Electron-Based Dissociation Is Needed for O-Glycopeptides Derived from OpeRATOR Proteolysis

Abstract

The recently described O-glycoprotease OpeRATOR presents exciting opportunities for O-glycoproteomics. This bacterial enzyme purified from Akkermansia muciniphila cleaves N-terminally to serine and threonine residues that are modified with (preferably asialylated) O-glycans. This provides orthogonal cleavage relative to canonical proteases (e.g., trypsin) for improved O-glycopeptide characterization with tandem mass spectrometry (MS/MS). O-glycopeptides with a modified N-terminal residue, such as those generated by OpeRATOR, present several potential benefits, perhaps the most notable being de facto O-glycosite localization without the need of glycan-retaining fragments in MS/MS spectra. Indeed, O-glycopeptides modified exclusively at the N-terminus would enable O-glycoproteomic methods to rely solely on collision-based fragmentation rather than electron-driven dissociation because glycan-retaining peptide fragments would not be required for localization. The caveat is that modified peptides would need to reliably contain only a single O-glycosite. Here, we use methods that combine collision- and electron-based fragmentation to characterize the number of O-glycosites that are present in O-glycopeptides derived from the OpeRATOR digestion of four known O-glycoproteins. Our data show that over 50% of O-glycopeptides in our sample generated from combined digestion using OpeRATOR and trypsin contain multiple O-glycosites, indicating that collision-based fragmentation alone is not sufficient. Electron-based dissociation methods are necessary to capture the O-glycopeptide diversity present in OpeRATOR digestions.

Graphical Abstract

INTRODUCTION

Mucin-type O-glycosylation is a prevalent post-translational modification on extracellular and secreted proteins that drives both biochemical and biophysical interactions at the cell surface.^1-5 This important modification, characterized by an initiating α-N-acetylgalactosamine (α-GalNAc) monosaccharide on serine and threonine (and sometimes tyrosine) residues is challenging to study due to several inherent features.^6-9 These include nontemplate driven elaboration of the α-GalNAc residue into four major core structures, the lack of a well-defined sequence motif, and occurrence in densely glycosylated regions rich in serine and threonine residues.¹⁰ The combination of these attributes leads to O-glycosylated sequences that are heterogeneous both in the glycosites that are occupied and the glycans that modify them, requiring site-specific characterization. Tandem mass spectrometry (MS) is the premier method for O-glycosite mapping, but O-glycopeptides are largely intractable using standard glycoproteomic approaches, which have focused almost entirely on N-glycans. To this end, recent efforts have focused on improving O-glycopeptide analyses,^6,7,11,12 ranging from developments in sample preparation,^13-17 data acquisition,^18-24 and postacquisition data analysis.^25-30

One promising avenue to emerge from these investigations is the description of O-glycoproteases that can generate O-glycopeptides more amenable to MS characterization.^31-34 One such enzyme is OpeRATOR from A. muciniphila, which cleaves N-terminally to O-glycosylated serine and threonines, generating O-glycopeptides that retain the modified residue at their N-terminus.^33-35 In the handful of studies to utilize OpeRATOR thus far, collision-based fragmentation, namely, higher-energy collisional dissociation (HCD), has been used to identify O-glycopeptides. The operating presumption in these studies is that site-specific O-glycosite localization can be reported by attributing the entire glycan mass identified for the peptide to an O-glycan modifying the N-terminal serine or threonine, instead of requiring glycan-retaining peptide backbone fragments. If true, this presumption separates O-glycopeptides derived through OpeRATOR cleavage into a unique class differing from canonical O-glycopeptides, where reliable site assignment can rarely be achieved using collisional activation. In order to preclude the need for electron-based fragmentation (such as electron transfer dissociation [ETD] and ETD with HCD supplemental activation [EThcD]), however, O-glycopeptides from OpeRATOR proteolysis must not have missed cleavages. Missed cleavages by this definition would mean that internal O-glycosites also exist within the peptide sequence rather than exclusively at the N-terminus, ultimately leading to misattribution of a portion of the identified glycan aggregate mass to the N-terminal O-glycosite under the operating presumption above.

Here, we use methods employing both HCD and ETD/EThcD fragmentation to investigate O-glycopeptides derived from OpeRATOR proteolysis of four O-glycoprotein standards (fetuin, monocyte differentiation antigen [CD14], fibronectin, and P-selectin glycoprotein ligand 1 [PSGL-1]). Our data show that over half of O-glycopeptide identifications in our sample harbor multiple glycosites. As such, O-glycosite localization exclusively at N-terminal residues cannot be assumed. This renders collisional activation inadequate for site-specific analysis for OpeRATOR-derived O-glycopeptides, placing them under the same analytical constraints as the more thoroughly studied O-glycopeptides derived from other proteases. Thus, electron-driven dissociation is needed for accurate site-specific analysis of O-glycopeptides generated using OpeRATOR.

EXPERIMENTAL SECTION

A mixture of O-glycopeptides generated by sequential OpeRATOR and trypsin digestion was generated using four glycoproteins: bovine fetuin (alpha-2-HS-glycoprotein, P12763), recombinant human CD14 (P08571), human fibronectin (P02751), and recombinant human P-selectin glycoprotein ligand 1 (PSGL1; Q14242). OpeRATOR proteolysis was coupled with sialidase cotreatment, and PNGaseF was used to remove N-glycans. Following digestion and desalting, peptides were combined in equal parts by mass for the four proteins and analyzed using 90 min LC-MS/MS product-dependent methods constructed using ETD or EThcD triggered scans (HCD-pd-ETD and HCD-pd-EThcD methods).^36-39 All raw data were searched using O-Pair Search implemented in MetaMorpheus (0.0.308), which is available at https://github.com/smith-chem-wisc/MetaMorpheus.³⁰ A glycan database representing 12 common O-glycans was used, and the “Maximum OGlycan Allowed” parameter was set to 5. Only level 1 and 1b identifications were retained for further analysis, meaning all identifications had spectral evidence in ETD or EThcD spectra for localized glycosites (or had only one plausible glycoform). For glycan modifications shown as H#N#, H represents the number of hexose residues (galactose in O-glycans), and N represents the number of N-acetylhexosamines, which can be N-acetylgalactosamine or N-acetylglucosamine in O-glycans. Spectral annotation was aided by the Interactive Peptide Spectral Annotator (IPSA, http://www.interactivepeptide-spectralannotator.com).⁴⁰ Graphs were generated using OriginPro 2018 with the exception of the alluvial diagram, which was generated using the RAWGraphs web app, licensed under CC BY-NC-SA 4.0.⁴¹ WebLogo was used for Logo plot generation (https://weblogo.berkeley.edu/logo.cgi).⁴² More details are available in theSupporting Information.

RESULTS AND DISCUSSION

O-glycosite localization in OpeRATOR-derived O-glycopeptides has thus far relied on HCD fragmentation and the assumption that the total glycan mass observed can be attributed to a single glycosylated serine or threonine at the N-terminus. Figure 1 provides an example of why collision-based fragmentation is insufficient for O-glycosite localization and instead why electron-driven dissociation is necessary for OpeRATOR-based workflows. An O-glycopeptide from fetuin, ²⁸²SAAGPPVASVVVGPSVVAVPLPLHR³⁰⁶, represents two well-known O-glycosites, Ser-282 and Ser-296, in addition to Ser-290, which was identified as O-glycosylated using OpeRATOR in the EXoO workflow.³³ Two glycopeptidoforms identified using an HCD-pd-EThcD25 method include the top example, with an O-glycosylated Ser-282 (orange) and unmodified Ser-290 and Ser-296 (green; Figure 1a), and the bottom example, with O-glycosylated Ser-282 and Ser-296 (orange) and unmodified Ser-290 (green; Figure 1b). In the top example, b-type peptide fragment ions in the HCD spectrum did not retain the O-glycan (as denoted by “~”), but the EThcD spectrum confirmed that the assumption of an N-terminal Ser-282 modified with H1N1 is correct. Here, HCD was adequate, albeit less than ideal, for glycosite localization; the H1N1 glycan mass is presumably a single core-1 GalNAc-Gal, a common mucin-type O-glycan, but O-mannose (a hexose) can modify similar regions as O-GalNAc on the same glycan (albeit less likely to be the case).⁴³ This distinction is possible with EThcD fragmentation, but not HCD.

Figure 1.

Bovine fetuin sequence ²⁸²SAAGPPVASVVVGPSVVAVPLPLHR³⁰⁶ is a tangible example of the benefits of electron-driven dissociation. Diagrams on the left present two glycopeptidoform possibilities of this sequence, which have both O-glycosylated (orange) and nonmodified (green) serine and threonine residues. These two O-glycopeptidoforms were identified in our data set using HCD-EThcD spectral pairs, including (a) a canonical OpeRATOR-derived O-glycopeptide with a single O-glycosite at the N-terminus and (b) a doubly glycosylated O-glycopeptide from a missed OpeRATOR cleavage that has both an N-terminal O-glycosite site and a second internal O-glycosite. Spectral information is included in Supplemental Data 1.

The doubly O-glycosylated peptide in Figure 1b, however, demonstrates a more problematic issue when relying on HCD. The total O-glycan mass observed is H2N2. Neither b- nor y-type fragments retain glycan modifications in the HCD spectrum (marked with “~”), leaving ambiguity as to which serine residues were modified. OpeRATOR digestion affords the assumption of the N-terminal serine as glycosylated, but H2N2 is equally as plausible of a glycan to observe at a single O-glycosite as H1N1. Even with the assumption that the N-terminal serine has the same H1N1 modification as the top example, ambiguity remained as to whether Ser-290 or Ser-296 was modified with the other H1N1 O-glycan. On the other hand, the EThcD spectrum in the bottom panel provided c- and z^•-type fragment ions that unambiguously localized the H1N1 modification to both Ser-282 and Ser-296.

To understand the prevalence of multiply O-glycosylated peptides present in OpeRATOR digests, we generated a data set comprising four different product-dependent methods with electron-driven dissociation as the triggered scan: HCD-pd-ETD, HCD-pd-EThcD15, HCD-pd-EThcD25, and HCD-pd-EThcD35 (collected in technical triplicate, 12 raw files total). To remove O-glycopeptide identifications resulting from non-OpeRATOR-involved cleavage (e.g., standard trypsin cleavage), identifications either had to have a serine or threonine residue at the N-terminus without the residue immediately preceding it being an arginine or lysine, or had to have the immediately C-terminal residue (following the C-terminal site) be a serine or threonine. Figure 2a provides the proportion of the total O-glycopeptide identifications that had one or multiple localized O-glycosites. Note that O-Pair Search returns a single identification representing two spectra, both an HCD and ETD/EThcD spectrum.³⁰ Approximately 46% of the glycopeptide spectral matches contained a single O-glycosite, indicating that over half of the identifications harbored two or more O-glycosites. We also note that our data were generated using a sialidase mixture to produce asialylated O-glycans, which is known to aid in OpeRATOR digestion. The number of missed cleavages and multiply O-glycosylated peptides would likely be higher if sialidase treatment was omitted. Also worth considering is the lack of perfect efficiency of sialidases (although their efficiency is generally high), which could affect missed cleavages as well. In this data set, we searched for sialylated glycans to understand the extent to which this may be the case; less than 5% of identifications contained sialylated O-glycans, indicating sialidase efficiency is likely a minor issue. Our data also corroborate a recent reanalysis of the EXoO data set using MSFragger-Glyco (a new search algorithm that functions similarly to the O-Pair Search strategy used here), where the authors observed both a substantial number of missed OpeRATOR cleavages and glycan mass modifications indicative of multiple O-glycosites.⁴⁴ In that study, Polasky et al. also identified the co-occurrence of phosphorylation sites on O-glycosylated sequences, furthering the need for electron-driven dissociation to properly localize multiple sites of modification.

Figure 2.

Characteristics of O-glycopeptides derived from OpeRATOR proteolysis. (a) A total of 1183 O-glycopeptide spectral pairs with confidently localized glycosites were detected using HCD-pd-ETD and HCD-pd-EThcD methods. The pie graph shows the proportion of identifications that had one or multiple O-glycosites. (b) The bar graph delineates the proportion of O-glycopeptide spectral pairs that can be grouped into three main classes: (1) identifications that have an O-glycosylated serine or threonine at the N-terminus (light pink; bottom), (2) O-glycopeptides from sequences upstream (i.e., N-terminal) of the OpeRATOR (OPR) cleavage site that indicate a missed cleavage (pink; middle), and (3) O-glycopeptides that do not appear to follow the OpeRATOR cleavage motif (dark pink; top). (c) The alluvial diagram relates the aggregate glycan compositions that were identified on O-glycopeptides with a varying number of localized O-glycosites. Numbers in parentheses provide the number of O-glycopeptide spectral pairs identified for each condition. For glycan modifications shown as H#N#, H represents the number of hexose residues (galactose in O-glycans), and N represents the number of N-acetylhexosamines, which can be N-acetylgalactosamine or N-acetylglucosamine in O-glycans.

Figure 2b examines what proportion of identifications represent O-glycopeptides that are either upstream/N-terminal or downstream/C-terminal from the OpeRATOR cleavage site. Note, “canonical” OpeRATOR glycopeptides with an O-glycosylated N-terminal residue are the downstream/C-terminal case. For O-glycopeptides with one O-glycosite, ~75% of identifications were canonical OpeRATOR peptides with a single N-terminal O-glycosylated serine or threonine residue, while nearly 20% of identified O-glycopeptides were from sequences that were upstream from an OpeRATOR cleavage site. These upstream O-glycosylated peptides would likely evade analysis in approaches that couple solid supports for enrichment and OpeRATOR cleavage for O-glycopeptide release, because they remain tethered to the support matrix. The presence of internal glycosites, whether they occur in canonical OpeRATOR peptides or they occur in peptides that are N-terminal to OpeRATOR cleavage, is supported by the presence of serine and threonine residues occurring 1 to >10 positions away from the cleavage site (both up- and downstream; Figure S1). Additionally, search strategies that only consider peptides starting with serine or threonine in their searches, such as those used in previous OpeRATOR studies, would either be blind to these upstream sequences or could mis-assign their glycosites. A small percentage of O-glycopeptides passed filtering criteria yet did not follow the expected OpeRATOR motif, instead showing only an internal O-glycosite (Figure 2b). These have serine or threonine as their N-terminal residue but are not the result of expected trypsin activity (i.e., not resulting from cleavage at lysine or arginine residues); however, the N-terminal residue is not glycosylated while an internal serine or threonine is glycosylated. Such peptides could be the presence of some nonglycosylation-dependent activity of OpeRATOR at serine and threonine residues or due to other nonspecific cleavage (e.g., chemical degradation or from trace amounts of unknown proteases). Such unexpected cleavage events are likely more prevalent in biofluids with several active endogenous proteases. These types of O-glycopeptides would have their O-glycosites falsely attributed to the N-terminal residue if not carefully examined using ETD-based fragmentation. Interestingly, approximately 48% of all identifications came from the mucin domain of PSGL-1, which could contribute to the proportion of multiply glycosylated sequences observed. That said, mucin O-glycoproteins were discussed as major components of the previous OpeRATOR data sets, making O-glycopeptides from densely O-glycosylated mucins a mainstay rather than an exception when digesting with OpeRATOR. Glycosite macroheterogeneity is also an important consideration. If every potential O-glycosite were occupied in a mucin domain, only the most C-terminal site would exist in a detectable O-glycopeptide because the rest would be cleaved off as single glycosylated amino acids (presuming full OpeRATOR efficiency). Differences in OpeRATOR cleavage could reveal variable partial occupancy of mucin domains, which may prove useful in evaluating mucin glycosylation that is subject to biological regulation. That said, regions with high levels of both macro- and microheterogeneity may lead to complications.

The alluvial diagram in Figure 2c maps the aggregate glycan masses that were identified for O-glycopeptides harboring a varying number of O-glycosites. The only two aggregate glycan masses that are exclusive to singly O-glycosylated peptides are N1 and H1N1, meaning that identification of any other glycan composition using HCD fragmentation can lead to ambiguity in O-glycosite assignment. As noted above, H2N2 is one such aggregate glycan mass with ambiguity, with possibilities including one glycan at a single glycosite or two glycans split between two sites. Here, the majority of identifications with an aggregate composition of H2N2 were the result of two H1N1 O-glycosites rather than the total glycan mass modifying a single residue. Once multiple glycosites were considered on a peptide, we saw a substantial increase in the glycoforms (i.e., glycosites per peptide) that can explain an observed aggregate glycan mass, highlighting the need to localize O-glycans to specific residues with electron-driven methods. The presence of sialic acids, beyond the reduced OpeRATOR efficiency discussed above, would further complicate this issue because it drastically increases the number of possible glycan structures that can be observed, as noted by Polasky et al.⁴⁴ One potential option to eliminate the presence of core-2 structures like H2N2 would be to evaluate oxonium ion ratios, which can indicate the presence of GlcNAc residues rather than exclusively GalNAc.^23,45

To test ETD and EThcD efficacy for OpeRATOR-based workflows, we compared identifications of the four ETD-centric methods; EThcD with collision energies between 25 and 35 nce performed best (Figure S2). Glycans and serine/threonine-rich sequences of O-glycopeptides often generate low charge density precursor ions, which can be challenging for electron-driven fragmentation. More than a third of EThcD25 and EThcD35 identifications were from doubly charged precursor ions (Figure S3), likely contributing to better performance. Benefits of supplemental activation can be even more pronounced for multiply glycosylated peptides that have low charge density but require extensive fragmentation for site localization. As such, other hybrid electron-driven methods may also be well-suited for such analysis,^46-48 although they remain largely untested for O-glycopeptide characterizations.

We also collected a collisional dissociation-centric data set to understand the prevalence of presumably singly O-glycosylated peptides (i.e., N1 and H1N1 modified) compared to presumably multiply O-glycosylated peptides (i.e., those with other aggregate glycan masses; Figure S4). Of those identifications purported to have one O-glycosite (approximately half of the total identifications), only 60% had glycan modification masses that exclusively indicate a single O-glycan (i.e., N1 or H1N1). In combination with Figure 2c above, this means that 40% of the O-glycosylated identifications reported to be singly glycosylated (without localization evidence) are likely to have ambiguous or incorrect O-glycosite assignments if relying on the assumption of a N-terminal O-glycosite. These data further demonstrate the challenges with ambiguity that accompany reliance exclusively on HCD fragmentation with OpeRATOR digestion, and they show that HCD-only methods fail to properly identify a significant proportion of O-glycopeptides from the sample.

CONCLUSIONS

O-glycopeptides often contain multiple potential O-glycosites that can be modified by a heterogeneous pool of labile O-glycans. Thus, site-specific analysis of O-glycosylation typically requires alternative MS/MS fragmentation methods (namely electron-driven dissociation such as electron capture dissociation (ECD), electron transfer dissociation methods (ETD), and their derivative methods^49-51) that can retain intact glycan moieties on peptide backbone fragments. Nevertheless, approaches that could make site-specific O-glycopeptide analysis amenable to collisional-dissociation-centric methods would be valuable. One potential benefit of the recently described O-glycoprotease OpeRATOR is the reported ability to localize O-glycosites using collisional dissociation due to the N-terminal location of modified sites within the sequence. Indeed, previous literature dismisses the need for electron-driven dissociation for site-specific analysis when using OpeRATOR. Here, we tested that claim by investigating how often O-glycopeptides generated from OpeRATOR proteolysis contain only one N-terminal O-glycosite versus multiple glycosites. We show that over half of O-glycopeptides from OpeRATOR digestion of our sample contain multiple glycosites, where the total glycan mass addition cannot be solely attributed to the N-terminal residue. This alone indicates that collision-based fragmentation (e.g., HCD) is not sufficient for site-specific characterization of OpeRATOR O-glycopeptides. Additionally, we identified cases where glycan compositions that could ostensibly be attributed to a single O-glycosite were actually several smaller glycans distributed among multiple O-glycosites in OpeRATOR O-glycopeptides, as confirmed by electron-based fragmentation.

It is worth noting here that these data do not invalidate the results of previous studies using OpeRATOR and HCD-only methods. Their results appear to contain quality identifications and confidently localized O-glycosites, especially considering the high proportion of reported H1N1 modifications. Also, HCD-only methods remain effective for generating O-glycopeptide identifications (from OpeRATOR proteolysis and beyond) that include peptide sequence and total glycan mass shift, which can be informative despite the lack of site-specific information. Creative alterations to workflows can improve O-glycosite localization for HCD-centric methods, e.g., the recently reported EXoO-Tn study that can indicate the number of O-glycans present in the sequence using heavy labeled monosaccharides.³⁵ Yet, ambiguities can still remain. Rather than discard previous work, this study points out (1) that ambiguity is possible when using OpeRATOR for O-glycosite localization in a considerable number of identifications when solely utilizing HCD and (2) that many multiply O-glycosylated peptides generated by OpeRATOR proteolysis go undetected when HCD-only methods are used to presume a single O-glycosite at the peptide N-terminus. As such, it is clear that use of the O-glycoprotease OpeRATOR is a valuable tool for O-glycosite mapping, but methods relying on this approach cannot forego electron-based dissociation, especially EThcD methods, for localizing O-glycosites.