Comprehensive characterization of complex glycosphingolipids in human pancreatic cancer tissues

Pancreatic ductal adenocarcinoma (PDAC) is one of the most common causes of cancer-related deaths worldwide, accounting for 90% of primary pancreatic tumors with an average 5-year survival rate of less than 10%. PDAC exhibits aggressive biology, which, together with late detection, results in most PDAC patients presenting with unresectable, locally advanced, or metastatic disease. In-depth lipid profiling and screening of potential biomarkers currently appear to be a promising approach for early detection of PDAC or other cancers. Here, we isolated and characterized complex glycosphingolipids (GSL) from normal and tumor pancreatic tissues of patients with PDAC using a combination of TLC, chemical staining, carbohydrate-recognized ligand-binding assay, and LC/ESI-MS2. The major neutral GSL identified were GSL with the terminal blood groups A, B, H, Lea, Leb, Lex, Ley, P1, and PX2 determinants together with globo- (Gb3 and Gb4) and neolacto-series GSL (nLc4 and nLc6). We also revealed that the neutral GSL profiles and their relative amounts differ between normal and tumor tissues. Additionally, the normal and tumor pancreatic tissues differ in type 1/2 core chains. Sulfatides and GM3 gangliosides were the predominant acidic GSL along with the minor sialyl-nLc4/nLc6 and sialyl-Lea/Lex. The comprehensive analysis of GSL in human PDAC tissues extends the GSL coverage and provides an important platform for further studies of GSL alterations; therefore, it could contribute to the development of new biomarkers and therapeutic approaches.

Pancreatic ductal adenocarcinoma (PDAC) is one of the most common causes of cancer-related deaths worldwide, accounting for 90% of primary pancreatic tumors with an average 5-year survival rate of less than 10%. PDAC exhibits aggressive biology, which, together with late detection, results in most PDAC patients presenting with unresectable, locally advanced, or metastatic disease. In-depth lipid profiling and screening of potential biomarkers currently appear to be a promising approach for early detection of PDAC or other cancers. Here, we isolated and characterized complex glycosphingolipids (GSL) from normal and tumor pancreatic tissues of patients with PDAC using a combination of TLC, chemical staining, carbohydrate-recognized ligand-binding assay, and LC/ESI-MS 2 . The major neutral GSL identified were GSL with the terminal blood groups A, B, H, Le a , Le b , Le x , Le y , P1, and PX2 determinants together with globo-(Gb 3 and Gb 4 ) and neolactoseries GSL (nLc 4 and nLc 6 ). We also revealed that the neutral GSL profiles and their relative amounts differ between normal and tumor tissues. Additionally, the normal and tumor pancreatic tissues differ in type 1/2 core chains. Sulfatides and GM 3 gangliosides were the predominant acidic GSL along with the minor sialyl-nLc 4 /nLc 6 and sialyl-Le a /Le x . The comprehensive analysis of GSL in human PDAC tissues extends the GSL coverage and provides an important platform for further studies of GSL alterations; therefore, it could contribute to the development of new biomarkers and therapeutic approaches.
Pancreatic ductal adenocarcinoma (PDAC) is the most prevalent type of primary pancreatic malignant tumors (accounting for more than 90% of all types of pancreatic cancer) with highly aggressive behavior and extremely poor prognosis (1)(2)(3). A major problem in the treatment of PDAC consists mainly of the difficult diagnosis of early stage (i.e., T1 and T2 tumors), which are usually asymptomatic. Most patients (80%) are diagnosed in advanced stages (i.e., T3 or T4 tumors with lymph node and distant metastases) and are not eligible for complete surgical resection and thus incurable (1,4). Another significant hallmark of PDAC is high resistance and low response rate to treatment with anticancer drugs and radiation (1,2,5). The high resistance of PDAC to available therapies, together with late detection, results in a 5-year overall survival rate of less than 10% and, particularly in metastatic PDAC, an overall 1-year survival rate of less than 20%. This makes PDAC the most lethal cancer (1)(2)(3)6). Therefore, novel diagnostic biomarkers for early cancer detection are urgently needed (2,5).
The carbohydrate antigen sialyl Lewis a (i.e., sLe a or CA 19-9) is one of the well-known and frequently used serological biomarkers for the clinical diagnosis of pancreatic (7,8), gastrointestinal, and other types of epithelial cancers (9). The determination of CA 19-9 test is routinely used to monitor treatment response in patients with advanced PDAC. However, the limited sensitivity and specificity does not allow to use CA 19-9 as a diagnostic biomarker for early stage tumors since CA 19-9 concentrations do not increase in a substantial percentage of patients with PDAC, and increased levels may be observed in patients with non-neoplastic disorders, despite high specificity for high cutoff values. Consequently, the CA19-9 assay is of limited utility for the diagnosis or monitoring of PDAC, preventing its use for early detection (10)(11)(12)(13). In a recent paper by Wolrab et al. (14), it was concluded that MS-based lipidomic profiling of human blood outperforms common clinical methods established for the monitoring of PDAC progression, including the CA 19-9 test.
Lipids have several key functions in human metabolism, such as constituting cell membrane components, signal molecules, energy supply, storage, and barriers (15)(16)(17). Specifically, glycosphingolipids (GSL) are ubiquitous constituents of eukaryotic plasma membranes and membrane-bound subcellular organelles that occur along with the most abundant phospholipids (15,18,19). GSL consist of a hydrophobic ceramide backbone bound to a hydrophilic carbohydrate part by a glycosidic bond, and both parts show immense structural diversity that makes them remarkably assorted compounds (18). Furthermore, GSL with blood group determinants is well known to be synthesized at high levels in the pancreas (20). Aberrant expression of GSL including alterations in the composition and concentrations of GSL and lipids is a typical hallmark of a wide range of cancers (7,14,(21)(22)(23)(24)(25), which has been extensively documented in cancer cell lines (22,(26)(27)(28)(29) or tissues (20,24,(30)(31)(32)(33)(34) and also reported in body fluids of cancer patients (35)(36)(37)(38). Several of the studies mentioned above concluded that the reported dysregulation of lipid metabolism in cancer cells is relevant to distinguish cancer patients from healthy controls, suggesting that changes in lipidomes are strongly associated with cancer progression (6).
Glycosylation occurs in all organisms and plays a crucial role in many cellular processes (39)(40)(41)(42). The disruption of glycosylation, such as aberrant glycan structure formation and alteration of glycosylation pathways, is probably intricately associated with a number of disorders including malignant transformation and tumor progression (19,40,42,43). This may also be accompanied by the expression of tumorassociated carbohydrate antigens (39). As a consequence, changes in lipid metabolism and glycosylation have received significant attention in recent decades and are commonly documented in cancer investigations (40). Alterations in glycan structures have been observed in many cancers (42,44,45). However, the complex biology of cancer development and progression is not yet fully understood. Investigations are specifically aimed at pathways linked to two main types of protein glycosylation, that is, N-linked and O-linked glycosylation, to reveal its role in cancer pathogenesis (39). Moreover, the results obtained by Zhang et al. (6) demonstrated that GSL-glycosylation and O-glycosylation play a more dominant role, in particular in pancreatic cancer, than N-glycosylation (46). However, targeted approaches that focus mainly on tumor cells and predefined metabolic pathways may not show the full extent of complex metabolic alterations (5). In addition, there are still major challenges that stem mainly from the lack of sensitive, accurate, and reliable methods for the separation of GSL isomers as well as for the detection, identification, and quantitation of less prevalent GSL species (47).
The aim of the present study is to characterize the GSL of human pancreatic tissues of patients with PDAC with a particular interest in minor complex GSL to expand the database of lipids that are routinely analyzed and to allow mutual comparison of GSL alterations in normal and tumor pancreatic tissues. The future perspective of this study is to incorporate these complex GSL into the screening method for PDAC based on body fluid analysis, as recently published by our research group (14).

Isolation of GSL for in-depth analysis
The GSL were isolated by a micro method ( Fig. 1) according to Barone et al. (48), which allows the isolation and purification of GSL with a wider range of carbohydrate units. This is of particular advantage for complex GSL that are found in biological materials in tiny amounts, and their effective isolation by conventional extraction methods, such as Folch (49), Bligh and Dyer (50), or Matyash (51), has not yet been described.
In total, 24 paired tissue samples of tumor and normal tissues were collected from 12 patients. After total lipid extraction, the extracts were subjected to mild alkaline methanolysis to remove acylglycerols and alkali-labile phospholipids. The purpose of the ensuing acetylation was to change the polarity of glycolipids from polar to nonpolar so that alkali-stable phospholipids (mainly sphingomyelins) were removed. Consequently, acetylated GSL were separated from the nonpolar compounds (e.g., ceramides) and alkali-stable phospholipids (especially sphingomyelins) using silica-based chromatography. After deacetylation, the GSL were separated into neutral GSL (N-GSL) and acid GSL (A-GSL) fractions using ion-exchange chromatography. In summary, 6.3 mg and 26.2 mg of N-GSL were obtained, together with 11.6 mg and 14.3 mg of A-GSL from pooled tumor and normal pancreatic tissues, respectively (Table 1).
Rhodococcus spp. recombinant endoglycoceramidase II (rEGCase II) was used for the hydrolysis of GSL, although the hydrolytic capacity of this enzyme to globo-series GSL and some gangliosides is restricted (28). In contrast, EGCase I has a broader substrate specificity and better reaction efficiency than EGCase II and III (52,53). However, the use of rEGCase II in this study was intentional because globotriaosylceramide and globotetraosylceramide (Gb 3 and Gb 4 ) are major GSL of many tissues, resulting in MS spectra being dominated by Gb 3 and Gb 4 ions. The main advantage of using rEGCase II in this study is that it allowed the detection of low abundant complex GSL.

Separation and structural characterization of GSL
We performed liquid chromatography electrospray ionization tandem mass spectrometry (LC/ESI-MS 2 ) analysis of intact GSL (both N-and A-GSL) and neutral GSL-derived oligosaccharides from human pancreatic cancer and surrounding normal tissues. The major mono-and di-hexosylceramides (i.e., GlcCer, GalCer, LacCer), globotriaosylceramides and globotetraosylceramides (i.e., Gb 3 and Gb 4 ), and (neo)lacto-GSL together with several ganglioside subclasses and sulfatides have been extensively investigated in various biological matrices, as thoroughly summarized in studies by Zhuo et al. (54) and Wolrab et al. (23). In contrast, only a few recent studies showed altered complex GSL in most tumor cells (6,33,34). Therefore, this study focuses mainly on tetrasaccharides and larger oligosaccharides with the goal of comparing the GSL profiles of normal and tumor pancreatic tissues and implementing the GSL database for lipidomic analysis.

LC/ESI-MS 2 of neutral GSL-derived oligosaccharides
Oligosaccharides released from total N-GSL fractions isolated from the tumor and surrounding normal tissues were analyzed by LC/ESI-MS 2 in the negative-ion mode (Fig. 2).
Base peak chromatograms (BPCs) from pooled normal ( Fig. 2A)  Most of these observed deprotonated molecules, particularly those most abundant, were additionally confirmed by the presence of sodium and potassium adducts (i.e., [M-2H+Na]and [M-2H + K] -) in ion profiles of deprotonated molecules, as depicted in Figure 2, C and D. A detailed interpretation of Table 1 Amounts of acid and neutral glycosphingolipids obtained from normal and tumor pancreatic tissues of PDAC patients and expressed in mg of glycosphingolipids per g of tissues in dry weight N-GSL and A-GSL denote total neutral and acid glycosphingolipids, respectively. T and N denote tumor and normal, respectively, and ND denotes not determined.    Characterization of glycosphingolipids in pancreatic cancer Characterization of glycosphingolipids in pancreatic cancer identification of P1 pentasaccharide (57) and considering that the α1,3-galactosyltransferase is not expressed in humans (58) (Fig. 5F). A type 2 core chain Galβ4GlcNAc was inferred from the fragment ions 0,2 A 3 at m/z 589.1, 0,2 A 3 -H 2 O at m/z 571.3, and 2,4 A 3 at m/z 529.3. Taken together, it was assigned as Galcα3(Fucα2)Galβ4GlcNAcβ3Galβ4Glc, i.e., a blood group B group type 2 hexasaccharide (B6-2).
Deprotonated molecule [M-H]at m/z 1160.3 consistent with the composition of Hex 4 HexNAc 1 Fuc 2 was eluted in both pooled normal and tumor tissues at 13.9 min and 14.1 min (Figs. 3, A and B and 6D), respectively. The oligosaccharide sequence was concluded from the MS 2 spectrum (Fig. 5G) based on the series of C-type fragment ions (C 2 at m/z 486.9, C 3 at m/z 836.0, and C 4 at m/z 998.3) together with cross-ring fragment ions 0,2 A 5 at m/z 1100.3, 0,2 A 5 -H 2 O at m/z 1082.1, and 2,4 A 5 at m/z 1040.3. The diagnostic fragment ion C 3 /Z 3β at m/z 672.9 provided the evidence of 4-substituted GlcNAc with Fuc at 3-position and, furthermore, affirms the terminal 3-linked branch of GalGal(Fuc)GlcNAc. Therefore, it was assigned as Galcα3(Fucα2)Galβ4(Fucα3)GlcNAcβ3Galβ4Glc, i.e., a blood group B type 2 heptasaccharide (B7-2).
In summary, the LC/ESI-MS 2 employed for the structural analysis of GSL-derived oligosaccharides on the porous graphitized carbon column provided a powerful platform that allowed the discrimination of isomeric glycan structures and allowed clear deduction of the carbohydrate sequence based on the typical series of C-and B-type fragment ions obtained by MS 2 analysis. Moreover, the diagnostic cross-ring 0,2 A/ 0,2 A-H 2 O and 2,4 A fragment ions of antepenultimate N-GlcNAc distinguished neolacto series (Galβ4GlcNAc) from lacto series (Galβ3GlcNAc) (60). For instance, the presence of fragmentation ions at m/z 427/409 ( 0,2 A 3 / 0,2 A 3 -H 2 O, Fig. 4, A and B) allowed the identification of linkage positions, i.e., type 1 or type 2 chain, and indicated that Hex 3 HexNAc 1 Hex 1 ion at m/z 852 was H5-2 rather than H5-1, which is in correlation with the previously published data (56,60,61). Furthermore, the characteristic diagnostic ions resulting from the double glycosidic cleavage of 3-linked branches supported the identification of type 1 and type 2 core chains as well as enabled the differentiation of A, B, Le a , Le b , Le x , Le y blood group epitopes. In case of Le a/b , the presence was supported by the fragmentation ions at m/z 348 (Fig. 4, C and D), while Le x/y was indicated by the fragmentation ions at m/z 364 (Fig. 4, C and F) and m/z 510 (Fig. 4E). Neolacto tetrasaccharides (i.e., nLc 4 , Fig. 5A) were further elongated (e.g., nLc 6 in Fig. 5B) or capped with blood group epitopes (e.g., A6-2 and B6-2 in Fig. 5, E and F, respectively). More interestingly, the presence of P1-5 (Fig. 5C) and PX2-5 (Fig. 5D) was only detected in the pooled tumor tissue. It should also be mentioned that double peak formation was observed in most GSL subclasses (Figs. 2,   A and B, and 6) and is most likely due to the existence of both α and β anomers of glucose at the reducing end. The identical composition of these double peaks was also confirmed by MS 2 analysis, as illustrated in Figure 7. The α-/β-anomers can be condensed by reduction of the samples. However, when analyzing reduced samples, the predominance of C-type fragment ions that allow a straightforward interpretation of the carbohydrate sequence is lost, and instead, a mixture of B, C, Y, and Z ions is obtained, making interpretation more difficult (55). Overall, a clear distinction between GSL profiles of normal and tumor pancreatic tissues was found. The neutral GSL-derived oligosaccharides identified and structurally characterized by LC/ESI-MS 2 in the N-GSL fractions obtained from tumor and normal pancreatic tissues of PDAC patients are summarized in Table 2.

LC/ESI-MS 2 of native GSL
Native total N-GSL and A-GSL fractions isolated from human normal and tumor pancreatic tissues of PDAC patients were separated by hydrophilic interaction liquid chromatography (HILIC) and subsequently analyzed by LC/ESI-MS 2 coupled with a capillary HILIC column in the negative ion mode, detected mostly as [M-H] -. The quality of HILIC runs was poor, even when rerunning the samples, and the intensity of the signal was generally low, which complicated the identification of GSL in the samples. The sensitivity issues caused that LC/ESI-MS and LC/ESI-MS 2 analyses of the total fraction of GSL of human pancreatic tissues did not provide too much information, which resulted in the identification of only a few species of GSL in human pancreatic tissues. The nomenclature and shorthand notation of individual lipid species follow the standardized system for reporting lipid structures, as described by Liebisch et al. (62).
Total N-GSL fractions-To obtain an overview of the ceramide composition of the native N-GSL fractions from the pooled pancreatic tissues, these fractions were analyzed by LC/ ESI-MS 2 using a HILIC column. This yielded very weak MS spectra that, together with subsequent MS 2 analysis, allowed the reliable identification of only a few GSL species. Among these N-GSL were nLc 4  Total A-GSL fractions- Figure 8 illustrates the BPC of total A-GSL fractions of pooled tumor (Fig. 8A) and normal (Fig. 8B) tissues. The pooled tumor sample contained dominant sulfatides and gangliosides, while the former ones were not detected in the pooled normal tissue (Fig. 8, A and B). Trace amount of other acidic GSL was also detected in pooled tumor tissues.
The presence of sulfatides was indicated by B 1 ions at m/z 241.1 or C 1 ions at m/z 259.1, demonstrating a terminal SO 3 -Hex in their MS 2 spectra (Fig. 9). The BPC obtained from LC-ESI/MS of the A-GSL fraction from pooled tumor tissues (Fig. 8A)    The gangliosides were detected in both pooled tumor and normal tissues. One of the major deprotonated ions from pooled tumor tissues was observed at m/z 1151.8 (Fig. 8A). The MS 2 spectrum of this ion yielded a series of Y/Z ions (i.e., Y 0 at m/z 536.7, Z 0 at m/z 518.5, Y 1 at m/z 698.6, Z 1 at m/z 680.6, and Y 2 at m/z 860.6), which implies an oligosaccharide with the composition of NeuAc 1 Hex 2 (Fig. 10A). Moreover, there was 0,2 X 2 fragment ion at m/z 930.  (Fig. 10, B and C), respectively.
Few other minor acidic GSL were also detected. The minor ion at m/z 1517.0 corresponds to a monosialylated neolactotetraosylceramide (i.e., Neu5Ac-nLc 4 Cer), as characterized by MS 2 sequencing (Fig. 11). The glycan sequence was deduced from a series of Y-/Z-type fragment ions (i.e., Y 0 at m/z 536.6, Y 1 at m/z 698.6, Y 2 at m/z 860.7, Y 3 at m/z 1063.7, Y 4 at m/z 1225.8, Z 1 at m/z 680.6, and Z 3 at m/z 1045.7). In addition, we found that the ion at m/z 1517.0 represents two GSL structures. These two GSL species were distinguished based on distinct retention times and the specific 0,2 X 4 fragment ion at m/z 1295.8 arising from the cross-ring cleavage of sialic acid. This fragment ion is highly abundant (>50 % of relative intensity) in α6-linked sialic acid, whereas it is low abundant or absent in α3-linked sialic acid (66). Collectively, these features were recognized as Neu5Acα3-nLc 4 Cer (eluting at 28.6 min, Fig. 11A) and Neu5Acα6-nLc 4 Cer (eluting at 30.5 min, Fig. 11B) with ceramide 18:1;O2/16:0. Neu5Acα3-nLc 4 Cer and Neu5Acα6-nLc 4 Cer are termed as iso-CD75s-and CD75sganglioside, which elevate in pancreatic tumor (67).
The BPC of the total A-GSL fraction from pooled normal pancreatic tissues was very weak, and we found only Characterization of glycosphingolipids in pancreatic cancer gangliosides (Fig. 8B). The main ions observed (Fig. S1

Chromatogram-binding assay
Next, the binding of antibodies, lectins, and bacteria to GSL fractions isolated from pooled tumor and normal pancreatic tissues was tested to substantiate the data obtained from LC/ESI-MS 2 . The results of binding assays clearly illustrate the differences in GSL expression in normal and tumor pancreatic tissues (Figs. 12 and 13). Thin-layer chromatography TLC with anisaldehyde detection of N-GSL fractions showed that the major bands migrated in the monoglycosylceramide to tetraglycosylceramide regions along with some minor slow-migrating compounds (exemplified by pooled normal and tumor pancreatic tissue in Figure 12A, lanes 1 and 2). TLC with detection of the resorcinol reagent of A-GSL fractions had several weak bands that confirmed the presence of neuraminic acid and/or its derivatives. Moreover, the TLC with anisaldehyde detection of A-GSL fractions showed the presence of Neu5Ac-GM 3 in both normal and tumor pancreatic tissue (Fig. 13A, lanes 1 and 2), as indicated by comigration with the reference Neu5Ac-GM 3 (Fig. 13A,  lane 3). Furthermore, several other slow-migrating and Neu5Ac-containing GSL were found (Fig. 13A, lanes 1 and 2). The appearance of double bands on the TLC chromatogram (Fig. 12A, lanes 1 and 2) is caused by ceramide heterogeneity.

Chromatogram-binding assay for N-GSL fractions
The binding of antibodies, lectins, and bacteria to N-GSL fractions is illustrated in Figure 12. The presence of globotriaosylceramide (Gb 3 ) and globotetraosylceramide (Gb 4 ) in both pooled normal and tumor tissues were demonstrated by the binding of 35 S-labeled Galα4Gal-recognizing P-fimbriated Escherichia coli strain 291-15 in the triglycosylceramide and tetraglycosylceramide regions (Fig. 12B, lanes 1 and 2). This result is consistent with the study published by Distler et al. (69).
Next, the Galβ4GlcNAc-binding lectin of Erythrina cristagalli provided a more intense staining in the pooled pancreatic tumor tissue fraction (Fig. 12C, lane 2) than the pooled normal tissue fraction (Fig. 12C, lane 1), which corresponds to higher amounts of neolactotetraosylceramides (nLc 4 Cer) found in pooled tumor tissue.
Furthermore, monoclonal antibodies directed against Le a (Fig. 12D) and Le b (Fig. 12E) determinants were mainly bound to the fractions obtained from pooled tumor tissues (Fig. 12, D  and E, lane 2), which confirmed the higher amounts of Le a pentosylceramide (Le a -5) and Le b hexosylceramide (Le b -6) detected by LC/ESI-MS 2 in the tumors. A considerably weaker binding of anti-Le a and anti-Le b antibodies was also observed in pooled normal tissues (Fig. 12, D and E, lane 1). Additionally, some compounds that migrate above and below the Characterization of glycosphingolipids in pancreatic cancer pentasaccharide region were recognized by anti-Le a antibodies (Fig. 12D, lane 2) indicating the presence of more complex GSL with the Le a epitope in the tumor tissue.
In contrast, the monoclonal antibodies directed against the blood group A determinants (Fig. 12F) and the Griffonia simplicifolia IB4 lectin recognizing Galα terminals, i.e., binding to blood group B determinants (Fig. 12G), were bound mainly to the fraction obtained from pooled normal pancreas tissues (Fig. 12, F and G, lane 1). A weak binding of G. simplicifolia IB4 lectin was observed indicating the presence of determinants of blood group B in the fraction obtained from pooled tumor tissues (Fig. 12G, lane 2), while no binding of anti-A antibodies was observed in pooled tumor tissues (Fig. 12F, lane 2). Additionally, several other compounds migrating below the hexasaccharide region were recognized by anti-A antibodies and indicate more complex GSL with blood group A determinants (Fig. 12F, lane 1) in the pooled normal pancreas tissues. Taken together, these results support the hypothesis that GSL with the determinants of blood groups A and B were predominantly present in the fraction obtained from pooled normal pancreas tissues.
A well-known problem with carbohydrate-binding ligands is that many of them are not as specific as it is claimed (70)(71)(72). They may show cross-reactive binding to other glycans or in some cases be nonbinding. However, the binding specificities of the ligands used in this study have been well characterized by us and others (71,73,74) and used in many previous studies.

Chromatogram-binding assay for A-GSL fractions
The binding of antibodies to the A-GSL fractions is illustrated in Figure 13. The antibodies directed against Neu5Acα3-nLc 4 determinants were bound to both pooled normal and tumor pancreatic tissue fractions (Fig. 13B, lanes 1  and 2), confirming the presence of sialyl-nLc 4 . No binding of anti-Neu5Acα3-Lc 4 was observed. Additionally, anti-Neu5Ac-Le a (Fig. 13C) and anti-Neu5Ac-Le x (Fig. 13D) antibodies were mainly bound to the fraction obtained from pooled tumor pancreatic tissue (Fig. 13, C and D, lanes 1 and 2), indicating higher amounts of sialyl-Le a (sLe a ) and sialyl-Le x (sLe x ) pentaosylceramides. The former one is also known as carbohydrate antigen 19-9 (CA 19-9), which is known as a pancreatic cancer marker suitable for the monitoring of disease progress but not suitable for early cancer detection. The presence of sLe a and sLe x in the fractions was also indicated by comigration with the reference A-GSL fraction obtained from lung cancer metastatic tissue (Fig. 13, C and D, lane 4) since it has previously been shown that these sialylated GSL play a role in lung cancer (75,76). In line with this, a higher amount of Le a -5 pentaosylceramides was also detected by LC/ESI-MS 2 analysis in tumor tissue (Table 2).

Discussion
The present work is a systematic and detailed investigation of mainly neutral GSL and further acid GSL isolated from human pancreatic tissues of patients with PDAC. The identification and structural characterization are accomplished with a combination of TLC, chemical staining, binding of carbohydrate-recognizing ligands (antibodies, lectins, and bacteria), and LC/ESI-MS 2 , with a major focus on complex GSL.
GSL patterns of pooled human pancreatic tissues revealed that glycan profiles of tumor pancreatic and surrounding normal pancreatic tissues differ in the region from four to seven sugar units. The lipid and glycan profiling investigated here revealed that the major N-GSL of tumor pancreatic tissues identified by LC/ESI-MS 2 were GSL with the blood group Le a and Le b determinants together with neolactotetraosylceramides (nLc 4 Cer) (Fig. 2B), while the predominant components of normal tissues were GSL with the blood group A and B determinants ( Fig. 2A). These findings are remarkable since the type 2 core chain of complex GSL was dominating in human normal pancreatic tissues compared to the complex GSL in human pancreatic cancer tissues, where type 1 core chain was mainly found. These results are also supported by the virtually identical results obtained with the binding assay, as illustrated in Figure 12. Furthermore, we found GSL with the blood group Le x , Le y , and H determinants and neolactohexaosylceramides (nLc 6 Cer) in both pooled normal and tumor pancreatic tissues. Moreover, PX2 and P1 pentaosylceramides alongside Le x heptosylceramides were characterized as minor components in pooled tumor pancreatic tissues. Additionally, the presence of globotriaosylceramides (Gb 3 ) and globotetraosylceramides (Gb 4 ) in both pooled samples was indicated by the binding assay (see Fig. 12B), although these were not identified and characterized by LC/ESI-MS 2 . The absence of globo-series GSL in MS spectra may be in line with the relative resistance of globo-series GSL to hydrolysis by rEGCase II, as previously reported (28,55,77,78).
In case of A-GSL fractions, we obtained very little information from both pooled tissue samples, since the MS spectra did not allow the identification of a larger number of GSL. Nevertheless, several sulfatides and GM 3 gangliosides were identified and characterized as the main components of the pooled tumor tissues together with other minor compounds such as monosialylated neolacto(tetra/hexa)osylceramides (Neu5Ac-nLc 4 Cer/nLc 6 Cer) (Fig. 8A). Sulfatides and GM 3 gangliosides with 34:1;O2 and 34:1;O3 ceramides were the most predominant GSL species observed (Figs. 9, A and B and  10, A and B). Additionally, Neu5Ac-Le a (i.e., sLe a or also CA 19-9 biomarker) and Neu5Ac-Le x GSL were identified by binding assay as well, despite not being characterized by mass spectrometry.
Importantly, the results presented in this report support that alterations in GSL composition, including aberrant glycosylation, sialylation, and/or fucosylation, are an integral part of malignant transformation and tumor progression (6,22,27,32,46,75,(79)(80)(81)(82). Interestingly, striking differences in fucosylation, representing one of the most important oligosaccharide modifications linked to cancer, have been previously reported in cell lines (81,83) and tumor tissues (79,80) and therefore appear to be a promising target for cancer diagnosis and therapy (84). The changes in glycan structures in PDAC are linked to the expression of glycosyltransferases and related to the formation of Lewis blood group antigens. Deregulations of fucosyltransferases (FUTs) in PDAC have previously been reported (85). Specifically, FUT1 preferentially fucosylate type 2 core chains, while FUT2 and FUT3 prioritize type 1 chains as a substrate (86). Here, we demonstrate that there is a predominance of fucosylated type 1 core GSL (i.e., Le a -5 and Le b -6) and nLc 4 Cer in pancreatic tumors, whereas the major compounds in the nontumor tissues are blood A and B GSL (i.e., A6-2, B6-2, and B7-2) on type 2 core chains. Thus, the overexpression of Lewis blood group antigens Le a and Le b in PDAC may be associated with the upregulation of FUT2 and/ or FUT3. Furthermore, the higher amount of nLc 4 Cer i.e., type 2 chain) in PDAC tissues may be due to the downregulation of FUT1, which by adding a Fuc to the terminal Gal of nLc 4 Cer creates a H type 2 determinant, which is the precursor for the subsequent action of a GalNAcT and a GalT creating the blood group A and B determinants. Clearly, further studies are needed to clarify these results. We should also note that the relative amounts of GSL in the N-GSL fractions (Fig. 6) were different between tumor and normal pancreatic tissues.
Furthermore, GSL with blood group A and B determinants are declined or practically eliminated compared to normal tissues of the same patient where they predominate. We can only speculate that individuals carrying blood groups A and B determinants may be more prone to develop pancreatic cancer based on the comparison of tissue samples, which is in agreement with previously published studies (87)(88)(89)(90). To our knowledge, there is only one previous study of GSL in normal human pancreas published by Breimer (91) in 1984, where the occurrence of both type 1 and type 2 core chain blood group ABH and Lewis glycolipids in pancreas is reported in two individuals with blood group A and B. However, more studies will be needed to clarify the value of these findings. The present work focuses on qualitative analysis and lipid profiling of mainly complex GSL in human pancreatic cancer, which are not commonly included in conventional lipidomic methods and extends the coverage of GSL commonly analyzed in cancer research. Therefore, future studies should also investigate whether the differences observed between normal and pancreatic tumor tissues translate into differences in GSL profile between PDAC patients and healthy subjects.

Reference GSL
N-GSL and A-GSL fractions were isolated as described by Karlsson (92). Individual GSL were isolated by repeated chromatography on silicic acid columns and by HPLC and further identified and characterized by mass spectrometry (55,93) and proton NMR spectroscopy (94).

Sample collection
Tissue samples including tumor and surrounding normal pancreatic tissues were obtained from 12 different patients with PDAC (see Table 3). The samples were collected at the University Hospital Olomouc and kept in a freezer at −80 C prior to further processing. The study was approved by the Regional Ethics Committee of University Hospital Olomouc, Czech Republic (reference number 57/15) following the Declaration of Helsinki and the General Data Protection Regulations. All patients received written and verbal information before signing an informed consent for inclusion in the study. The complete list of samples with clinicopathological information is described in "Table S1" in Supporting information. The information on blood groups is not available.

Isolation and preparation of GSL
Samples obtained from 12 PDAC patients were pooled separately for tumor and adjacent nontumor tissues, and lyophilized. The nontumor tissue is further annotated as "normal tissue". The initial amounts of the tissue samples (i.e., before lyophilization) used for the isolation of GSL are listed in Table 3.
Due to the limited amount of starting material that restricted the experiments performed, we used the micro Table 3 Initial amounts of the tissue samples (i.e., before lyophilization) used for the isolation of GSL Characterization of glycosphingolipids in pancreatic cancer method described by Barone et al. (48), which is based on the method originally introduced by Prof. Karlsson, for the isolation of total N-GSL and A-GSL. The only modification was the use of Soxhlet extraction at the beginning of the experiment. The scheme of the procedure used for the preparation of total N-GSL and A-GSL is shown in Figure 1, and a detailed description of the protocol is described in "Protocol S1" in Supporting information. The obtained total GSL fractions (i.e., N-GSL and A-GSL) were characterized by a combination of TLC, binding of carbohydrate-recognizing ligands in chromatogram-binding assays, and LC/ESI-MS 2 as described below.
Thin-layer chromatography TLC was performed continuously throughout the whole extraction protocol to control each step of the procedure. The TLC was accomplished on aluminum-backed or glass-backed silica gel 60 high performance TLC plates (Merck). GSL mixtures (40-80 μg) and/or pure GSL (4 μg) were applied to high performance TLC plates and chromatographed with a solvent system composed of CHCl 3 /MeOH/H 2 O (60:35:8, v/v/v). The developed plates were air-dried and subsequently chemically detected using the anisaldehyde staining reagent for both GSL fractions (i.e., anisaldehyde/acetic acid/H 2 SO 4 in proportions 1:98:2, v/v/v) (72) or the resorcinol staining reagent (95,96) for total A-GSL fractions (i.e., 0.

Chromatogram-binding assays
Binding of monoclonal antibodies to GSL separated on thinlayer chromatograms was performed as described by Barone et al. (48,72). A detailed description of the binding procedure is described in "Protocol S2" in the Supporting information. The binding of 35 S-labeled Galα4Gal-binding P-fimbriated E. coli, 125 I-labeled E. crista-galli lectin, G. simplicifolia lectin IB4, and anti-Neu5Ac-nLc 4 /Lc 4 to GSL in thin-layer chromatograms was performed as previously reported (73,74,97,98). The specifications of carbohydrate-recognizing ligands tested for binding to the GSL of human PDAC tissues are listed in Table 4.

Endoglycoceramidase digestion
rEGCase II from R. spp. (Takara Bio Europe S.A.) was used for the digestion of N-GSL as described (57). A detailed description of the whole procedure is listed in "Protocol S3" in Supporting information. The neutral oligosaccharides released from GSL were resuspended in 50 μl of deionized water prior to analysis.
Detailed descriptions of the LC/ESI-MS 2 conditions for the analysis of native GSL and GSL-derived oligosaccharides are listed in the Supporting information in "Methods S1" and "S2", respectively.

Data processing
Thermo Scientific Xcalibur software (Version 2.0.7) was used for data processing. Assignment of the glycan sequence and GSL structures was done manually based on the knowledge of mammalian biosynthetic pathways together with the help of the GlycoWorkbench tool (Version 2.1, https:// glycoworkbench.software.informer.com/download/) (99), Lipid Maps MS analysis tools (https://www.lipidmaps.org/ tools/ms/). The characteristic fragmentation patterns of the identified GSL subclasses follow general rules and nomenclature for the cleavages of linear and branched oligosaccharides (100) (see Fig. 14) and have previously been well described (56,61,(63)(64)(65)68). Structures were verified by comparison of retention times and in-depth examination of relevant MS 2 / MS 3 spectra of GSL or GSL-derived oligosaccharides from reference GSL (55).
Supporting information-This article contains supporting information.