Is metabolism spatially optimized? Structural modeling of consecutive enzyme pairs reveals no evidence for spatial optimization of catalytic site proximity.

This study utilizes structural modeling and computational analysis of 107 consecutive enzyme pairs in *E. coli* to demonstrate that, despite a tendency for these enzymes to interact, their catalytic sites are not systematically positioned in spatially optimized configurations to facilitate metabolite transfer.

The Gist

The Big Question: Do Factory Workers Sit Next to Each Other?

Imagine a massive, bustling factory (your cell) where thousands of workers (enzymes) are building products. In this factory, one worker finishes a part and hands it off to the next worker to finish the job.

Scientists have long suspected that to make the factory run super fast, these workers might stand right next to each other. This would allow them to pass the "parts" (metabolites) directly from hand to hand without the parts getting lost or taking a long walk across the factory floor. This idea is called metabolic channeling or forming a "metabolon."

The big question this paper asks is: Is this true for all the workers in the factory, or just a few special teams? Do the workers who do step 1 and step 2 of a process actually stand close together, or are they just randomly scattered around?

The Experiment: A Digital Simulation

Since we can't easily see these tiny workers moving in real-time inside a living cell, the authors used a super-powerful computer simulation. Think of it like a high-tech video game where they built 3D models of 107 different pairs of "consecutive workers" (enzymes that work one after another) in E. coli bacteria.

They used four different "AI architects" (AlphaFold2, AlphaFold3, ESMFold, and HDOCK) to predict how these pairs would look if they shook hands (interacted).

The Findings: The "Handshake" vs. The "Hand-off"

The researchers found two very interesting things:

1. They do shake hands, but not for the reason we thought.
The computer models showed that consecutive enzymes do tend to interact physically more often than random pairs of enzymes. It's like finding that the worker on the assembly line is more likely to stand near the person who comes after them than a random person from the cafeteria.

2. But they don't stand close enough to pass the baton.
Here is the twist. Even though these pairs interact, their "workstations" (the catalytic sites where the chemical magic happens) were not positioned closer together than you would expect by pure chance.

• The Euclidean Distance (The "Bird's Eye View"): If you measure the distance in a straight line through the air (like a bird flying), the workstations looked a bit closer.
• The SASP Distance (The "Real Walk"): The authors realized that a metabolite can't fly through the air; it has to walk around the outside of the proteins, like a person walking around a building. When they calculated this "walking distance" (called the Shortest Accessible Space Path or SASP), the workstations were not any closer than random pairs.

The "Cave" Analogy: Why the Distance Tricked Us

Why did the straight-line measurement look close? The authors explain this with a clever analogy.

Imagine the "workstation" is a cave dug into the side of a mountain (the enzyme).

• If you pick two random spots on the surface of two mountains, they might be far apart.
• But if you pick two "caves" (workstations), they are naturally dug into the mountain. Because they are recessed, the distance between the entrances of the caves is naturally shorter than the distance between two random spots on the flat surface.

The study found that the apparent "closeness" of the enzymes was mostly just a geometric trick because workstations are usually hidden in little pockets on the protein surface, not because the proteins are actively arranging themselves to be neighbors.

The One Exception: A New Discovery

While the general rule was "no special arrangement," the team did find one specific pair of enzymes (MetB and MetC) that seemed to break the rule. The AI models predicted they stand very close together with a clear path between them. This suggests a brand-new, previously unknown "hand-off" team in the bacteria that scientists haven't discovered yet.

The Bottom Line

Do enzymes organize themselves to pass products directly?
Probably not generally.

The study concludes that while enzymes might hang out together in groups, they don't seem to arrange their "workstations" in a special, optimized way to speed up the passing of materials. The factory floor is likely a bit more chaotic than we hoped. The speed of the factory is probably determined by how fast the workers can do their jobs, not by how close they stand to each other.

Why does this matter?
This study is important because it uses the latest AI tools to test a fundamental biological theory. It tells us that while "teamwork" (physical interaction) happens, "perfect positioning" (spatial optimization) might not be the universal rule we thought it was. It also gives us a new toolkit (the SASP method) to measure these distances more accurately in the future.

Technical Summary

1. Problem Statement

Metabolic pathways are often hypothesized to benefit from spatial organization, specifically through metabolic channeling or metabolon formation, where consecutive enzymes physically interact to facilitate substrate transfer and minimize diffusion times. While this phenomenon is well-documented in specific, stable multi-enzyme complexes, it remains unclear whether it represents a general organizational principle of metabolism or is restricted to specific pathways.

The central hypothesis tested in this study is: Do enzymes catalyzing consecutive metabolic steps, when physically interacting (even transiently), exhibit structurally optimized arrangements that minimize the distance between their catalytic sites?

2. Methodology

The authors employed a large-scale computational framework to test this hypothesis using Escherichia coli as a model organism.

• Dataset Construction:

  • Consecutive Pairs: Identified 107 unique enzyme pairs catalyzing sequential reactions in E. coli by integrating data from the Catalytic Site Atlas (CSA) and KEGG pathway annotations.
  • Control Groups: Created a "Randomized Pairs" set by randomly shuffling the enzymes from the consecutive set to serve as a null hypothesis control.
  • Benchmark Set: Curated a dataset of 112 low-affinity (weak/transient) protein dimers from PDBbind (dissociation constant $K_d > 1 \mu M$) to validate prediction methods.

• Structural Modeling:

  • Generated 3D interaction models for all pairs using four distinct prediction frameworks:
    1. AlphaFold2 (AF2)
    2. AlphaFold3 (AF3)
    3. ESMFold
    4. HDOCK (rigid-body docking)
  • Evaluated model quality using DockQ scores against the benchmark set.

• Distance Metrics:

  • Euclidean Distance: Calculated the straight-line distance between the geometric centers of mass of catalytic residues.
  • Shortest Accessible Space Path (SASP): Developed a novel computational procedure to calculate the shortest physically possible path between catalytic sites along the solvent-accessible surface and through the surrounding space, avoiding steric clashes with the protein interior. This provides a more realistic estimate of diffusion paths than Euclidean distance.

• Statistical Analysis:

  • Compared observed distances against a baseline of distances between catalytic sites and randomly sampled surface points on the partner enzyme.
  • Correlated distance metrics with interaction confidence scores (e.g., ipTM, ipSAE, VoroIF-GNN).

3. Key Contributions

• Development of SASP Metric: Introduced a robust method to calculate the Shortest Accessible Space Path, addressing the limitation of Euclidean distance which ignores steric hindrance and protein topology.
• Benchmarking Weak Interactions: Systematically evaluated the performance of state-of-the-art AI models (AF2, AF3, ESMFold) and classical docking (HDOCK) on low-affinity/transient interactions, identifying ipTM, ipSAE, and VoroIF-GNN as the most reliable confidence metrics.
• Large-Scale Structural Analysis: Applied these methods to a curated set of 107 consecutive metabolic enzyme pairs in E. coli, a scale previously unattainable with experimental structural biology alone.

4. Key Results

• Prediction Performance:

  • AlphaFold-based methods (AF2/AF3) outperformed ESMFold and HDOCK in recovering correct geometries for weak interactions.
  • ipTM showed the highest correlation with DockQ scores for AF2/AF3, while VoroIF-GNN was the most versatile metric across all four modeling platforms.
  • Consecutive enzyme pairs were predicted to interact with higher confidence scores than random pairs, suggesting a tendency for physical association, though many interactions remained weak/transient.

• Spatial Proximity Findings:

  • Euclidean Distance: Consecutive enzyme pairs showed shorter catalytic site distances compared to random expectations. However, this trend was also observed in random enzyme pairs. The authors attribute this to the geometric fact that catalytic sites are often located in invaginated (concave) surface pockets, naturally reducing the distance between surface points on interacting spheres regardless of biological optimization.
  • SASP Distance (Critical Finding): When using the physically realistic SASP metric, no evidence of spatial optimization was found. The distribution of SASP distances for consecutive pairs was statistically indistinguishable from random pairs.
  • Confidence Correlation: There was no systematic correlation between high interaction confidence scores and reduced catalytic site distances. High-confidence predictions did not necessarily yield closer active sites.

• Specific Case Study:

  • The study identified a potential novel interaction between MetB and MetC (methionine biosynthesis) with high confidence across multiple models, suggesting a specific instance of channeling, but this was an exception rather than the rule.

5. Significance and Conclusion

• Refutation of Universal Optimization: The study provides strong evidence that metabolism is not universally optimized for minimal diffusion distances between catalytic sites of consecutive enzymes. The apparent proximity observed in simple Euclidean metrics is largely a geometric artifact of catalytic site burial rather than evolutionary selection for channeling.
• Kinetic Implications: The findings align with kinetic theories suggesting that for most enzymes operating far from the diffusion limit, the time required for substrate diffusion between active sites is negligible compared to the catalytic turnover time ($k_{cat}$). Therefore, evolutionary pressure to minimize diffusion paths may be weak or non-existent for general metabolic pathways.
• Methodological Impact: The work highlights the potential and limitations of current structure prediction tools for studying transient metabolic interactions. It establishes a rigorous computational workflow (incorporating SASP) for analyzing the physical organization of metabolism, moving beyond binary interaction networks to atomic-level spatial analysis.
• Future Directions: The authors suggest that while universal spatial optimization is unlikely, localized instances (like the MetB/MetC pair) may exist for specific high-flux pathways, warranting further context-dependent investigation.