Full-Length Structural Modeling of Mitofusins with AlphaFold Reveals a Novel Cross-Type Dimerization and Insights into Oligomerization

This study utilizes AlphaFold to generate full-length structural models of mitofusins, revealing a novel cross-type dimerization mode mediated by heptad repeat domains and proposing a new mechanism for outer mitochondrial membrane fusion.

The Gist

The Big Picture: The Power Plant's Dance Floor

Imagine your cells are bustling cities, and the mitochondria are the power plants that keep the lights on. For a city to function, these power plants need to be healthy, connected, and able to share resources. Sometimes, they need to merge (fusion) to combine their energy, and sometimes they need to split (fission) to multiply.

The "construction workers" responsible for merging these power plants are proteins called Mitofusins. Think of Mitofusins as giant, flexible cranes that reach out, grab a neighboring power plant, and pull them together until they fuse into one.

However, for a long time, scientists have been trying to build a 3D blueprint of these cranes, but the blueprints were always incomplete. We knew what the top of the crane looked like, but the part that actually grabs the other plant (the "arms" and the "legs" that touch the ground) was a mystery. Without the full blueprint, we didn't fully understand how they did their job, and we knew that when these cranes broke, it caused diseases like Parkinson's and Alzheimer's.

The New Tool: The AI Architect

In this paper, the researchers decided to stop waiting for a perfect physical blueprint. Instead, they used a super-smart AI architect called AlphaFold. Think of AlphaFold as a master builder who has read every single instruction manual for every protein in existence. It can look at a protein's genetic "recipe" and predict exactly what its 3D shape should be.

The team asked AlphaFold to build the full-length models of these Mitofusin cranes, including the parts that were previously missing.

The Surprising Discovery: The "Crossed-Arms" Dance

Here is the twist: When the AI built these models, it didn't look like the old theories suggested.

• The Old Theory: Scientists thought the cranes would stand side-by-side, parallel to each other, like two people shaking hands with straight arms.
• The New Reality: The AI models showed the cranes doing something completely different. They found a "Cross-Type Dimerization."

The Analogy: Imagine two people trying to hug.

• Old Idea: They stand face-to-face and wrap their arms around each other in a straight, parallel embrace.
• New Discovery: The AI showed them doing a complex dance move where they cross their arms over each other, like a figure-eight or a knot. One crane's "arm" (a specific part called the HR domain) reaches across and grabs the other crane's arm in a crossed pattern.

This "crossed" position is a brand-new way these proteins interact that no one had seen in a lab before. It's like discovering that the way people hug is actually a complex, crossed-arm dance move, not a simple side-by-side squeeze.

The Helpers: The "Glue" and the "Anchors"

The researchers didn't just look at the cranes alone; they also modeled them with their helpers.

• The Helpers (Ugo1/SLC25A46): These are like the glue or the scaffolding that helps the cranes stay in place. The AI showed that when the cranes work with these helpers, they stand much straighter and more stable.
• The Ground (The Membrane): The cranes have "legs" (transmembrane domains) that stick into the cell wall. The AI showed that without the helpers, the legs were wobbly and bent. But with the helpers, the legs stood firm, anchoring the crane perfectly.

The New Story of How Fusion Happens

Based on these new AI blueprints, the authors proposed a new story for how mitochondria merge:

1. The Approach: Two mitochondria approach each other. The cranes (Mitofusins) on one side reach out to the cranes on the other side.
2. The First Grab: They grab each other using their "heads" (the GTPase domain), forming a temporary bridge across the gap.
3. The Cross-Over: Then, they perform that new "crossed-arm" dance. Their bodies twist and cross over, pulling the two membranes tight against each other.
4. The Fusion: This twisting motion creates the tension needed to melt the two membranes together, fusing the power plants into one.
5. The Aftermath: Once fused, they stay in this "crossed" position, stabilized by their helper glue, until they are ready to let go.

Why This Matters

This is a big deal because:

• It fills the gaps: We finally have a picture of the whole machine, not just the top part.
• It explains the diseases: If you know how the crane is supposed to dance, you can see exactly what goes wrong when a mutation (a typo in the recipe) happens. This helps us understand why diseases like Parkinson's occur.
• It guides future drugs: Now that we have this new "crossed-arm" blueprint, scientists can design drugs that either fix broken cranes or stop cancer cells (which often have too many cranes) from fusing.

In a nutshell: Scientists used AI to build a complete 3D model of the protein that merges mitochondria. They discovered that instead of standing side-by-side, these proteins do a complex, crossed-arm dance to pull membranes together. This new view helps us understand how cells stay healthy and why things go wrong in serious diseases.

Technical Summary

1. Problem Statement

Mitochondrial dynamics (fission and fusion) are critical for cellular health, and their dysfunction is linked to neurodegenerative diseases (Parkinson's, Alzheimer's, Charcot-Marie-Tooth type 2A) and various cancers. Mitofusins (Mfn1, Mfn2 in mammals; Fzo1 in yeast) are large GTPase transmembrane proteins responsible for tethering and fusing the outer mitochondrial membrane (OMM).

Despite their importance, the precise molecular mechanism of mitofusin-mediated fusion remains unclear due to a lack of complete high-resolution experimental structures. Existing structures (e.g., PDB: 5YEW, 5GOF, 9KFE) are partial, missing critical transmembrane (TM) domains and large portions of heptad repeat (HR) domains. Previous computational models relied on homology to bacterial dynamin-like proteins (BDLP) and often assumed specific conformational states (open/closed) that lacked experimental validation for full-length proteins.

2. Methodology

The authors employed a multi-step computational approach to generate and analyze full-length structural models of mitofusins:

• AlphaFold Predictions:
  • Used AlphaFold 3 (AF3) and ColabFold (AlphaFold 2) to predict structures of Mfn1, Mfn2, and Fzo1.
  • Generated models for monomers, homodimers, heterodimers, and tetramers.
  • Included complexes with fusion partners: Ugo1 (yeast) and SLC25A46 (human), which are mitochondrial solute carriers.
  • Incorporated fatty acids and GTP/GDP into simulations to mimic the membrane environment and nucleotide states.
• Custom Multiple Sequence Alignment (MSA):
  • To improve TM domain prediction, a custom MSA of 47 fungal sequences was generated using PSI-BLAST and T-Coffee. This was fed into AlphaFold 2 to bias sampling toward specific TM structural features.
• Validation and Refinement:
  • HADDOCK3: Used for energy minimization and CAPRI evaluation (calculating i-RMSD, l-RMSD, and FNAT) to compare predicted models against available experimental structures (PDBs: 5YEW, 5GOF, 5GOM, 6JFK, 9KFE, 9KFF, 9KFD).
  • Molecular Dynamics (MD): Coarse-grained MD simulations (using MARTINI2 force field in GROMACS) were performed on a heterotetramer (2 Fzo1 + 2 Ugo1) embedded in a lipid bilayer to assess the stability of predicted interfaces over 10 µs.
• Contact Analysis:
  • Inter-residue distances and salt bridges were analyzed using MDAnalysis to identify stable interaction interfaces.

3. Key Contributions

• Full-Length Modeling: First comprehensive structural models of full-length mitofusins (including TM and HR domains) generated using AlphaFold, overcoming the limitations of previous partial experimental structures.
• Discovery of Cross-Type Dimerization: Identification of a novel "cross-type" dimerization mode where the heptad repeat (HR) domains of opposing mitofusins interact in a crossed arrangement, rather than the parallel alignment seen in previous BDLP-based models.
• Role of Solute Carriers: Demonstration that the presence of solute carriers (Ugo1/SLC25A46) and fatty acids significantly influences the conformation of the TM domains, promoting stable helical structures and specific oligomerization interfaces.
• Hypothetical Fusion Mechanism: Proposal of a new mechanistic model for OMM fusion that integrates these structural findings, transitioning from trans-dimers to cis-dimers via cross-type interactions.

4. Key Results

A. Monomeric Structures

• Open Conformation: Default AF3 predictions for monomers (Mfn1, Mfn2, Fzo1) consistently adopted an "open" conformation.
• TM Domain Issues: Default models showed low pLDDT scores and kinks in the TM domains, failing to predict the expected two-helix bundle.
• Custom MSA Success: Using a custom MSA for Fzo1 successfully predicted a stable two-helix TM domain with inter-helical interactions, consistent with previous coarse-grained simulations and experimental data regarding the Lys716 residue.

B. Oligomeric Structures (Dimers and Tetramers)

• Cross-Type Dimerization: The most significant finding is a novel dimerization mode involving cross-interactions between the GTPase and HR coiled-coil domains. Unlike previous models where HR domains ran parallel, these models show the HR regions interacting in a crossed fashion.
• Solute Carrier Influence:
  • Models containing Ugo1 (yeast) or SLC25A46 (human) showed improved TM domain organization and compactness.
  • The solute carriers localized to the membrane plane, interacting with the disordered N-terminal regions of mitofusins and stabilizing the TM helices.
  • Salt Bridges: Stable interfaces were maintained via salt bridges in the HRN, GTPase, and HR2 domains.
• Tetramers: Predicted tetramers showed high conformational diversity, including both cis (same membrane) and trans (opposite membranes) configurations. Some models suggested a "ring-like" assembly capable of bringing opposing membranes into proximity.

C. Validation Against Experimental Data

• CAPRI Metrics: The AF3 models aligned well with existing experimental structures (e.g., PDB 5YEW, 9KFD), achieving medium-to-high quality scores (e.g., FNAT > 0.6, i-RMSD < 2.7 Å).
• Reproduction of Known Contacts: The models reproduced >75% of the native contacts found in the recently solved Fzo1 structure (PDB 9KFD), validating the predicted cross-type interface.
• MD Stability: Coarse-grained MD simulations confirmed that the predicted cross-type interfaces and salt bridges remained stable over time, supporting their biological relevance.

D. Proposed Fusion Mechanism

The authors propose a multi-step mechanism:

1. Trans-Dimerization: Mitofusins on opposing membranes bind via their GTPase domains upon GTP loading.
2. Tethering: Initial tethering brings membranes closer.
3. Cross-Type Transition: Conformational changes induce the formation of cross-type cis-dimers where HR domains interact across the membrane interface.
4. Fusion: This reorganization, potentially stabilized by solute carriers (Ugo1/SLC25A46), destabilizes the lipid bilayer to facilitate fusion.
5. Post-Fusion: The complex stabilizes as a cis-dimer on the fused membrane.

5. Significance

• Bridging the Structural Gap: This work provides the first structural hypotheses for the full-length architecture of mitofusins, specifically addressing the elusive TM and HR domains that are critical for membrane fusion but missing from crystal/cryo-EM structures.
• Challenging Existing Paradigms: The discovery of the "cross-type" dimer challenges the prevailing view that mitofusin HR domains interact in a parallel, BDLP-like manner, suggesting a more complex and flexible mechanism.
• Therapeutic Implications: By elucidating the structural basis of mitofusin dysfunction, these models offer new targets for understanding and treating neurodegenerative diseases and cancers associated with mitochondrial dynamics.
• Methodological Advance: The study demonstrates the power of combining AlphaFold with custom MSAs and membrane-mimetic environments (fatty acids/solute carriers) to model complex transmembrane protein assemblies that are difficult to capture experimentally.

In conclusion, the paper leverages advanced AI-driven structural biology to propose a refined, testable mechanism for mitochondrial outer membrane fusion, highlighting the critical role of solute carriers and a novel cross-type dimerization interface.