A comprehensive reference database to support untargeted metabolomics in Pseuudomonas putida

To address the lack of comprehensive metabolomics resources for the model organism *Pseudomonas putida* KT2440, the authors developed the publicly available PPMDB v1, a reference database that integrates curated and computationally predicted metabolites with analytical properties and pathway annotations to facilitate untargeted metabolomics and biological interpretation.

The Gist

Imagine Pseudomonas putida (let's call him "Penny") as a super-powered, tiny biological factory. Scientists love Penny because he can eat almost anything, survive in harsh environments, and be reprogrammed to turn waste into useful things like biofuels, plastics, and medicines. He is the "champion athlete" of the synthetic biology world.

However, there was a major problem: We didn't have a complete instruction manual for Penny's internal chemistry.

The Problem: A Library with Missing Books

Think of metabolomics (the study of all the tiny chemicals inside a cell) as trying to solve a massive jigsaw puzzle. To solve it, you need a picture on the box to know what the pieces look like.

For other famous organisms like E. coli or humans, scientists have huge, well-organized libraries (databases) containing pictures of every chemical piece. But for Penny, the library was a mess. It was like trying to solve a puzzle in the dark, using a flashlight that only lit up a few corners. Scientists had to hunt for chemical information in scattered books, old papers, and different websites, and even then, they were missing thousands of pieces. This made it very hard to understand exactly what Penny was doing when scientists tried to engineer him for new jobs.

The Solution: Building the "Penny Metabolome Database" (PPMDB)

The authors of this paper decided to build a centralized, super-library specifically for Penny. They called it PPMDB v1.

Here is how they built it, using some creative analogies:

1. The "Scavenger Hunt" (Curation)
First, they went on a massive scavenger hunt. They gathered every chemical they knew Penny produced from three main sources:

• BioCyc & BiGG: These are like the official "textbooks" of Penny's known biology.
• Scientific Literature: They read hundreds of recent research papers to find chemicals that scientists had engineered into Penny recently.
• Result: They collected about 2,000 known chemical "pieces."

2. The "Crystal Ball" (Computational Prediction)
But 2,000 pieces weren't enough. Penny is complex, and scientists are constantly inventing new ways to make him produce new things. So, the team used a "crystal ball" (a computer program called BioTransformer) to predict what other chemicals Penny might encounter or produce.

• They asked the computer: "If Penny eats this weird plastic, what tiny chemical pieces will he break it down into?"
• This added another 4,000+ predicted pieces to the library.

3. The "Fingerprinting Station" (Analytical Properties)
Just listing the names of chemicals isn't enough to find them in a real experiment. You need their "fingerprints."

• Imagine trying to find a specific person in a crowd. Knowing their name isn't enough; you need to know their height, weight, and what their voice sounds like.
• The team used advanced computer tools to predict these "fingerprints" for every chemical in the library:
  • Mass: How heavy is the molecule?
  • Collision Cross-Section (CCS): How "bouncy" or shaped is it when it hits air?
  • Spectra: What does it look like under a super-microscope (Mass Spec) or a special infrared camera?
• Now, when a scientist runs an experiment, they can match the "fingerprint" they see against this massive library to instantly identify what they found.

4. The "Road Map" (Pathways)
Finally, they didn't just list the chemicals; they drew the road map. They connected the chemicals to show how they turn into one another (reactions) and which "streets" (pathways) they travel on. This helps scientists understand the story of what Penny is doing, not just the list of ingredients.

Why This Matters

Before this paper, studying Penny's chemistry was like trying to navigate a new city with a torn, incomplete map. You might get lost, or miss the most interesting parts.

Now, scientists have a GPS.

• Faster Discovery: They can instantly identify what chemicals are present in their experiments.
• Better Engineering: They can see exactly where to tweak Penny's factory to make more of the good stuff and less of the waste.
• Open Source: The best part? They put this entire library online for free. Anyone can use it to help build a better, greener future using Penny.

In short, this paper gave the scientific community the ultimate reference guide to unlock the full potential of one of nature's most versatile microscopic factories.

Technical Summary

1. Problem Statement

Pseudomonas putida strain KT2440 is a premier model organism for synthetic biology, bioremediation, and bioeconomy applications due to its metabolic robustness and adaptability. However, a critical gap exists in its research infrastructure: there is no comprehensive, organism-specific metabolomics database comparable to those available for E. coli (ECMDB), yeast (YMDB), or humans (HMDB).

Current limitations include:

• Fragmented Resources: Researchers must rely on disparate databases (e.g., MassBank of North America) or general computational tools that lack P. putida-specific context.
• Limited Scope: Existing resources (like BioCyc or BiGG) focus primarily on genomic reconstructions and central metabolism, missing hundreds of metabolites relevant to engineered strains and niche biotechnological applications.
• Annotation Barriers: The lack of curated analytical reference data (e.g., mass spectra, collision cross-sections) hinders untargeted metabolomics, making it difficult to annotate experimental observations and derive biological insights.

2. Methodology

The authors developed the P. putida Metabolome Reference Database (PPMDB v1) through a multi-step pipeline combining manual curation, literature mining, and computational prediction.

A. Data Curation and Integration

• Primary Sources: Metabolites were sourced from:
  • BioCyc (PPUT160488CYC v29.0): 2,078 initial compounds.
  • BiGG Database: Metabolites from two metabolic reconstructions (iJN746 and iJN1463).
  • Literature Mining: 58 peer-reviewed publications (2019–2024) focusing on engineered P. putida metabolism yielded 305 unique metabolites.
• Cleaning: Metadata (InChI, SMILES, molecular weight) was populated via PubChem. Duplicate InChI keys were merged, and entries lacking essential metadata were excluded.
• Final Curated Set: 2,022 unique literature-sourced metabolites were retained after deduplication and merging.

B. Computational Expansion

To address the "chemical space" gap, the authors used BioTransformer (v3) to predict environmental microbial transformations.

• Process: 2,166 curated metabolites were submitted to BioTransformer using the "Environmental Microbial Transformation" rule set.
• Output: This generated 4,500 predicted metabolites. After filtering for redundancy (retaining only neutral structures for charged variants), 4,131 predicted metabolites were added.

C. Analytical Property Prediction

To support mass spectrometry (MS)-based annotation, the database was augmented with four key analytical properties, sourced from existing literature or predicted using containerized machine learning tools:

1. Collision Cross-Sections (CCS): Predicted using CCSbase and SigmaCCS (ML-based, <2% error).
2. Tandem Mass Spectra (MS/MS): Predicted using QC-GN2oMS2 and GrAFF-MS at collision energies of 10, 20, and 40 eV for various adducts ([M+H]+, [M-H]-).
3. Gas-Phase Infrared (IR) Spectra: Predicted using a modified Graphormer-IR implementation for [M+H]+, [M+Na]+, and [M-H]- adducts.
4. Implementation: Tools were containerized using Apptainer to ensure reproducibility across different systems.

D. Database Architecture

• Format: SQLite relational database.
• Structure: Hierarchical design linking metabolites to molecular properties, reactions, enzymes, and pathways.
• Functional Data: Integrated reaction and pathway information from KEGG and BioCyc, linking metabolites to enzymes and biological pathways.

3. Key Results

Database Composition

• Total Metabolites: 6,153 unique compounds (2,022 literature-derived + 4,131 computationally predicted).
• Chemical Space:
  • Dominated by Organic acids and derivatives (27.5%) and Lipids (20.8%).
  • Significant representation of amino acids, peptides, prenol lipids, and fatty acyls.
  • PaCMAP Visualization: Confirmed that predicted metabolites do not drastically shift the chemical landscape but rather increase the "depth" of coverage within existing chemical clusters, effectively broadening the identification aperture without introducing unrealistic chemical diversity.

Analytical Property Coverage

• Coverage: Literature-derived compounds show the highest overlap of all properties (m/z, CCS, MS/MS, IR).
• Distribution: The database is dominated by small molecules (50–400 m/z).
• Correlations: Strong correlations were observed between precursor m/z and CCS values, validating the utility of the database for Ion Mobility Spectrometry (IMS) coupled with MS.
• Spectral Data: The database includes predicted MS/MS spectra for multiple collision energies and IR spectra for different adducts, providing a multi-dimensional fingerprint for identification.

Functional Context

• Pathways: 972 metabolites are linked to 1,050 reactions and 944 pathways.
• Key Pathways: Top pathways include amino acid processing, lipid metabolism, cofactor biosynthesis, and energy production.
• Enzymes: High connectivity was found for enzymes involved in Fe-S cluster systems and lipid metabolism.

4. Significance and Impact

• Filling a Critical Gap: PPMDB v1 is the first comprehensive, organism-specific metabolome database for P. putida, bridging the gap between genomic models and metabolomic data.
• Enabling Untargeted Metabolomics: By providing reference data for CCS, MS/MS, and IR, the database allows researchers to move beyond targeted analysis to untargeted, exploratory studies, crucial for discovering novel engineered pathways.
• Synergy with Genome-Scale Models: Unlike gene-centric models (e.g., iJN1462) which optimize for biomass or specific product yields, PPMDB supports the interpretation of complex, untargeted molecular data, complementing existing macro-level phenotypic models.
• Community Resource: The database is publicly available (Zenodo) with source code and build scripts, encouraging community contributions to expand its scope as P. putida engineering advances.

Conclusion

The PPMDB v1 represents a foundational infrastructure upgrade for Pseudomonas putida research. By consolidating curated literature data with high-confidence computational predictions and multi-dimensional analytical properties, it empowers the synthetic biology community to perform deep, molecular-level exploratory analyses, accelerating the development of P. putida as a platform for bioengineering and bioremediation.