Classification of aggressive and classic mantle cell lymphomas using synchrotron Fourier Transform Infrared microspectroscopy

Mantle cell lymphoma (MCL) is regarded as an incurable neoplasm, even to the novel drug strategies. It is known MCL has two morphological variants- classic and aggressive. Aggressive MCL is characterized by a higher mitotic index and proliferation rate, and poor overall survival in comparison to classic subtype. The insight into the detailed biochemical composition of MCL is crucial in the further development of diagnostic and treatment guidelines for MCL patients; therefore Synchrotron radiation Fourier Transform Infrared (S-FTIR) microspectroscopy combined with Principal Component Analysis (PCA) was used. The major spectral differences were observed in proteins and nucleic acids content, revealing a classification scheme of classic and aggressive MCLs. The results obtained suggest that FTIR microspectroscopy has reflected the histopathological discrimination of both MCL subtypes.

and overall survival, when compared to classic variants and according to current guidelines aggressive MCL patients are allocated into high-risk groups 2,5 .
Although the MCLs classification is well-established with histopathological assessment, the insight into the molecular/biochemical information would be invaluable for a better description of both entities and would be a fundamental proof for the validity of their classification.
Fourier Transform Infrared (FTIR) microspectroscopy offers a novel approach for the assessment of biochemical changes such as healthy and cancerous tissue differentiation and the determination of cancer subtypes without the use of any additional reagents 8,9 . Synchrotron radiation (SR) sources provide bright, broadband infrared light, enabling the analysis of micron-sized samples with higher than the conventional signal to noise than is possible with conventional IR sources 10 . The purpose of the current study was to determine differences between classic and aggressive mantle cell lymphomas using S-FTIR microspectroscopy combined with PCA analysis of the acquired spectroscopic data.

Results
Histopathological micrographs of control and malignant lymph node tissues, classic and aggressive MCL, are shown in Fig. 1.

FTIR spectral description. Representative absorbance (A) and Extended Multiplicative Scattering
Corrected second derivative (B) average spectra of the healthy control and two MCL subtypes are presented in Fig. 2. Since the visual inspection of peak positions on raw spectra was difficult, spectra were transformed into the second derivative to enhance the features of overlapping bands (Fig. 2b). The second derivative spectrum gives a negative value for every band located in the absorbance spectrum and allows for more accurate identification of individual peaks in complex spectra. The averaged spectra of every patient included in the further analysis can be found in Supplementary Figs S1-S4.
The absorbance minima determined for all tissues in the protein region are vibrations of amide I (1700-1630 cm −1 ) and amide II (1580-1500 cm −1 ) functional groups [10][11][12] . The peaks localised at 1695 cm −1 , 1682 cm −1 and 1639 cm −1 are characteristic for aggregated β-sheet, β-turn, and β-sheet structures, respectively 11,13 . Typically, the minimum found at 1655 cm −1 is attributed to α-helix structures of amide I 13 . The (N-H) bending coupled to (C-N) symmetric stretching vibrations assigned to amide II are localised at 1571 cm −1 and 1541 cm −1 12,14 . Of note is a peak found at 1515 cm −1 , typically attributed to (C-H) bending vibrations of tyrosine 15 . Of interest are absorbance intensity changes observed for malignant tissues. Both classic and aggressive MCL represents an increase in absorbance intensity noted in peak attributed to α-helix (1655 cm −1 ) structure of amide I, more pronounced in aggressive MCL (Fig. 2b). Absorbance intensity increase noticed in aggressive subtype has also been observed for minima attributed to amide I β-sheet (1639 cm −1 ), amide II (1541 cm −1 ) and tyrosine (1515 cm −1 ) (Fig. 2b). It was already reported that features associated with an aggressive clinical course of MCL included overexpression of the p53 protein 6 . This protein plays an important role in the regulatory control of the cell cycle and its mutations have been associated with the progression to more aggressive forms of the disease 6,16 .
The absorbance minima found at 1425 cm −1 and 1330 cm −1 are typically responsible for asymmetric and symmetric CH 3 and CH 2 bending vibrations of lipids and proteins 8,10 . The peak localised at 1404 cm −1 is responsible for (CH 3 ) bending vibrations of proteins 13 .
The other prominent peaks occur in the lower wavenumber region. The peak found in control tissue at 1231 cm −1 , assignable to asymmetric stretching vibrations of − PO 2 in DNA 17 is shifted towards higher wavenumber by 3 cm −1 and 7 cm −1 in classic and aggressive MCL respectively. Moreover, the absorbance intensity of this peak is increased in aggressive MCL, which coincide with available knowledge about cyclin D1 overexpression 18 . The absorbance intensity of minimum localised at 1172 cm −1 , attributed to symmetric stretching vibrations of PO 2 − in DNA 17 is slightly decreased in both MCL subtypes, indicating DNA fragmentation associated with apoptotic cell death 19 . The detailed assignments of the minima found in 2 nd derivative spectra are shown in Table 1.
Principal component analysis. The PCA results were obtained with three spectral ranges 1720-1495 cm −1 , 1440-1400 cm −1 and 1360-1160 cm −1 covering spectral features characteristic for proteins, lipids, carbohydrates and nucleic acids functional groups. Initially PCA was performed to differentiate MCL tissues from healthy control and results are presented on Fig. 3a,b. The PC loading plots show the amide band region attributable to proteins (1700-1500 cm −1 ) was heavily loaded for PC1 revealing separation of healthy control from both malignant tissues with 56% explained variance (green ellipse, Fig. 3a). Spectra of control can be distinguished from MCL tissues by having negative PC1 scores (Fig. 3a), which can be explained by strong positive loading observed at 1630 cm −1 attributed to β-sheet structures of amide I functional group (Fig. 3b). Positive loadings at 1559 cm −1 and 1541 cm −1 , attributable to amide II protein conformers, also separated healthy control cluster from malignant tissues. Moreover, both cancer tissues spectra are separated by positive PC1 scores (purple ellipse, Fig. 3a), explained by strong negative loadings indicated at 1661 cm −1 (amide I) 14 , 1566 cm −1 ((COO-) asymmetric stretching vibrations of amide II) 20 and 1551 cm −1 (amide II α-helix structures) 20 (Fig. 3b). These outcomes suggest that the amide I and II structures are most responsible for discrimination of healthy control from lymphoma tissues, which confirms the conclusions drawn from the examination of the average spectra (Fig. 2). Other differences with an impact on classification involve the negative loadings from rocking vibrations of CH 2 of distributed cis-olefins (1419 cm −1 ) and C=O stretching from polysaccharides (1195 cm −1 , 1167 cm −1 ) as well as positively loaded peak responsible for symmetric stretching vibrations of PO 2 − (1185 cm −1 ) (Fig. 3b). The PC2 loading plot reveals components mainly responsible for healthy control spectral data dispersion.
Subsequently, PCA was performed including only the lymphoma tissue spectral datasets. Results presented in Fig. 3c reveals separation of spectral clusters of classic (blue ellipse) and aggressive (pink ellipse) MCL. Classic MCL distinction is explained by negative loadings at 1661 cm −1 , 1641 cm −1 and 1530 cm −1 assignable to amide I (2019) 9:12857 | https://doi.org/10.1038/s41598-019-49326-3 www.nature.com/scientificreports www.nature.com/scientificreports/ and II protein conformers (Fig. 3d). Aggressive MCL cluster can be distinguished by positive loadings observed at 1650 cm −1 , 1624 cm −1 and 1541 cm −1 . This outcomes clearly corresponds to changes in absorbance intensity of protein moiety of described average spectra: in the aggressive subtype protein level is higher than in classic MCL (Fig. 2). Of interest is a PC2 positive loading observed at 1647 cm −1 and assigned to α-helix amide I structures. This loading seems to be responsible for distinction of spectra obtained from two patients, for whom the treatment was not successful (black ellipse, Fig. 3c).

Discussion
This research has demonstrated that the histopathological subtyping of MCL into classic and aggressive forms has its strong background in the biochemical landscape of both subtypes. It should be emphasised that this is the first study which has reported the combination of S-FTIR and PCA analysis for the assessment of MCL subtypes. We previously reported the usefulness of the presented approach in distinction of lung cancer subtypes and estimation of chemotherapy efficacy in breast cancer 8,9 . Our present results showed an absorbance increase in peaks attributed to amide I, amide II and nucleic acids noticed in both malignant tissues, much more pronounced in aggressive MCL. The shift of wavenumber was    Table 2.
The H&E slides were scanned using UltraFast Scanner (Philips IntelliSite Solution, USA) with DigiPath ™ Professional Production Software (Xerox, Norwalk, CT, USA) and representative areas of each case were selected and microtomed into 8 µm thick sections and mounted onto 1-mm-thick calcium fluoride (CaF 2 ) windows (Crystran, UK).   www.nature.com/scientificreports www.nature.com/scientificreports/ S-FTIR measurements and spectral analysis. The S-FTIR measurement was performed in transmission mode using a Bruker Vertex V80v FTIR spectrometer coupled with a Hyperion 2000 FTIR microscope (Bruker Optik GmbH, Ettlingen, Germany) equipped with a liquid nitrogen-cooled narrow-band mercury cadmium telluride (MCT) detector, at the Australian Synchrotron IR Microspectroscopy Beamline (Victoria, Australia). The spectral acquisition was performed using a 36× IR objective (NA = 0.50; Bruker Optik GmbH, Ettlingen, Germany) with the aperture size adjusted to 6.9 µm diameter beam size, and the spectra were acquired at a 4 µm step interval between pixels. The S-FTIR transmission maps were then acquired to cover an area of 200 μm × 200 μm on the MCL tissue. For each pixel, the S-FTIR spectrum was recorded within a spectral range of 3800-700 cm −1 using 4-cm −1 spectral resolution and 8 co-added scans. In all cases, Blackman-Harris 3-Term apodization, Power-Spectrum phase correction, and zero-filling factor of 2 were set as default acquisition parameters using OPUS 8.0.19 software suite (Bruker). Background spectra were collected from sample-free clean areas on the same CaF 2 substrate, following the acquisition of every 50 single spectra of the tissue, using 64 co-added scans and the same default parameters.
Before spectral pre-processing atmospheric compensation function (OPUS 8.0.19 software, Bruker) was applied to remove CO 2 and water vapour interference features. FTIR spectra embedded in acquired chemical maps were extracted and pre-processed using CytoSpec ™ version 1.4.02 (Cytospec Inc., Boston, MA, USA) as follows. Prior to Hierarchical Cluster Analysis (HCA), the spectra were quality screened based upon a minimum signal-to-noise (S/N) ratio of 100 measured over the spectral ranges of 1720-1495 cm −1 , 1440-1400 cm −1 and 1360-1160 cm −1 . Next, quality-screened spectra were pre-processed using noise-reduction algorithm, followed by second derivatization using 13-point Savitzky-Golay algorithm to eliminate the broad baseline offset and curvature and to enhance the features of hidden and overlapping bands. Subsequently, spectra were vector-normalised to account for pathlength differences between samples. HCA based on five clusters was applied on the pre-processed spectra using three spectral regions of 1720-1495 cm −1 , 1440-1400 cm −1 and 1360-1160 cm −1 to exclude the paraffin bands typically found at 2920 cm −1 , 2850 cm −1 , 1470 cm −1 and 1465 cm −1 [10]. As a result, obtained average absorbance and second derivative spectra were used for cluster selection for further analysis.
Principal component analysis (PCA) was performed using The Unscrambler ® 10.4 software package (CAMO Software AS, Oslo, Norway). Prior to PCA selected representative second derivative spectra were corrected using Extended Multiplicative Scatter Correction (EMSC) in order to correct spectral artefacts commonly found in FTIR spectra of biological samples. The PCA approach was first applied to three individual groups: healthy control, classic MCL and aggressive MCL to eliminate outliers from samples in the same group.
After the selection of representative spectra, the EMSC-normalised second derivative spectral datasets of all groups were combined into one single set. PCA was subsequently performed on the entire combined dataset to investigate similarities and differences of biochemical makeups between healthy and malignant tissues. To exclude the bands associated with paraffin, the PCA was calculated using three spectral regions of 1720-1495 cm −1 , 1440-1400 cm −1 and 1360-1160 cm −1 .