Glyco

Title : Large-scale glycoproteome characterization enabled by AI-ETD

Abstract :

Given the AI-ETD method is fast and easily automated, we reasoned the technique could provide analysis of the glycoproteome at a large-scale
To test this hypothesis we extracted proteins from mouse brain lysate, digested them with trypsin, enriched for glycosylated peptides , and performed high-throughput LC–MS/MS analysis using AI-ETD scans triggered by the presence of oxonium ions in HCD scans
In total, we identified 5662 unique N-glycopeptides (24,099 glycopeptide spectral matches) mapping to 1545 unique N-glycosites on 771 glycoproteins with 117 different glycan com positions, which were included in a database compiled from literature on previous mouse and rat brain glycosylation studies
These data are the result of several steps of post-Byonic search filtering, which were performed because caveats still exist in automated glycopeptide identification—as evidenced by the current HUPO glycoproteomics initiative (https://hupo.org/ HPP-News/6272119)
Note, we do not offer any fundamentally new approach to address such challenges here, but rather we present AI-ETD data for large-scale glycoproteomics using the tools that are currently available in the field
See the Methods for discussion of the six post-Byonic search filtering steps we performed
Following post-search filtering, no decoy peptides remained in the dataset
All the data reported here comprise tryptic N-glycopeptides carrying only one glycan modification and have a DeltaMod score that indicates the correct glycosite has been identified within the confidence range suggested by Byonic
With this extensive dataset in hand we next characterized several Figures of Merit
First, we examined the percentage of cleaved bonds observed relative to total possible backbones bonds (for both peptide and glycan backbones, Fig. 1b)
Here we achieve 89% median peptide backbone sequence coverage and 78% median glycan sequence coverage with AI-ETD, which significantly outperforms HCD (Supplementary Fig. 5a)
Figure 1c presents the average distribution of explainable signal amongst fragment ion types in AI-ETD spectra
On average AI-ETD produces relatively equal proportions of signal in Y-type and peptide backbone fragments (41% and 45%, respectively), compared to HCD which has less signal in peptide backbone fragments and more in Y-type and oxonium ions (Supplementary Fig. 5b)
This is congruent with observations presented in Supplementary Fig. 2 (discussed above)
Approximately 29% and 46% of total ion current could be explained on average in AI-ETD and HCD spectra, respectively, but we only considered the following fragment ion classes: (1) unmodified peptide backbone fragments (i.e., b/y/c/z-type), (2) peptide backbone fragments with intact glycan still attached, (3) peptide backbone fragments with only a HexNAc moiety still attached, (4) Y-type ions (intact peptide plus glycan fragments ), and (5) oxonium ions/glycan B-type ions
It is possible that photoactivation generated some degree of glycan fragmentation on peptide backbone fragments , which could provide more explainable signal in AI-ETD spectra, and this will be the subject of future work
Even so, 71% of AI-ETD spectra (compared to <4% of HCD spectra) contained fragments with the intact glycan species retained on peptide backbone fragments
This percentage would likely increase (especially for AI-ETD) by extending the m/z range of MS/MS scans above 2000 Th
Note, intact glycopeptides have considerably larger precursor m/z distributions (Supplementary Fig. 6), as compared to unmodified peptides , and these low-charge density precursors (z ≤ 3) can be a challenge to dissociate
Even so, AI-ETD provided robust fragmentation that often facilitated identification of these challenging glycopeptides (Fig. 1d)
AI-ETD spectra provided evidence for 4680 unique N-glycopeptides (83%) and 1361 (88%) of the glycosites reported in this study, with the remaining identifications/glycosites supported only by HCD spectral evidence
In all , these data represent the one of the most in-depth N-glycoproteome characterization to date to rely on intact glycopeptide identifications (Fig. 1e)
Importantly, this method is amenable to in vivo sources, as demonstrated here, for applicability to practically any mammalian system (workflow in Supplementary Fig. 7)
The ability to profile glycosites with intact glycan modifications at this scale provides opportunities to investigate system-wide glycosylation patterns
Despite differences in enrichment strategies and fragmentation methods, the overlap in identified glycosites is relatively high between this study and two other recent intact glycopeptide studies in mouse brain (Fig. 2a) and deglycoproteomic datasets (Supplementary Fig. 8)
Supplementary Figure 9 compares one example of overlapping glycosites between this study and the Liu et al. dataset, demonstrating that similar glycosites and glycan heterogeneity were identified on integrin alpha-1 in mouse brain, for example, and that AI-ETD methods further add to the number of glycosites and glycosite-glycan combinations observed
Figure 2b demonstrates that ~69% of the glycosites identified in this study are annotated as glycosites in the UniProt database
Of the 1065 UniProt-annotated glycosites , the majority of them were assigned via ‘sequence analysis,’ which indicates a prediction based on presence of the N-X-S/T sequon rather than by experimental observation
In total, we provide experimental evidence for over 850 UniProt-predicted glycosites , in addition to identifying nearly 200 previously observed glycosites
Expected N-X-S and N-X-T sequons were observed in our identified glycosites (Fig. 2c), with ~59% of the glycosites having the N-X-T sequon
Figure 2d displays the percentage of glycosites containing high mannose glycans, fucosylated glycans, or sialylated glycans, which resembles previous studies (although Liu et al. observed significantly higher proportions of fucosylated glycopeptides )
Note, this calculation did not consider glycosites exclusively, so one site can count toward multiple types if multiple glycans were identified at that site
Gene ontology (GO) enrichments of functional category terms from identified glycoproteins are available in Supplementary Fig. 10, which shows expected enriched terms such as glycoprotein , several membrane terms, and extraceullar/secreted protein related terms
See the Discussion section for more about differences in glycosylation profiles between this study and other published datasets, where we also discuss the implications of our lectin enrichment strategy compared to other strategies

Output (sent_index, trigger, protein, sugar, site):

15. ions, , -, -, fragments
15. ions, , -, -, peptide
19. glycopeptides, , -, -, glycopeptides
2. glycosylated, , -, -, peptides
20. glycopeptides, , -, -, glycopeptides
21. N-glycopeptides, , -, -, N-glycopeptides
21. N-glycopeptides, , -, -, glycosites
21. glycosites, , -, -, glycosites
21. identifications/glycosites, , -, -, identifications/glycosites
22. glycopeptide, , -, -, glycopeptide
24. glycosites, , -, -, glycosites
25. glycopeptide, , -, -, glycopeptide
25. glycosites, , -, -, glycosites
26. glycosites, , -, -, glycosites
27. glycosites, , -, -, glycosites
28. glycosites, , -, -, glycosites
29. glycosites, , -, -, glycosites
3. N-glycopeptides, , -, -, N-glycopeptides
3. N-glycosites, , -, -, N-glycosites
3. glycopeptide, , -, -, glycopeptide
3. glycoproteins, , glycoproteins, -, -
30. glycosites, , -, -, glycosites
31. fucosylated, , -, -, glycopeptides
31. glycopeptides, , -, -, glycopeptides
31. glycosites, , -, -, glycosites
32. glycosites, , -, -, glycosites
33. glycoprotein, , glycoprotein, -, -
33. glycoproteins, , glycoproteins, -, -
4. glycopeptide, , -, -, glycopeptide
8. N-glycopeptides, , -, -, N-glycopeptides
8. glycosite, , -, -, glycosite

Output(Part-Of) (sent_index, protein, site):

27. database, glycosites
3. glycoproteins, N-glycosites
3. glycoproteins, positions,

*Output_Site_Fusion* (sent_index, protein, sugar, site):