Longitudinal monitoring of immunoglobulin A glycosylation during pregnancy by simultaneous MALDI-FTICR-MS analysis of N- and O-glycopeptides

Immunoglobulin A (IgA) is a glycoprotein of which altered glycosylation has been associated with several pathologies. Conventional methods for IgA N- and O-glycosylation analysis are tedious, thus limiting such analyses to small sample sizes. Here we present a high-throughput strategy for the simultaneous analysis of serum-derived IgA1 N- and O-glycopeptides using matrix-assisted laser/desorption ionisation Fourier transform ion cyclotron resonance (MALDI-FTICR) mass spectrometry (MS). Six non-fucosylated diantennary complex type glycoforms were detected on the Asn144-containing glycopeptide. Thirteen distinct glycoforms were identified for the Asn340-containing tailpiece glycopeptide, mainly of the diantennary complex type, and low amounts of triantennary glycoforms. Simultaneously with these N-glycopeptides, 53 compositional glycoforms of the hinge region O-glycopeptide were profiled in a single high resolution MALDI-FTICR spectrum. Since many pregnancy associated changes have been recognized for immunoglobulin G, we sought to demonstrate the clinical applicability of this method in a cohort of 29 pregnant women, from whom samples were collected at three time points during pregnancy and three time points after delivery. Pregnancy associated changes of N-glycan bisection were different for IgA1 as compared to IgG-Fc described earlier. We foresee further applications of the developed method for larger patient cohorts to study IgA N- and O-glycosylation changes in pathologies.

(LC-MS). Sample throughput can be improved after reducing sample complexity using sialidases 3 , however at the expense of important information.
MS strategies based on matrix-assisted laser/desorption ionisation (MALDI) allow a high-throughput analysis of a complex mixture with the additional benefit of preventing carryover between samples. Most commonly, MALDI is coupled to time-of-flight (TOF) mass spectrometers. However, for simultaneous MALDI-MS analysis of N-and O-glycopeptides from polyclonal serum IgA high resolving powers (> 50000) are pivotal. Moreover, sialic acids are instable in commercially available MALDI-TOF systems. Therefore, MALDI has hitherto hardly been applied for IgA O-glycopeptide analysis 32 . Fourier transform (FT) MS methods provide the highest resolving power and mass accuracy within the different mass spectrometers 33 . Previously, we have applied MALDI-FT-ion cyclotron resonance (ICR)-MS to map singly-charged proteins up to about 17000 Da and use the accurate mass differences for identification purposes 34,35 . These properties allow both the resolution of near isobaric species and the identification of possible post-translational modifications (PTMs). Another advantage of MALDI-FTICR-MS is the fact that the intermediate pressure in the MALDI source leads to a significant reduction of the sialic acid fragmentation 36 . Even though a MALDI approach will not allow for full discrimination of all macro-and microheterogeneity, it is suitable for high-throughput profiling in a clinical setting.
Therefore, in this study serum IgA glycopeptides were enriched with high specificity using a combination of purification techniques including a 96-well plate IgA affinity purification, trypsin digestion and two sequential HILIC solid-phase extractions. The enriched N-and O-glycopeptides were measured with isotopic resolution in an m/z-range from 3499 to 10000 using a 15T MALDI-FTICR mass spectrometer. This approach allowed for the first time the specific, high-throughput analysis of both N-and O-glycosylation of IgA1 by ultrahigh resolution MALDI-FTICR-MS. We observed 53 different compositions for the O-glycopeptide, 6 glycoforms for the Asn144 containing N-glycopeptide, and 13 glycoforms for the Asn340 containing glycopeptides. Confident assignment of the glycoforms was achieved making use of the high resolution and mass accuracy, together with fragment ion spectra. For immunoglobulin G it is known that many glycosylation changes occur with pregnancy, and therefore we applied our method on a pregnancy cohort (n = 29 individuals, 6 time points) to explore whether the same holds true for immunoglobulin A. The application of this method showed several site specific differences in N-glycosylation between pregnancy and the postpartum state. O-glycosylation showed some minor pregnancy associated alterations with respect to galactosylation and the ratio of galactoses per GalNAc.

Results
IgA capturing and digestion. We established a workflow for the high-throughput profiling of serum IgA1 N-and O-glycosylation by MALDI-FTICR-MS. IgA was affinity-captured from human serum in a 96-well format resulting in a highly enriched IgA sample (see Supplementary Table S1). The captured IgA was digested using trypsin after reduction and alkylation in order to generate specific glycopeptide fragments.

MALDI-FTICR-MS profiling of IgA glycopeptides.
To be able to accurately detect all glycoforms present in the IgA glycopeptide mixture ultrahigh resolution MALDI-FTICR-MS measurements were performed using 4-chloro-α -cyanocinnamic acid matrix after HILIC enrichment of the IgA glycopeptides. HILIC enrichment was optimized to obtain the IgA1 hinge-region O-glycopeptides and the IgA1 Asn340 N-glycopeptides in one single spectrum after purification from 70% ACN, while the second IgA1 N-glycopeptide cluster (Asn144) was subsequently HILIC enriched from the flow-through starting with 80% ACN (see Methods section for the details) and obtained in a second spectrum (Fig. 2).

Peak list generation.
Manual inspection of the spectra showed no IgA2 specific glycopeptide clusters. For IgA1 several glycoforms were observed for the N-glycopeptides containing Asn144 (peptide moiety 127 LSLHRPALEDLLLGSEANLTCTLTGLR 153 ) or containing Asn340 (peptide moiety 332 LAGKPTHVNVSV VMAEVDGTCY 353 ), and for the hinge region O-glycopeptide ( 89 HYTNPSQDVTVPCPVPSTPPTPSPSTPPTPSP SCCHPR 126 ). In addition, some apparent mass deviations were observed, that could be explained by the loss of C-terminal tyrosine from the Asn340 containing glycopeptide, as was previously observed by others 18,30 , and by oxidation of the methionine in the Asn340 glycopeptides. Furthermore, we did observe some sodium adducts in the spectra, however these were generally minor and therefore not included in the analysis. In some spectra we observed contaminants and these were excluded from further analysis. The high resolving powers (> 50000) that are characteristic of MALDI-FTICR-MS were crucial for simultaneous profiling of the N-and O-glycopeptides in a single spectrum. Moreover, the high mass accuracies (< 2 ppm) provided by the system allowed confident assignment of the IgA glycopeptides ( Supplementary Fig. S3).
An analyte list containing the observed putative glycopeptides was curated on the basis of overall abundance and QC-values. The identity of some O-glycopeptide species was confirmed by ESI-FTICR-MS(/MS), and N-glycopeptide confirmation was obtained by C18-nanoLC-quadrupole-TOF-MS/MS. The major glycovariants of each glycopeptide cluster were thus identified (Fig. 3, Supplementary Table S2, and Supplementary Fig. S2). Of note, for the O-glycopeptides no site assignment has been performed. The N-glycopeptide moieties were confirmed by performing a glycan release on trypsin digested IgA, and analysing these digests on LC-ESI-MS/MS, followed by a proteomics data analysis where deamidation was set as a variable modification (Supplementary Methods, Supplementary Table S1). Finally, N-glycan structures were confirmed by analysing ethyl esterified released glycans using MALDI-TOF-MS (Supplementary Methods, Supplementary Fig. S4), and O-glycan structures by HPLC 7 . The released N-glycan analysis additionally indicated that the sialic acids were almost exclusively α 2,6-linked.
The combined MS and MS/MS data for both glycopeptides and glycans furthermore indicated the absence of difucosylated species, which would, if present, in part overlap with sialylated species. Most of the analysed samples showed a partial oxidation of the glycopeptides covering site Asn340. While the assignment of most of the peaks was straight-forward, oxidized, monofucosylated species, which are isomeric with species that instead contain an additional hexose, were slightly more challenging. However, by using the released glycan information in most cases unambiguous assignment was achieved, which was illustrated by highly similar oxidation ratios for all observed oxidized/non-oxidized glycopeptide pairs (data not shown).
Altogether, this resulted in a final list of 53 O-glycopeptide conformations, 6 glycoforms for the Asn144 containing N-glycopeptide, and 13 glycoforms for the Asn340 containing glycopeptides, of which 3 glycoforms were observed for both the truncated and the non-truncated peptide variant, and 10 only for the truncated peptide (Table 1; Supplementary Table S3). For the O-glycopeptide cluster (approximately m/z 5350-8000) an average MALDI-FTICR-MS spectrum selection. The glycopeptide clusters were subsequently subjected to a quality check (see Methods section for details). Robustness of the method was then tested using a standard plasma sample along with the clinical samples. The standard samples were included at least in triplicate on each 96-wells plate. For all glycopeptide clusters the inter plate variation of the standard was below 20% (Supplementary Table S4), while the biological variation was on average almost twice as high.
After spectra curation the final dataset contained 329 hinge region O-glycosylation profiles, 292 Asn144 N-glycosylation profiles, and 329 Asn340 N-glycosylation profiles. When two profiles of the same glycopeptide cluster were available for a sample, the highest intensity variant was used for the analysis. Only patients for whom spectra of all time points were available were included in the final statistical analysis. Calculated O-glycosylation traits could be analysed pairwise for 27 individuals, traits from Asn144 for 25 individuals, and from Asn340 glycopeptides the traits could be analysed pairwise for 23 (non-truncated peptide) and 25 individuals (truncated peptide).
Sialylation of Asn144 was higher during pregnancy than after delivery (63% vs. 59%, p = 0.0002; Table 2, Fig. 4e). The decrease occurred rapidly between the last time point of pregnancy and 6 weeks postpartum, after which it remained lowered. In contrast, for Asn340 no changes in the level of sialylation (> 89%) were observed.
The tailpiece glycan fucosylation showed a minor, but significant, decrease during pregnancy, and an increase after delivery (p < 0.002, Fig. 4g). This was not reflected on the Asn144 N-glycosylation site, where no fucosylated glycans were observed.  The larger variety of N-glycans on Asn340 compared to Asn144 was further illustrated by the presence of triantennary glycans. The abundance of these glycans increased during pregnancy (5 to 6%, p = 0.0017, Fig. 4h). After delivery the levels dropped to 5% (p < 0.008).
Both N-glycosylation sites of IgA1 showed complete or near complete levels of galactosylation, for which no changes were observed during pregnancy, nor after delivery.
Values for all calculated N-and O-glycosylation traits at all time points, and the corresponding standard errors, are depicted in Supplementary Table S5.

Pregnancy associated changes in IgA O-glycosylation.
The O-glycosylation of the IgA1 hinge region remained fairly stable throughout pregnancy and the time after delivery. The numbers of GalNAc residues and sialic acids did not change, and neither did the number of sialic acids per galactose. Very small yet significant increases of the number of galactoses (3.96 to 3.98, p = 0.0095; Table 2, Fig. 4i) and the related ratio of galactoses per GalNAc (0.82 to 0.83, p = 0.0109, Fig. 4j) during pregnancy were observed.

Discussion
In the current manuscript we describe a method for the simultaneous analysis of IgA1 N-and O-glycosylation using affinity purification and a simple trypsin digestion, followed by a two-step HILIC enrichment of the glycopeptides, and MALDI-FTICR-MS detection. The high-throughput technique is estimated to allow for the analysis of 384 serum samples within 24 h. Fifty-three O-glycopeptide compositions, 6 Asn144 glycoforms and 13 Asn340 glycoforms can be distinguished with high mass accuracy (average < 2 ppm error). The technique was applied on a small pregnancy cohort, showing several changes during pregnancy and/or after delivery.
Only a limited number of groups performed MALDI-MS analysis of IgA glycosylation before 5,6,11 . These MALDI-TOF-MS experiments were performed in low resolution linear mode, and mass precision was not stated 5 , or large mass deviations were allowed: ± 1 Da (140-250 ppm; m/z window 3500-8000) for O-glycopeptide analysis 6 , and 0.3% (3000 ppm) for sialidase treated N-glycopeptides 11 . Of note, Tarelli et al. did not analyse tryptic peptides; therefore only the compositions can be compared to our data, and not the observed masses. Nevertheless, a large body of correctly annotated glycopeptide was thus gathered (Table 1; Supplementary Table S3). Generally, we were able to gain more information in the high mass region, whereas previous publications observed more low mass O-glycopeptides. We observed five O-glycopeptide compositions that have not been published previously. This may be caused by different sample preparation, for example a slight skewing of the spectra to higher masses because of the HILIC-SPE, or due to limited in-source fragmentation in our intermediate pressure FTICR mass spectrometer.
In addition, the use of an ultrahigh resolution MALDI-FTICR-MS circumvented LC steps -generally taking 30 to 120 min per sample, e.g. 6 -prior to the MS measurement. For the MALDI approach a simple 2-step HILIC was necessary, taking only 60 minutes for 96 samples. The second step of the HILIC was necessary to obtain data concerning the Asn144 containing glycopeptide. However, this manual step can be automated to reduce hands-on time. Moreover, the intermediate pressure in the FTICR prevents the loss of sialic acids observed in MALDI-TOF-MS 32,36 . Generally, the measurements and relative quantitation of IgA1 O-glycosylation in MALDI mode has been performed on de-sialylated glycopeptides 17 , or on low-level sialylated myeloma species 37 , thus losing the valuable information about the sialic acids. MALDI-MS analysis of the intact N-glycopeptides to obtain site-specific information has only been published once 11 . Again, sialidase treated glycopeptides were analysed. The technique described in the current manuscript allows to analyse intact N-and O-glycopeptides bearing sialic acids in a high-throughput MALDI-MS approach. This allowed for the discovery of eight glycopeptides containing the Asn340 site, generally with higher sialylation compared to those reported in literature (Supplementary  Table S3). Among these are some triantennary structures. Furthermore, a few non-fucosylated glycoforms were newly detected. Interestingly, the described technique without sialidase treatment allowed for the specific detection of non-sialylated species on Asn144 as well.
We observed the Asn340 containing glycopeptide with a truncated C-terminus, which is in accordance with previous reports 18, 30 . In our data the truncated version is more abundant than the non-truncated version, which is reflected in the higher spectral count in our proteomics analysis (42 vs. 16 counts). We assume that this truncation is naturally occurring on IgA from the circulation, and is not induced during storage or sample preparation. However, it is unclear which protease would be responsible for this processing, and whether it is of functional significance.
Both IgA subclasses (i.e. IgA1 and IgA2 isoforms) were captured in the here described procedure, as concluded from bottom-up protein identifications (Supplementary Table S1). However, IgA2 specific glycopeptides were not detected in our MALDI-FTICR-MS spectra. Likely, this is due to the rather low abundance of this subclass (~10%). Moreover, the expected masses of glycopeptides related to the IgA2 Asn205 containing tryptic peptide are outside the detection window.
In the current analysis of pregnancy-associated glycosylation changes on the glycopeptide level, we show changes in N-glycosylation and -to a lesser extent -O-glycosylation traits for IgA1. The site specific approach showed a decrease in sialylation after delivery (63% to 59%) for Asn144, whereas sialylation on Asn340 did not change. On the other hand, there was a pronounced increase in bisection on Asn340 from 1 st trimester up to 6 weeks postpartum (52% to 58%), while the same process was observed on Asn144 to lower extent. Notably, our earlier pilot study by Ruhaak   N-glycans, did not reveal pregnancy-associated glycosylation changes, which may be due to the fact that only samples from 6 individuals were analysed 9 . Whether the changes are biologically important needs to be further investigated. Surprisingly, IgA bisection shows a different time course than IgG bisection. While IgA bisection increases during pregnancy, and even further after delivery, IgG bisection remains at the same low level during pregnancy and only increases after delivery 38 . After the first time point postpartum IgG bisection remains at the same level while IgA bisection decreases after this time point. The differences between IgA and IgG bisection time courses may have to do with their different biological roles and sites of action, yet this will require further research. In general, the role of bisection in modulating antibody function still needs to be clarified for both IgG and IgA.
For IgG-Fc N-glycans increased levels of galactosylation and sialylation during pregnancy compared to the non-pregnant state are well described in both healthy and diseased women [38][39][40][41][42] . On IgG these changes are restricted to the Fc portion, whereas the Fab portion, carrying more processed glycans, does not exhibit these pregnancy associated changes in galactosylation and sialylation, although for the latter a trend is visible 38 . Similarly, these changes are not observed for the diantennary glycans on Asn340 of IgA1, which are near completely processed with levels of galactosylation and sialylation of > 99% and > 89%, respectively.
Sialic acids on N-glycans of the standard serum IgA samples analysed in this study were almost exclusively α 2,6-linked (> 95%), which is in line with previous reports 4,10,25 . With regard to the level of sialylation of N-glycans, Orczyk-Pawilowicz et al. reported no significant changes after delivery 25 . However, in the current study we have found that sialylation levels at Asn144 were lower after delivery. These differences could be explained by the chosen time points: 6 weeks postpartum in our study versus 2 weeks after delivery in the earlier report. In addition, in this earlier study the glycosylation of IgA was detected by using lectins that are not site-specific. Differences in sialylation occurring at one site (e.g. 4% change at Asn144) can be obfuscated by the high levels (> 89%) of sialylation on Asn340, as well as by the highly sialylated O-glycans.
In conclusion, we have developed a high-throughput method based on ultrahigh resolution MALDI-FTICR-MS that enables simultaneous profiling of IgA N-and O-glycosylation at the glycopeptide level. The ultrahigh resolving power allow for accurate detection of more than 50 O-glycopeptides and 6 and 13 glycoforms for the Asn144 and Asn340 N-glycopeptides, respectively. The suitability of the method to profile clinical samples was demonstrated by the application on a pregnancy cohort to study potential serum IgA glycosylation changes. In addition, this method can be used in conjunction with remodelling of IgA glycosylation and in vitro IgA bioactivity assays to study the functional relevance of IgA glycosylation with respect to antibody dependent cellular cytotoxicity and other effector mechanisms.

Methods
Human serum and plasma samples. Twenty-nine healthy Caucasian women who participated in the PARA (Pregnancy-induced Amelioration of Rheumatoid Arthritis) study donated serum at the 1 st , 2 nd and 3 rd trimester of pregnancy and at six, twelve and > 26 weeks postpartum 40 . The PARA study was in compliance with the Helsinki Declaration and was approved by the Ethics Review Board at the Erasmus University Medical Centre, Rotterdam, The Netherlands, and experiments were performed accordingly. Informed consent was obtained from all subjects.
In addition, EDTA plasma from a healthy donor was used as a technical control throughout the measurements.
Capture of polyclonal IgA. Human polyclonal IgA was captured from 10 μ L human serum or plasma using CaptureSelect IgA beads. Twenty microliter bead slurry was applied to each well of a 96-well Orochem filter plate (10 μ m pore size). The beads were pre-washed on a vacuum manifold with three times 200 μ L PBS. Subsequently, the samples were diluted in 100 μ L PBS on the plate, and incubated on a multiwell plate shaker for 1h. The beads with captured IgA were washed on the vacuum manifold with PBS and MQ water (3 × 200 μ L), followed by elution using 100 μ L 100 mM FA. The eluates were collected by centrifugation (1 min, 50 × g) into V-bottom plates. Finally, the IgA samples were dried by vacuum centrifugation.
Trypsin digestion. The dried samples were reconstituted in 100 μ L 20 mM ammonium bicarbonate buffer (pH 8) and reduced with 2 μ L 125 mM DTT for 30 min at 60 °C. Sequentially alkylation with 2 μ L 200 mM IAM was performed for 30 min at room temperature, followed by overnight digestion of IgA by TPCK treated trypsin at 37 °C, using 600 ng enzyme per well.
HILIC enrichment and spotting. Obtained trypsin digests of IgA samples were enriched for glycopeptides by a two-step microtip cotton HILIC SPE, using cotton thread as the solid phase, as described before 38 , with minor modifications in the used solution volumes and composition. Briefly, 15 μ L of the digest was transferred to a 96 wells V-bottom plate, and adjusted to 70% ACN by addition of 35 μ L ACN. The cotton thread was washed three times with 20 μ L MQ, and conditioned three times with 20 μ L 70% ACN, using a 12-channel micropipette (2-20 μ L). The sample was then loaded on the cotton by pipetting up-and-down 20 times. Subsequently, the cotton was washed 3 times with 20 μ L 70% ACN containing 1% TFA and three washes with 20 μ L 70% ACN, followed by elution in 5 μ L MQ. Twenty-five microliter ACN was then added to the tryptic digest to bring the sample to 80% ACN for a sequential HILIC enrichment with 80% instead of 70% ACN, to capture glycopeptides that were not detectable in the 70% HILIC eluates. The eluates from both HILIC experiments were mixed separately with 15 μ L 4-chloro-α -cyanocinnamic acid matrix (Cl-CCA; 0.33 mg/mL in acetone:ethanol 1:2) and 1 μ L of each eluate was spotted on a Bruker AnchorChip plate with 800 μ m anchors. All glycopeptide samples were spotted in duplicate, resulting in 384 spots per 96 serum samples. Two duplicate spots were measured for each spotted sample. The FTICR system was externally calibrated using a dextran polysaccharide hydrolysate. The MALDI-FTICR-MS method introduced in this manuscript was applied to 174 serum samples from the cohort, 10 standard plasma samples from a healthy volunteer, and 12 blanks. All samples were spotted in duplicate and the whole workup was performed twice. In total 1536 MALDI spectra were obtained. DataAnalysis Software 4.0 SP4 (Bruker Daltonics) was used for the visualization and conversion of the MALDI spectra into .xy files.
For the characterization of some of the O-glycopeptides direct infusion ESI-FTICR-MS experiments were carried out using the settings specified in the Supplementary Methods section. Prior to MS/MS experiments the FTICR system was externally calibrated using a commercially available tune mix (Agilent). The in-house developed tool sequentially integrated the data for all isotopic peaks that theoretically cover 99% of the analyte intensity, and determined the signal-to-noise ratios. In addition, a quality control (QC) value gave an indication of the quality of each analyte signal by analysing the deviation of the observed isotopic pattern from the theoretical pattern. The QC value calculation was adapted not to include noise in the equation.

LC-ESI
Generation of analyte list. Although most literature depicted the N-glycans of IgA1 to be highly galactosylated and sialylated diantennary complex type, some also reported high mannose structures, as well as tri-and tetraantennary glycans. The O-glycans were found previously to be of various variants of the core 1 type. Therefore we started our analysis with a non-curated list of N-and O-glycopeptides. The list contained 65 potential N-glycoforms for each N-glycopeptide. Three N-glycopeptide sequences were included, namely LSLHRPALEDLLLGSEANLTCTLTGLR for Asn144, and LAGKPTHVNVSVVMAEVDGTCY for Asn340, the latter with and without a truncated tyrosine (Y) at the C-terminus. Finally, for the Asn340 glycopeptides oxidation of the methionine (Met345) was included as a potential variable. In addition 182 potential O-glycopeptides (ranging from 1 HexNAc up to 6 HexNAc + 6 Hex + 8 SA) were included for extraction, considering a maximum of six occupied sites with core 1 O-glycans on the peptide HYTNPSQDVTVPCPVPSTPPTPSPSTPPTPSPSCCHPR. In total the list of potential analytes contained 507 (182+ 65+ 65+ 65+ 65+ 65) structures. The theoretical m/z values of the truncated Asn340 containing glycopeptide (both oxidized and non-oxidized) bearing a Man5 or H3N4 glycan were outside the measuring range and were therefore excluded.
In order to curate the list of analytes the following steps were taken: analytes showing signal-to-noise greater than six in less than 50% of the spectra were removed from the analysis. In addition, the 25% quartile boundary of the QC value distribution had to be smaller than 0.03 for an analyte to be included in the final extraction. When two analytes were overlapping, the analyte with the lowest QC value was selected. In case of small differences in QC values, the expectancy based on the released N-glycan relative abundance was used to select the analyte. Finally, the internally calibrated spectra for the technical controls and for the healthy pregnant women were summed, resulting in two sum spectra for the analysis of the O-and the Asn340 N-glycopeptide clusters, and two for the Asn144 N-glycopeptide cluster. Again, the QC values for the analytes had to be < 0.03 for analytes to be included. Finally, the mass accuracy errors of the observed species were inspected to confirm the identity of the extracted species. A median error of less than 5 ppm was set as an inclusion threshold. In the final analyte list overlapping N-glycopeptides were curated based on the relative abundances of released glycans, which were obtained as described in the Supplementary Methods. Quality threshold for MALDI-FTICR-MS spectra. In order to obtain high quality data, a curation step was performed on the obtained MALDI-FTICR-MS spectra. The total intensity per glycopeptide cluster was calculated based on the final analyte list. In addition, the percentage of glycopeptides with a signal-to-noise ratio (S/N) greater than six was calculated for each of the clusters. The total intensity was then plotted against the percentage with S/N > 6 (example in Supplementary Fig. S1). This plot was used as a guide to set cut-off values for both total intensity and the percentage. As a readout the average relative standard deviation per time point was monitored, as well as the percentage of spectra passing the cut-off threshold for which approximately 90% was considered as the lower limit. Finally, from each sample the highest intensity spectrum of the duplicates was selected.
Scientific RepoRts | 6:27955 | DOI: 10.1038/srep27955 Calculation of glycosylation traits. O-glycosylation features were calculated for each sample: the number of GalNAcs, galactoses, sialic acids, as well as the ratios for sialic acids per galactose and galactose per GalNAc. In addition, we checked the relative abundances of peptides with more GalNAcs than galactoses, indicative for the presence of Tn-antigen, and those with more sialic acids than galactoses. For all the N-glycopeptides sialylation and bisection of diantennary glycans was calculated. For the truncated Asn340 containing peptide additionally galactosylation and fucosylation of diantennary glycans, and the abundance of triantennary glycans were calculated. The formulas used to calculate all the derived traits are shown in the Supplementary Methods. Statistical analysis. The Wilcoxon signed-rank test was performed for paired samples without assumptions about normality. To correct for the three tests within each calculated glycosylation trait a Bonferroni correction for multiple testing was performed, with p < 0.017 considered as significant.