Imaging Mass Spectrometry–Based Lipidomic Approach to Classification of Architectural Features in Nevi

Abstract

Cutaneous melanocytic lesions represent a collection of heterogeneous entities, ranging from completely benign melanocytic nevi to malignant melanomas, the correct diagnosis of which is sometimes challenging (Brenn, 2012Brenn T. Pitfalls in the evaluation of melanocytic lesions.Histopathology. 2012; 60: 690-705Crossref PubMed Scopus (22) Google Scholar, Hawryluk et al., 2012Hawryluk E.B. Sober A.J. Piris A. Nazarian R.M. Hoang M.P. Tsao H. et al.Histologically challenging melanocytic tumors referred to a tertiary care pigmented lesion clinic.J Am Acad Dermatol. 2012; 67: 727-735Abstract Full Text Full Text PDF PubMed Scopus (34) Google Scholar, Nobre et al., 2013Nobre A.B. Pineiro-Maceira J. Luiz R.R. Analysis of interobserver reproducibility in grading histological patterns of dysplastic nevi.An Bras Dermatol. 2013; 88: 23-31Crossref PubMed Scopus (13) Google Scholar). Here, we explore the use of matrix-assisted laser desorption/ionization imaging mass spectrometry (IMS) (MALDI-IMS) to monitor lipid disease-specific alterations in situ, linking histology with spatial distribution of molecules in tissues (Fernández et al., 2011Fernández J.A. Ochoa B. Fresnedo O. Giralt M.T. Rodríguez-Puertas R. Matrix-assisted laser desorption ionization imaging mass spectrometry in lipidomics.Anal Bioanal Chem. 2011; 401: 29-51Crossref PubMed Scopus (71) Google Scholar, Gessel et al., 2014Gessel M.M. Norris J.L. Caprioli R.M. MALDI imaging mass spectrometry: spatial molecular analysis to enable a new age of discovery.J Proteom. 2014; 107: 71-82Crossref PubMed Scopus (202) Google Scholar, Gustafsson et al., 2011Gustafsson J.O.R. Oehler M.K. Ruszkiewicz A. McColl S.R. Hoffmann P. MALDI imaging mass spectrometry (MALDI-IMS)-application of spatial proteomics for ovarian cancer classification and diagnosis.Int J Mol Sci. 2011; 12: 773-794Crossref PubMed Scopus (80) Google Scholar, Stoeckli et al., 2001Stoeckli M. Chaurand P. Hallahan D.E. Caprioli R.M. Imaging mass spectrometry: a new technology for the analysis of protein expression in mammalian tissues.Nat Med. 2001; 7: 493-496Crossref PubMed Scopus (1005) Google Scholar, Walch et al., 2008Walch A. Rauser S. Deininger S.O. Hüfler H. MALDI imaging mass spectrometry for direct tissue analysis: a new frontier for molecular histology.Histochem Cell Biol. 2008; 130: 421-434Crossref PubMed Scopus (255) Google Scholar). Lipids play very specialized roles in tissues and cells in general, and in the skin in particular (Feingold, 2009Feingold K.R. The outer frontier: the importance of lipid metabolism in the skin.J Lipid Res. 2009; 50: S417-S422Crossref PubMed Scopus (123) Google Scholar, Hart et al., 2011Hart P.J. Francese S. Claude E. Woodroofe M.N. Clench M.R. MALDI-MS imaging of lipids in ex vivo human skin.Anal Bioanal Chem. 2011; 401: 115-125Crossref PubMed Scopus (78) Google Scholar). We previously demonstrated that the expression of lipids is tightly regulated in the colon crypts, changing as colonocytes differentiate (Bestard-Escalas et al., 2016Bestard-Escalas J. Garate J. Maimó-Barceló A. Fernández R. Lopez D.H. Lage S. et al.Lipid fingerprint image accurately conveys human colon cell pathophysiologic state: a solid candidate as biomarker.Biochim Biophys Acta. 2016; 186: 1942-1950Google Scholar, López et al., 2018López D.H. Bestard-Escalas J. Garate J. Maimó-Barceló A. Fernández R. Reigada R. et al.Tissue-selective alteration of ethanolamine plasmalogen metabolism in dedifferentiated colon mucosa.Biochim Biophys Acta. 2018; 1863: 928-938Crossref PubMed Scopus (22) Google Scholar). Thus, matrix-assisted laser desorption/ionization IMS can detect perturbations in maturation, even at very early stages, enabling early detection of a neoplasia. Likewise, imaging the distribution of lipids may be also suitable to aid the early diagnosis of skin lesions. The procedure for mapping lipids in nevi sections starts with a cryosection of freshly frozen tissue. The sample is covered with an appropriate matrix that helps desorbing the lipids (Supplementary Figure S1 online and Supplementary Materials and Methods online) and then, is explored following an array of spatial coordinates, recording a mass spectrum at each of them. The species (lipids) appear as sharp peaks, which are afterwards integrated and represented against the acquisition coordinates using a false color scale, to obtain the final maps. Thus, MALDI-IMS is, by definition, a digital technique that facilitates further statistical analysis of each sample, or even of multiple samples. Figure 1a shows a comparison between the distributions of three representative lipids along a section of a nevus biopsy. The images clearly show that while phosphatidylinositol 38:5 is preferentially detected in melanocytes, phosphatidylinositol 34:1 is more abundant in epidermis, and sphingomyelin d42:3 is almost exclusively found in the dermis. These maps, together with those presented in Supplementary Figure S2 online, demonstrate that each lipid species presents a characteristic distribution. Accordingly, every single lipid mapped could be considered as a specific dye in a digital molecular staining experiment. The data generated in this way can be analyzed by applying a classification algorithm, such as K-means, to group the pixels according to the similarity of their lipid fingerprints. In the example shown in Figure 1a, the algorithm formed four groups that faithfully represent the architecture of the tissue. Comparison between the optical images of six additional sections from benign intradermal melanocytic nevi from different individuals (Supplementary Table S1 online) and the corresponding IMS experiments is shown in Figure 1b–1g. The optical images show melanocytes in the dermis separated from the epidermis by a connective zone and no mitotic activity is appreciated. Despite the morphological variations between the samples, similar results were obtained in the IMS experiment. In some sections (Figure 1d, 1f, 1g), collagen-rich areas appear as a new cluster (yellow) and there is an area in one of the samples (light blue in Figure 1d) that corresponds to a nest of superficially located melanocytes. Interestingly, the clustering algorithm classified melanocytes into two subpopulations (bright green- and dark green-colored areas) in two of the sections (Figure 1b, 1c), indicating significant, though modest, anatomical differences in lipids. Whether such differences may correspond to melanocytes with a different degree of maturation warrants deeper analysis. Because melanocytes “mature” with depth in benign but not in malignant lesions, a metabolic profile study of melanocyte clusters would be of great interest for the diagnosis. To demonstrate that each of the histologic regions identified has a unique lipid fingerprint that does not differ significantly between individuals (n = 7), we carried out a principal components analysis using the average fingerprint over the regions identified (Supplementary Figure S3 online). Then, to identify the lipid species that mark the differences among epidermis, dermis, and melanocytes, we performed one-way analysis of variance with Bonferroni/Games-Howell post hoc correction (a list can be found in Supplementary Table S2 online). The results are represented using a hierarchical dendrogram in Figure 2a and as a heat map in Supplementary Figure S4 online. Clear differences in some lipid species were observed. It is remarkable the abundance of poorly unsaturated sphingomyelin species particularly in dermis, highly unsaturated phosphatidylethanolamine and phosphatidylethanolamine plasmalogens (phosphatidylethanolamine ether [alkyl-acyl]/ phosphatidylethanolamine ether [alkenyl-acyl]) in melanocytes and mono- and di-unsaturated molecular species of phosphatidylinositol in epidermis. To validate the characteristic lipid fingerprints found, we have analyzed a second set of samples, the validation group (Supplementary Table S3 online and Supplementary Figure S5 online, n = 7). Principal component analysis illustrates in Figure 2b that the three histologic areas are consistently defined in this validation group by using the biomarker candidates shown in Supplementary Table S2. We employed logistic regression to build a classifier model with the discovery data and then test it using the validation data set. The classifier model yielded a specificity of 89.0% and a sensitivity of 82.9% (Figure 2c, 2d). The ensuing conclusion is that IMS enables a tissue to be identified based on its molecular (lipid) composition in a robust, systematic, and semi-automated way. Each histologic region presents a unique lipid profile that is consistent between individuals, presumably because it is related both to the functions these lipids play in each cell type and to their particular environment. This study therefore opens the door for MALDI-IMS to be used to diagnose skin diseases. Some of the data obtained suggest that some changes in the lipid fingerprint of melanocytes may be related to their degree of maturation, which might be particularly useful for the early detection of malignant alterations that might precede the appearance of a melanoma. However, further studies will be necessary to test this possibility. One of the most appealing characteristics of this application is that tissues are classified based on their lipid composition. Thus, the technique offers the chance to detect not only early stages of melanoma, but also the metabolic changes that are produced in connection with the malignancy transformation, helping us to understand the metabolic bases of this deadly disease. This study was approved by the Euskadi Ethics Committee (OncoImage, 14-10) and written informed consents were obtained from all the subjects. Collection of nevi and melanoma frozen samples is recorded at the Basque Biobank (https://www.biobancovasco.org/). Jone Garate: http://orcid.org/0000-0003-2258-7116 Sergio Lage: http://orcid.org/0000-0002-7029-5428 Roberto Fernández: http://orcid.org/0000-0002-3424-9758 Verónica Velasco: http://orcid.org/0000-0003-3759-2639 Beatriz Abad: http://orcid.org/0000-0001-6945-5730 Aintzane Asumendi: http://orcid.org/0000-0002-2156-2323 Jesús Gardeazabal: http://orcid.org/0000-0001-5938-0835 Yoana Arroyo-Berdugo: http://orcid.org/0000-0003-3268-1616 María Ángeles Rodríguez: http://orcid.org/0000-0001-7746-3044 Juan Luis Artola: http://orcid.org/0000-0003-2125-7684 Iganacio Zabalza: http://orcid.org/0000-0001-7892-4447 Begoña Ochoa: http://orcid.org/0000-0002-1580-7160 José Andrés Fernández: http://orcid.org/0000-0002-7315-2326 María Dolores Boyano: http://orcid.org/0000-0001-6051-1054 The authors state no conflict of interest. We are grateful to the Basque Biobank for the collection, of tissue samples, and to SGIker (Universidad del País Vasco, Euskal Herriko Unibertsitatea [UPV/EHU], Ministerio de Ciencia e Innovación [MICINN], Gobierno Vasco [GV/EJ], European Science Foundation [ESF]) for technical and human support in ultra-high pressure liquid chromatography-mass spectrometry (UHPLC-MSE) analysis. We are most grateful to Anna Crespo (Basque Biobank) for her technical support on histologic sections and to David Fernández and Cristian Mankoc (Noray Bioinformatics, SLU, Bizkaia, Spain) for their help with the statistical analysis. This work was supported by grants from the Basque Government (KK2016-036 and KK2017-041 to MDB, and IT971-16 to BO) and UPV/EHU (GIU17/066 to MDB). JG holds a predoctoral fellowship from the UPV/EHU. Conceptualization: MDB, JAF, BO; Data Curation: JG, SL, RF; Formal Analysis: JG, SL, RF, YA-B, BA, AA; Funding Acquisition: MDB, JAF, BO; Investigation: JG, SL, RF; Methodology: JG, SL, RF, YA-B, BA, AA; Project Administration: MDB; Resources: VV, JG, MAR, JLA, IZ; Supervision: MDB, JAF, BO; Validation: JG, SL, RF; Visualization: JG, SL, RF; Writing - Original Draft Preparation: MDB, JAF, BO; Writing - Review and Editing: MDB, JAF, BO Fourteen patients were enrolled in this study. The discovery group was composed by seven samples from seven different patients: two males and five females; their ages ranging from 28 to 63 years (Supplementary Table S1). “Cloth discomfort” was the main reason for consultation and they all gave their written informed consent to participate in the study. The validation group was composed of 10 samples from seven different patients (Supplementary Table S3). The study protocol conformed to the tenets of the Helsinki Declaration (Version Brazil, 2013) and it was approved by the Euskadi Ethics Committee (OncoImage, 14-10: see workflow in Supplementary Figure S1). Biopsies were obtained at the Department of Dermatology at Cruces University Hospital (discovery group) or Galdakao-Usansolo Hospital (validation group) using surgical scalpel blades (Swann-Morton Limited, Sheffield, UK) and immediately after resection, the biopsies were snap-frozen in liquid nitrogen and stored at –80ºC at the Basque Biobank for Research (www.biobancovasco.org). Cryosections (∼10 μm thick: ThermoFisher, San Jose, CA) were prepared in the absence of cryoprotectant or embedding material at –20°C and recovered on plain glass microscope slides. All skin lesions were histologically diagnosed as intradermal nevi. Staining with hematoxylin and eosin (Sigma Aldrich Química, Madrid, Spain) was carried out for all biopsies once MALDI-IMS experiments were completed and matrix removed (Olympus BX51). A detailed description of the imaging mass spectrometry data acquisition and analysis can be found elsewhere (Garate et al., 2015Garate J. Fernández R. Lage S. et al.Imaging mass spectrometry at increased resolution using 2-mercaptobenzothiazole and 2,5-diaminonaphtalene matrices: application to lipid distribution in human colon.Anal Bioanal Chem. 2015; 407: 4697-4708Google Scholar). Briefly, nevi sections were scanned at a spatial resolution of 25 μm in negative-ion mode using the orbitrap analyzer of an LTQ-Orbitrap XL (ThermoFisher) equipped with an N2 laser (100 μJ max power, elliptical spot, 60 Hz repetition rate). Two microscans of 10 laser shots were recorded at each pixel. Due to the large dimensions of the laser spot, the images were acquired using the oversampling method (Jurchen et al., 2005Jurchen J.C. Rubakhin S.S. Sweedler J.V. MALDI-MS imaging of features smaller than the size of the laser beam.J Am Soc Mass Spectrom. 2005; 16: 1654-1659Crossref PubMed Scopus (220) Google Scholar). Mass resolution was set at 30,000 for data recording, and scanning ranged from 550 to 1,200 Da. 1, 5-diaminonaphtalene (Sigma Aldrich Química, Madrid, Spain) was chosen as a matrix for the negative-ion mode (Thomas et al., 2012Thomas A. Charbonneau J.L. Fournaise E. Chaurand P. Sublimation of new matrix candidates for high spatial resolution imaging mass spectrometry of lipids: enhanced information in both positive and negative polarities after 1,5-diaminonapthalene deposition.Anal Chem. 2012; 84: 2048-2054Crossref PubMed Scopus (246) Google Scholar). Matrix deposition on the tissue sections was performed with the aid of a glass sublimator (Ace Glass 8023, Vineland, NJ) (Garate et al., 2015Garate J. Fernández R. Lage S. et al.Imaging mass spectrometry at increased resolution using 2-mercaptobenzothiazole and 2,5-diaminonaphtalene matrices: application to lipid distribution in human colon.Anal Bioanal Chem. 2015; 407: 4697-4708Google Scholar). The amount of matrix deposited on each tissue section was such that it enabled us to obtain a uniform thin layer so that the matrix peaks did not to interfere with the lipid detection, while enabling us to scan samples for several hours. The IMS experiments were analyzed with dedicated imaging software (MSIAnalyst, Noray Bioinformatics, SLU, Bizkaia, Spain) and in house algorithms generated in Matlab (MathWorks, Natick, MA) to organize, calibrate, and align the spectra. The data were normalized using the total ion current and the spectra in each imaging experiment were aligned using the Xiong algorithm during the parsing stage, assuming a maximum misalignment of 0.04 Da (Xiong et al., 2012Xiong X. Xu W. Eberlin L.S. Wiseman J.M. Fang X. Jiang Y. et al.Data processing for 3D mass spectrometry imaging.J Am Soc Mass Spectrom. 2012; 23: 1147-1156Google Scholar). The peaks in the spectra with an intensity <0.5% of the maximum were discarded during data handling to reduce the number of m/z values and to speed-up the analysis. The regions of interest within each section were analyzed by means of the K-means algorithm in the statistical package provided with the imaging software. The regions of interest were classified with the principal component analysis algorithm (Unscrambler, version 9.7; Camo Software AS, Oslo, Norway) with the variables (peaks) normalized to their standard deviation in order to compensate the otherwise excessive weight of the changes on the strongest peaks (Astigarraga et al., 2008Astigarraga E. Barreda-Gomez G. Lombardero L. Fresnedo O. Castaño F. Giralt M.T. et al.Profiling and imaging of lipids on brain and liver tissue by matrix-assisted laser desorption/ionization mass spectrometry using 2-mercaptobenzothiazole as a matrix.Anal Chem. 2008; 80: 9105-9114Crossref PubMed Scopus (112) Google Scholar, Fonville et al., 2013Fonville J.M. Carter C.L. Pizarro L. Steven R.T. Palmer A.D. Griffiths R.L. et al.Hyperspectral visualization of mass spectrometry imaging data.Anal Chem. 2013; 85: 1415-1423Google Scholar). IBM SPSS Statistics for Windows, version 23.0 (IBM, Armonk, NY) was used for Levene-test, one-way analysis of variance, and Bonferroni/Games-Howell post hoc. Levene test determines the homogeneity (H0 = groups have equivalent variance) to choose the post hoc method: Bonferroni if Levene P ≥ 0.05 and Games-Howell if Levene P ≤ 0.05. To conduct hierarchical clustering and regression logistic validation we used Matlab and WEKA (Witten et al., 2016Witten I.H. Frank E. Hall M.A. Pal C.J. The WEKA workbench [online appendix]. Data Mining: Practical Machine Learning Tools and Techniques. 4th ed. Morgan Kaufmann, Cambridge, MA2016Google Scholar), respectively. Lipid assignment was based upon the comparison between the experimental m/z and the species in the software’s database (>33,000 lipid species and their adducts), and those in the LIPID MAPS database (www.lipidmaps.org). Mass accuracy was always better than 9 ppm and it was typically better than 3 ppm in the individual spectra. In this type of analyses, mass accuracy depends to some extent on the peaks’ intensity and thus, higher intensity m/z values have a better mass accuracy. When multiple candidates were found for a given m/z, confirmation of the species’ identity was attained by comparison with the data from ultra-high performance liquid chromatography tandem mass spectrometry experiments. More details regarding the lipid extraction from nevi sections and ultra-high performance liquid chromatography tandem mass spectrometry conditions can be found elsewhere (Fernández et al., 2017Fernández R. González P. Lage S. Garate J. Maqueda A. Marcaida I. et al.Influence of the cation adducts in the analysis of matrix-assisted laser desorption ionization imaging mass spectrometry data from injury models of rat spinal cord.Anal Chem. 2017; 89: 8565-8573Google Scholar). Lipid names and abbreviations were assigned according to the Lipid Maps nomenclature (http://www.lipidmaps.org). The MALDI-IMS technique determined molecular species composition as total number of carbon atoms and double bonds, and it differentiated the alkyl-acyl or alkenyl-acyl (ether-linked) from the diacyl phospholipids. Thus, the numbers between parentheses refer to the total number of carbons of the fatty acyl chains followed by the number of double bonds of all the chains. The ether linkage is represented by the “O” (alkyl-acyl) or “P (alkenyl-acyl, plasmalogen) prefix; as in PE(O-38:7) and PE(P-38:6).Supplementary Figure S2Distribution of some lipid species in sections of different skin nevi. PE, phosphatidylethanolamine; PE-P, PE ether (alkenyl-acyl) or plasmalogens; SM, sphingomyelin; PI, phosphatidylinositol. Scale bar = 200 μm.View Large Image Figure ViewerDownload Hi-res image Download (PPT)Supplementary Figure S3Principal component analysis for the discovery group. Principal component analysis showing full discrimination between epidermis (red), dermis (blue), and melanocytes (green) lipid fingerprints. Only those species with a standard deviation <80% of their average intensity within the spectrum of each type of tissue were reproducible and included in the analysis.View Large Image Figure ViewerDownload Hi-res image Download (PPT)Supplementary Figure S4Heat maps of all the lipid species detected by imaging mass spectrometry that classify the architectural features of nevi. Statistical significance was assessed using Student t test: ∗P ≤ 0.05, ∗∗P ≤ 0.01, ∗∗∗P ≤ 0.001, ∗∗∗∗P ≤ 0.0001. PC, phosphatidylcholine; PE, phosphatidylethanolamine; PE-O, PE ether (alkyl-acyl); PE-P, PE ether (alkenyl-acyl) or plasmalogens; PG, phosphatidylglycerol; PI, phosphatidylinositol; PS, phosphatidylserine; SM, sphingomyelin.View Large Image Figure ViewerDownload Hi-res image Download (PPT)Supplementary Figure S5Validation of the panel of biomarkers extracted from the discovery group. Comparison between H&E images and those from the K-means analysis of the imaging mass spectrometry data—using only the lipid biomarkers obtained from the discovery group—demonstrates that the algorithm automatically distinguishes four histologic areas: stratum corneum (pink), epidermis (red), dermis (blue) and melanocytes (green). Different green colors indicate populations of melanocytes with slightly different lipid profiles. Likewise, different blue colors indicate dermis with slightly different lipid profiles, but all were classified as dermis in the principal component analysis in Figure 2b of the main text. Yellow colors indicate areas with higher abundance of collagen. Finally, white areas correlate with the presence of physical damage of the tissue. Two sections of two of the samples were scanned (marked with asterisks in the image). Scale bar = 1 μm.View Large Image Figure ViewerDownload Hi-res image Download (PPT)Supplementary Table S1Characteristics of patients and lesions, discovery groupPatient No.SexAge, yLesion LocationDiagnosisMotive for ConsultationEvolutionP1F38BackIntradermal melanocytic nevus—Compound nevusEpidermal cystP2F33Right inframammary skinIntradermal melanocytic nevus—OkP3F38NeckIntradermal melanocytic nevusCloth discomfortIntradermal melanocytic neviP4F63NeckIntradermal melanocytic nevusCloth discomfortInflammatory lesion (post-inflammatory hyperpigmentation)P5M34ChestIntradermal melanocytic nevus—OkP6F28ScalpIntradermal melanocytic nevusCloth discomfortOkP7M47ChestIntradermal melanocytic nevusCloth discomfortOkAbbreviations: F, female; M, male. Open table in a new tab Supplementary Table S2Statistical analysis of the lipid fingerprints of the discovery groupVariableLevene TestOne-Way ANOVAEpidermis vs DermisEpidermis vs MelanocytesDermis vs MelanocytesPair-Wise Minimum Significance[Lyso-PI 18:0 - H]-0.363680.42722———NS[Lyso-PI O-18:2 - H]-0.570080.77967———NS[PC 32:0 - CH3]-0.119410.029020.204171.000000.02935∗[PC 33:2 - CH3]- & [PE 35:2 - H]- & [PE P-36:1 - H]- & [PC P-34:2-CH3]-1Games-Howell.0.020560.019380.075430.150260.72349NS[PC 35:1 - CH3]- & [PE 37:1 - H]-0.202420.90664———NS[PC O-38:8 - CH3]- / [PC P-38:7 - CH3]- & [PE O-40:8 - H]- / [PE P-40:7 - H]-0.101230.14191———NS[PE 34:1 - H]-0.745810.36869———NS[PE 34:2 - H]-0.405040.000000.000000.000030.01127∗∗∗∗[PE 36:1 - H]-0.557130.16975NS[PE 36:2 - H]-0.389410.000610.001140.003151.00000∗∗[PE 36:3 - H]-0.408850.000010.000010.011010.00848∗∗∗∗[PE 36:4 - H]-0.101690.002610.388110.069260.00214∗∗[PE 38:1 - H]-0.240740.004270.834750.004200.04856∗∗[PE 38:3 - H]-0.206720.05270———NS[PE 38:4 - H]-0.211740.000410.008570.000400.54704∗∗∗[PE 38:5 - H]-0.211510.000010.146850.000010.00076∗∗∗∗[PE 40:4 - H]-0.337470.000020.000010.098490.00201∗∗∗∗[PE 40:5 - H]-0.414100.000110.000090.006480.19332∗∗∗∗[PE O-34:2 - H]- / [PE P-34:1 - H]-0.569780.71210———NS[PE O-34:3 - H]- / [PE P-34:2 - H]-0.118720.000000.000000.000000.20388∗∗∗∗[PE O-36:3 - H]- / [PE P-36:2 - H]-0.235080.000110.000130.001610.78726∗∗∗[PE O-36:5 - H]- / [PE P-36:4 - H]-0.908870.000010.166290.000000.00032∗∗∗∗[PE O-38:5 - H]- / [PE P-38:4 - H]-0.124820.000700.062740.000510.12487∗∗∗[PE O-38:6 - H]- / [PE P-38:5 - H]-0.982390.022740.699310.020750.25887∗[PE O-38:7 - H]- / [PE P-38:6 - H]-0.738650.017910.183450.016720.79570∗[PE O-40:5 - H]- / [PE P-40:4 - H]-0.106180.000000.000010.000090.52263∗∗∗∗[PE O-40:7 - H]- / [PE P-40:6 - H]-0.418300.000470.020530.000370.25963∗∗∗[PG 34:1 - H]-0.969570.59320———NS[PG 36:1 - H]-0.336070.047101.000000.052710.22772NS[PG 36:2 - H]-0.136070.000140.580260.000150.00287∗∗∗[PG 36:3 - H]-0.156400.000000.235860.000000.00003∗∗∗∗[PI 34:1 - H]-0.197350.000000.000000.000001.00000∗∗∗∗[PI 34:2 - H]-0.146790.000000.000000.000001.00000∗∗∗∗[PI 36:1 - H]-0.437010.000030.000100.000121.00000∗∗∗∗[PI 36:2 - H]-0.194860.000010.000010.000140.86754∗∗∗∗[PI 36:3 - H]-0.075310.000970.002151.000000.00360∗∗[PI 36:4 - H]-0.240540.11879———NS[PI 37:4 - H]-0.380080.23937———NS[PI 38:3 - H]-0.656420.14446———NS[PI 38:4 - H]-0.591190.000000.000950.000000.01665∗∗∗∗[PI 38:5 - H]-1Games-Howell.0.005600.000000.000480.000160.01016∗∗∗[PI 40:4 - H]-0.385040.14813———NS[PI 40:6 - H]-1Games-Howell.0.006220.13105———NS[PS 36:1 - H]-1Games-Howell.0.000350.000080.005050.000010.75288∗∗∗∗[PS 36:2 - H]-0.625480.001870.001490.063650.31148∗∗[PS 38:4 - H]-1Games-Howell.0.049000.000020.001990.000580.49679∗∗∗[PS 38:5 - H]-0.851560.002690.002180.076080.36336∗∗[PS O-38:2 - H]- / [PS P-38:1 - H]-0.909030.025301.000000.056570.05005NS[SM d32:1 - CH3]-0.111240.012931.000000.030980.02703∗[SM d33:1 - CH3]-0.149570.000360.089260.000250.04578∗∗∗[SM d34:1 - CH3]-0.203520.000031.000000.000310.00005∗∗∗∗[SM d34:2 - CH3]-0.200830.021530.131891.000000.02323∗[SM d35:1 - CH3]-0.121240.000010.043530.000010.00172∗∗∗∗[SM d36:1 - CH3]-0.059920.000400.001560.000971.00000∗∗∗[SM d36:2 - CH3]-0.084000.05880———NS[SM d38:1 - CH3]-0.516990.002850.145940.201560.00219∗∗[SM d40:1 - CH3]-0.074410.000000.000000.868460.00000∗∗∗∗[SM d40:2 - CH3]-0.187580.001790.002301.000000.01334∗∗[SM d41:1 - CH3]-0.136460.001460.030220.514680.00130∗∗[SM d41:2 - CH3]-0.208710.001410.002931.000000.00529∗∗[SM d42:1 - CH3]-0.206580.000000.000000.290690.00000∗∗∗∗[SM d42:2 - CH3]-0.057910.000000.000000.009020.00000∗∗∗∗[SM d42:3 - CH3]-1Games-Howell.0.036890.000210.004290.029760.06705∗∗[Sulfatide d34:3 - H]-0.763280.376451.000001.000000.53016NSOne-way ANOVA and multiple testing correction. The table shows the P-values for Levene test to determine the homogeneity (H0 = groups have equivalent variance) to choose the post hoc method (Bonferroni or Games-Howell); one-way ANOVA P-values; Bonferroni or Games-Howell P-values for the three pair-wise comparisons and the minimum significance in the pair-wise comparison. ∗P ≤ .05, ∗∗P ≤ .01, ∗∗∗P ≤ .001, ∗∗∗∗P ≤ .0001.Abbreviations: ANOVA, analysis of variance; NS, not significant; PI, phosphatidylinositol; PS, phosphatidylserine; SM, sphingomyelin.1 Games-Howell. Open table in a new tab Supplementary Table S3Characteristics of patients and lesions, validation groupPatient No.SexAge, yLesion LocationDiagnosis1Number in parentheses indicates the number of samples shown in Supplementary Figure S5.Motive for ConsultationEvolutionP8M19Lower limbCompound nevus (1)DiscomfortOKP9M26BackIntradermal melanocytic nevus (1)Concern about malignancyOKP10M56AcralIntradermal melanocytic nevus (1)Compound nevus (2)Compound nevus (3)DiscomfortOKP11M25HeadIntradermal melanocytic nevus (1)—OKP12F35BackCongenital nevocellular nevus (1)Congenital nevocellular nevus (2)Cloth discomfortOKP13F72BackIntradermal melanocytic nevus (1)—OKP14M22BackJunction nevus (1)Concern about malignancyOKAbbreviations: F, female; M, male.1 Number in parentheses indicates the number of samples shown in Supplementary Figure S5. Open table in a new tab Abbreviations: F, female; M, male. One-way ANOVA and multiple testing correction. The table shows the P-values for Levene test to determine the homogeneity (H0 = groups have equivalent variance) to choose the post hoc method (Bonferroni or Games-Howell); one-way ANOVA P-values; Bonferroni or Games-Howell P-values for the three pair-wise comparisons and the minimum significance in the pair-wise comparison. ∗P ≤ .05, ∗∗P ≤ .01, ∗∗∗P ≤ .001, ∗∗∗∗P ≤ .0001. Abbreviations: ANOVA, analysis of variance; NS, not significant; PI, phosphatidylinositol; PS, phosphatidylserine; SM, sphingomyelin. Abbreviations: F, female; M, male.