Alterations of the Human Skin N- and O-Glycome in Basal Cell Carcinoma and Squamous Cell Carcinoma

The glycome of one of the largest and most exposed human organs, the skin, as well as glycan changes associated with non-melanoma skin cancers have not been studied in detail to date. Skin cancers such as basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) are among the most frequent types of cancers with rising incidence rates in the aging population. We investigated the healthy human skin N- and O-glycome and its changes associated with BCC and SCC. Matched patient samples were obtained from frozen biopsy and formalin-fixed paraffin-embedded tissue samples for glycomics analyses using two complementary glycomics approaches: porous graphitized carbon nano-liquid chromatography electro spray ionization tandem mass spectrometry and capillary gel electrophoresis with laser induced fluorescence detection. The human skin N-glycome is dominated by complex type N-glycans that exhibit almost similar levels of α2-3 and α2-6 sialylation. Fucose is attached exclusively to the N-glycan core. Core 1 and core 2 type O-glycans carried up to three sialic acid residues. An increase of oligomannose type N-glycans and core 2 type O-glycans was observed in BCC and SCC, while α2-3 sialylation levels were decreased in SCC but not in BCC. Furthermore, glycopeptide analyses provided insights into the glycoprotein candidates possibly associated with the observed N-glycan changes, with glycoproteins associated with binding events being the most frequently identified class.

inTrODUcTiOn Skin is the largest human organ covering an area of up to two square meters fulfilling vital functions such as protection against pathogens, prevention of water loss, and provision of a sensation platform. It is a widely accepted fact that frequent exposure of skin to ultraviolet radiation is a major risk factor in the development of skin cancers (1,2) one of the most prevalent types of cancer in Caucasians (3). Basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) are the more frequently occurring types of skin cancer next to the highly aggressive malignant melanoma type skin cancers. There are an estimated three million people diagnosed with some form of BCC or SCC each year in the US alone (4) posing a serious health burden as they are causing severe local damage if left untreated.
A cell's transformation into a malignant state is accompanied by numerous changes on a molecular level including changes in protein glycosylation (5), one of the most common types of post translational modifications. Protein glycosylation is the result of a complex enzymatic network that is regulated by numerous intrinsic and extrinsic factors. Changes in glycosyltransferase levels or activities and monosaccharide substrate availability are just a few examples out of the numerous factors that ultimately result in altered glycosylation patterns that are frequently reported for malignant cells (6). As glycans play important roles in cell adhesion, cell signaling as well as in adaptive and innate immune responses (7,8), aberrant protein glycosylation patterns represent a recurring feature reported for all types of cancer studied to date (5).
Changes in glycosylation-associated pathways are often reflected in up-or down regulation of glycan structure features such as sialic acid (9), bisecting N-acetylglucosamine, Lewis structures, core-type fucose, or β1-6-linked branching (10,11). Changes in glycosylation can alter the functional properties of their individual protein carriers affecting tumor growth, formation of metastases, escaping immune recognition or conferring resistance to therapeutic interventions (12)(13)(14). The introduction of a β1-6 branch can alter subsequent glycan processing by mediating the addition of poly-N-acetyllactosamine residues. Due to the configuration of the β1-6 branch interactions with the polypeptide chain are possible that alter protein properties (15). An increase in the levels of sialic acid (16) or Lewis structures can affect lectin interactions, which can significantly alter vital functions such as cell signaling, cell adhesion or cell motility. O-glycans on the other hand are often observed as truncated structures only consisting of one or two monosaccharide units such as the Tn and T antigen (17). Both these antigens play a major role as possible anticancer vaccines, and increased Tn/Tantigen levels often correlate with poor disease prognosis. Other cancer-associated glycan changes have also been recognized for their potential diagnostic value in tumor detection and evaluation of tumor growth and progression (18)(19)(20). This makes glycans and their specific tumor signatures highly interesting to gain a better understanding of the molecular events associated with tumor onset and progression. Even though skin represents the most exposed organ surprisingly little information is available concerning the protein glycosylation repertoire of healthy human skin and its alterations in non-melanoma skin cancer. To fill this gap, we used porous graphitized carbon liquid chromatography (PGC-LC) coupled with electrospray ionization (ESI) ion-trap tandem MS (ESI-MS/MS) detection and multiplexed capillary gel electrophoresis with laser induced fluorescence detection (xCGE-LIF) (21) to investigate the healthy human skin N-and O-glycome and the alterations associated with BCC and SCC. Moreover, to obtain insights into the identity of the glycoproteins carrying these N-glycans hydrophilic interaction chromatography (HILIC) enriched (glyco)peptides fractions were analyzed by nanoLC ESI Orbitrap tandem mass spectrometry (MS/MS).

MaTerials anD MeThODs consumables
All materials were purchased in the highest possible purity from Sigma-Aldrich (St. Louis, MO, USA) unless stated differently. Glycerol-free peptide-N-Glycosidase F (PNGase F) was obtained from Roche Diagnostics GmbH (Mannheim, Germany). Water was taken from a MilliQ-Direct 8 system (Merck KGaA, Darmstadt, Germany).

Patient samples
The study was approved by the institutional review board of the medical faculty at the University of Leipzig (No. 127- . Patients referred to the Department of Dermatology at the Leipzig University Medical Center for surgical treatment of BCC or SCC were asked to participate on a voluntary base and were included after written informed consent. To allow correct dermato-pathological investigation, only patients were selected that exhibited a well-circumscribed BCC or SCC of at least 5 mm diameter. Following written informed consent, basal cell or SCCs were routinely excised (fusiform excision) in local anesthesia with micrographic control of the margins. Punch biopsies were taken from tumor and normal, healthy skin of the excised tissue, snap frozen in liquid nitrogen, and then stored at −80°C. The remaining material was subjected to the routine dermato-pathological work up. Biopsies were transferred on dry ice to the Max Planck Institute of Colloids and Interfaces for further analyses.
In addition to the fresh frozen samples, formalin-fixed paraffin-embedded (FFPE) tissue from 20 patients with BCC and 15 patients with SCC was obtained following routine dermatopathological work up. Tissue slides containing tumor tissue and slides of the tumor free margins (=healthy skin) were provided respectively. More detailed patient information can be found in Tables S4-S6 in Supplementary Material.

Biopsy sample homogenization and Protein extraction
Biopsy tissue samples were washed three times with 1 mL 70% ethanol (EtOH) and three times with 50 mM NH4HCO3. The sample was homogenized in 1 mL of homogenization buffer [50 mM NH4HCO3, 1 M urea, 10% ACN, and 0.1% sodium dodecyl sulfate (SDS)] for (3× 30 s) using a homogenizer (Ultraturrax, IKA, Staufen, Germany) with subsequent sonication for 30 s. Samples were kept at 0°C during the entire homogenization process before any protease, and glycosidase activity was terminated by incubating them for 5 min at 96°C. To increase protein/peptide solubility, 4 µg of trypsin was added, and the samples were incubated over night at 37°C before insoluble particles were removed by centrifugation at 12,000 rpm for 30 min. The supernatant was collected and delipified by adding 400 µL chloroform, rigorous vortexing, and centrifugation at 12,000 rpm for 5 min. The lipid phase at the interphase was discarded. The supernatant and the chloroform phase were combined, lyophilized, and subsequently stored at −20°C until further use. For dot blotting onto PVDF membranes, freeze-dried samples were resuspended in 200 µL of 8 M urea.

Protein extraction From FFPe histopathological slides
Antigen retrieval from FFPE slides was performed as described earlier (22). In brief, tissue samples were washed directly on the glass slides using a washing chamber with three alternating washing steps using xylene, followed by EtOH (2 min each). The tissue section was then detached with a sharp blade, transferred into a sample tube and subsequently 200 µL buffer (4% SDS in 0.1 M Tris-HCl, pH 8, 0.1 M dithiothreitol) were added. The suspension was sonicated on ice for 15 s at level 1.5 (Branson Sonic Power Company, Danbury, CT, USA) and subsequently incubated in a thermoshaker at 99°C at 600 rpm for 60 min. The sample was allowed to set to room temperature before it was centrifuged at 2,000 g for 20 min. The supernatant was transferred into a new vial and dried in a SpeedVac concentrator (Thermo, Bremen, Germany). Prior dot blotting onto PVDF membranes, samples were dissolved in 20 µL 8 M urea.

glycan release
The extracted proteins/peptides were blotted onto a PVDF membrane (0.2 µm pore size, Millipore, Tullagreen, Ireland) in five 1 µL steps allowing the droplet to dry between each blotting step. N-and O-glycans were sequentially released, reduced, and purified as described in earlier work (23).
In brief: protein/peptide sample solution was dot blotted onto a PVDF membrane. The membrane was washed in methanol and water for 15 min each. Spots were cut out and blocked for 5 min in 100 µL 1% polyvinyl pyrrolidone solution (PVP40, v/v). PVP was removed, and the membrane spots were washed three times with water. N-glycans were released by addition of 3 U of PNGase F (Roche, Mannheim, Germany) in 10 mM NH4HCO3. After removal of the N-glycans the O-glycans were released by β-elimination in 0.5 M NaBH4 in 50 mM KOH. N-glycans were reduced for 3 h at 50°C in 20 µL of 1 M NaBH4 in 50 mM KOH. The reaction was quenched by the addition of 2 µL glacial acetic acid.
N-and O-glycans were desalted using filter TopTip (Glygen, Columbia, MD, USA) filled with Dowex cation exchange resin (Dow, Midland, USA). The remaining borate was removed by several rounds of co-evaporation using 100 µL methanol in the SpeedVac concentrator.
neuraminidase Digest for Pgc-lc esi-Ms/Ms analysis The glycan sample was reconstituted in 10 µL 25 mM ammonium acetate buffer (pH 5.5) and 25 U of a α2-3-specific neuraminidase (NEB, Ipswitch, MA, USA) were added. The sample was incubated for 3 h at 37°C. The reaction was quenched by heating to 70°C for 5 min followed by a carbon clean up as described earlier (23).

glycopeptide sample Preparation
Patient samples were pooled for glycoproteomics analysis, and the peptide/protein concentration was estimated by measuring the absorbance at a wavelength of 280 nm in a Nanodrop ND-1000 spectrophotometer (Thermo Fisher Scientific) according to the manufactures instructions.
Proteins from pooled snap frozen biopsies were digested with trypsin using two approaches: in liquid as well as after having been blotted onto PVDF membrane. Pooled FFPE samples were digested on PVDF membrane. Peptide recovery from the PVDF membrane was achieved by three extractions with 50% ACN, one with 100% ACN followed by a final extraction with 5% formic acid (FA). The steps were repeated, and the extracts were combined. Glycopeptides were enriched using self-packed zwitterionic hydrophilic interaction liquid chromatography stagetips (10 µm particle size, 100 Å pore size, SeQuant AB, Sweden). The column was washed three times with 0.1% trifluoroacetic acid (TFA) and equilibrated three times with 80% ACN/0.1% TFA. Samples were reconstituted in 10 µL 0.1% TFA and brought to a concentration of 80% ACN/0.1% TFA by slowly adding 40 µL of 100% ACN/0.1% TFA. Each sample was applied twice to the column. Samples were washed twice with 50 µL of 80% ACN/0.1% TFA and eluted twice with 50 µL of 0.1% TFA followed by 50 µL of 80% ACN/0.1% TFA.
Before liquid chromatography (LC)-MS analysis the sample were run through a C18-ZipTip (Merck Millipore, Tullagreen, Ireland) for sample clean up and to avoid sample overload. The amount of protein in the eluted fraction was assumed to be equal to the 5 µg binding capacity of ZipTip. The eluted fraction was reconstituted in 10 µL 0.1% FA, and 3 µL was injected for LC-MS analysis.

Porous graphitized carbon (Pgc)-nanolc esi-Ms/Ms glycomics
Porous graphitized carbon-nanoLC ESI-MS/MS glycomics was performed using an amaZon ETD speed ion trap (Bruker, Bremen, Germany) coupled to an Ultimate 3000 UHPLC system (Dionex/Thermo). The deprotonated ions were detected in negative ion mode, and the instrument was tuned to perform CID fragmentation on the three most intense precursors of each MS scan. An m/z range from 350-1,800 was used for data-dependent precursor scanning. MS as well as MS/MS data were recorded in the instrument's "ultra-scan mode." Detailed information on the MS settings is found in Table S3 in Supplementary Material. N-glycans were loaded onto a PGC precolumn (Hypercarb™ KAPPA 30 × 0.32 mm, 5 µm particle size) and separated on an analytical PGC column (Hypercarb™ PGC Column, Thermo, 100 mm × 75 µm, 3 µm particle size). The samples were loaded onto the precolumn at a flow rate of 6 µL/min in 98% buffer A (10 mM NH4HCO3). The analytical column was equilibrated in 3% buffer B (10 mM NH4HCO3 in 60% ACN) at a flow rate of 1 µL/min. Buffer B increased from 3 to 15.8% (5-6 min), then to 40.2% B (6-67.25 min) and then to 55% B (67.25-78 min). The concentration was further increased to 95% B over 1 min and held at 95% B for 5 min (79-84 min) while the precolumn was flushed with 90% buffer C (10 mM NH4HCO3 in 90% ACN) at a flow rate of 6 µL/min before both columns were re-equilibrated for 5 min in 98% buffer A before the next sample was loaded. Generally, the MIRAGE guidelines were followed (24,25).

xcge-liF analysis and sample Preparation
As glycans need to be unreduced so that the fluorescent label can be attached via reductive amination for subsequent xCGE-LIF analyses only N-glycans were analyzed by this method. Sample preparation and xCGE-LIF measurement was performed as described earlier (21,26,27). A detailed description of sample preparation and xCGE-LIF measurement conditions is listed in Supplementary Material.

lc-Orbitrap glycoproteomics
Glycopeptide analyses were performed in the positive mode on an Orbitrap Fusion™ Tribrid™ mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA) coupled to an Ultimate 3000 UHPLC system (Dionex/Thermo). The instrument was tuned to perform HCD-product dependent (pd)-CID MS 2 fragmentation, namely additional ion-trap CID MS 2 was only triggered when glycopeptide oxonium ions (204.0867 and 366.1395) were detected in the HCD analyses. An m/z range from 400 to 2,000 Da was used for data-dependent precursor scanning. Peptides were directly loaded onto the analytical column (Acclaim PepMap RSLC C18, 2 µm, 100 Å, 75 µm × 25 cm) in 100% buffer A (99.9% v/v water, 0.1% v/v FA). A 90-min gradient of increasing buffer B (99.9% v/v ACN, 0.1% FA) was developed at a flow rate of 500 nL/min: 5% at 2 min, 35% at 62 min followed by an increase to 85% buffer B at 65 min. The column was held at 85% for another 10 min before it was re-equilibrated in 100% buffer A for 15 min. More detailed information on all instrument settings can be found in Supplementary Material.

glycan analysis and Quantification
Mass spectrometry N-and O-glycan spectra were evaluated manually by Compass DataAnalysis version 4.2 (Bruker). The data were then loaded into QuantAnalysis version 2.2 (Bruker) where quantitation was performed on extracted ion chromatograms (XIC) including all detected charge states of the respective glycan. A smoothing width of 3 s (Gauss algorithm) was applied on each XIC.
Data from xCGE-LIF aquisition data was analyzed using the Java-based glycan analysis software glyXtool™ (glyXera, Magdeburg, Germany). Data were normalized to an internal standard, transforming electropherograms to "N-glycan fingerprints" (see Figure S9 in Supplementary Material). Automated peak picking, integration, and relative quantification were performed, followed by N-glycan assignment to peaks via migration time matching to an in-house N-glycan database. N-glycan sequences and linkages were confirmed by exoglycosidase digests (see Supplementary Material).

statistical evaluation
Statistical analysis was partly performed using "R" version 3.0.2 (2013-09-25) using the "gplots" package 1 version 2.16.0 as well as the "pROC" package 2 version 1.9.1. The heatmap was generated using the logarithmic relative abundance values to calculate the Euclidean distance matrix and complete linkage clustering. Values in the heatmaps were scaled individually for each N-glycan structure. Statistical test were performed with performed with SigmaPlot software version 13.0. Samples were tested for normal distribution by Shapiro-Wilk test, significance was determined by t-test or Mann-Whitney test (if samples were not normal distributed). p-Values below or equal to 0.04 were considered significant. Bean and violin diagrams were created with the free online tool BoxPlotR (Version 3.1 and shiny-server 1.2).

resUlTs
A detailed workflow overview is described in Figure S1 in Supplementary Material.

healthy human skin glycoproteins carry Mainly sialylated N-glycans
Skin biopsies obtained from healthy skin areas of 14 BCC patients were used to establish the glycome reflecting healthy human skin. We used a PGC-nano-UHPLC ESI-MS/MS-and the orthogonal xCGE-LIF-based glycomics workflow. Both approaches allow for the separation of glycan structure isomers enabling differentiation of features such as α2-3, α2-6 linkages of sialic acid or core and arm fucosylation (Figures S6 and S7 in Supplementary Material) (21,26,(28)(29)(30). The human skin N-glycome consisted of 107 identified structures that were very similar within the analyzed patient cohort ( Figure 1A; Table S1 in Supplementary Material). Expected patient-specific variability  (Table S1 in Supplementary Material). (B) Relative N-glycan class abundances found in healthy skin biopsies. Sialylated and fucosylated structures were the major components representing the human skin N-glycome. (c) Relative abundances of different structure features found on sialylated N-glycans. Blue bars represent sialylated N-glycans with and without core fucose depending on their sialic acid linkage, showing that core fucosylation was a more abundant feature on N-glycans carrying one or two α2-3 linked N-acetylneuraminic acid (NeuAc) residues (if both, α2-3 and α2-6 linkages were present on one N-glycan, this N-glycan was considered in both linkage categories). Overall, α2-6-linked NeuAc was a slightly more abundant feature compared with α2-3-linked NeuAc. Most of the sialylated N-glycans carried two NeuAc residues. Tri-antennary species were below 3% and rather low abundant using porous graphitized carbon, while multiplexed capillary gel electrophoresis with laser induced fluorescence detection (xCGE-LIF) detected slightly higher levels of tri-and tetra-antennary N-glycans (see Supplementary file "xCGE-LIF quant.xlsx" in Supplementary Material).
in individual N-glycan abundances was in particular found for the high abundant N-glycans ( Figure 1A). Biantennary, sialylated N-glycans were the dominating structures, comprising on average about 82% of the human skin N-glycome ( Figure 1B). Human skin N-glycans exhibited an equal distribution of α2-3and α2-6-linked N-acetylneuraminic acid (NeuAc) residues ( Figure 1B), which is in contrast to human serum where N-glycans are mainly carrying α2-6-linked NeuAc on the non-reducing end (23,31). Oligomannose type (5%), hybrid type (3%), bisected (2.6%), and paucimannosidic (1.4%) N-glycans overall were just minor components of the human skin N-glycome (Table S1 in Supplementary Material). Overall, the data between the PGC-LC ESI-MS/MS-and the xCGE-LIF-based glycomics were in good agreement with respect to identified N-glycan structures, sialic acid and fucose linkage, and relative quantitation.

core Fucosylation is Markedly increased on α2-3-neuac carrying N-glycans
Approximately half of all detected human skin N-glycans were fucosylated with the fucose residue being almost exclusively attached to the N-glycan core. While this modification occurred roughly on a quarter of all α2-6 sialylated N-glycans it was present on 75% of all α2-3 sialylated N-glycans ( Figure 1B). Our data indicate that in human skin core-fucosylated N-glycans are the preferred substrate for α2-3 sialyltransferases, while α2-6 sialyltransferases appear to prefer non-core-fucosylated N-glycan substrates ( Figure 1C). Roughly 90% of neutral N-glycans were core-fucosylated, indicating that core fucosylation occurs before sialylation and that a competitive interplay between α2-6 sialyltransferase and core-fucosyltransferase activity might take place in the Golgi, subsequently resulting in the observed skin N-glycome patterns.
The healthy human skin O-glycome consists of simple core 1 and 2 Type O-glycans After the enzymatic removal of N-glycans O-glycans from nine healthy skin samples were released by reductive β-elimination. With just 13 detected structures of the core 1 and core 2 type, the skin tissue O-glycome was clearly less diverse (Figure 2). Sialylation was a major structural feature with roughly 96% of the glycans carrying at least one NeuAc residue ( Figure 2B). Three structures (ID 5; ID 3 and ID 12, Table S2 in Supplementary Material) contributed to more than 60% of the total skin O-glycome. Interestingly, the presence of a di-sialic acid residue on a core 1 O-glycan was confirmed on a triply sialylated O-glycan (ID 6, Figure S2 in Supplementary Material). Treatment with α2-3-NeuAc-specific neuraminidase removed a single NeuAc residue, but did not result in the loss of the diagnostic m/z = 581.05 B-ion signal that is indicative for a NeuAc-NeuAc disaccharide fragment ( Figure S2 in Supplementary Material). These exoglycosidase digestion data also indicated that the di-sialo epitope was attached in the C6 position to the reducing end core GalNAc. Its resistance to α2-3-specific neuraminidase digestion also indicated that these two NeuAc residues are linked via an α2-8 or α2-9 linkage.
Oligomannose N-glycan levels increase in Bcc The results obtained for healthy skin were compared with the glycome of BCC from the same patients to minimize effects derived from individual variations. Relative abundance was assessed for 65 most abundant N-glycans isolated from snap frozen biopsies. Representation of the data by volcano plot show several N-glycan structures that significantly changed in cancer ( Figure 3A). Oligomannose type N-glycans exhibited the highest fold change (2×) followed by an α2-3/2-6-sialylated structure (ID 74), a bisected, α2-3/2-6-sialylated (ID 48), and a paucimannosidic N-glycan (ID 36). By contrast, the abundances of three complex singly sialylated N-glycans (ID 23, 30, and 32) decreased in BCC compared with healthy skin. Virtually all oligomannosidic  (Table S1 in Supplementary Material). Based on the N-glycan data, the algorithm classified the majority of the healthy and tumor skin samples into two major groups. Two glycan clusters (marked by green and black bar) were observed in particular exhibiting increased levels for oligomannose type N-glycans. structures showed increase in abundance resulting in the whole class being differentially expressed in BCC ( Figure 3C). A consistent change of all members within other major glycan classes such as for instance paucimannosidic structures (ID 32) could not be observed (Figure 4; Figure S3 in Supplementary Material). Furthermore, an unsupervised cluster analysis was performed to evaluate if the glycan data allow distinguishing between healthy and BCC skin tissue ( Figure 3B). Three major patient groups became apparent: 2 groups of 12 samples each reflected largely healthy (10 healthy/2 tumor) and tumor skin samples (11 tumor/1 healthy), while the third group consisting of four samples was derived from tissue biopsies classified as N-glycans such as Man9 + Glc, Man9, and Man8, which are also associated with early stages of glycoprotein biosynthesis, exhibited a stronger increase compared with structures representing the next processing stages (Man7, Man6, and Man5).
To determine the distinguishing power of the individual glycan species, we calculated the area under the curve of the receiver operating characteristic (ROC, Figure S8 in Supplementary Material) of each of the glycans showing significant abundance differences. Oligomannosidic N-glycans ID 8, 9, 10, and the sialylated complex N-glycan 74 exhibited an AUC value above or equal to 0.9 and might provide either alone or in combination valuable diagnostic targets (Figure 3D).
Bcc N-glycan signatures are confirmed in FFPe Tissue sections These initial glycomics results obtained for the tissue biopsies were further validated using an independent sample cohort consisting of FFPE histopathological tissue representing corresponding healthy skin and BCC sections from 20 patients. The total number of identified N-glycans was lower in the FFPE samples. The results obtained for the FFPE sample cohort confirmed the previous findings of increased oligomannose N-glycan levels in BCC. Fourteen out of 20 patient specimens exhibited an average 1.5-fold upregulation of Man9, Man8, Man7, and Man6 in BCC tissue ( Figure S4 in Supplementary Material).
Similar to earlier reports (32), minor differences in relative abundance were detected for individual N-glycans when comparing the results from fresh tissue biopsies and FFPE histopathological tissue sections. While the relative abundance for α2-3 doubly sialylated fucosylated N-glycans (ID 76) was increased, the levels for α2-6 doubly sialylated biantennary non-fucosylated N-glycans (ID 56) decreased. In summary, in 70% of the analyzed BCC FFPE tissue sections an increase in oligomannose N-glycan levels was confirmed.
xcge-liF analyses confirm N-glycan alterations in Bcc identities were determined by correlation of the migration time with an in-house database and confirmed using exoglycosidase digestion. Notably, the number of identified N-glycans was only slightly lower in the xCGE-LIF analyses. The xCGE-LIF analyses confirmed that significantly higher oligomannose N-glycan levels were present in BCC tissue (1.4-fold increase in fresh and 1.3-fold in FFPE tissue). Overall, the abundance levels for the individual N-glycan structure classes were comparable between MS-and xCGE-LIF-based analytical approaches. However, the ratio found for neutral and sialic acid containing N-glycans showed a notable difference (Figure 4B). While roughly 16% less sialylated N-glycans (66%) were found in xCGE-LIF the amount of neutral structures was increased (25%). Partial desialylation may occur during the acidic labeling process (33), which could in fact account for the observed reduced amounts of sialylated N-glycans. On the other hand, xCGE-LIF detected some tri-and tetra-antennary N-glycans that were not detectable by the PGC-LC ESI-MS/MS setup used in this study, clearly showing the importance of having orthogonal glycomics approaches available for a comprehensive analysis (Table S1 and Supplementary file "xCGE-LIF quant.xlsx" in Supplementary Material).

core 2 Type O-glycan levels are increased in Bcc
The BCC tissue O-glycan profile showed significant level changes for various structures (Table S7 in Supplementary Material). Core 1 type O-glycan levels were reduced while core 2 type O-glycans experienced upregulation from 38% in healthy tissue to 53% in BCC ( Figure 5). As a consequence, the increase of core 2 type O-glycans affected the relative abundance of the NeuAc linkages. Levels of α2-3-NeuAc were increased while α2-6-NeuAc levels were reduced since the latter linkage was exclusively found on core 1 type O-glycans (including the most abundant structure ID: 5). Other structural features such as levels of total sialic acid, fucosylation or sulfate did not exhibit significant changes in BCC. Increase of core 2 type O-glycans in BCC was also confirmed in all of the 20 analyzed FFPE tissue sections ( Figure S5 in Supplementary Material).

Oligomannose and α2-3 sialylated N-glycans are affected in scc
The results obtained for the BCC-associated glycome changes encouraged us to apply our glycomics methods to FFPE histopathological tissue sections obtained from 15 SCC patients to compare possible protein glycosylation signature differences between the two cancer types. To identify potentially up or down regulated glycan structures we generated a heatmap based on the 40 most abundant N-glycans ( Figure 6A). Clusters indicated structures experiencing level changes in SCC tumor tissue. In SCC oligomannose carrying N-glycans showed increased levels, while α2-3-NeuAc N-glycans were less abundant compared with healthy tissue. With an average 1.56-fold upregulation in tumor samples, oligomannose N-glycans Man9, Man8, Man7, and Man6 showed a similar trend in SCC as demonstrated for BCC. The total abundance of sialylated N-glycans was reduced from 70% in healthy tissue to 64% in SCC (Figure 6B). This effect was mainly due to a significant reduction of α2-3 sialylated N-glycans (from 36% down to 26%) while the levels of α2-6 sialylated N-glycans remained essentially unaffected.

O-glycan alterations in FFPe histopathological Tissues of scc
The SCC O-glycome could be determined for 10 patient samples revealing a trend similar to the one observed for BCC where core 2 type O-glycan levels were increased, albeit to a lesser extent. Core 1 type O-glycans remained the most abundant structures in both healthy and tumor tissue (Figure 7). Other structure features such as sialic acid linkage or fucosylation did not show any significant alterations. glycoproteins associated With the gO-Term "Binding" are the Major glycoproteins identified in skin Tissue To gain insight into the glycoproteins carrying oligomannose structures as well as sialylated glycan structures, we performed a qualitative glycoproteomics analysis. Tryptic (glyco)peptides generated from the pooled protein extracts were split into two  fractions: one non-enriched, but enzymatically deglycosylated fraction and one fraction containing glycopeptides enriched by HILIC. (Glyco)peptides were analyzed by reversed phase nano-LC-MS/MS, whereas glycopeptide data for the HILIC fraction were acquired under HCD-pd-CID mode.
In the non-enriched fraction, over 4,000 proteins were identified, of which 407 are listed as glycoproteins in the UniProt database. The majority of all identified proteins are located in the cytoplasm, nucleus, and in membranes, whereas the glycoproteins are particularly located in membranes, the ER or found secreted ( Figure 8A). Glycopeptides were identified in the HILIC enriched fraction using Byonic software, and a library based on the 407 initially identified glycoproteins. Only glycopeptide hits exhibiting a score greater than 100 were considered, while glycopeptide spectra with a score below 100 were only considered after manual validation if other glycoforms of the same peptide were identified with a score above 100. Overall, glycopeptides from 78 glycoproteins were identified (Table S8 in Supplementary Material). The intention of this experiment was to identify glycoproteins including their sitespecific glycosylation, but it was not possible to quantitatively distinguish between healthy and tumor tissue since the sample material was limited. The majority of the identified glycoproteins were involved in binding activities such as receptor binding, protein complex binding, or cell adhesion molecule binding ( Figure 8B).
Among the 78 glycoproteins, 33 were found to be sialylated ( Table 2). Glycoproteins most commonly detected with core fucose, and one and two terminal sialic acids were again decorin and two isoforms of collagen VI. While sialylated N-glycans from collagen were only detected together with core fucosylation, decorin was also found to be singly sialylated without core fucosylation. The most frequently detected, non-fucosylated sialylated glycoproteins were vitronectin, alpha-1-antitrypsin, torsin-1A-interacting-protein-1, and metalloreductase STEAP3 ( Table 3).

DiscUssiOn
Little is known about the human skin glycome as detailed glycomic and glycoproteomic investigations have not been undertaken. Skin cancer studies were based on cell culture material (34,35), limiting the value of information that can be obtained from these studies as the skin glycoproteome is likely to be highly dependent on the developmental stage, cell environment and the organism. Skin consists of various layers with different organism-specific properties and a complex extracellular matrix network. Outcomes of glycomics studies in an artificial environment as present in cell culture might result in detection of culture condition artifacts that deviate from the actual in vivo glycome. We therefore focused on tissue biopsies to investigate diseaseassociated glycoproteome signatures.
More than 82% of the N-glycans found in human skin carried at least one or more NeuAc residue. In skin tissue, α2-3 and α2-6-NeuAc linkage to N-glycans occurs in almost equal levels. The high occurrence of sialic acid containing N-glycans was particularly interesting since an earlier study (36) using a hydrazine-capture MALDI-TOF-MS approach reported mainly neutral and singly sialylated (~12% epidermis/~35% dermis) species as well as lower amounts of doubly sialylated N-glycans (~5% epidermis/~20% dermis) to be present in human skin. Some of the sialic acids could have been lost during the vari ous preparation steps. Uematsu and coworkers separated dermis and epidermis by heat treatment, which could also have contributed to the reduced global sialylation levels. Since our results clearly indicated that the global N-glycome was overall very comparable between the individual patient samples, interperson variations are an unlikely explanation for the sialylation level differences. Cancer-associated glycosylation differences were generally observed for both, BCC and SCC. The extent of observed N-glycan level changes showed some patient-specific variations and could not be detected in six FFPE sample sets. Such variation is expected for complex biological samples as numerous individual factors can influence the outcome of such studies. Healthy skin biopsies were taken in direct proximity to the tumor tissue. MALDI imaging studies have revealed that tumor-associated molecular differences can in some cases extend in a decreasing manner up to 1.5 cm into the surrounding healthy tissue (37). Thus, the observed N-glycosylation changes could be even more prominent if compared with more distal tissue. The observed N-glycan level differences were more prominent in the biopsies than in FFPE prepared samples. This could be due to the overall lower sample amount that was obtained from FFPE histopathological slides and the resulting general lower glycan yield. Moreover, the frozen biopsies were specifically prepared for this study by "punch excision" after removal from the patients' skin whereas the FFPE tissue was prepared using a routine dermato-pathological workup possibly containing more surrounding tissue. The different sampling methods for the tissue biopsies and the FFPE tissue slides could have influenced the ratio and amount of tumor and surrounding tissue. Despite the fact that overall the same pattern of N-glycan level changes was observed in FFPE tissue, it could account for the fact that these differences were less profound in the FFPE sample set.
Altered glycosylation is a hallmark of cancer (38), which raises the questions whether such alterations are simply side effects of the malignant transformation, if cancer cells actively benefit from this development or even both. Answers to these questions are likely to be specific for each cancer type. In this study, oligomannose type N-glycan levels were found to be increased in BCC and SCC both being epithelial cancers. The same pattern was also reported to occur in breast cancer patient (39) as well as in mouse sera of implanted head and neck tumors which also arise from epithelium (40). Recently published mass spectrometry imaging studies reported an increase of the Man8 N-glycan in tumor tissue from ovarian cancer (41) and hepatocellular carcinoma (42). In the ER oligomannose N-glycans are crucial for N-glycan processing and protein folding. Increased cell division and hence protein expression in cancer tissue can lead to incomplete N-glycan processing manifesting itself in incomplete trimming of oligomannose structures. As glycosylation is protein-and site-specific (43) changes from complex to oligomannose structures can significantly impact the stability, binding affinities or functions of the respective glycoprotein (44,45). Oligomannose N-glycans are also known ligands for C-type lectins such as DC-SIGN and langerin (46). Both are found on antigen presenting immune cells (47). Thus, an increase in oligomannose structures would suggest an upregulation of immune responses; however, studies have shown that dendritic cells in non melanoma skin cancer seem to be impaired in their ability to stimulate T-cell proliferation (48,49). This condition could then allow an increased presence of these structures without risking augmented immune attention.
Our glycoproteomics study identified the presence of oli gomannose N-glycans on several extracellular matrix-associated glycoproteins such as laminin, serpin, and microfibrilassociated glycoprotein-4. Oligomannose N-glycans on extracellular matrix glycoprotein are well known to participate in cell adhesion processes and can represent ligands for monocyte adhesion in endothelial cells, which associates these with inflammation processes at the tumor site (50). While upregulation of immune responses would rather point toward a reaction in response to tumor progression, cancer-associated inflammation could be beneficial for the tumor by driving its progression (51).
In agreement with literature, we found complement C3 glycosylated with oligomannose structures (52). Complement C3 usually plays a crucial role in the immune system's defense mechanisms against malignant cells and pathogens by initiating the membrane attack complex (53). By contrast, several studies have also demonstrated a tumor promoting role of complements. Complement activation may drive chronic inflammation, promote an immunosuppressive microenvironment, induce angiogenesis, and activate cancer-related signaling pathways (54). Recently, significant elevated mRNA levels of complement C3 have been reported to be associated with the progression of SCC (55). Among others complement activation can be triggered by oligomannose N-glycans via the mannan binding pathway (56). Therefore, it might be conceivable that tumor-associated increased oligomannose N-glycan levels could derive from an increased expression of oligomannose carrying glycoproteins and the increased presence of complement C3. Both BCC and SCC show increased core 2 O-glycan levels. Upregulation of core 2 O-glycans can originate from increased core 2 β-1,6-N-acetylglucosaminyltransferase (C2GnT) expression, leading to an augmented conversion of core 1 to core 2 O-glycans. Reducing the amount of core 1 and increasing core 2 O-glycans changes options for subsequent modifications potentially reducing molecular and cellular interactions while giving rise to others. Increased C2GnT expression has also been described for colorectal and lung carcinomas also correlating with increased occurrence of Lewis X structures (14,57,58). Another study reported upregulation of C2GnT and increased core 2 glycosylation in close relation to bladder cancer progression and the cancer cells ability to evade NK immunity (14,59).
In our study differences in the sialome were only detected in SCC. A previous study compared the abundance of sialic acid linkages between SCC and BCC by lectin histochemistry and reported increased α2-6-NeuAc levels in SCC (60). As both α2-6 and α2-3-sialyltransferases share common substrates, α2-6 sialyltransferase overexpression can result in decreased α2-3sialylation. We detected reduction of α2-3-NeuAc in SCC only accompanied by a non-significant increase of α2-6-NeuAc. Our data point toward an α2-3 sialyltransferase down regulation as the causative effect behind the change in NeuAc linkage levels. Differences in the two studies might originate from the fact that lectin assays consider not only N-and O-glycans but the sum of all glycan bearing structures such as glycolipids and glycosaminoglycans. Lectin assays are a powerful tool and commonly utilized but have to be carefully evaluated as lectines often have binding preferences to not only one but several glycan structures (61)(62)(63).
As changes in N-glycan-associated α2-3-NeuAc levels did not have any detectable effect on O-glycosylation it seems likely that the observed changes only affect specific proteins. Our N-glycome data revealed a link between α2-3-NeuAc linkage and core fucosylation. Interpolating these findings to the glycoproteomics data it appears that in the skin extracellular matrix-associated proteins like collagen, decorin, vitronectin are more likely to carry α2-3-NeuAc modified N-glycans. The observed reduction of α2-3-linked NeuAc could thus have a considerable impact on the functionality of these proteins and thus on the entire intercellular communication of tumor cells.
There is a long history in studying sialylation (16) and associated transferases (64) in the context of cancer. Being the terminating monosaccharides they are prominently exposed on the most outer part of the glycan and vital for numerous specific interaction events with lectins such as siglecs (sialic acid binding immunoglobulin-like lectins) or selectins (65). Many of these interactions seem to be related to immunoevasion pathways by inhibiting activation of immune cells (66,67) or promote metastasis progression (68). Furthermore, increase in sialic acid levels also results in increase of negative charges on the cell surface, which is believed to decrease cell adhesiveness thus driving metastasis (69). Interestingly, cell culture studies of the more aggressive malignant melanoma reported increase in sialylated β1-6-branched N-glycans (70) as well as increased α2-3 linked NeuAc levels correlating with aggressiveness (71). With BCC and SCC being less aggressive and metastatic than melanoma the absent overexpression of α2-3-linked NeuAc and β1-6 branches in our study hence, could reflect some of the potential factors accounting for their lower aggressiveness. Many changes in glycosylation have been reported focusing on specific glycoproteins. Global glycomics might be unable to pick up such proteinspecific changes, in particular if just subtle changes are present that affect specific glycoproteins. Future studies investigating protein and protein site-specific glycosylation patterns will be vital in uncovering the role such altered glycosylation signatures have on cancer onset and progression, which will inevitably lead to the identification of novel diagnostic, prognostic and treatment strategies.
cOnclUsiOn Our detailed study of the skin N-and O-glycome provided the basis for investigations of cancer-associated N-glycan changes. Oligomannose N-glycan levels were elevated in BCC and SCC. Expression of α2-3 sialylated N-glycan decreased in SCC while core 2 O-glycans increased in both BCC and SCC. Several glycoproteins were identified that might potentially be affected by the N-glycosylation changes. The exact biological causes for the observed upregulation together with their involvement in disease onset and progression cannot be explained yet. Our findings represent, however, a basis for studies toward a better understanding of the various molecular aspects of nonmelanoma skin cancers.

eThics sTaTeMenT
The study was approved by the institutional review board of the medical faculty at the University of Leipzig (No. 127-11-18042011).
aUThOr cOnTriBUTiOns UM, DK, PS, JS, and SG planned the project. UM and DK wrote the manuscript. JS, SG, and HV provided all samples. UM performed sample preparation and glycan analysis. UM, C-WK, and K-HK performed the glycopeptide analysis and sample preparation. RH and ER performed xCGE-LIF analysis. UM and FS performed statistical analyses. All the authors contributed during the review process with their individual input and ideas.