Expanding Glycopeptide Identification with Match-Between-Glycans in FragPipe

This paper introduces Match-Between-Glycans (MBG), a method integrated into FragPipe that expands glycopeptide identification by leveraging MS1 signals displaced by monosaccharide units from known identifications, thereby recovering low-abundance or complex glycopeptides without drastically increasing the search space.

The Gist

The "Missing Puzzle Piece" Detective: How MBG Finds Hidden Sugar Tags on Proteins

Imagine your body is a massive, bustling city. The proteins are the workers, the buildings, and the machines keeping everything running. But these proteins aren't just plain; they are often decorated with sugar tags (called glycans). These sugar tags act like ID badges, delivery addresses, or "Do Not Disturb" signs that tell the protein where to go, how long to stay, and what to do.

Scientists use a high-tech microscope called Mass Spectrometry to take pictures of these protein-sugar combinations. However, there's a big problem: the sugar tags are incredibly complex and vary wildly. It's like trying to identify a specific car in a parking lot where every car has a slightly different color, a different roof rack, and a different bumper sticker.

The Problem: The "Stuttering" Camera

In the past, when scientists tried to identify these sugar-tagged proteins, they relied on a method called DDA (Data-Dependent Acquisition). Think of this like a camera that takes a photo of a protein, then tries to smash it open to see the pieces inside (the sugar and the protein) to figure out what it is.

But here's the catch:

1. Low Battery: If a protein is rare (low abundance), the camera might not get a good enough "smash" photo to see the details.
2. Too Many Options: If a protein has many different sugar variations, the camera gets confused and might skip taking a photo of the rarer ones entirely.

As a result, scientists were missing a huge chunk of the story. They knew a protein existed, but they didn't know which sugar version of it was present. It's like knowing a person is in the room, but not knowing if they are wearing a red hat, a blue hat, or no hat at all.

The Solution: Meet MBG (Match-Between-Glycans)

The authors of this paper introduced a new tool called MBG. You can think of MBG as a super-smart detective who doesn't just look at the smashed pieces; they look at the pattern of the parking lot.

Here is how MBG works, using a simple analogy:

The "Sugar Ladder" Analogy
Imagine the sugar tags are like rungs on a ladder.

• Rung 1: A protein with a small sugar.
• Rung 2: The same protein with one extra sugar unit.
• Rung 3: The same protein with two extra sugar units.

In the real world, these "rungs" (glycoforms) usually line up perfectly in time. If you see a protein with a small sugar at 10:00 AM, the version with one extra sugar will almost always show up at 10:02 AM. The version with two extra sugars will show up at 10:04 AM.

How MBG Solves the Mystery:

1. The Anchor: MBG starts with the proteins that were successfully identified by the old camera method (the "Anchors").
2. The Prediction: It knows the "rules of the road." It knows that adding one sugar unit usually shifts the arrival time by exactly 2 minutes.
3. The Search: It looks at the raw data for a signal that arrived at 10:02 AM. Even if the camera didn't smash that specific signal to get a photo, MBG sees the signal and says, "Hey, this arrived exactly 2 minutes after the Anchor. It must be the 'one extra sugar' version!"
4. The Verification: It checks if the signal is strong enough and fits the pattern. If it does, it adds it to the list of discoveries.

Why This is a Big Deal

The paper tested this detective on three different "cities" (datasets):

1. The Simple City (Yeast): In yeast, the sugar tags are very uniform. MBG found 23.6% more sugar-protein combinations than before. It was like finding a whole new neighborhood of houses that the old map missed.
2. The Complex City (Human Blood): Human blood is messy and full of complex sugars. MBG found more sialylated (a specific type of sugar) and fucosylated tags. These are crucial because changes in these tags are often signs of diseases like cancer. MBG helped spot these "low-abundance" signals that were previously invisible.
3. The Mystery City (Brain Tumors): In a study of brain tumors, MBG found extra sugar tags that helped scientists see the difference between healthy tissue and tumors more clearly.

The "Bonus" Superpower: Finding the Unusual

Usually, if a protein has a weird chemical attachment (like a metal ion or a rare phosphate group), scientists have to tell the computer to look for it specifically, which slows everything down.

MBG is clever enough to say, "Wait, this signal looks exactly like the 'Anchor' protein, but it's slightly heavier. Maybe it has a metal ion attached?" It can find these rare, weird modifications without needing a pre-made list of them. It's like a detective finding a suspect wearing a disguise without needing a photo of that specific disguise beforehand.

The Bottom Line

MBG is a "one-click" upgrade for protein science.

It doesn't require new equipment or new experiments. It just takes the data scientists already have and uses a clever logic trick (looking for patterns in time and mass) to fill in the blanks. It turns a blurry, incomplete picture of our body's sugar tags into a high-definition, complete map.

This means scientists can now see the "hidden" parts of the glycoproteome, leading to better understanding of diseases and potentially better ways to diagnose them. It's like finally getting the full instruction manual for the body's most complex machinery.

Technical Summary

1. Problem Statement

Glycosylation is a critical yet highly complex post-translational modification (PTM). Mass spectrometry (MS)-based glycoproteomics faces significant challenges in comprehensive identification due to:

• Stochastic Sampling: In Data-Dependent Acquisition (DDA), low-abundance glycopeptides are often missed because they are not selected for fragmentation (MS2), or the resulting spectra are of insufficient quality for confident assignment.
• Signal Dilution: A single peptide backbone can carry multiple glycoforms (differing by monosaccharide units). This heterogeneity dilutes the ion signal for any specific glycoform, making detection difficult compared to standard proteomics.
• Search Space Limitations: Traditional database searches require predefined glycan compositions. Including rare adducts (e.g., metal ions) or rare modifications (e.g., phosphorylation) drastically expands the search space, reducing sensitivity and increasing false discovery rates (FDR).
• Incomplete Coverage: Current workflows often fail to recover the full spectrum of glycoforms at a specific glycosite, leading to missing values in quantitative studies and incomplete biological insights.

2. Methodology: Match-Between-Glycans (MBG)

The authors introduce Match-Between-Glycans (MBG), a modular, MS1-based inference method integrated into the FragPipe ecosystem. It operates as a post-processing step following a standard high-confidence database search (e.g., MSFragger-Glyco).

Core Workflow:

1. Seed Selection: MBG starts with high-confidence glycopeptide identifications (Parent PSMs) that meet specific criteria (minimum PSM count, glycan q-value, and minimum distinct glycans per site).
2. Glycoform Inference: Based on the known mass differences of monosaccharides (e.g., Hex, HexNAc, NeuAc) or adducts (e.g., NH₄⁺, Fe³⁺), MBG generates potential "neighboring" glycoforms that differ from the seed by one or more units.
3. MS1 Feature Matching:
   • Mass: Searches for MS1 features with the calculated mass shift.
   • Retention Time (RT) & Ion Mobility (IM): Utilizes learned median RT and IM shifts observed in the dataset for specific glycan transitions. If specific shifts are unavailable, it uses general monosaccharide unit shifts.
   • Tolerance: Matches are evaluated within user-defined RT/IM and m/z tolerance windows.
4. Decoy Generation & Scoring:
   • For every target candidate, a decoy is generated with the same RT/IM but a mass shifted by +11 Da (to avoid isotopic interference).
   • LDA Modeling: Seven features are used to score candidates: RT/IM shift, mass error, precursor intensity, Y0/Y1 ion intensity, isotopic envelope shape (Kullback-Leibler divergence), and shift frequency.
   • FDR Control: Target-decoy FDR is applied at the precursor level to ensure statistical rigor.
5. Output: Validated inferred glycopeptides are written back to the PSM table for downstream quantification and analysis.

Key Parameters:

• Maximum glycan shift steps (chain depth).
• Minimum PSMs and glycans per site for seed selection.
• Specific mass shifts allowed (e.g., Hex, Fuc, NeuAc, Adducts).

3. Key Contributions

• Workflow Integration: MBG is seamlessly integrated into FragPipe, allowing "one-click" operation without disrupting existing glycoproteomics pipelines.
• MS1-Driven Expansion: It recovers glycopeptides that lack MS2 spectra or have poor quality spectra by leveraging the physicochemical consistency of glycan elution behavior.
• Adduct/Modification Discovery: It identifies glycans with adducts (NH₄⁺, Fe³⁺, Na⁺) or rare modifications (e.g., Mannose-6-Phosphate) without requiring them to be included in the initial database search, thus avoiding search space explosion.
• Robust FDR Control: Unlike previous methods that offered only "confidence levels," MBG implements a rigorous target-decoy FDR framework at the precursor level.

4. Results and Validation

The method was validated across four diverse datasets:

• Fission Yeast (Benchmark):

  • Result: MBG increased unique glycopeptide identifications by 23.6% at 5% FDR.
  • Validation: 92 of 202 new IDs were confirmed in other runs; 135 had supporting MS2 spectra. Spectral similarity to parent spectra was significantly higher than random controls.
  • Specificity: An entrapment search showed a false positive rate of only 0.6% for biologically implausible glycans (NeuAc in yeast).
  • Sensitivity: Recovered low-abundance glycopeptides missed by standard searches.

• Human Plasma (PASEF):

  • Result: Increased identifications by 14.6% (1,144 to 1,311 unique glycopeptides).
  • Insight: Successfully recovered complex sialylated and fucosylated glycans. Notably, it recovered low-abundance fucosylated species critical for biomarker discovery.
  • RT/IM Behavior: While sialic acid (NeuAc) showed variable RT shifts, Ion Mobility (IM) shifts remained tightly distributed, validating IM as a robust filter.

• Glioblastoma (GBM) TMT Dataset:

  • Result: Identified 740 additional glycopeptides exclusively via MBG.
  • Application: Enabled differential analysis revealing clear separation between tumor and normal tissue. Confirmed known biological trends (downregulation of fucosylation, upregulation of sialylation).
  • Constraint: For isobaric labeling (TMT), MBG requires the inferred glycopeptide to have a corresponding MS2 spectrum to extract reporter ions, ensuring quantitative validity.

• Adduct and Modification Identification:

  • Adducts: Identified NH₄⁺ and Fe³⁺ adducts in mouse liver data without prior database inclusion. Validated Fe³⁺ assignments via characteristic isotope patterns (mass defect) and diagnostic B-ions.
  • Modifications: Detected Mannose-6-Phosphate (M6P) in mouse brain data, inferring 4 new phosphorylated compositions not in the original database. High-precision mass analysis confirmed phosphorylation over sulfation.

5. Significance

• Completeness: MBG significantly closes the gap between observed and expected glycoproteome coverage, providing a more complete quantitative profile of glycosylation at each glycosite.
• Efficiency: It achieves these gains with minimal computational overhead compared to expanding the primary database search space.
• Biological Relevance: By recovering low-abundance species and rare modifications, MBG enhances the ability to detect subtle biological variations and disease biomarkers (e.g., in cancer or plasma proteomics).
• Accessibility: As a module within FragPipe, it makes advanced glycopeptide inference accessible to the broader community without requiring custom coding or complex workflow adjustments.

Limitations:

• Accuracy depends on the density of observed glycan transitions in the dataset.
• RT shifts for sialylated species can be variable due to isomerism and linkage position.
• Does not currently account for complex structural isomers that share the same mass and RT/IM.

In conclusion, MBG represents a robust, data-driven strategy to expand glycopeptide identification, leveraging the predictable physicochemical properties of glycan transitions to overcome the limitations of stochastic MS sampling.