Abstract
Phosphorus is essential for life and critically influences marine productivity. Despite geochemical evidence of active phosphorus cycling in deep-sea cold seeps, the microbial processes involved remain poorly understood. Traditional sequence-based searches often fail to detect proteins with remote homology. To address this, we developed a deep learning model, LucaPCycle, integrating raw sequences and contextual embeddings based on the protein language model ESM2-3B. LucaPCycle identified 5,241 phosphorus-cycling protein families from global cold seep gene and genome catalogs, substantially enhancing our understanding of their diversity, ecology, and function. Among previously unannotated sequences, we discovered three alkaline phosphatase families that feature unique domain organizations and preserved enzymatic capabilities. These results highlight previously overlooked ecological importance of phosphorus cycling within cold seeps, corroborated by data from porewater geochemistry, metatranscriptomics, and metabolomics. We revealed a previously unrecognized diversity of archaea, including Asgardarchaeota, anaerobic methanotrophic archaea and Thermoproteota, which contribute to organic phosphorus mineralization and inorganic phosphorus solubilization through various mechanisms. Additionally, auxiliary metabolic genes of cold seep viruses primarily encode the PhoR-PhoB regulatory system and PhnCDE transporter, potentially enhancing their hosts' phosphorus utilization. Overall, LucaPCycle is capable of accessing previously 'hidden' sequence spaces for microbial phosphorus cycling and can be applied to various ecosystems.
Introduction
Phosphorus (P) is a fundamental requirement for life, serving as the backbone of DNA and the primary currency of cellular energy (ATP). In the extreme environments of deep-sea cold seeps, geochemical data suggests active phosphorus cycling is occurring, yet the specific biological "machinery" responsible has remained elusive. The primary bottleneck in identifying these proteins is remote homology. Traditional search tools like BLAST rely on high sequence similarity, which often fails when evolution has significantly altered the amino acid sequence while preserving the protein's 3D shape and function. Because deep-sea microbes have evolved in isolation for eons, their P-cycling proteins often differ substantially from those cataloged in standard databases, making them "invisible" to traditional bioinformatic pipelines.
Methodology: The LucaPCycle Framework
To address the challenge of remote homology, we developed LucaPCycle, a dual-channel deep learning architecture designed to capture both the "text" and the "context" of a protein sequence. The model integrates two primary components:
- Protein Language Model (PLM) Channel: Utilizing ESM2-3B, a model trained on millions of protein sequences, LucaPCycle learns the underlying "grammar" of biology. This allows the tool to infer function based on biochemical context, even for novel sequences.
- Transformer-Encoder Channel: This channel processes raw amino acid sequences to capture local structural features.
The model was trained on PCycDB, a curated phosphorus-cycling database containing approximately 214,000 positive and 853,000 negative samples. It performs two distinct tasks: a binary classification (identifying P-cycling proteins) and a 31-class categorization (mapping proteins to specific steps in the P-cycle).
Results and Benchmarking
LucaPCycle demonstrated superior performance compared to industry standards. In benchmarking against Diamond Blastp and KofamScan, LucaPCycle achieved a precision of 86.13% and a recall of 85.63%. In contrast, Diamond Blastp and KofamScan showed significantly lower precision (51.98% and 58.24%, respectively) when identifying remote homologs.
Applying LucaPCycle to a global cold seep dataset of 147 million genes and 1,878 metagenome-assembled genomes (MAGs) led to the identification of 5,241 phosphorus-cycling protein families that were previously unannotated. Functional confirmation was further supported by molecular docking, which showed that these remote homologs possessed high binding affinities for various phosphoester compounds.
Biological Insights and Ecological Implications
The discovery of these "hidden" protein families significantly expands our understanding of marine nutrient cycles. Key findings include:
- Novel Alkaline Phosphatases: We identified three new families of alkaline phosphatases (ALPs). Notably, the ALP1 family features a unique domain organization, lacking the standard N-terminal domain but possessing a unique C-terminal $\beta$-sheet, likely an adaptation to high-pressure deep-sea environments.
- Archaea and Viral Roles: The study reveals that Asgardarchaeota and anaerobic methanotrophic archaea (ANME) are major contributors to organic phosphorus mineralization. Furthermore, cold-seep viruses were found to carry Auxiliary Metabolic Genes (AMGs), such as the PhoR-PhoB regulatory system, potentially allowing them to reprogram host metabolism to enhance phosphorus scavenging during infection.
Conclusion
LucaPCycle represents a significant leap in our ability to access the "dark matter" of functional genomics. By moving beyond simple sequence similarity to leverage the contextual power of protein language models, we have uncovered a vast, previously unrecognized diversity of phosphorus-cycling machinery in deep-sea cold seeps. This framework is not limited to phosphorus or cold seeps; it can be adapted to study other critical nutrient cycles and diverse ecosystems, providing a new lens through which to view global biogeochemistry.
Summary updated