A glycopeptide is one molecule built from two polymers whose chemistries disagree. Fmoc solid-phase assembly subjects the growing chain to repeated secondary-amine treatment and finishes with a strongly acidic cleavage cocktail; the glycan has to survive both, and not every glycan does. The analytical problem then inverts the same asymmetry. Under HCD the sugar comes off at energies below those that break the backbone [12], so the spectrum that tells you which glycan was attached is frequently not the spectrum that tells you where it sat.
What follows separates the decisions that are genuinely independent: how the glycan is attached, how the anomeric centre gets set, which route joins glycan to peptide, and what the resulting mixture will and will not admit to under LC-MS.
Two attachment chemistries, two different downstream problems
N-linked glycans are joined to an asparagine side chain through an amide bond, which is why the convergent chemical route to them is an acylation rather than a glycosylation. Wang and colleagues built that route as a one-flask aspartylation in which long peptide fragments carrying proximal pseudoproline functionality are merged with complex glycan domains, after which acid-mediated global deprotection reveals the elaborated glycopeptide [6].
Mucin-type O-glycosylation is a true glycosidic linkage — glycans attached through O-linked N-acetylgalactosamine to serine and threonine residues — and it is initiated not by one or two enzymes but by a family of up to 20 homologous genes encoding UDP-GalNAc:polypeptide GalNAc-transferases, described as the largest glycosyltransferase enzyme family covering a single known glycosidic linkage [1]. That multiplicity is what gives mucin-type O-glycosylation, on the authors' reading, the greatest potential for differential regulation in cells and tissues — though the same review notes that the large number of GALNT genes does not amount to full functional redundancy [1].
The analytical consequence lands immediately. Unlike N-linked glycans, the O-glycosylation site is not within a known consensus sequence, and O-glycans are heterogeneous with numerous potential modification sites [13]. You cannot shrink the candidate set by inspecting the sequence, so every serine and threonine in the peptide stays live until a fragment ion rules it out.
Setting the anomeric centre
Forming 1,2-trans-glycosides by neighbouring-group participation — a stereodirecting ester at the 2-position of the glycosyl donor — is, as Basu and Crich put it, widely held to be a robust and reliable protocol, yet examples abound of cases in which less-than-perfect selectivity is obtained [2]. Their mechanistic account places a competing SN2-like glycosylation of the activated covalent donor alongside the participation pathway, and the two respond differently to how concentrated the reaction is. Across three case studies targeting both 1,2-trans-equatorial and 1,2-trans-axial glycosides, higher concentrations — reached either by increasing the acceptor:donor stoichiometry or by increasing concentration at fixed stoichiometry — mostly led to erosion of 1,2-trans-selectivity as the competing SN2-like reaction with the covalent donors became increasingly important [2]. The suggested move is to change concentration regime before surveying alternative esters. Hold the boundary in view: these are model glycosylation reactions, and the conclusion is about 1,2-trans selectivity, not about couplings onto a resin-bound peptide.
The mucin α-GalNAc-Thr linkage is 1,2-cis, so the participating C-2 ester that reliably delivers 1,2-trans is exactly what the donor must not carry. Practical building blocks reflect that. Liu, Barany and Live assembled a mucin-2-like glycopeptide bearing three consecutive O-glycosylated threonines from Fmoc-L-Thr(Ac3-α-D-GalN3)-OPfp, a 2-azido-2-deoxy-α-D-galactopyranosyl threonine pentafluorophenyl ester in which the C-2 nitrogen is masked as an azide throughout assembly [3]. Unmasking it later, they found direct treatment with thioacetic acid preferable to the two-step sequence of dithiothreitol reduction followed by N-acetylation [3].
The building-block route, and what each step costs
Two numbers from that same study are worth carrying into your own protocol. The glycosylated building block coupled efficiently at a relatively low molar excess of approximately 1.5 equiv in DMF [3] — which matters because the glycosyl amino acid is the expensive component and standard peptide protocols assume a large excess you cannot afford here. And O-deacetylation of the finished glycopeptides in solution was achieved with sodium methoxide at 10–15 mM (approximately 5 equiv) in methanol, while on-resin deacetylation used sodium methoxide at 6–10 mM in DMF–methanol (17:3) or hydrazine at 70 mM in methanol; the on-resin DMF–methanol route gave yields similar to the solution conditions [3].
Two observations from the same work bear on how you will read your own chromatograms and spectra. Free glycopeptides eluted earlier than the corresponding O-acetylated derivatives, and retention times varied systematically with the number of sugar moieties [3]. Separately, 1H NMR in water showed an increase in conformational organization of the glycopeptides with increased density of glycosylation [3]. The glycan is not an inert tag hanging off the backbone.
What the base step actually does
The standing worry in Fmoc O-glycopeptide synthesis is that repeated base exposure will trigger β-elimination of the glycosyl-serine or glycosyl-threonine, giving a dehydroamino acid and a free sugar. Meldal and co-workers tested it directly rather than assuming it. They synthesised a glycosylated pentapeptide under three deprotection regimes of differing base strength — 50% morpholine, 10% piperidine, or 2% DBU, each in DMF — and independently prepared the dehydropentapeptide Nα-Ac-Thr-Thr-ΔAba-Val-Thr-NH2 that β-elimination would produce, so that it could serve as an HPLC standard. That product was not formed under any of the deprotection conditions applied [4]. A 23-amino-acid glycopeptide from intestinal mucin prepared with 2% DBU as the Fmoc-cleavage base likewise showed no β-elimination, and the products tolerated prolonged treatment with sodium hydroxide in methanol/water without significant byproduct formation [4].
Read that result for what it is. It licenses the claim that these specific sequences, under these specific bases, did not eliminate detectably against a real synthetic standard. It does not license a general claim that base-mediated elimination never happens for any glycosylated residue in any sequence context, and it says nothing about acid.
What the acid step actually does
Acid is where the losses are real, and sialic acid is where they show first. Takeda and colleagues prepared sialic-acid-bearing amino acids protected as the 4-methylbenzyl (MBn) and the 4-methoxybenzyl (MPM) ester and carried both through Fmoc solid-phase synthesis of a glycopeptide. The α-sialyl linkage of the MBn-protected unit was partially cleaved under the final TFA deprotection, and removal of several MBn groups was itself incomplete; the MPM-protected unit gave the desired glycopeptide without decomposition of the α-sialyl linkage [5]. The protecting group on the sialic acid carboxyl is therefore not a bookkeeping detail — it determines whether your final cleavage returns the compound you designed or a partially desialylated mixture of it.
Convergent aspartylation, and the aspartimide it invites
Activating an aspartyl side chain in the middle of a peptide in order to acylate a glycosylamine sets up the same intramolecular chemistry that plagues Asp-containing sequences generally. The convergent method of Wang and co-workers addresses it structurally: temporary pseudoproline functionality installed proximal to the reactive aspartate serves to suppress formation of aspartimide side products both during solid-phase synthesis and during the aspartylation itself, with the whole aspartylation/deprotection performed in one flask [6]. The appeal of the convergent route is that the glycan never sees the elongation chemistry — but the aspartylation step inherits the aspartimide problem in exchange.
The chemoenzymatic route
The third option removes the glycan from the synthesis entirely and adds it afterwards. Build a peptide carrying a single GlcNAc at the target asparagine by ordinary Fmoc chemistry, then let an endo-β-N-acetylglucosaminidase transfer a whole oligosaccharide onto it. Li and colleagues made this efficient by using synthetic oligosaccharide oxazolines — mimics of the presumed oxazolinium ion intermediate formed in a retaining mechanism of substrate-assisted catalysis — as donors and GlcNAc-peptides as acceptors, obtaining 73–82% yields of the corresponding glycopeptides in a regio- and stereospecific manner, regardless of the size of the peptide portions [7]. Because the enzyme sets the linkage, the anomeric and regiochemical problems of the chemical route do not arise at that bond at all.
Scaling the Endo-M version of this route to practical application has two documented bottlenecks: low yield of the transglycosylation product, and difficulty obtaining the activated sugar oxazoline substrate, particularly the sialo-complex type [8]. Umekawa and co-workers attacked both, using a glycosynthase-like N175Q mutant of the Mucor hiemalis endoglycosidase Endo-M — found to possess enhanced transglycosylation activity with sugar oxazoline as donor — together with a preparation of sialo-complex-type sugar oxazoline from natural sialoglycopeptide in egg yolk; the mutant transferred the sialo-complex donor onto peptide acceptors and onto bovine ribonuclease B, each sialylated compound obtained in significantly high yield [8]. The enzyme's specificity is the constraint that replaces the chemist's: Endo-M transfers the oligosaccharide en bloc onto acceptors carrying an N-acetylglucosamine residue [8], so an acceptor that does not present that handle gives the enzyme nothing to act on.
Microheterogeneity is the analytical problem
Whatever the route, a natural or cell-derived glycoprotein does not present one structure per site. Thousands of possible glycan structures and compositions create potential for tremendous site heterogeneity [9]. Riley and colleagues captured that heterogeneity at scale using activated ion electron transfer dissociation, characterising 1,545 N-glycosites and more than 5,600 unique N-glycopeptides from mouse brain tissue, and reported that N-glycosylation profiles can differ between subcellular regions and structural domains and that site heterogeneity manifests in several different forms, including dramatic differences between glycosites on the same protein [9]. Two sites on one protein are not two samples of one distribution.
This is why releasing the glycans and profiling them separately answers a different question from the one you usually asked. It gives you the glycan pool; it destroys the site assignment that made the pool interesting.
Oxonium ions tell you a glycan is there
Under collisional activation the glycosidic bonds go first, and the low-mass region fills with saccharide oxonium ions. Peterman and Mulholland exploited exactly this with source CID on a hybrid linear ion trap/FT-ICR instrument: because source CID is non-selective it activates all charge states and co-eluting glycoforms simultaneously, generating greater ion abundance for the m/z 204 and/or 366 oxonium ions, and a pseudo-MS3 stage then adds selectivity by removing chemical noise before the tandem acquisition [10]. Applied to a bovine fetuin digest, the combined workflow identified four N-linked and two O-linked glycopeptides without enzymatic cleavage of the sugar moiety or release of the sialic acids beforehand, alongside over 95% of the total protein sequence [10].
How much of the glycan versus the peptide you fragment is a dial, not a fixed property. Cao and colleagues varied the normalized collision energy for HCD from 30 to 60% and found that lower NCE values preferentially fragmented the sugar chains, generating a ladder of neutral losses of monosaccharides that enables putative glycan structure characterization, while higher NCE values preferentially fragmented the peptide backbone and so supplied the information needed for confident peptide identification; detection of the oxonium ions enabled unambiguous differentiation of glycopeptides from non-glycopeptides [12]. Alternating the NCE within one run gave 38 unique intact glycopeptides corresponding to eight glycosylation sites — six N-linked and two O-linked — from a standard protein mixture [12].
Oxonium ions carry more than a yes/no. Analysing beam-type and ion trap CID spectra of synthetic and natively derived N- and O-glycopeptides, Halim and colleagues found that oxonium ion fragmentation patterns characteristically differ between glycopeptides terminally substituted with GalNAcα1-O-, GlcNAcβ1-O-, Galβ3GalNAcα1-O-, Galβ4GlcNAcβ-O- and Galβ3GlcNAcβ-O- structures, and concluded that the difference may thus be used to distinguish among these glycan structures [11]. Keep the hedge and keep the list: the discrimination was demonstrated for those terminal structures under those activation conditions, and it is a profile comparison, not a formula you evaluate on a single peak.
A mass is not a composition, and a composition is not a structure
Open glycan search in MSFragger reports a glycopeptide as a peptide sequence plus the mass of the attached glycan. In samples with complex glycosylation patterns, converting that mass to a specific glycan composition is not straightforward, because many glycans have similar or identical masses [16]. Polasky and colleagues built a multi-attribute method for N-linked glycopeptides fragmented by collisional or hybrid activation that discriminates between candidates using several sources of information from the spectrum at once — observed glycan B-type (oxonium) and Y-type ions together with mass and precursor monoisotopic selection errors — and paired it with false discovery rate estimation for the glycan assignment specifically, producing a pipeline with FDR control of both the peptide and the glycan components [16]. The design admits the underlying point: peptide-level confidence and glycan-level confidence are separate quantities, and a well-controlled peptide FDR says nothing about whether the composition attached to it is right.
Site localisation needs electron-driven dissociation
The property that makes collisional activation good at reading the glycan makes it bad at placing it. In ETD the glycan side chain remains intact and the glycosylation site can be localized on the basis of the c and z fragment ions [13]. Zhu and colleagues derived fragmentation rules for O-glycopeptides under ETD, implemented them in a scoring algorithm, and correctly assigned site-specific O-glycopeptide composition in every case across glycopeptides from mucin, erythropoietin, fetuin and a viral envelope glycoprotein [13].
EThcD exists because neither dissociation type alone gives both halves. Yu and colleagues applied a redefined electron-transfer/higher-energy collision dissociation scheme to incorporate glycan and peptide fragments in a single spectrum, validating on fetuin with 19 glycopeptides and the corresponding five glycosylation sites identified [15]. Their framing of the problem is the honest one: conventional workflows lean on identifying enzymatically released glycans and their peptide backbones separately, because no efficient fragmentation method produced unbiased glycan and peptide product ions simultaneously in one spectrum [15].
Do not generalise from that to "always use EThcD". Riley and colleagues compared conventional HCD, stepped collision energy HCD, ETD and EThcD head to head on both glycopeptide classes. For N-glycopeptides, HCD and sceHCD generated similar numbers of identifications with sceHCD generally providing higher quality spectra, and both significantly outperformed EThcD methods in identifications — indicating that ETD-based methods are not required for routine N-glycoproteomics even though they can generate higher quality spectra. Conversely, ETD-based methods and EThcD in particular were indispensable for site-specific analyses of O-glycopeptides, and their data showed that O-glycopeptides cannot be robustly characterized with the HCD-centric methods that suffice for N-glycopeptides [14]. The sequon does the work on the N-linked side; on the O-linked side, with no consensus to lean on, the c and z ions are the only thing standing between you and an arbitrary assignment.
Exoglycosidase sequencing, and what it trades away
When you need linkage and anomeric detail that fragmentation will not give you, the enzymatic route reads the glycan from the outside in. Sequential exoglycosidase digestion followed by LC-FLD, LC-MS or CE analysis provides detailed carbohydrate structural information, because highly specific exoglycosidases cleave monosaccharides from the non-reducing end and thereby yield information about the linkage, the stereochemistry and the configuration of the anomeric carbon [17]. A representative workflow releases the glycans with PNGase F, labels them directly with procainamide or 2-aminobenzamide, cleans up, and runs a 3 h enzymatic digestion, the whole sequence designed to fit an 8 h window ahead of overnight LC-FLD, LC-MS or CE [17].
The trade is explicit in the first step. Releasing the glycan is what makes clean exoglycosidase interpretation possible, and it is also what discards the site information that intact glycopeptide analysis exists to preserve. Run both on the same material and the exoglycosidase result constrains which structures are plausible candidates for the compositions your intact-glycopeptide search reported — which is a different and much stronger claim than either measurement supports alone.