A Robust and Versatile Automated Glycoanalytical Technology for Serum Antibodies and Acute Phase Proteins: Ovarian Cancer Case Study

: The direct association of the genome, transcriptome, metabolome, lipidome and proteome with the serum glycome has revealed systems of interconnected cellular pathways. The exact roles of individual glycoproteomes in the context of disease have yet to be elucidated. In a move towards personalized medicine, it is now becoming critical to understand disease pathogenesis, and the traits, stages, phenotypes and molecular features that accompany it, as the disruption of a whole system. To this end, we have developed an innovative technology on an automated platform, “GlycoSeqCap”, which combines N -glycosylation identify aberrant glycosylation. In our sample cohort, we exhibit improved selectivity and specificity over the currently used biomarker for ovarian cancer, CA-125, for early stage ovarian cancer. This technology will establish a new state-of-the-art strategy for the characterization of individual serum glycoproteomes as a diagnostic and monitoring tool which represents a major step towards understanding the changes that take place during disease. significant difference (HSD) ANOVA. Peak areas compositional data (presented a % of the total area under the graph) CSC the was controlled for age confounder linear regression model (ANCOVA). Correction for multiple testing was performed using Benjamini-Hochberg procedure with a 10% false positive rate. This study presents the development of an elegant glycoanalytical platform for the detailed characterisation and investigation of a sequence of glycoprotein N-glycosylation: antibodies IgG, IgM and IgA and acute phase proteins Trf, Hpt and A1AT. In the course of this analysis, besides achieving a detailed glycoprofile for each glycoprotein through the identification and profiling of more than a hundred glycan structures, we were able to identify novel glycan motifs and traits that have not been outlined to date. Its utility in the context of ovarian cancer is highlighted-Trf and Hpt glycosylation can be exploited as biomarker tools for ovarian cancer and may play a cardinal role. More importantly, this comprehensive and reproducible glycoprofiling technology could provide significant insights and serve as a baseline for the identification of biomarkers and their regulation in a series of diseases.

identify aberrant glycosylation. In our sample cohort, we exhibit improved selectivity and specificity over the currently used biomarker for ovarian cancer, CA-125, for early stage ovarian cancer. This technology will establish a new state-of-the-art strategy for the characterization of individual serum glycoproteomes as a diagnostic and monitoring tool which represents a major step towards understanding the changes that take place during disease.

Introduction:
Glycosylation is the most common and complex type of post-translational modification and glycoscience is rightly recognised as a current frontier. Over 1% of total human genome codes for approximately 300 functional glycogenes known to exist in humans 1 . Post-translational modifications (PTMs) of a protein are critical late events in protein biosynthesis and glycosylation is one of the most important. It reflects the genome of the individual person and also the many other influences on the cellular pathways. In fact, the glycoproteome can be thought of as the final readout of the genome that confers function on the gene products. A range of human genetic disorders, including congenital disorders of glycosylation 2,3 , MODY type diabetes 4 , galactosemia 5 and muscular dystrophies 6 have been directly linked to or shown to involve faulty glycosylation. Cancer associated aberrant gene regulation also results in alterations of glycan structures and has been well studied both in the serum glycome and specific serum glycoproteins [7][8][9] . In addition, chronic inflammatory diseases display altered glycosylation 10 .
Immunoglobulin G (IgG) N-glycome is well characterised using liquid chromatography and mass spectrometry methods, [11][12][13] but N-glycosylation analysis of other glycoproteins has been studied to a far lesser extent. For example, N-glycosylation analysis of acute phase proteins, such as transferrin (Trf), alpha-1-antitrypsin (A1AT) and haptoglobin (Hpt), has been conducted in our laboratory 8,14 using complex methodologies such as isoelectric focussing (IEF) or 2D-gel electrophoresis for glycoprotein separation. In addition, N-glycosylation of serum immunoglobulin M (IgM) 15 and immunoglobulin A (IgA) 16 have been characterized previously using normal phase high performance liquid chromatography (NP-HPLC) a decade ago with only a proportion of the glycans identified, a limitation of the technology at the time.
More recently, IgA site specific glycosylation (N-and O-glycopeptides from IgA1) was elucidated in serum of patients with rheumatoid arthritis (RA) during pregnancy using matrix-assisted laser/desorption ionisation Fourier transform ion cyclotron resonance (MALDI-FTICR) mass spectrometry 17 . Of note, the team additionally performed released N-glycan analysis and could identify major N-glycans that were not identified in the characterized IgA1 N-glycopeptide fractions. As such, a detailed structural characterization of serum IgA N-glycome analysis remains to be undertaken. In a separate study, protein-specific differential glycosylation of immunoglobulins (IgG, IgM and IgA) were studied in serum of ovarian cancer patients 18 using multiple reaction monitoring on a triple quadrupole mass spectrometer. This technology presents similar limitations as to the total structural elucidation of the N-glycome, but is highly revealing, whereby the authors hypothesize that within the total serum glycome profile, which is largely dominated by the highest abundance proteins (such as antibodies IgG, IgM and IgA or acute phase proteins Trf, Hpt, A1AT) that protein-and site-specific glycosylation profiles will be likely to provide further insights into protein specific alterations in glycosylation of the glycans related to ovarian cancer as well as serve as more specific biomarkers for ovarian cancer than current tools 18 .
Although some advances have been made in the sample purification and target glycoprotein enrichment 19 , there is a growing necessity for the development of more robust, consistent, sensitive and versatile methods that can be automated and applied to personalized medicine approaches. To address this need, we optimised and automated a glycoanalytical technology to capture and glycoprofile six abundant individual glycoproteins by serial extraction-IgG, IgM, IgA, Trf, Hpt and A1AT. We performed a detailed analysis of the N-glycomes from these six glycoproteins using 50 µl of pooled normal human serum (NHS). Then we applied this technology to an ovarian cancer patient cohort, consisting of 7 healthy controls, 6 borderline and 21 metastatic ovarian cancer patients and tested its potential to be used as a diagnostic tool for earlier detection of ovarian cancer. This study is a follow-up of our previously reported sample preparation and chromatography technologies for glycan analysis 11,20,21 and biomarker discovery.

Experimental: Serum samples
Normal Human Serum (NHS) samples were used as described previously 20 Tables S9)   following ethical approval and obtaining informed consent. After allowing the blood to   clot for 30-60min, serum was obtained by centrifugation at 2,000g for 10min and stored at -80 o C until analysis.

Chemicals and Reagents
All chemical reagents and solvents were purchased from Sigma-Aldrich. Pre-packed Protein G (PTH93-20-02) 300µL tip columns obtained from PhyNexus Inc. Albumin

Phytip Glycoprotein Affinity Purification
Glycoproteins from 50μL of whole serum sample were captured in the following sequence: Albumin, Trf, IgG, IgM, IgA, Hpt and A1AT on the automated liquid handling station (Hamilton Starlet) in a 96-well format using the custom Phytips (Figure 1). In this method Phytip equilibration, capture, wash and elution refers to cycling a solution through the resin bed for a fixed number of aspirations and dispense steps termed cycles at defined flow rates. In a sequential fashion Albumin and the glycoproteins were captured from serum or from serum depleted of the preceding glycoproteins. For example, in the case of Trf, the Albumin depleted serum was used for the affinity chromatography. In the case of IgG affinity, the serum was depleted of both Albumin and Trf prior to IgG purification.

1D SDS-PAGE
Affinity purified glycoproteins were reduced and loaded on SDS-PAGE gels (NuPAGE® Novex 4-12% Bis-Tris) and separated in a XCell SureLock™ Mini-Cell (Invitrogen, Carlsbad, CA) for 90min at 120V using a MES running buffer. 5μg of total protein was loaded in each lane of the gel. Once electrophoretic separation was completed, proteins were visualized by staining with Coomassie blue staining solution followed by destaining with multiple changes of deionised water (Figure 1).

Automated glycoprotein denaturation and N-Glycan release
As described previously 22 , briefly N-glycan analysis was performed using a pooled serum sample from 100 apparently healthy male and female adult blood donors (U.K. Blood Transfusion Service). The antibodies IgG, IgA, IgM and the acute phase proteins Trf, Hpt and A1AT were purified from the pool using the respective affinity resin (Life Technologies) packed in Phynexus Phytips as previously described (in above section). On an automated platform, the glycoprotein samples were dispensed into two 384-well ultrafiltration plates (maximum loading: 60μg protein, 10kDa). Ultrafiltration was performed by centrifugation (3700g, 30min, room temperature) or on a vacuum manifold (> 25 in Hg vacuum, 30min). Denaturation buffer (25μL per well, 100mM sodium bicarbonate, 50mM dithiothreitol (DTT), 0.1% sodium dodecyl sulfate (SDS)) was dispensed into 2 × 384 well ultrafiltration plates. After 10min incubation at room temperature, the samples were mixed 10 times (mixing volume: 15μL, flow rate: 10μL/s) and transferred to a 384-well PCR plate (Armadillo). The plate was placed into a robotic incubation chamber at 95°C for 10min. The plate was removed from the incubator and equilibrated to room temperature for 10min. 1M iodoacetamide (IAA), 10μL, was dispensed into each well of the ultrafiltration plate and the samples (25μl) were transferred back into the 384 well PCR plate. The samples were mixed 5 times (mixing volume 20μL, flow rate, 10μL/s). After 10min incubation at room temperature, the ultrafiltration plate was stacked onto a 240μL collection plate (Corning block) and centrifuged (3700g, 30min, room temperature) and the supernatant was removed.
Next, 10μL of a 25mM sodium bi-carbonate solution was dispensed into each well and the ultra-filtration plate was centrifuged (3700g, 30min, room temperature) and the supernatant was removed.
Finally, 10μL of 25M sodium bicarbonate solution was dispensed into each well of the ultrafiltration plate (still stacked onto the PCR plate) and the assembly was centrifuged (3700g, 10min, room temperature). The released N-glycans in the PCR plate were subsequently fluorescently labelled. For glycan labelling, 5μL of glycan sample (PCR plate) was transferred to a 95μL Corning block and 11.6μL of AQC (3mg/mL MeCN) was added. 3μL of this crude mixture was directly injected into the UPLC system.
Alternatively, after the PNGase F release, glycans can be frozen and are stable at -20°C for labelling later.

Ultra-Performance Liquid Chromatography (UPLC)
As previously described 22  volume of 3μL prepared in 70% v/v MeCN was used throughout. Samples were maintained at 5°C prior to injection, and the separation temperature was 40°C. The FLD excitation /emission wavelength were λex = 245nm and λem = 395nm, respectively. The system was calibrated using an external standard of hydrolyzed and 2-AB-labeled glucose oligomers to create a dextran ladder, as described previously 11 .
A fifth-order polynomial distribution curve was fitted to the dextran ladder to assign glucose unit (GU) values from retention times (using Empower software from Waters).

Liquid Chromatography-Mass Spectrometry (LC-MS)
Online coupled fluorescence (FLR)-mass spectrometry detection was performed using a Waters Xevo G2 QTof with Acquity® UPLC (Waters Corporation, Milford, MA, USA) and BEH Glycan column (2.1 x 150mm, 1.7μm particle size). For MS acquisition data the instrument was operated in positive-sensitivity mode with a capillary voltage of 3kV. The ion source block and nitrogen desolvation gas temperatures were set at 120°C and 350°C, respectively. The desolvation gas was set to a flow rate of 800L/h. The cone voltage was maintained at 40V. Full-scan data for glycans were acquired over m/z range of 300 to 2000. Data collection and processing were controlled by MassLynx 4.1 software (Waters Corporation, Milford, MA, USA). The fluorescence detector settings were as follows: λexcitation: 245nm, λemission: 395nm; data rate was 10pts/second and a PMT gain = 20. Sample injection volume was 10μL (75% MeCN). The flow rate was 0.400mL/min (unless specified) and column temperature was maintained at 60°C; solvent A was 50mM ammonium formate (pH 4.4) and solvent B was MeCN. A 60min linear gradient was used and was as follows: 25-46% A for 35min, 46-80% A for 8min (flow rate at 0.2mL/min), 80-25% A for 27min. To avoid contamination of system, flow was sent to waste for the first 1.2min and after 55min.

Computational procedures
Statistical analysis: Variables age, menopause status, logCA125, C-reactive protein levels (CRP), protein titre and glycan peak areas between patients and controls were compared using a Tukey honest significant difference (HSD) test with ANOVA. Peak areas are compositional data (presented as a % of the total area under the graph) and therefore the constant-sum constraint (CSC) occurs. The CSC means individual variables do not vary independently, violating common assumptions upon which standard statistical analyses are performed. This was avoided by performing a log transform, log(Peaki)/(100-Peaki), on all peak areas 24  Therefore, correlated variables appear close together in the hieratical dendrogram produced by the clustering. were packed in a series of PhyNexus phytips and each serum sample was passed successively through each resin in a sequential order Trf, IgG, IgM, IgA, Hpt and A1AT.

Results and Discussion
The purified proteins, >98% pure as determined by 1D-SDS page, were then subjected to PNGaseF release, ultrafiltration and aminoquinoline carbamate (AQC) fluorescent labelling of the N-glycan pools using the Hamilton Starlet robotic system using a previously published protocol (Figure 1) 11 . These pools were subjected to hydrophilic ultra-high performance liquid chromatography (HILIC-UPLC) followed by exoglycosidase array glycan analysis according to the established procedures in our laboratory 23 . The UPLC chromatograms for the fluorescently labelled released Nglycans of affinity purified glycoproteins (IgG, IgM, IgA, Trf, Hpt and A1AT) from human serum and the corresponding released N-glycans from total human serum 21 are presented with annotations for the major N-glycans ( Figure 2).

Advantages and limitations of GlycoSeqCap technology
This technological advancement provides N-glycosylation information for IgG, IgM, IgA, Trf, Hpt and A1AT and serves as a template for many other glycoproteins of interest which may find use as a personalized medicine tool for future applications. A major advantage of the developed workflow is the ability to have multiplexed glycomics information from a single clinical source. In addition, the Phytips are reusable and show good reproducibility over three consecutive runs and after 1 month of storage.
However, one limitation regarding the extension of the technology to include other glycoproteins is the requirement for an affinity re sin with high selectivity and specificity for the targeted glycoprotein (e.g. antibodies or lectins) and in certain cases high amounts of biomaterial (in this case serum) may be required for lower abundant glycoproteins. In addition, the analyst must be cognisant of the purity of the targeted glycoprotein prior to enzymatic release of the N-glycans.
For example, a recent publication cautions the users to account for the ubiquitous presence of varying levels of other contaminating plasma glycoproteins 27 . In our study, we were careful to assess the purity of the glycoproteins by SDS-PAGE ( Figure   1) and systematically altered the number of capture, binding and wash steps to maximise purity prior to PNGase F release of N-glycans. As in the case of any affinity purification, it is possible that small traces of other glycoproteins have contributed to the N-glycan structures identified, albeit in tiny proportions.
The glycosylation data for six abundant glycoproteins in normal human serum aligns nicely with the total glycan pool identified for human serum previously ( Figure 2) 21,28 .
All major glycans from human serum are identified in one/more of the individual glycoproteins that were characterized. In a large part, this identification on an individual glycoprotein level was enabled on such a small amount of serum (50µL) due to the use of a fluorescent label AQC which shows a twenty fold increase in fluorescent detection compared to the traditional label, 2-AB for released N-glycans which we have previously described for human IgG previously 11,23 . Various high-throughput large-scale studies have been conducted on a serum N-glycome level since 2009 [29][30][31] .
These studies have yielded significant contributions to understanding the role of glycosylation in many diseases but cannot pinpoint the exact glycosylation processing pathways involved. On an individual glycoprotein level, large scale studies have focussed on IgG N-glycome only to date, 32 leaving plenty of scope for future investigations on an individual glycoprotein level. In addition, it may also be possible to link significant alterations from the serum N-glycome to a specific glycoprotein by extrapolation using the major glycans identified in this work.

Profiling and detailed N-glycan analysis of serum IgG, IgA and IgM
In order to execute glycoprofiling to compare glycosylation of two distinct populations  Table S1). They are in agreement with literature 11,32 (with the addition of a high mannose species (M7) not previously identified) and are presented in Supporting Table S6.
The equivalent data for serum IgM and IgA N-glycomes were split into 24 (M1-M24) and 25 (A1-A25) GPs respectively and are shown in Figure 3 (reproducibility data in Supporting  Table S2 and S3) and are presented in Supporting   Table S6.

Profiling and detailed N-glycan analysis of serum Trf, Hpt and A1AT
The acute phase proteins, Trf, Hpt and A1AT were purified from human serum by

Comparison of glycosylation between antibody classes: IgG, IgM and IgA
We probed the glycosylation of selected antibodies to gain an insight into the structurefunction relationship of antibody classes. One important point to consider is that serum glycoproteins are a combination of active components and waste products and this may complicate interpretations. Structural similarities and differences in antibody glycosylation of IgG, IgM and IgA are presented for pooled normal human serum (Figure 5a,b) respectively) and no outer arm fucose were identified for the serum antibody series.
This observation is consistent with other reports for serum antibodies 15,17,32 . IgA contains the highest abundances of bisecting GlcNAcs (52%) and galactose residues (93%) and a very small proportion of triantennary structures (0.9%) not observed for the other antibodies. IgM exhibits the lowest amount of biantennary structures (64%) and galactosylation (65%) but significantly higher amounts of high mannose structures (31%).

Acute phase protein glycosylation traits: Trf, Hpt and A1AT
Unlike the selected antibodies which share structural similarities in terms of their protein structure, the selected acute phase proteins (Trf, Hpt and A1AT) have less commonality and exhibit more varied glycosylation traits. These structural similarities and differences are presented for pooled normal human serum (Figure 5a,c). As for the antibodies, a high degree of galactosylation was observed (˃95%) for selected acute phase proteins but relatively low amounts of total fucosylation (˂20%) were identified, accounting for a combination of core fucose and the outer arm fucose residues (not identified in the selected serum antibodies in this study). Sialyl Lewis x

Glycosylation traits for functional studies
The quantification of glycosylation traits for a series of glycoproteins is presented here for the first time from a single source of human serum ( Figure 5). One can envisage this information as being a starting point for functional studies to investigate the specific role of glycosylation in our immune system and has the added advantage over  37,38 , and looking at levels of antibody fucosylation in human serum, the striking difference between IgG (93%) and IgM or IgA core fucosylation (58% and 28% respectively) may hint that serum IgA recruit effector cells for ADCC more efficiently compared to IgG or IgM. Importantly, it must be considered that we have measured only in human serum samples and this data does not necessarily reflect on a tissue/cell specific antibody glycosylation level. In addition to core fucosylation, we reported no outer arm fucose for the antibody classes. Noteworthy, outer arm fucosylation has been reported for secretory component (SC) N-glycans of IgA, but none for the corresponding J or H chain N-glycans 34 . It is known that serum IgA is predominantly monomeric and does not contain the J chain and the SC and as such we report no conflicting findings in this study. Another interesting differentiation between the glycosylation traits of the antibody isotypes is the high abundance of sialylation for IgA (85%) relative to IgM (60%) and IgG (22%). We propose that serum for serum IgM. This proportion of high mannose was higher than the value of 23% previously reported for serum IgM and the corresponding amount of galactosylation was lower (65%) compared to the literature value of 84% 15 . We suggest that advances in modern technology have allowed for a more accurate and precise assignment of Nglycans in our study-however it cannot be ignored that the human serum populations used in the two studies may have been different.
In contrast with immunoglobulins, which are mainly produced by B-cells, major plasma glycoproteins including Trf, Hpt and A1AT originate from hepatocytes, which express only very low levels of the FUT8 fucosyltransferase and thus contain a low percentage of core fucosylated glycans 32 . As anticipated, we observed lower levels of core fucose for the acute phase proteins in comparison to the antibody classes in our study ( Figure   5). Another interesting point relates to the increased complexity of the acute phase protein N-glycosylation compared to immunoglobulin classes ( Figure 5), with tetraantennary glycans present in these glycoproteins not isolated from the antibody classes. We cannot find a biological rationale for the reasons as to why the N-glycans are more branched in these acute phase proteins but it may be related to their inflammatory properties. Also, these glycoproteins contain SLex motifs. SLex are known to play a vital role in cell-cell recognition and other processes and are known to be overexpressed in cancer cells 39 . As such, this feature is particularly important in the context of our ovarian cancer cohort.

Glycoprofiling and statistical significance of selected glycoproteins in ovarian cancer
Having comprehensively assigned N-glycans and calculated derived glycosylation  Table S9) as a test group were analysed resulting in a total of 204 processed glycoprofiles. The profiles were split into glycan peaks (GPs) according to the individual glycoproteins (e.g. G1-G25 for IgG) and the values are presented as a proportion of total % peak area in Supporting Tables S10-S15 for glycoproteins IgG, IgM, IgA, Trf, Hpt and A1AT respectively. The derived glycosylation traits were also calculated and were derived as described earlier (Supporting Table S8 GPs, glycosylation traits and clinical parameters for the patients are presented in Figure 6 for the glycoprotein series. Boxplots (6c) and the major glycan (6b) for the statistically significant GPs/glycosylation traits are also presented.
From the glycoproteins selected, Trf shows the best discrimination between normal patients (n=7) and borderline (n=6) or metastatic patients (n=21). Most notably, Trf glycosylation can distinguish between normal and borderline samples using four individual GPs (T11, T16, T17 and T18), whereby CA125, the gold standard for ovarian cancer detection does not show the same statistical significant separation.
Fucosylation also stands out as a feature that may be exploited for this discrimination.
The GPs/glycosylation features of Trf cannot differentiate between metastatic and borderline samples in this cohort however, whereas CA125 does show a significance.
Several studies have presented evidence implicating Trf in ovarian cancer biology 41 .
In addition, alterations in Trf glycosylation in the context of cancer and other inflammatory diseases have been reviewed in the literature 36,42 . Taken together, Trf sialylation is often altered in disease states and is no exception in this study, whereby sialic acid (α2,6) is altered in the metastatic cancer cohort compared to normal controls. Since desialylated Trf has faster clearance, it may be an evolutionary tactic of bacteria/pathogens to contribute to oncogenesis. What is more striking in our study is the alterations of Trf fucosylation. Trf fucosylation has been shown to be dysregulated in diseases such as classical galactosemia 43 but to the best of our knowledge has not been reported in the context of ovarian cancer to date. Future studies are warranted to investigate further.
In addition, haptoglobin glycosylation is significantly altered in metastatic cancer (n=18) compared to normal patients (n=7) in this cohort by six distinct GPs (H2, H3, H11, H20, H21 and H22) as well as glycosylation features fucosylation and SLex motif.
This data is consistent with literature reports 44 , whereby the main glycosylation alterations of Hpt in cancer appear to be the presence of aberrantly fucosylated and sialylated structures as well as increased branching.

Discrimination and cluster analysis in ovarian cancer
A discrimination analysis was explored for the probabilistic classification of healthy  Table S23, S24 and S25 respectively). The individual models performed well with regards to distinguishing normal from borderline patients, with the most promising result using derived traits from Hpt (AUC=1.000, SEN=1.000 and SPE=1.000) ( Figure   7a). This finding, if validated could have major ramifications for early detection of ovarian cancer using Hpt glycosylation. Similarly, good discrimination was observed for normal vs metastatic patients. Again Hpt showed perfect discrimination for all peaks, derived traits and combinations thereof. However, CA125 also allows perfect discrimination so this negates the glycome data in this instance. No glycosylation feature could discriminate between borderline and metastatic patients and superior results were observed for CA125 in this case.

Glycoanalytical diagnostic tools for ovarian cancer
The availability of diagnostic tools for ovarian cancer remains somewhat elusive.
CA125 is currently the best diagnostic tool for ovarian cancer on the market, but is not reliable for diagnosing early stage ovarian cancer 46,47 . Early stage detection and subsequent treatment of ovarian cancer is an attractive approach to reduce morbidity from ovarian cancer. In this study, we corroborate these literature findings and show that CA125 cannot differentiate between borderline (n=6) and normal (n=7) samples ( Figure 6). For IgG, IgA, IgM, Hpt and A1AT, again we cannot see any significant alterations between these classes but find instead that Trf glycosylation can discriminate using GPs T11, T16, T17 and T18 ( Figure 6) in our small clinical cohort.
If these findings could be replicated and validated, Trf glycosylation could be exploited as an early detection tool for ovarian cancer. To probe the potential role of Trf glycosylation in ovarian cancer on a very basic level, we measured the upregulation/downregulation of the relative abundance of N-glycans for the statistically significant GPs (Figure 6b) and hypothesize that the dominant glycosylation features may be significant in the progression of the disease. Two of the major glycans that are upregulated in the borderline samples (n=6) contain core fucose-FA2G2 (T11) and FA2G2S(6)1 (T16) whereas there is a corresponding decrease in afucosylated structure A2G2S(3,6)2 (T18). Taking the classification and clustering data into consideration Trf glycosylation does not provide superior differentiation between classes compared to the other glycoproteins-no clear PCA differentiation is observed (Supporting Figure S10) but the hierarchical clustering does show its promise for resolving power between the groups (Supporting Figure S16).
With respect to clustering and classification analysis, the best performer is Hpt glycosylation (Figure 7) which displays clear groupings in the PCA analysis. In addition, regarding the statistical findings for Hpt glycosylation, there is a notable increase in two core fucosylated species FA2G2 (H11) and FA2G2S(6,6)2 (H20) in metastatic cancer patients (n=18) compared to normal controls (n=7) and a decrease in the afucosylated complex glycan A3G3S2 (H22). Taken collectively, this data suggests that core fucosylation (the FA2 Series) is significantly upregulated in this ovarian cancer cohort in both Trf and Hpt. Importantly, FA2 was previously found to be significantly altered (increased) in sera of ovarian cancer patients using an independent N-glycoanalytical technology 46 . We propose these changes reflect either differences in the expression levels of the α(1-6)-fucosyltransferase (FUT8) or donor substrate (GDP-fucose) in the medial-Golgi in ovarian cancer specific cells.