A protein interactome for the last eukaryotic common ancestor illuminates the biochemical basis of modern genetic diseases

1. Problem Statement

The Last Eukaryotic Common Ancestor (LECA), which existed approximately 1.5–1.8 billion years ago, possessed a complex molecular toolkit that gave rise to all modern eukaryotes. While phylogenomic studies have reconstructed the gene content of LECA, there has been no integrated picture of how these ancient proteins interacted to form higher-order macromolecular complexes (the interactome).

• The Gap: Genomic reconstructions alone cannot reveal the physical assembly of proteins, which is critical for understanding cellular function. Furthermore, the evolutionary conservation of these ancient protein networks has not been systematically leveraged to predict human disease mechanisms, particularly for genes with unknown functions or those involved in complex traits like bone density and renal failure.
• The Challenge: Reconstructing an ancestral interactome requires distinguishing conserved interactions from lineage-specific ones across a vast evolutionary timescale, dealing with incomplete data, and overcoming biases in current proteomic datasets (which are heavily skewed toward mammals).

2. Methodology

The authors employed a multi-step computational and experimental pipeline to reconstruct the LECA interactome and validate its utility in human genetics.

A. Inferring the LECA Gene Set

• Orthogroup Definition: Using the eggNOG database, they defined orthologous groups (OGs) across 156 reference proteomes (spanning eukaryotes, archaea, and bacteria).
• Ancestral Reconstruction: They applied Dollo parsimony (assuming gene loss is irreversible) to model gene duplication and loss, identifying 10,091 candidate ancestral gene families.
• Robustness Filtering: To mitigate biases and account for horizontal gene transfer (HGT), they cross-validated these candidates using Wagner parsimony (allowing gains/losses) and phylogenetic reconciliation (DTL models). They restricted biological interpretation to the "core" set of 6,429 OGs supported by at least two methods.

B. Constructing the Interactome via Mass Spectrometry

• Data Integration: The study integrated >26,000 mass spectrometry experiments (including Co-fractionation Mass Spectrometry [CFMS], Affinity Purification MS [APMS], proximity labeling, and RNA-pulldown) from 31 diverse eukaryotic species. This dataset covered ~1.8 billion years of evolution, including unicellular and multicellular taxa from all major supergroups (Amorphea, Excavata, TSAR, Archaeplastida).
• New Data Generation: They generated new CFMS data for four minimally characterized unicellular eukaryotes (Brachionus rotundiformis, Euglena gracilis, Phaeodactylum tricornutum, Tetrahymena thermophila) and ciliary tissues from pig and frog to ensure broad phylogenetic coverage.
• Machine Learning Pipeline:
  1. Feature Extraction: They constructed elution vectors for protein pairs based on co-elution profiles across 149 fractionations.
  2. Training: Using a gold standard of known complexes (from CORUM and Complex Portal), they trained supervised machine learning classifiers (ExtraTrees, LinearSVC, SGDClassifier) to predict interactions.
  3. Stratification: To prevent overfitting, they implemented a custom data stratification approach, grouping complexes into "supergroups" to ensure the model learned general interaction principles rather than memorizing specific complex compositions.
  4. Clustering: Predicted interactions (filtered at 10% False Discovery Rate) were clustered into hierarchical protein complexes using the Walktrap community detection algorithm.

C. Disease Prediction and Validation

• Network Propagation: They applied a "guilt-by-association" algorithm to the LECA interactome. Known disease-associated genes were used as seeds to propagate scores to neighboring proteins, predicting novel gene-disease links for 109 Mendelian disorders.
• Experimental Validation:
  • EFHC2: Validated via whole-exome sequencing of a patient with renal failure and localization studies in Xenopus cilia.
  • ATP6V1A: Validated using heterozygous CRISPR-Cas9 knockout mice from the KOMP2 program.
  • GLG1: Validated via morpholino knockdown and live imaging in Xenopus multiciliated cells.

3. Key Contributions

1. First LECA Interactome: The paper presents the first high-resolution map of protein-protein interactions (PPIs) for the Last Eukaryotic Common Ancestor, comprising 109,466 pairwise interactions among 3,193 unique orthogroups, organized into hierarchical complexes.
2. Functional Annotation of "Unknowns": By mapping ancient interactions, the authors assigned tentative functions and subcellular localizations to ~1,000 OGs previously categorized as "function unknown" by standard annotation algorithms (eggNOG), particularly in cytoskeletal and ciliary pathways.
3. Evolutionary Insights:
   • Vesicle Tethering: Revealed that complexes like GARP, EARP, and TRAPP are ancient, but identified unexpected ancient members (e.g., EIPR1, TRAPPC12) and lineage-specific losses (e.g., TRAPPC10 in TSAR).
   • Phagocytosis: Provided strong evidence that LECA possessed the full molecular machinery for phagocytosis (including ARP2/3, coronins, and capping proteins), supporting the hypothesis that the ancestor was a phagocytosing cell.
4. Disease Mechanism Discovery: Demonstrated that the LECA interactome is a powerful tool for identifying the molecular basis of modern human diseases, specifically linking ancient protein networks to renal failure, bone density disorders, and ciliopathies.

4. Key Results

• Gene Content: The core LECA gene set contains ~6,400 OGs. Approximately 50% of human genes map to these ancient families. Notably, humans have lost many genes related to the glyoxylate cycle (e.g., malate synthase), explaining the mechanism of primary hyperoxaluria.
• Complex Reconstruction: The model successfully recapitulated known complexes (e.g., spliceosome, COPI, 20S proteasome) and discovered novel assemblies. For instance, they found that the HOPS complex in LECA may have been a single entity before splitting into HOPS and CORVET in modern lineages.
• Case Study 1: EFHC2 and Renal Failure: A patient with end-stage renal disease and microcephaly was found to have a hemizygous mutation in EFHC2 (R133H). The LECA interactome placed EFHC2 in a ciliary motility cluster. Experimental validation showed the mutant protein failed to localize to cilia in Xenopus, linking a ciliary defect to renal failure despite motile cilia being absent in mammalian kidneys.
• Case Study 2: ATP6V1A and Osteopetrosis: Network propagation predicted ATP6V1A (a V-ATPase subunit) as a candidate for osteopetrosis. Heterozygous Atp6v1a knockout mice exhibited significantly increased bone mineral density, confirming the prediction.
• Case Study 3: GLG1 and Short-Rib Thoracic Dysplasia (SRTD): The algorithm predicted the Golgi protein GLG1 as a candidate for SRTD. Knockdown of GLG1 in Xenopus caused cilia loss and accumulation of Intraflagellar Transport (IFT) proteins, revealing a novel mechanism where Golgi function is required for IFT trafficking.

5. Significance

• Evolutionary Biology: This work provides a "biochemical snapshot" of the eukaryotic cell at its origin, resolving debates about the complexity of LECA (e.g., confirming the presence of phagocytosis machinery and complex vesicle trafficking systems).
• Medical Genetics: It establishes a novel paradigm for ancestral network medicine. By focusing on deeply conserved protein interactions, researchers can bypass the noise of lineage-specific adaptations to identify core disease mechanisms. This approach successfully predicted disease genes that were previously uncharacterized or linked to unexpected phenotypes (e.g., Golgi proteins causing ciliopathies).
• Methodological Advance: The study demonstrates that integrating massive, cross-species proteomic datasets with machine learning can overcome the limitations of single-species interactome mapping, providing a robust framework for studying protein evolution and function.

In conclusion, the paper successfully bridges deep evolutionary history with modern clinical genetics, proving that the protein interaction networks established 1.8 billion years ago remain the fundamental scaffold for understanding human disease.