Importance of taking Single Amino Acid Variant and accessory proteome variability into account in Data Independent Acquisition Proteomics: illustrated with Legionella pneumophila analysis

This study demonstrates that integrating single amino acid variants and accessory proteome variability into a DIA-NN workflow significantly enhances protein identification and coverage in *Legionella pneumophila* proteomics, enabling more accurate bacterial proteotyping and a deeper understanding of allelic diversity.

The Gist

The Big Picture: Finding the "Real" Legionella

Imagine you are a detective trying to identify a specific criminal (a bacteria called Legionella pneumophila) at a crime scene. Usually, police rely on a standard "Wanted Poster" (a reference database) that shows what the criminal should look like.

However, in the real world, criminals often wear disguises, change their hairstyles, or have slight scars. If you only look for the person exactly as they appear on the standard poster, you might miss them, or worse, you might arrest the wrong person who just happens to look similar.

This paper is about upgrading the police force's toolkit. The researchers developed a new method to catch Legionella bacteria not just by their "standard" look, but by noticing their unique, individual quirks (genetic variations).

The Problem: The "One-Size-Fits-All" Database

In the world of protein science (proteomics), scientists use a technique called DIA (Data Independent Acquisition) to take a "snapshot" of all the proteins inside a bacteria. To make sense of these snapshots, they compare them against a digital library of known proteins.

The Old Way (The Reference Database):
Scientists used to use a single "Master Blueprint" (the reference strain) for all Legionella.

• The Flaw: If a specific bacteria had a tiny mutation (like a single letter change in its DNA code), the Master Blueprint didn't have that version.
• The Result: The computer would either miss the protein entirely, or it would force a match with the closest-looking protein on the list, leading to mistakes. It's like trying to identify a person with red hair using a database that only has photos of people with brown hair. You might guess it's a brown-haired person with bad lighting, but you'd be wrong.

The Solution: The "Customizable" Database

The team created a new workflow that builds a custom library for every single bacteria sample they analyze.

1. Sequencing the DNA: First, they read the full genetic code of 15 different Legionella strains.
2. Grouping the Variants: They realized that while these bacteria are different, they are still related. They used a smart algorithm to group similar proteins together. Think of this like a family tree.
   • The "Canonical" Protein: The "Head of the Family" (the standard version).
   • The "Variant" Proteins: The cousins with slightly different features (mutations).
3. The "Chimeric" Trick: To make the computer run faster, they created "Frankenstein" sequences (chimeric proteins). They took all the unique parts of a family of proteins and stitched them into one long string. This allowed the computer to scan the data much faster without losing any detail.

The Results: Catching More Criminals

When they tested this new method against the old one:

• More Hits: They found significantly more proteins. In some cases, they identified 23% more proteins than the old method.
• Spotting the Differences: They could tell the difference between two bacteria that looked almost identical but had a tiny genetic mutation. This is crucial for understanding why some bacteria are more dangerous or resistant to antibiotics than others.
• Accuracy: They didn't just find more proteins; they found the right proteins. The "false positive" rate (mistaken identity) remained very low.

A Real-World Example: The Ribosomal Protein

The paper gives a great example involving a protein called "30S ribosomal protein S1."

• The Scenario: One specific bacteria (Isolate 10) had a mutation where one amino acid (a building block of protein) changed from Serine to Threonine.
• The Old Method: The computer looked at the data, didn't see the "Serine" version in its library, and falsely concluded the bacteria had the standard "Threonine" version. It was a case of mistaken identity.
• The New Method: Because the new library included the "Threonine" mutation, the computer correctly identified the bacteria as having the mutated version. It was like finally seeing the red hair in the photo and saying, "Aha! That's the guy!"

Why Does This Matter?

This isn't just about counting proteins; it's about Proteotyping.

Just as police use fingerprints to distinguish between two people with the same name, this method allows scientists to distinguish between different strains of Legionella based on their unique protein "fingerprints."

• Better Medicine: If we can see exactly which version of a bacteria is causing an infection, we can understand if it's likely to be resistant to treatment.
• Faster Science: By using their "Frankenstein" (chimeric) libraries, they made the computer analysis much faster, meaning scientists can get answers sooner.

The Takeaway

The authors have built a smarter, more flexible way to look at bacteria. Instead of forcing every bacteria to fit into a single, rigid mold, they built a system that appreciates the unique details of every individual strain. This leads to a clearer picture of how these bacteria work, how they evolve, and how we can fight them.

Technical Summary

1. Problem Statement

In Data-Independent Acquisition (DIA) proteomics, peptide identification relies heavily on comparing experimental spectra against a database or spectral library. Standard workflows typically use a reference proteome (a single consensus sequence per protein). This approach presents two major limitations when analyzing bacterial strains with high genetic diversity:

• Missed Variants: It fails to identify Single Amino Acid Variants (SAAVs) or accessory proteins unique to specific strains, leading to false negatives.
• False Positives: When a sample contains a variant peptide not in the database, the search algorithm may incorrectly match it to a similar peptide from the reference sequence (a "neighbor peptide" or "homeometric peptide"), resulting in false positive identifications.
• Database Trade-off: Simply adding all possible genomic variants to a database increases its size, which raises the statistical threshold for identification (False Discovery Rate control), potentially reducing sensitivity and increasing false negatives.

The authors aimed to develop a workflow that integrates allelic variability and accessory proteome diversity into DIA analysis without compromising identification accuracy or computational efficiency.

2. Methodology

The study utilized 15 Legionella pneumophila isolates (including reference strains and clinical isolates) to develop and validate a custom bioinformatics workflow.

A. Genomic Data Processing & Clustering

• Sequencing: Whole Genome Sequencing (WGS) was performed using Illumina and Nanopore (both "old" and "new" chemistry) technologies.
• Clustering Strategy: Instead of using a single reference, the authors used MMseqs2 to cluster protein sequences from all 15 isolates.
  • Parameters: Sequences were grouped into "homology clusters" based on 80% sequence coverage and 80% sequence identity (determined as optimal after testing 90% identity and different Nanopore chemistries).
  • Canonical vs. Variant: Within each cluster, one sequence was designated as the Canonical Protein, and the others were Variant Sequences.
  • Database Construction: This resulted in a Variable Database (varDB) containing 5,021 canonical proteins and ~18,000 variant sequences, compared to the standard Reference Database (refDB) with ~3,200 proteins.

B. Spectral Library Generation & DIA Analysis

• In-silico Digestion: Both refDB and varDB were digested in-silico (Trypsin) to generate peptide lists.
• Spectral Libraries: Libraries were generated using DIA-NN (v1.9).
  • refSL: Based on the reference proteome.
  • varSL: Based on the variable proteome.
  • varSL-Chim (Optimization): To reduce computational time, a "chimeric" library was created by concatenating all peptides from a protein group into a single sequence. This reduced redundancy without losing peptide information, as protein inference is handled post-DIA-NN.
• DIA Acquisition: Samples were analyzed using a Sciex ZenoTOF 7600 system in SWATH mode with 65 variable windows.

C. Protein Inference Logic

The authors developed a custom inference logic to handle the complex peptide specificity:

1. Peptide Categorization:
   • Variant-specific: Unique to a single variant sequence.
   • Canonical-specific: Shared by all/many variants within a cluster but not other clusters.
   • Non-specific: Shared across different canonical proteins.
2. Identification Criteria:
   • Canonical Protein: Identified if $\ge$ 2 canonical-specific peptides are detected.
   • Variant Sequence: Identified if $\ge$ 1 variant-specific peptide AND 1 canonical-specific peptide are detected (or 2 variant-specific peptides).

3. Key Contributions

1. Novel Workflow: A pipeline that integrates genomic variability (SAAVs and accessory genomes) directly into DIA spectral libraries, moving beyond the "one reference per species" paradigm.
2. Optimized Clustering: Validation of specific clustering parameters (80% identity/coverage) that balance biological relevance with the reduction of sequencing-error-induced artifacts.
3. Chimeric Library Strategy: A method to drastically reduce spectral library size (by ~3x) and processing time by using chimeric sequences for library generation while retaining the full varDB for final protein inference.
4. Proteotyping Capability: Demonstrating that proteomic data, when analyzed with variability-aware databases, can accurately discriminate bacterial strains, mirroring genomic phylogenies.

4. Key Results

• Increased Identification: Using the varDB increased the number of identified proteins by an average of 6% across isolates, with gains up to 23% in highly divergent strains (e.g., Isolate 1).
• Variant Detection: The workflow successfully identified 28% to 77% of variant-specific sequences in each isolate.
• False Positive Rates:
  • Canonical proteins: Very low false positive rate (0.06% – 0.16%).
  • Variant sequences: Slightly higher (1% – 2.5%) due to less stringent identification criteria (relying on a single variant-specific peptide), but still robust.
• SAAV Resolution: In a case study of the "30S ribosomal protein S1," the varDB correctly identified specific SAAVs (Serine vs. Threonine; Glutamic vs. Aspartic acid) that were misidentified as the reference sequence when using the refDB. This proved the method's ability to resolve "neighbor peptides."
• Proteotyping Accuracy: Hierarchical clustering (Jaccard distance) of proteomic data using varDB produced dendrograms that closely matched proteogenomic (genomic) clustering, correctly grouping isolates into four distinct clusters. The reference database approach failed to distinguish these groups effectively.
• Efficiency: The varSL-Chim approach reduced library generation time by 2x and total sample reprocessing time by ~4 hours for 45 samples, with no loss in identification performance compared to the full varSL.

5. Significance

This study demonstrates that ignoring allelic variability in DIA proteomics leads to significant information loss and potential misinterpretation of bacterial phenotypes.

• Biological Insight: It enables a more comprehensive view of the bacterial proteome, capturing accessory genes and specific mutations that drive virulence or antibiotic resistance.
• Strain Typing: It establishes a robust method for bacterial proteotyping, allowing for rapid strain differentiation based on protein expression and sequence variation, which is crucial for epidemiological tracking of pathogens like L. pneumophila.
• Scalability: The proposed workflow is adaptable to other bacterial species and can be expanded to include variability from public databases (e.g., UniProt), offering a scalable solution for precision microbiology.

In conclusion, the authors provide a validated, efficient, and accurate framework for integrating genomic diversity into proteomic analysis, significantly enhancing the reliability and depth of bacterial proteomics.