Lipid and Metabolite Profiles of Human Brain Tumors by Desorption Electrospray Ionization - Mass Spectrometry

Transcription

1 Lipid and Metabolite Profiles of Human Brain Tumors by Desorption Electrospray Ionization - Mass Spectrometry Alan K. Jarmusch, a Valentina Pirro, a Zane R. Baird, a Eyas M. Hattab, b Aaron A. Cohen-Gadol, c and R. Graham Cooks a a Department of Chemistry and Center for Analytical Instrumentation Development, Purdue University, West Lafayette, IN b Department of Pathology and Laboratory Medicine, Indiana School of Medicine, Indianapolis, IN c Department of Neurological Surgery, Indiana University School of Medicine, Indianapolis, IN SUPPLEMENTAL MATERIAL AND METHODS Specimens Cryopreserved human neurological specimens were obtained from 74 patients (58 patients of whom were included in the sample cohort) by the Biorepository of Methodist Research Institute. The samples were purchased from the Biorepository of Methodist Research Institute after Purdue IRB determined IRB review was not required (# ). Tissue specimens were used to produce sections and smears. Details on the patient cohort are reported in Table S4 - S5. The specimens were codified based on pathologic evaluation as grey matter (N=23 patients), white matter (N=14), gliomas (N=12) which consisted of 10 high-grade and 2 low-grade gliomas, meningiomas (N=13), pituitary tumor (N=14), and brain parenchyma containing reactive astrocytes (N=2). Subsequent to tissue sectioning and analysis, histopathologic evaluation determined that a number of samples contained multiple regions with different pathologies, hence the total number of patients for all classes does not equal the total number in the patient cohort. Tissue sections were made using a cryotome (Cryotome FSE, Thermo Scientific), 15 µm thickness, and thaw mounted onto glass microscope slides (Superfrost Plus, 25x75x1mm, Electron Microscopy Sciences). Subsequently, tissue smears were prepared for each specimen by removing a small piece of frozen tissue (~10-50 mm 3 ), placing the tissue onto a glass microscope slide, allowing for the tissue to come to room temperature (20-25 C), and then smeared using a custom 3D printed device (Fig S22). Tissue sections and smears were stored at -80 C prior to analysis. Materials were properly disinfected or autoclaved prior to disposal. DESI-MS Imaging The analysis of tissue sections and smears were performed on a linear ion trap mass spectrometer, model Finnigan LTQ (Thermo Electron Corporation, USA). The only modifications to the instrument were the installation of a custom DESI source, which includes a source override adapter, an external cable for the application of high voltage, and an extended ion transfer capillary. The extended ion transfer capillary measuring a total length of 180 mm was constructed from stainless steel tubing (0.02 inner diameter and 1/16 outer diameter), the length protruding from the MS vacuum system was ~87.5 mm. Connection of the capillary with the threaded MS inlet was accomplished using Swagelok stainless steel fittings machined so that the capillary can be place all the way through the fitting and secured on one side with a ferrule and metal nut. DESI-MS was performed using dimethylformamide-acetonitrile (1:1 v/v) which preserved tissue morphology for subsequent pathology.(1) Dimethylformamide and acetonitrile were purchased from Mallinckrodt Chemicals and Sigma-Aldrich, respectively. Additional source parameters are as follows: solvent flow rate, 1.0 µl min -1 ; pressure of nitrogen gas, 160 PSI; applied high voltage, 5 kv; incident angle, 52 ; spray-to-surface distance, 2-3 mm; spray-to-inlet 1

2 distance, 5-7 mm. Sections and smears were subjected to two sequential negative mode DESI- MS image acquisitions. The first image acquired data from m/z with the mass spectrometer tuned for maximum transmission of m/z 786. The moving stage was then reset to the origin position, allowing for a subsequent image acquired from m/z (MS tuned for m/z 174). Further detail of the different tune methods and corresponding instrument conditions and settings are reported in Table S6. DESI-MS imaging was performed upon tissue sections and smears by affixing the material to glass microscope slides and fixing these onto a custom two-dimensional precision moving stage. Images are collected as rows by coordinating linear motion of the moving stage with MS acquisition rate, defining resolution of 250 µm in x, upon completion of a row the moving stage resets to the original x position while stepping 250 µm in y. Upon completion of a sufficient number of y steps to cover the entire tissue surface, the moving stage was reset to the origin and a second image was acquired. The first image collected data from m/z with an MS scan time of 0.85 s (automatic gain control, AGC, was disabled), 350 ms injection time with 2 microscans. The corresponding speed of the moving stage was µm s -1. The second image acquired data from m/z with an MS scan time of 0.62 s, 125 ms injection time with 4 microscans - a faster injection time was necessary to compensate for greater MS abundance and lack of AGC. The linear motion of the moving stage was increased to µm s -1, compensating for the change in MS scan time, maintaining an average of one MS scan per 250 microns in the x dimension. DESI High Resolution MS Data Acquisition High mass resolution measurements were performed using a Thermo Exactive Orbitrap mass spectrometer (Thermo Scientific, USA). DESI parameters were as follows: solvent, dimethylformamide-acetonitrile (1:1 v/v); solvent flow rate, 1.0 µl min -1 ; pressure of nitrogen gas, 160 PSI; applied high voltage, 5 kv; incident angle, 52 ; spray-to-surface distance, 2-3 mm; spray-to-inlet distance, 5-7 mm. Data were acquired from multiple spots from representative specimens of grey and white matter, glioma, meningioma, and pituitary tumor. Theoretical mass (monoisotopic) were calculated using IsoPro 3.1 (freeware) and mass error was calculated in parts-per-million. Pathology Tissue sections and tissue smears analyzed by DESI-MS were subsequently stained by hematoxylin and eosin. Slides were chemically fixed in methanol (100%) for 2 minutes and rinsed in nanopure water (10 dips). Then slides were stained in modified Harris hematoxylin solution (Sigma-Aldrich) for 1.5 minutes, rinsed in nanopure water (10 dips), quickly dipped into Bluing Reagent (Protocol), and dipped 10 times in nanopure water. Slides were then placed in Intensified Eosin Y (Protocol) for 8 seconds and rinsed in ethanol (180 proof) in two sequential steps (10 dips each). Finally, the slides were dipped a total of 12 times into xylenes. The slides were allowed to air dry and were subsequently mounted using mounting media (Organo/Limonene Mount, Sigma-Aldrich) and glass coverslips ( C, Fisher Scientific). Blind histopathologic evaluation of tissue sections and smears provided the diagnosis (e.g. glioma), grading information (if possible), and an estimated tumor cell concentration. Some of the stained tissue sections and smears were digitally scanned for creation of publication quality figures using the Aperio whole slide digital imaging system (Leica Biosystems, CA). The system imaged all slides at 20x. The scan time ranged from 1.5 minutes to a maximum of 2.25 minutes. Aperio ImageScope v was used to view and save images of the sections and smears. DESI-MS Data Handling, Processing, and Statistics 2

3 Data Handling For each specimen, section and smear MS images were acquired using XCalibur 2.0 (.raw). The data were converted into.csv files using an in-house program, and then imported into MATLAB (MathWorks, Inc., Natick, MA, USA) to reconstruct the corresponding hyperspectral datacubes, which are composites of a spatial domain ( x and y ), and a spectral domain, comprising the m/z value and the corresponding m/z intensity. (2) For the tissue sections, each datacube was structured with pixels in x and y and the m/z domain. The m/z domain was comprised of 9600 variables for the lipid image (mass range over m/z ), and 1440 variables for the metabolite image (mass range over m/z ), both having an acquisition step of m/z The datacubes for smear MS images were compiled in the same manner. In-house MATLAB routines were used to process all MS data, as detailed below, and generate the corresponding plots, except the averaged mass spectra that were exported from MATLAB and plotted in Origin 2015 for improved graphical quality. Figures were created and exported using Adobe Illustrator Univariate Statistics The Kruskal-Wallis non-parametric hypothesis test was chosen to verify the occurrence of statistically significant differences of NAA between classes. NAA signal was normalized to the total ion current prior to Kruskal-Wallis. The null hypothesis H0 affirms that there are no significant differences between the independent populations under examination. A significant level (a twotailed P-value) of 0.05 (CI = 95%) was chosen for the statistical test. When the experimental P- value is lower than the critical P-value, the hypothesis H0 is rejected. As a post hoc test, Mann- Whitney U hypothesis test was performed. The statistical tests were performed in OriginPro 2015 using built-in statistics package upon data exported from MATLAB. OriginPro 2015 was used to generate the box and whisker plots. The median line is shown, the box corresponds to the interquartile range, whiskers represent ±1.5 standard deviation, and outliers are annotated. ROC curve analysis was performed to estimate sensitivity and specificity of NAA, determining the cut-off value (i.e. TIC normalized ion abundance, percentage) of 1.5% - corresponds to the value displayed in the y-axis of the box and whisker plot for NAA. Further, the overall accuracy (area under the curve) in discrimination of brain parenchyma and glioma was calculated. Principal Component Analysis Principal component analysis (PCA), using the nonlinear iterative partial least squares algorithm, was used to explore the DESI-MS data in an unsupervised fashion and to visualize groupings of samples based on chemical similarity. PCA was first performed on each single image (lipid and metabolite) acquired for each sample (tissue and smears) in order to carefully explore the chemical and spatial information via an interactive brushing procedure that automatically connects regions of PCA score space to pixels of the 2D DESI-MS image. (2) Note that the mass range for the lipid profile was truncated (m/z , which equals 3601 m/z variables) for statistical analysis, as it was found to contain less analytical variability, providing more consistent and clear separation with regard to disease state. Background pixels of each ion image were excluded a priori using the summation of the absolute intensity of m/z 281, 303, 788, 834, 885, present in tissue, regardless of their specific ratio within the mass spectrum. Only pixels corresponding to tissue were used for data exploration via interactive brushing and data selection. The lipid and the metabolite images were collected in such a way that they were superimposable, thus the same background pixels excluded based on the m/z previously mentioned were automatically excluded in the corresponding metabolite images. PCA was performed on MS spectra normalized by the standard normal variate (SNV) transform, 3

4 (3) so as to correct for both baseline shifts and global intensity variations, and then columncentered. No background signal correction, smoothing filters, or data binning were applied. For each tissue section, multiple regions of interest (ROI) from the MS image were selected based on histopathologic evaluation. For each ROI selection, an area of 1 mm 2 (4 pixels by 4 pixels = 16 pixels total) was averaged into one mass spectrum, for lipid and metabolite profiles. The number of such selections per tissue section varied between two and more than a dozen, according to the dimension and pathologically defined regions in the tissue. Four new training matrices were compiled possessing different numbers of objects and variables: 447 objects (samples) x 3601 variables (m/z values) and 447 x 1440, respectively for lipid and metabolite profiles of grey matter, white matter, and glioma samples; 423 x 3601 and 423 x 1440, respectively for lipid and metabolite profiles of glioma, meningioma, and pituitary tumors. We will refer to the first two matrices as brain reference training set, and to the latter two matrices as tumor reference training set. PCA was performed on the individual reference matrices, after SNV transformation and column-centering. PCA was also performed with a mid-level data fusion approach, in order to comprehensively visualize relationships among samples and variables coming from multi-block experiments (lipid and metabolite datacubes). (4) PCA acts as an unsupervised latent-variable compression technique, informative features extracted from the raw signals of each block of measurements were combined into new (reduced) single sets, merged column-wise, and then processed again by PCA. In more detail, PCA was applied separately to the lipid and metabolite reference matrices; for both, the first principal components (PCs) that accounted for about 90% of the total variance were selected (i.e. the scores of these PCs), merged column-wise together into a new dataset, and then further analyzed using PCA after column autoscaling of the new data (scores from the individual PCAs). The procedure was done separately for the brain and the tumor reference training sets. The score plots resulting from PCA using mid-level fusion to examine MS spectra from other DESI-MS images (not included in the training set) or smears and their similarity with the reference dataset. For each pixel in an image or the average MS spectrum of a smear, PC scores were computed SNV row-pretreatment and column-centering based on the parameters computed for the training dataset (lipids and metabolites considered separately) by multiplying the corrected values for the PC loadings obtained for the training dataset.(5) The computed scores were merged column-wise and PC scores for the fusion approach computed no row pretreatment and column-autoscaling based on the parameters computed for the training dataset by multiplying the corrected values for the PC loadings obtained for the training dataset (fused matrix of score values). The test sample points were plotted as black points in the PCA score plot of the training sets. For the DESI-images, the image points were plotted using a false-color scale in the corresponding 2D DESI spatial domain, in order to explore both chemical and spatial relationships. Linear Discriminant Analysis Linear discriminant analysis (LDA) was performed as a supervised discriminant classification technique on the reference dataset. Discriminant methods look for a delimiter that divides the global multidimensional data domain into a number of regions, each assigned to one of the sample classes. This delimiter identifies an open region for each class and such regions determine the assignment of the samples to one of the classes. Model validation (i.e. evaluation of the predictive ability of the model) was performed using cross-validation (CV) on the reference database built for the tissue sections (five deletion groups were used). CV prediction rates (i.e. the percentage of correct predictions on the objects in the CV evaluation sets) were computed. The CV confusion matrix shows how many samples belonging to a certain category were correctly/incorrectly assigned by the classification rule to that category. Indeed, in this matrix, each element gives the number of samples of the row category assigned to the column category. When the matrix is 4

5 diagonal (entries outside the main diagonal are all zero) there is a perfect prediction of all the samples.(6) For this study, LDA was applied after unsupervised data compression using PCA with a mid-level fusion approach, thus using only selected PCs (i.e. scores of the samples for these PCs resulting from PCA on lipid and metabolite profiles) fused together column-wise and used as new variables, instead of the original MS dataset. Two LDAs were performed on the brain and tumor reference training sets: 447 rows (samples) and four fused PCs (scores of the first two PCs resulting from PCA on the lipid and metabolite profiles merged column-wise) to discriminate between glioma, grey and white matter; 423 rows and 10 fused PCs (scores of the first five PCs resulting from PCA on the lipid and metabolite profiles merged column-wise) to discriminate between glioma, meningioma, and pituitary tumors. Details of the methodology are reported in (6). Five cross-validation deletion groups were selected, meaning that all samples were divided 5 times systematically in a training set (objects used for building the classification model) and a test set (remaining objects used to evaluate the predictive ability of the model), with all the samples being in the evaluation set only once. Eventually, the final model was built with the objects all. Canonical Correlation Analysis Canonical correlation analysis (CCA) is a way of measuring the linear relationship between two blocks of multidimensional variables observed on the same specimens.(7) In this study, the two datasets were comprised of DESI-MS spectra from tissue sections and smears of 48 specimens (only specimens with matching section and smear pathology were considered, see Table S4). Groupings of adjacent pixels, i.e. regions of interest (ROIs), representing areas of tumor or normal tissue were selected in the 2D DESI-MS spatial domain ( x by y ), according to pathological evaluation, and the corresponding mass spectra were averaged. Only smears and sections with matching pathologic evaluation were included. Regions of smeared tissue were selected based on the summation of signal as discussed previously, eliminating background pixels. The DESI mass spectra obtained from tissue sections were considered as reference spectra. CCA rotates the original variables in the two blocks, to obtain pairs of variables (one for each block), called canonical variables, that maximize correlation between the two blocks. The canonical variables are linear combinations of the original autoscaled (i.e. unitary variance) variables. The correlation coefficients between the canonical variables of the two blocks of measurements are termed canonical correlation coefficients. In order to overcome the high inter-correlations across m/z values, CCA was performed after PCA, which acts as an unsupervised data compression technique. Separate PCAs were run on the section and smear DESI mass spectral data. The first PCs, explaining about 95% of total data variation, were selected and used for CCA. CCA was performed using the free chemometric package V-PARVUS 2010 (University of Genova, Italy, Tissue Smear Device Plastic smear devices were fabricated by fused deposition modeling (FDM) on a Makerfarm i3v 3D printer outfitted with a 250 µm nozzle. To prepare for printing, a 3D model was exported as stereolithography file format (.stl), and machine toolpaths were generated using Slic3r. Each device used in this study was printed in a polyamide composite material (Taulman Bridge) onto a borosilicate glass sheet coated with a thin layer of polyvinyl alcohol. First layer temperatures for the extruder and print bed were 255 C and 75 C, respectively. For all other layers the temperatures of the extruder and bed were decreased to 250 C and 55 C, respectively. General print settings were 100 µm layer height, 60 mm s -1 maximum print speed, 3 perimeters, 4 solid top and bottom layers, and 20% hexagonal infill. Using these parameters, a single device could be manufactured in approximately minutes. A rendering of the smear device is shown in Fig S22 along with a photograph of the final 3D printed product. 5

6 SUPPLEMENTAL TABLES Table S1. Tabulated DESI high resolution mass spectral data and ion identification Grey Matter a White Matter a Glioma a Meningioma b Pituitary Tumor c Name Ion Theo Mass Mass Measured Δ Mass (ppm) Mass Measured Δ Mass (ppm) Mass Measured Δ Mass (ppm) Mass Measured Δ Mass (ppm) Mass Measured Δ Mass (ppm) Lactic acid [M-H] hydroxyglutaric acid [M-H] N-acetyl-aspartic acid [M-H] C6 Sugar [M-H] Ascorbic acid [M-H] Palmitic acid [M-H] Oleic acid [M-H] Stearic acid [M-H] Arachidonic acid [M-H] Docosahexaenoic acid [M-H] PG 34:1 [M-H] PE 38:4 [M-H] PC 32:0 [M+Cl] PS 36:1 [M-H] PC 34:1 [M+Cl] PS 40:6 [M-H] PI 38:4 [M-H] (3 -sulfo)galcer 24:1 [M-H] (3 -sulfo)galcer 24:1(OH) [M-H] a specimen 51 b specimen 17 c specimen 5 Table S2. Data Fusion PCA-LDA CV Results for Brain Parenchyma versus Glioma 1 Actual Prediction Grey Matter White Matter Glioma Grey Matter White Matter Glioma Sensitivity Specificity PCs, 5 deletion groups Table S3. Lipid Profile PCA-LDA CV Results for Tumor Type 1 Actual Pituitary Meningioma Glioma Prediction Pituitary Meningioma Glioma Sensitivity Specificity

7 1 10 PCs, 5 deletion groups Table S4. Additional specimen information Specimen Sample Type Histopathology Diagnosis Grade Density INF. % #ROI PCA CCA Section Infiltrated, White y - Section Infiltrated, Grey y - Smear Glioma High Medium Section Pituitary Low High - 7 y y Smear Pituitary Low High y Section Glioma High High - 13 y y Smear Glioma High Medium y Section Glioma High Medium - 25 y y Smear Glioma High Low y Section Pituitary Low High - 7 y y Smear Pituitary High Medium y Section White Matter y - Section Grey Matter y - Section Infiltrated, Grey y y Smear Infiltrated, Grey y Section Pituitary Low High - 17 y y Smear Pituitary Low Medium y Section Infiltrated, Grey y y Section Infiltrated, White e Smear Infiltrated, Grey y Section Glioma a,c High Medium Smear Glioma a,c High Low Section Infiltrated, Grey y - Smear Grey Matter Section Infiltrated, Grey y y Smear Infiltrated, Grey y Section Infiltrated, Grey y - Section Glioma High Medium - 18 y - Section Glioma High High Smear Glioma High Medium Section Infiltrated, White y - Section Infiltrated, Grey y y Smear Infiltrated, Grey y Section Glioma High Medium - 16 y y Smear Glioma High Low y Section Metastatic Smear Metastatic Section Pituitary c Low High - 7 y y Smear Pituitary c Low High - y Section Meningioma Low High - 13 y y Smear Meningioma Low Medium y Section Infiltrated, Grey y y Section Infiltrated, White y - Smear Infiltrated, Grey y Section Glioma Low Medium - 6 y - Smear Section Meningioma Low High Section Reactive Grey y - Section Meningioma b Low High - 8 y - Smear Grey Matter

8 Section Meningioma Low High - 15 y - Smear Meningioma f Low Medium Section Pituitary Low High - 6 y - Smear Section Meningioma Low Medium - 7 y y Section Meningioma b Low Medium Smear Meningioma Low Low y Section Glioma High High - 4 y y Section Grey Matter y - Smear Glioma High Medium y Section Infiltrated, White y - Section Infiltrated, Grey y y Smear Infiltrated, Grey y Section White Matter y - Section Grey Matter y y Smear Grey Matter y Section Pituitary Low High - 15 y y Smear Pituitary Low High y Section Pituitary Low High - 15 y y Smear Pituitary Low High y Section Pituitary Low High - 7 y y Smear Pituitary Low High y Section Infiltrated, White y - Section Infiltrated, Grey y y Smear Infiltrated, Grey y Section Meningioma Low Medium - 7 y y Smear Meningioma Low Medium y Section Pituitary Low High - 16 y y Smear Pituitary Low Low y Section Pituitary Low High - 13 y y Smear Pituitary Low Medium y Section Meningioma d Low High - 11 y y Smear Meningioma Low low y Section Infiltrated, Grey y y Section infiltrated, White y - Smear Infiltrated, Grey y Section Pituitary Low High - 12 y y Smear Pituitary Low High y Section Glioma High High - 20 y y Smear Glioma High High y Section Infiltrated, White y - Section Infiltrated, Grey y y Smear Infiltrated, Grey y Section Infiltrated, Grey y y Smear Infiltrated, Grey y Section Infiltrated, Grey y y Section Infiltrated, White - - <5 9 y - Smear Infiltrated, Grey y Section Glioma High High - 20 y y Smear Glioma High High y Section Infiltrated, Grey y - Section Infiltrated, White y - Smear Glioma High Medium Section Pituitary Low High - 12 y y Smear Pituitary Low Medium y 44 Section White Matter y y 8

9 Section Grey Matter y - Smear White Matter y Section Glioma High High - 8 y y Smear Glioma High High y Section White Matter y Section Grey Matter y y Smear Grey Matter y Section Grey Matter y y Section Reactive, Grey y - Smear Grey Matter y Section Glioma a High High Smear Glioma a High High Section Glioma Low Medium - 5 y - Section Infiltrated, Grey y - Smear Glioma Low Low Section Pituitary Low High - 15 y y Smear Pituitary Low Medium y Section Glioma High High - 8 y - Section White Matter y - Section Grey Matter y y Smear Grey Matter y Section Meningioma Low Medium - 4 y y Smear Meningioma Low Medium y Section Glioma a,c High High Smear Glioma a,c High High Section Grey Matter y y Smear Grey Matter y Section Pituitary Low High - 10 y y Smear Pituitary Low High y Section Grey Matter y y Smear Grey Matter y Section Glioma c,d High High - 15 y y Smear Glioma c,d High Medium y Section Meningioma Low Medium - 6 y - Smear Meningioma f Low Medium Section Meningioma Low High - 11 y y Smear Meningioma Low Medium y Section Meningioma Low Medium - 14 y y Smear Meningioma Low Medium y Section Glioma f High Medium Smear Normal Section Meningioma Low Medium - 7 y y Smear Meningioma Low Medium y Section Infiltrated, Grey f Smear Grey Matter Section Meningioma Low Medium - 8 y y Smear Meningioma Low Medium - - y Section Glioma Low Medium Section Infiltrated, White Smear Grey Matter Section Pituitary a Smear Pituitary a Section Meningioma f Low Low Smear Meningioma f Low Low Section Meningioma c Low Medium Smear Meningioma c Low Medium

10 Section Glioma a Smear Glioma High Low Section Glioma a High Low Section Grey Matter Smear Glioma High Low Section Glioma a High Section Grey Matter Smear Glioma High Medium Section Pituitary d Smear Pituitary Section Meningioma c Smear Meningioma f Section Pituitary c Smear Pituitary a necrosis present b cauterized areas present c hemorrhage present d calcification present e regions too small for selection f excluded to due low DESI signal g excluded from unsupervised statistics Table S5. Additional Pathologic Information on Glioma Specimens Specimen Pathologic Diagnosis WHO Grade Anaplastic oligoastrocytoma, 45 recurrent III 65 Oligoastrocytoma II 53 GBM IV 41 Oligodendroglioma II 42 Oligodendroglioma II 69 Pilocytic astrocytoma I 3 Oligodendroglioma, recurrent III 48 GBM IV 9 Oligodendroglioma II 4 Anaplastic astrocytoma III 71 Oligoastrocytoma, recurrent III (at least) 14 Oligoastrocytoma II 57 GBM IV 49 Oligoastrocytoma II 70 GBM IV 12 Anaplastic oligodendroglioma, recurrent/residual III 19 Oligoastrocytoma II 51 Anaplastic oligodendroglioma II 1 GBM, recurrent/residual IV 37 GBM IV 24 Anaplastic oligodendroglioma III 10

11 Table S6. Source Conditions, Tuned Ion Optical and Ion Detection System Settings Conditions/Setting m/z m/z Spray Voltage kv kv Capillary Voltage V V Capillary Temperature C C Tube Lens V V Multipole 00 Offset 1.00 V 2.00 V Lens V 0.00 V Multipole 0 Offset 4.75 V 4.50 V Lens V 7.00 V Gate Lens V V Multipole 1 Offset 8.00 V 8.50 V Multipole RF V p-p V p-p Front Lens 5.20 V 5.50 V Front Section Offset 9.00 V 9.00 V Center Section Offset V V Back Section Offset 7.00 V 7.00 V Back Lens 0.00 V 0.00 V Dynode kv kv Multiplier Multiplier SUPPLEMENTAL FIGURES Fig S1. DESI-MS ion images of specimen 40 with selected lipid and metabolite ions: m/z 89, 124, 174, 281, 303, 788, 834, 885, and 888. Ion images are displayed in false-color, black to white (smallest to greatest intensity), and normalized to the maximum intensity of the ion plotted in each panel. H&E stain of analyzed tissue is shown. Differences between regions associated with grey and white matter are notable, particularly the finger-like projections of white matter extending from the major area on the right side of the tissue. Note also the decrease in m/z 788 around the white matter projections which corresponds to the molecular layer of the cortex. 11

12 Fig S2. MS/MS product ion spectra for m/z 788, 794, 885, and 834 detected in the negative ionization mode. Characteristic losses used in determining the lipid class, e.g. -87 (m/z 788 -> 701, head group loss of PS) and -50 (m/z 794 -> 744, loss of methylchloride indicative of a PC chloride adduct). Further, acyl chain could be determined based on fatty acid product ions, e.g. m/z 303, arachidonic acid. Fig S3. Representative HRMS lipid profile mass spectra with phospholipid region inset of (A) grey matter, (B) white matter, and (C) glioma. Illustrative ions important in distinguishing grey matter, white matter, and glioma are tabulated in Table S1. 12

13 Fig S4. Plot of average peaks with standard deviation. (A) m/z 174, N-acetyl-aspartic acid, (B) m/z 794, PC 34:1+Cl -, (C) m/z 834, PS 18:0_22:6, and (D) m/z 888, (3 -sulfo)galcer 24:1, of grey matter (green), white matter (blue), and glioma (red). Mean denoted by solid line with ± standard deviation illustrated by the shaded area between the dotted lines. Fig S5. Chemical difference noted between neuroanatomical regions in specimen 44. (A) DESI- MS ion images of m/z 788, 834, and 885. Ion images are displayed in false-color, black to white (smallest to greatest intensity), and normalized to the maximum intensity of the ion plotted in each panel. (B) H&E stain with annotated pathology, regions of white matter are annotated by pathologist in black marker. (C) Unsupervised multivariate recognition of the molecular layer of the cortex (red), grey matter region (green), and white matter region (blue) via interactive brushing (colored image). (D) Average DESI-MS spectra obtained from interactive brushing selections. 13

14 Fig S6. Illustrative mass spectrum from specimen 38 of mixed grey and white matter composition. (A) Mass spectrum with ions relevant to distinguishing grey and white matter annotated, and corresponding to (B) the point indicated in the lipid profile PCA score plot. Fig S7. DESI-MS ion images of specimen 42, containing 40% tumor cell concentration in white and grey matter regions. H&E image of the analyzed tissue with pathologist annotated delineation of grey and white matter regions. Fig S8. Representative mass spectra of grey and white matter regions containing 40% tumor cell concentration in specimen 42. (A) Metabolite profile of white matter region and (B) metabolite profile of grey matter region. (C) Lipid profile of white matter and (D) grey matter region. 14

15 Fig S9. Representative HRMS metabolite profile mass spectra with N-acetyl-aspartic acid region (amplified by a factor of 10) of (A) grey matter, (B) white matter, and (C) glioma. Fig S10. Receiver operating characteristic (ROC) curve of NAA for brain parenchyma and glioma. The area under the curve was

16 Fig S11. (A) H&E stain of specimen 51 with illustrative regions inset. (B) Region of grey matter at 5x magnification and (C) 20x magnification. (D) White matter regions at 5x magnification and (E) 20x magnification. (F) High grade glioma region at 5x magnification and (G) 20x magnification. Fig S12. (A) DESI-MS ion images for specimen 47 and H&E image (regions of normal grey matter annotated by pathologist) illustrating the lack of change in the lipid information and change in NAA (m/z 174). Ion images are displayed in false-color, black to white (smallest to greatest intensity), and normalized to the maximum intensity of the ion plotted in each panel. (B) Lipid MS profile subjected to PCA of low-grade glioma (beige triangles) and samples with reactive astrocytes (black circles). Low-grade glioma was correctly grouped within the glioma class, but the lipid profile did not differentiate samples with reactive matter from their respective background parenchyma. (C) PCA of the metabolites MS profile indicated that the low-grade glioma falls on a spectrum between brain parenchyma and glioma (due largely to NAA). The reactive matter samples were also differentiated from brain parenchyma while appearing more like the glioma class (i.e. diseased). 16

17 Fig S13. (A) Product ion scan of m/z 147 for 2HG in specimen 51, a high-grade glioma with positive IDH mutation status confirmed by pathology. (B) Sequential product ion scan for 2HG in specimen 51. Fragmentation matches previously reported pattern.(8) Fig S14. (A) Data fusion PCA score plot using metabolite and lipid information and (B) corresponding PCA loading plot. 17

18 Fig S15. Supporting plots related to Fig 3, specimen 65. (A) Projection of individual pixels of specimen 65 (black) onto the fusion PCA score space. (B) The TIC normalized NAA abundance of grey matter, white matter, and glioma samples plotted versus tumor cell concentration. (C) Natural log of the TIC normalized NAA abundance versus tumor cell concentration with a line of regression (red line). The equation of the line was calculated as y = x with a Pearson s r of Fig S16. Average DESI mass spectra obtained from tissue sections for different tumor types. (A) Average lipid MS and (B) metabolite MS for glioma (red), pituitary tumors (purple), and meningiomas (grey). 18

19 Fig S17. Specimen 20, DESI-MS ion images for m/z 89, 174, 281, 303, 788, 834, and 885. Ion images are displayed in false-color, black to white (smallest to greatest intensity), and normalized to the maximum intensity of the ion plotted in each panel. H&E shows a region of cauterized tissue on the left and invasive meningioma on right with adjacent grey matter in between. Inset depicts the boundary between grey matter and meningioma. 19

20 Fig S18. Examples of tissue smears analyzed and subsequently H&E stained: (A) specimen 51, (B) specimen 24, (C) specimen 20, (D) specimen 18, and (E) specimen

21 Fig S19. (A) Average DESI-MS lipid and (B) metabolite profile acquired from smears for grey matter (green), white matter (blue), pituitary (purple), meningioma (grey), and glioma (red). 21

22 Fig S20. Box and whisker plot for NAA from the DESI-MS analysis of tissue smears. The box represents the interquartile range with median line and whiskers ±1.5 standard deviation with outliers represented by closed diamonds. Fig S21. Correlation plot resulting from CCA performed between tissue sections and tissue smears upon (A) the metabolite MS profile and (B) the lipid MS profile. Fig S22. (A) Rendering of custom 3D printed smear device and (B) photograph of smear device use resulting in tissue smear. 22

23 SUPPLEMENTAL REFERENCES 1. Eberlin LS, et al. (2011) Nondestructive, histologically compatible tissue imaging by desorption electrospray ionization mass spectrometry. ChemBioChem 12(14): Pirro V, Eberlin LS, Oliveri P, & Cooks RG (2012) Interactive hyperspectral approach for exploring and interpreting DESI-MS images of cancerous and normal tissue sections. Analyst 137(10): Fearn T (2009) The effect of spectral pre-treatments on interpretation. NIR news 20(6): Pirro V, et al. (2014) Lipid characterization of individual porcine oocytes by dual mode DESI-MS and data fusion. Analytica Chimica Acta 848: Bagnasco L, Zotti M, Sitta N, & Oliveri P (2015) A PCA-based hyperspectral approach to detect infections by mycophilic fungi on dried porcini mushrooms (boletus edulis and allied species). Talanta 144: González-Serrano AF, et al. (2013) Desorption electrospray ionization mass spectrometry reveals lipid metabolism of individual oocytes and embryos. PloS one 8(9):e Doeswijk T, et al. (2011) Canonical correlation analysis of multiple sensory directed metabolomics data blocks reveals corresponding parts between data blocks. Chemometrics and Intelligent Laboratory Systems 107(2): Santagata S, et al. (2014) Intraoperative mass spectrometry mapping of an oncometabolite to guide brain tumor surgery. Proceedings of the National Academy of Sciences 111(30):