The amyloid packing difference: a pairwise comparison metric for amyloid structures

This paper introduces the amyloid packing difference (APD), a metric that quantifies structural variations between amyloid filaments based on unique cross-beta interactions and side chain orientations, demonstrating its ability to effectively distinguish disease-associated protofilament folds of proteins like alpha-synuclein and tau through clustering and specific value thresholds.

The Gist

Imagine you have a long, flexible rope made of beads. In a healthy body, this rope is usually coiled up neatly like a ball of yarn. But in certain diseases (like Alzheimer's or Parkinson's), this rope gets tangled into a stiff, rigid, twisted ladder called an amyloid filament.

Here's the problem: The same type of rope (the same protein) can twist itself into many different shapes of ladders. Some of these twisted shapes cause one specific disease, while others cause a different disease. Scientists need a way to tell these shapes apart to understand which disease is which.

The Old Way: The "Superposition" Problem

Previously, scientists compared these twisted ladders by trying to lay them on top of each other, like stacking two different origami cranes. They measured how much the atoms didn't line up (called RMSD).

The Analogy: Imagine trying to compare two different houses by stacking them on top of each other. If House A has a front door on the left and House B has it on the right, but the rest of the house is identical, the "stacking" method gets confused. It might say, "These houses are totally different!" just because you rotated one slightly. Or, if the houses have the same floor plan but different paint colors (side chains), the stacking method might say, "They are identical!" even though the paint matters a lot.

The New Solution: The "Amyloid Packing Difference" (APD)

The author of this paper, Sjors Scheres, invented a new tool called APD. Instead of trying to stack the ladders on top of each other, APD looks at how the beads touch each other.

The Analogy: Imagine two groups of people holding hands in a circle.

• Method A (Old): You try to make the two circles overlap perfectly. If one person is standing a few feet to the left, the whole circle looks "wrong."
• Method B (APD): You just ask, "Who is holding hands with whom?"
  • In Circle 1, Person A holds hands with Person B.
  • In Circle 2, Person A holds hands with Person C.
  • APD says: "These two circles are different because the hand-holding patterns are different."

It doesn't matter if the circle is rotated, flipped, or shifted; it only cares about the connections (the "packing") between the beads.

How It Works (The "Recipe")

The paper describes a computer program that does three things:

1. Scans the structure: It looks at every bead (amino acid) in the ladder.
2. Checks the neighbors: It asks, "Which beads are touching? Are they touching the same way as in the other ladder?"
3. Calculates the "Difference Score": It counts how many beads are holding hands with the wrong person or are facing the wrong direction.
   • 0% APD: The ladders are identical in how they pack together.
   • 100% APD: They are completely different.

What Did They Discover?

The author tested this new ruler on famous disease proteins (like Tau, Alpha-synuclein, and Prions). Here is what they found:

• The "Disease Threshold":

  • If two ladders have an APD score below 20%, they are likely the same disease. (Think of them as the same model of car, just different colors).
  • If two ladders have an APD score above 40%, they are definitely different diseases. (Think of them as a sedan vs. a pickup truck).
  • The "Gray Zone" (20-40%): This is tricky. It might be a slightly different version of the same disease, or a very similar disease.

• Checking the "Copycats":
  Scientists often try to grow these disease ladders in a lab to study them. Sometimes they claim, "Look! We made a lab version that looks exactly like the disease version!"

  • Using the old "stacking" method, some scientists claimed a lab-made ladder looked like a disease ladder.
  • Using the new APD, the author showed that the lab ladder was actually 80% different from the real disease ladder. The "hand-holding" patterns were totally wrong. The APD exposed that the lab version wasn't a good copy at all.

Why Does This Matter?

Think of amyloid structures like locks. The shape of the lock determines which key (the disease) fits into it.

• If you have a lock that looks similar but has the wrong internal pins (the packing), the wrong key might fit, or the right key might not work.
• The APD is a tool that checks the internal pins of the lock without needing to rotate the lock to see it.

The Bottom Line

This paper gives scientists a new, reliable "ruler" to measure amyloid structures. It stops them from getting confused by rotation or slight shifts. It helps them answer the question: "Is this new structure I found in the lab actually the disease I'm studying, or is it a fake?"

By using this new metric, researchers can better understand why different diseases happen and ensure that the models they build in the lab are actually accurate copies of the real thing.

Technical Summary

1. Problem Statement

• Structural Diversity of Amyloids: While most globular proteins adopt a single fold, amyloid filaments formed by a single protein sequence can adopt multiple distinct folds (polymorphism). Specific folds are often associated with distinct neurodegenerative diseases (e.g., different tauopathies or synucleinopathies).
• Limitations of Current Metrics: Traditional structural comparison methods, such as Root Mean Square Deviation (RMSD) of C-alpha atoms or Template Modeling (TM) scores, rely on global structural superposition.
  • Rigid Body Issues: These metrics fail when structures have distinct relative orientations of sub-domains (e.g., hinge movements), leading to high RMSD values even if local folds are similar.
  • Side-Chain Blindness: RMSD focuses on backbone atoms and may yield low values for structures with similar backbones but different side-chain packing, which is the primary driver of distinct amyloid folds.
  • Ambiguity: The lack of a robust, quantitative metric has led to ambiguous language in literature regarding whether two structures are "similar" or "distinct," complicating the replication of disease-specific folds in the lab.

2. Methodology: The Amyloid Packing Difference (APD)

The author proposes a new metric, Amyloid Packing Difference (APD), which quantifies structural divergence based on side-chain interactions rather than global superposition. The calculation is implemented via a three-script Python pipeline:

• Step 1: helix.py (Auxiliary): Applies helical symmetry operators to an input PDB/CIF file to generate multiple layers of the filament, ensuring sufficient context for inter-layer contacts.
• Step 2: contacts.py: Analyzes a single structure to generate a contact list.
  • Contact Definition: Two residues form a contact if non-hydrogen side-chain atoms (or C-alpha for Glycine) are within 6.5 Å.
  • Exclusions: Adjacent residues or those separated by one residue in the same chain are excluded.
  • Geometric Filtering: A reference plane (defined by N, C, and the N of the next layer) ensures contacts are only counted if side chains are on the same side of the backbone.
  • Output: A CSV file listing contacts, residue pairs, and the orientation of side chains relative to their N-C-N plane.
• Step 3: compare.py: Compares two structures using their CSV contact lists.
  • Common vs. Unique Contacts: Contacts are "common" if they exist in both structures with a distance <4.5 Å in one and <6.5 Å in the other. "Unique" contacts exist in one but not the other.
  • Side-Chain Orientation: Residues with flipped side-chain orientations relative to their N-C-N plane are flagged.
  • Calculation:
    $$APD = \frac{N_{different} + N_{extra}}{N_{common} + N_{extra}} \times 100\%$$
    Where $N_{different}$ includes residues with unique contacts or flipped orientations, and $N_{extra}$ accounts for residues present in only one structure (penalizing different ordered cores).
  • Variants:
    • XY-APD: Ignores Z-offsets (layer shifts); treats contacts as common regardless of vertical layer alignment.
    • XYZ-APD: Requires contacts to have the same relative Z-offsets (integer multiples of ~4.75 Å).
  • Visualization: Generates a schematic diagram showing common residues (white), unique contacts (orange/dark orange), and side-chain flips (orange fill).

3. Key Contributions

• A New Metric: Introduction of APD, a superposition-independent metric specifically designed for amyloid polymorphism.
• Validation: Demonstrated that APD-based clustering of $\alpha$-synuclein folds recapitulates existing classifications based on RMSD and TM-scores, validating APD as a robust measure of fold similarity.
• Disease Correlation Thresholds: Established empirical thresholds for interpreting APD values across five major amyloidogenic proteins (Prion, Tau, $\alpha$-synuclein, TDP-43, TAF15):
  • < 20%: Likely the same disease-associated fold (or very close variants).
  • 20% – 40%: Ambiguous; may represent variants within a disease or distinct diseases.
  • > 40%: Strongly suggests distinct diseases or fundamentally different folds.
• Tool Availability: Released open-source Python scripts (contacts.py, compare.py, helix.py) under the MIT license.

4. Key Results

• Prion Protein (Mouse vs. Hamster):
  • Two mouse strains (aRML and a22L) with a hinge movement showed an APD of 38%, correctly identifying them as distinct but related.
  • Mouse vs. Hamster strains showed an APD of 61%, reflecting significant structural divergence.
  • Comparison of two independent structures of the same strain (7QIG vs. 7TD6) showed an XY-APD of 8%, highlighting that minor reconstruction artifacts (e.g., spurious connections in cryo-EM maps) can inflate XYZ-APD but are minimized in XY-APD.
• $\alpha$-Synuclein Clustering:
  • Hierarchical clustering using pairwise APD values successfully separated known clusters (e.g., MSA types A/B, Lewy body folds) without manual superposition.
  • Distinct disease folds separated at APD > 30–80%, while variants within the same fold separated at lower values.
• Re-evaluating Literature Claims:
  • Tau (PHF1-4E vs. PSP): A study claimed a recombinant tau fold "resembled" the PSP fold. APD analysis revealed an 80% difference (no common cross-$\beta$ packing), refuting the claim of similarity.
  • Seed Amplification (MSA): A study claimed seed amplification preserved MSA structural properties. APD showed 38–56% differences in protofilaments and 100% difference in inter-protofilament packing, indicating the structures were not preserved.
  • In Vivo Seeding: Conversely, a mouse model study claiming mimicry of MSA fold A was validated, showing an APD of only 8% with the human MSA fold A.

5. Significance

• Standardization: Provides a quantitative, objective standard for comparing amyloid structures, replacing subjective visual assessments or misleading RMSD values.
• Disease Mechanism Insight: By correlating APD thresholds with disease states, the metric helps distinguish between structural variants that are biologically relevant (different diseases) versus those that are minor conformational fluctuations.
• Experimental Validation: Crucial for the field of synthetic biology and drug discovery, allowing researchers to rigorously verify if their in vitro or in vivo replication of amyloid filaments truly matches the disease-relevant structure.
• Future-Proofing: As cryo-EM resolution improves and reconstruction artifacts (like local minima) decrease, the XYZ-APD variant will become more reliable, but the current XY-APD offers a robust baseline for immediate use.

In summary, the APD metric shifts the focus from global backbone alignment to the specific side-chain packing interactions that define amyloid polymorphism, offering a critical tool for understanding the structural basis of neurodegenerative diseases.