimal and straight-forward. A tissue section, cut to a thickness of usually $4-8 mm using a microtome, is oated onto an appropriate transparent substrate, such as calcium uoride or barium uoride. The sample can then be probed by spectroscopy. Choosing the method of initial sample preparation, however, is an issue. Ultimately, the more sample preparation that is performed before the sample is analysed, the less biologically relevant the biochemical signatures become.
Freezing tissue, for example, preserves the biochemistry without the use of solvents. The freezing processes however can cause damage; If the freezing process is too slow then ice crystals can form, puncturing cell membranes. 1,2 Additionally, the thawing processes in frozen tissue samples will cause a time-dependent sample degradation gradient. There are, however, a number good recommendations to avoid or inhibit these processes, as described by Stitt et al. 3 Sometimes it is not possible to control the sample processing step. Retrospective investigation of diseased tissue oen requires archived tissue from Bio-Banks where the favoured protocol may be formalin-xed paraffin embedding (FFPE).
Fresh tissue is rstly immersed in an aqueous formalin solution, which acts as the xative. The tissue is then dehydrated in washes of xylene and ethanol. Finally, an embedding medium of paraffin wax is added to the tissue, creating a protective barrier.
When the tissues are ready for spectroscopic analysis, efficient removal of paraffin wax is desired in order to probe the full biochemical range available. Digital dewaxing, 4 i.e. measuring the sample without physical dewaxing followed by digital removal of the paraffin absorption bands, is a possible option, but this is a further complication to spectral processing and it is yet to be seen how reliable the recovered lipid signals are aer such a process. Doubt still remains, however, over the physical removal of paraffin during the dewaxing process and to what effect different solvents, such as hexane (C 6 H 14 ) or xylene (C 8 H 10 ), may contribute. 5,6 A common worry is that by removing paraffin wax with hydrocarbon solvents we also run the risk of removing increasing proportions of the native residual hydrocarbons from the tissue. At the primary washing stage (before tissue is paraffin-embedded) unsaturated inherent hydrocarbons such as free unbound, free fatty acids and triglycerides are inevitably leached from the tissue; the presence of these lipid types can only be inferred by their absence in the form of unstained 'holes' that are le behind during staining. 7,8 To remove paraffin from the tissue prior to analysis, the same solvents are used to remove the large hydrocarbon chains of unbound paraffin wax. At this point any free unbound tissue lipids have already been removed from the tissue prior to paraffin embedding. The only tissue-related hydrocarbons that remain are the solvent-resistant type. There are a large variety of solventresistant lipids. 9 Those that are most resistant are the lipid components of complexes which are bound to tissue proteins during the process of formalin xation, methylene bridges form between proteins. It is known that lipids, nucleic acids and carbohydrates can be trapped in a matrix of these insoluble and cross-linked proteins. 10 Partial binding of these lipids enables them to be xed sufficiently to resist routine processing and embedding in paraffin wax. 11 This predominantly describes the plasma membrane which is a continuous double-layer of phospholipids, interweaved with cholesterol and proteins. The plasma membranes also constitute the structure of the nuclear envelope. 12 Myelin and lipofuscin are good examples whereby lipid-relaxed components can be stained in FFPE sections. 13,14 The plasma membrane also contains many lipid ras. These ras are known to exhibit high concentrations of sphingolipids, a type of phospholipid with long-chain saturated fatty acid tails. 15 The exact nature of the bound lipids present aer the formalin xation process has yet to be determined. Yet if these solvent-resistant, bound hydrocarbons were not removed in the solvent washing process prior to the addition of paraffin, it seems difficult to imagine by what chemical mechanism they would be leached during paraffin removal, under the same solvent conditions.
In light of this fact, the following study was performed to explore the dewaxing process issue in the context of prostate cancer tissue sections by systematic acquisition of spatially resolved infrared images by using a focal plane array (FPA) detector. By this approach, spatial information is retained and associated with the spectra throughout the classication process, presenting a more reliable way to compare tissue sections that are oen heterogeneous.

Reagents and materials
Paraffin wax embedded prostate cancer tissue blocks were collected under ethical approval (Trent MREC 01/4/061). The standard histological sample collection protocol involved chemical xation of the tissue sample in 4% formalin for 24 hours. A Thermo Shandon Excelsior Processor was then used to pass varying concentrations of ethanol (20%, 90% and four aliquots of 100%) for an hour each through the tissue sample. This was followed by three, 1 hour xylene washes at 40 C and three 1 hour washes with Paraplast at 62 C. The tissue sample was then immersed in molten paraffin wax and cooled using an ice plate for 1 hour before the embedded tissue was storageready. 16

Sample preparation
Three 4 mm serial sections were cut and two of the sections were oated onto calcium uoride slides (1.2 mm thick, 12 mm diameter). The third section was placed on a standard histology glass slide. The glass slide specimen was then dewaxed under standard local histological protocol (5 minutes submersions in 3 different glass troughs lled with xylene). This section was then stained with Haematoxylin and Eosin (ESI, Fig. S1 †).

Solvent immersion protocol
The two sections on the infrared substrates were each designated for either xylene or hexane dewaxing solvents. The solvent selection would remain constant for the specic section throughout the experiment. Both solvents were ordered from Sigma-Aldrich Co Limited (hexane, (reagent-plus >99%) Cat. 139386 and Xylene, (histological grade) Cat. 534056). Due to the toxic nature and volatility of these solvents, all dewaxing activity was carried out in a fume hood.
The protocol was kept consistent for each time point. Tissue samples were dipped 10 times (in a 20 mL beaker) holding 3 mL of the respective solvent. Aer the 10th dip, the sample was then le immersed within the solvent for the specied amount of time. The sample was removed and dipped 10 times (in a 20 mL beaker) holding 3 mL of ethanol and then le to air dry at room temperature for 30 minutes. Aer each procedure, the infrared image was acquired. For time point one, 'T1', the tissue was immersed in solvent for 5 minutes (Fig. 1, ESI, Fig. S2 †). Following T1, the immersion time was increased by varying intervals up to an immersion of 24 hours (T6). Immersions times greater than 15 minutes required the beaker containing the solvent and sample to be covered with Paralm™ to prevent any of the solvent from evaporating off prematurely. The varying times of immersion in the solvent and the notation conventions are shown in Table 1.

Data acquisition and pre-processing
Transmission mode FTIR imaging spectroscopy was performed on a Varian 670-IR spectrometer coupled with a Varian 620-IR imaging microscope (Agilent Technologies, CA) equipped with a 128 Â 128 pixel liquid nitrogen-cooled Mercury-Cadmium-Telluride (MCT) focal plane array detector. Spectra were collected in the 1000-3800 cm À1 range, at a spectral resolution of 4 cm À1 , with the co-addition of 128 scans for sample spectra, and 256 scans for the background spectra. At a pixel resolution of $5.5 mm, a tissue sampling area of $700 Â 700 mm was captured per hyperspectral image. Each image was subject to RMieS-EMSC correction with 20 iterations using a Matrigel™ spectrum as the initial reference point. 17,18 The infrared hyperspectral images were pre-processed according to the protocol reported in Hughes et al. 19 Briey, images were qualitycontrolled to remove any noise-intensive signals related to non-tissue related artefacts (Fig. 2a). Remaining spectra were transformed to the rst derivative before PCA was performed for each hyperspectral image, retaining the scores for 10 PCs for   dimension reduction, corresponding to 95% explained dataset variance (Fig. 2b).

Data analysis
K-means cluster analysis was performed (with three replicates to avoid convergence of local minima) on the principal component scores using a cosine distance metric to dene the image by two partitioned clusters (Fig. 2c). Subsequent pseudo-colour cluster maps were constructed, assigning a specic colour to each spectral cluster and compared to the respective H&E section for denitive histopathological evaluation ( Fig. 1 and 3). The two classes (class 1, blue and class 2, red) describe different tissue structures, such as stroma (class 1) and epithelium (class 2).
Using the dened cluster classes as a mask, the corresponding original spectra were extracted (Fig. 2d). Finally, the extent of methylene bond vibrations were analysed by the calculation of two ratios (Fig. 4). These ratios were quantied in each IR image by the proportion of n as (CH 2 ) methylene stretching ($2917 cm À1 ) and d(CH 2 )methylene bending ($1462 cm À1 ) peak heights, normalised over n s (OH) ($3289 cm À1 ) or n as (PO)($1236 cm À1 ) peak heights (eqn (1) and (2)).

Results and discussion
The pseudo-colour classication images shown in Fig. 3 for the tissue sections immersed in either solvent for varying time periods show virtually no change as a function of time. This is important since it implies that there was no substantial difference in the wax retention from different types of tissue as classication of tissue type remained stable. The spectral signal of paraffin includes high absorptions in a number of regions of the wavenumber mid-range (ESI, Fig. S3 †). If residual paraffin had been present, large differences in spectral intensities would have obscured tissue classication (ESI, Fig. 4 †). This may have caused class 2 spectra to be classed as class 1 spectra, for  The molecular vibrations associated with paraffin can never be unique and associated to exclusively in either Raman and IR spectra as the structural bonding present is also present in abundance in biological molecules. There is a view, however, that FTIR spectroscopy is contextually less sensitive than Raman spectroscopy for the analysis of paraffin signal contamination. 20 The basis for this viewpoint stems from comparative Raman and IR studies where investigations have been limited to the ngerprint region of the IR spectrum (800-1800 cm À1 ). 21 There are also comparatively more assignable (sharp) peaks in the Raman spectrum of paraffin. 22 If a fullranged mid-IR spectrum of paraffin is considered, however, there are dramatically sharp methyl and methylene-related stretching mode peaks with intensive absorbance values that extend beyond the scale of typically-prepared biological tissue (ESI, Fig. S3 and S4 †). Fig. 5a displays an unprocessed single pixel spectrum from an area of tissue Sample 2 in paraffin, prior to dewaxing. The paraffin ranged in thickness across the sample to the point of extreme saturation (ESI, Fig S4 †). If the paraffindominated peaks present at $1462, 2846 and 2954 cm À1 with absorbance values of 0.38, 1.43 and 0.54 are considered, the peaks at 2846 cm À1 and 2954 cm À1 are approximately Â 3.7 and Â 1.42 larger than the peak at 1462 cm À1 . Fig. 5b displays the mean tissue spectra for Sample 2 at T7 and aer spectral processing. If, hypothetically, it was suspected that the spectrum was paraffin-contaminated due to observance of the peak at 1462 cm À1 (0.1 absorbance) then the values at 2846 cm À1 and 2954 cm À1 (0.13 and 0.12 respectively) should be closer to $0.37 and 0.14, to be consistent with the approximate ratios of the paraffin, especially in terms of the proportionality of methylene stretching to bending vibrations. Thus, any major issue with residual paraffin in the tissue spectrum should be evident.
Infrared FPA imaging enables the collection of spatiallyresolved spectra. 23 Not only can different tissue structures be identied by their infrared signal, large quantities of identied tissue class spectra can be extracted and used for subsequent evaluation. This is a major benet, over single-point infrared spectroscopy, for example. For each tissue section and each time point, over 100 000 spectra were analysed by calculating the peak height ratios described in Fig. 3 (Tables 2 and 3 contain the detail). In the order of $17 000 spectra were analysed per ratio. The ratios provided a normalised assessment of the proportion of hydrocarbon vibrations across the two tissue classes aer increasing solvent immersion times.
If the solvent-action against paraffin wax was not instantaneous and ultimately stubborn to fully remove, one would expect to observe a clear, decreasing trend in the intensity of methylene bond vibrations versus increasing time. In reality, the actual intensity values remained fairly consistent, suggesting that the vibrations have reached a baseline level (Fig. 6). The overall %RSD values across each of the 6 time point ratios are summarised in Table 4. Despite the tissue sections being serial with slightly different images and treated with different   solvents, the %RSD of peak height ratios in each class are surprisingly similar.
In both serial sections, methylene vibrations were consistently more abundant in class 1 tissue spectra. For this reason, mean spectra per time point were calculated to give the best chance of a qualitative assessment of any possible observable differences (Fig. 7). The mean spectral signals of the xylenewashed tissue specimen appeared consistent. This could suggest that to a detectable limit, all of the paraffin had been removed aer the rst time point. In the hexane specimen, only slight observable differences were noted between time points T1 and T2 (5 and 10 minutes solvent immersion). This can be seen by the difference in peak heights between n as (CH 2 ) ($2917 cm À1 ) and d s (OH) ($3289 cm À1 ) visually expressed by the gradient between them, as well as a slight change in the shape of the peak shoulder of d(CH 2 ) at $1462 cm À1 . This may be a case of detecting residual paraffin at a relatively low percentage, where differences in the smaller absorbing band at 1462 cm À1 are more difficult to detect due to the height of the baseline, rather than the hypothetical case shown in Fig. 5b. Nevertheless, it seems that aer 10 minutes even slight changes were not detectable, indicating a steady-state in accordance to that of chemically bound tissue.
Within the limitations of this study, it may be possible that a slightly longer immersion time (approximately 5 minutes) is required with hexane to remove the paraffin by comparison with xylene. Typically the paraffin from FFPE tissue sections is removed by solvents such as xylene or hexane. Previous studies 6 have reported to monitor paraffin removal by the disappearance of the 'paraffin' band at 1462 cm À1 (ESI, Fig. 3 †). This, however, is not a peak that is unique to paraffin as it is a bending (scissoring) vibration of methylene (CH 2 ). Biological molecules themselves will have abundant naturallyoccurring methylene bending vibrations of similar strength. 24-26 A more contextual example can be seen in the spectra of human prostate cells that have been formalin-xed (and not treated with paraffin) that evidentially display the distinct d(CH 2 ) 1462 peak shoulder. 27 Analysis of the dewaxing efficiency of xylene with Raman spectroscopy has been reported, but with conicting evidence. Previously, xylene has been shown to be less effective than hexane, but only by observation of selective paraffin-associated peaks. 28 One would expect to see a decreasing trend in all paraffin-associated peaks to interpret the result as a single consequence of paraffin removal. If this is not the case, it could be reasoned that there are other contributing factors involved. Changes in paraffin temperature, for example, can illicit Table 3 Summarised values for the peak height ratio calculations and subsequent barchart plots (Manuscript Fig. 5b) Tables 2 and 3.The overall %RSD of peak height ratios per class are shown in Table 4.  conformational structural modications which can in turn cause differences in band intensity. 29 More recently, however, dewaxing of a 20 mm tissue section with xylene was reported to be effective when analysed by Raman spectroscopy, 5 which is supportive of the observations found in this study. A potential indication for the discrepancy between studies may be that the efficacy of a dewaxing protocol may differ with various tissue types. While this cannot be ruled out, the ndings of this IR study of prostatic tissue that compared the results of two subtypes from a single organ (epithelial and stromal) tissue, indicated no difference in dewaxing efficacy. The only observably difference was inherent abundance of bound lipids, which remained consistent throughout the study. In the case of previous Raman studies, other possible reasons for the discrepancy could be due to differences in the protocol execution, or the fact that without spatially-resolved data analysis it is difficult to monitor the exact position of single-point data acquisition between tissue sample manipulation steps. Most recently, a comprehensive single-point Raman study has investigated tissue of different organ type and tumourrelated progression in terms of Histoclear dewaxing efficiency using a number of different substrates. 20 Interestingly, there is convincing evidence to suggest the choice of substrate may play a role in poor dewaxing which is something to equally consider for infrared experiments. Signicant lipid-associated spectral differences were also found between normal and metastatic brain tumours. This evidence suggests that there are added complications when comparing tissue different sites of the body and one would have to carefully consider the differences in residual paraffin versus the differences related to inherent tumour biology. For example, progression of malignant brain tumours is dependent upon vascularity and is associated with altered ganglioside composition and distribution 30 (gangliosides are cell surface-enriched glyco-sphingolipids) which would also cause inherent differences in lipid contributions.

Conclusions
In conclusion, we make the suggestion that paraffin removal by hydrocarbon solvent washing removes only those hydrocarbons present in free, unbound paraffin. The analysis of biologically native methylene vibrations should not be treated with suspicion, provided that the protocol provides evidence that paraffin is removed to the point that the methylene vibrations are at a stable baseline. A further limitation is that biochemical interpretation from such vibrations should only allude to chemically bound-lipid-complex associations. Frozen sections, however, remain the primary choice of sample preparation in order to probe the full complement of lipid structures within biological tissue.
In the connes of this experiment, we found no major difference between the use of the aliphatic hydrocarbon hexane versus the aromatic hydrocarbon solvent xylene. Although both solvents are particularly harmful in a health and safety context, hexane is more ammable than xylene (ash points of solvents stated in Section 2.3 are cited from their respective MSDS as À26 C versus 25 C). Therefore as a recommendation, we suggest that FFPE tissue should be de-waxed for a minimum of 5-10 minutes with xylene as it an industry standard for tissue processing. It may be advisable to adopt a more cautious protocol; however, that given there is no evidence of tissue damage as a function of immersion time. Such a protocol could include several 5-10 minutes immersions in fresh solvent.