TL;DR
This paper tackles the immense, largely unexplored chemical diversity of plants by estimating the size of "plant chemical space." Using large-scale metabolomics and literature data, they propose that current mass spectrometry data alone may already contain more unique plant compounds than previously estimated, suggesting millions of plant metabolites remain undiscovered.
What Problem Does This Paper Address?
Plants, with their vast diversity of species (nearly 400,000 known), produce an astonishing array of chemical compounds. These range from essential primary metabolites to specialized secondary metabolites with potential applications in medicine, agriculture, and biotechnology. This vast chemical landscape is often referred to as "plant chemical space." However, a significant challenge is that only a tiny fraction of plant species has been thoroughly studied, leaving much of this chemical diversity unexplored. Previous estimates of the total number of plant-derived compounds have been based on limited data and assumptions. This paper aims to provide a more robust estimation of the size of this plant chemical space by leveraging extensive metabolomics and literature datasets.
Methods
The researchers employed a multi-pronged approach to estimate plant chemical space:
- Literature Data Analysis: They analyzed existing literature datasets (COCONUT and LOTUS) to map the "known" chemical space, creating saturation curves to see how many unique compounds are discovered as more species are added.
- Mass Spectrometry (MS) Data Analysis: They utilized a large metabolomics dataset (ENPKG) from over 1,000 plant species. To estimate the number of unique chemical structures within this dataset, they applied four complementary computational approaches:
- Formula Prediction (SIRIUS): Predicting molecular formulas from MS/MS spectra.
- De Novo Prediction (MS2Mol): A machine learning model that predicts novel structures directly from MS/MS spectra, venturing into unexplored chemical regions.
- Hybrid Approach (CSI:FingerID + MS2Mol): Combining library-based searches for known compounds with de novo prediction for unknown ones.
- MS2 Clustering: Grouping similar MS/MS spectra to identify unique chemical structures without relying solely on predictive models.
- Extrapolation: Power law models were fitted to the saturation curves generated from the MS data to project the estimated total plant chemical space for the entire plant kingdom (estimated at 400,000 species).
Datasets
The study leveraged several key datasets:
- Literature Databases:
- COCONUT: A collection of open natural products.
-
LOTUS: An initiative for open knowledge management in natural products research. These datasets were used to define the "known" plant chemical space.
-
Metabolomics Databases:
- ENPKG (Experimental Natural Products Knowledge Graph): A large, harmonized knowledge graph containing LC-MS features from over 1,000 plant species. This was the primary dataset for estimating the chemical space captured by current mass spectrometry data.
- Korean Pharmacopoeia Plants Extracts & Pierre Fabre Research Institute Library (RFRIL): These were used for comparative analysis of metabolite coverage and consistency.
Key Insights and Ideas
- Current MS Data is Richer Than Expected: The study suggests that the mass spectrometry data alone, from just over 1,000 plant species, may already contain more unique chemical structures than previously estimated for the entire plant kingdom. This implies that our current understanding of plant chemistry is significantly lagging behind what is detectable.
- Literature vs. Metabolomics Coverage: While literature databases capture known compounds, they are sparse. Metabolomics data, even from a limited number of species, can reveal a much larger number of unique structures, highlighting the power of experimental profiling.
- The "Unknown" is Vast: A significant portion of the chemical structures identified in the metabolomics dataset were classified as "unknown" (based on CSI:FingerID confidence scores), underscoring the vastness of unexplored chemical space.
- Sampling Depth Matters: The study demonstrates that increasing the number of samples analyzed per plant species significantly increases the coverage of its chemical metabolome, suggesting that current estimates might be conservative due to limited sampling.
- Biases in Data Collection: The research acknowledges that biases exist in both literature and metabolomics datasets, stemming from factors like extraction methods, plant part sampling, and taxonomic focus, which can influence the perceived chemical space.
Key Results
- The "known" plant chemical space, based on literature, saturates around 124,000 unique structures from approximately 32,000 species.
- Analysis of metabolomics data from over 1,000 plants revealed that even conservative estimation methods (like formula prediction) suggested at least 1.5 million unique compounds.
- More advanced methods (MS2Mol, MS2 clustering) estimated the chemical space within this dataset to be in the range of 15 to 25 million unique structures.
- Projecting these findings to an estimated 400,000 plant species suggests the total plant chemical space could span millions, potentially tens of millions, of unique compounds.
- The overlap between literature-reported compounds and those potentially detectable by MS2 spectra was found to be around 25% on average for many plants, indicating limited coverage.
Limitations and Future Work
The authors candidly acknowledge several limitations:
- Dataset Coverage: The estimations are based on a small fraction of plant species, leading to potential biases.
- Extraction and Chromatographic Bias: Different extraction methods and incomplete anatomical sampling can miss specific metabolites.
- MS2 Spectrum Prediction Accuracy: The accuracy of predictive models can influence estimates.
- Disparate Solvents: Variations in solvents used for extraction can lead to inconsistencies.
- Adducts Considered: Focusing only on [M+H]+ adducts might miss other unique metabolites.
- Missed Compound Classes: Volatile and lipid compounds may not be captured by standard LC-MS methods.
- False Positives: Annotation errors and contaminants can inflate counts.
- Projection Modeling: Power law projections might overestimate space if actual data curves plateau unexpectedly.
Future work should focus on:
- Expanding metabolomics datasets to include more species and diverse sampling strategies.
- Developing more accurate and comprehensive predictive models.
- Integrating data from different analytical techniques (e.g., GC-MS for volatiles).
- Leveraging functional genomics to explore silent or cryptic biosynthetic pathways.
- Continuously updating estimations as more public data becomes available.
Summary updated