Introducing non-enzymatic crosslinks into atomistic simulations of collagen fibrils

This paper introduces an extension to the ColBuilder framework that enables the generation of atomistic collagen fibril models incorporating non-enzymatic advanced glycation end-product (AGE) crosslinks, providing validated force-field parameters and demonstrating their distinct structural and mechanical impacts compared to enzymatic crosslinks.

The Gist

The Big Picture: Building a Better Model of "Biological Glue"

Imagine your body is a massive construction site. The most important building material isn't steel or concrete; it's collagen. Collagen is the protein that gives your skin its bounce, your bones their strength, and your tendons their ability to snap back.

Think of collagen molecules as long, twisted ropes (triple helices). To make a strong rope ladder, you don't just lay the ropes next to each other; you tie them together with knots. These "knots" are called crosslinks.

For a long time, scientists have been trying to build a perfect digital model of these collagen ropes on a computer to see how they work. But they've been missing a crucial piece of the puzzle: the "rusty" knots that form as we age or when we have diabetes.

The Problem: The "Fresh" vs. "Rusty" Knots

There are two types of knots in collagen:

1. The Fresh Knots (Enzymatic): These are made by the body's own workers (enzymes) when you are young and healthy. They are precise, strong, and placed exactly where they should be.
2. The Rusty Knots (AGEs): As we get older, or if we have high blood sugar (diabetes), sugar molecules accidentally stick to the collagen ropes and bake them into new, messy knots. Scientists call these AGEs (Advanced Glycation End-products).
   • The Analogy: Imagine a brand-new rope ladder. The fresh knots are like tight, perfect knots tied by a master climber. The rusty knots are like random pieces of duct tape or gum stuck to the ropes by a clumsy child. They make the ladder stiffer, but also more brittle and prone to snapping.

Until now, computer models could only simulate the "Fresh Knots." They couldn't simulate the "Rusty Knots" because scientists didn't have the right digital blueprints (parameters) to describe how these messy sugar-knots behave.

The Solution: Introducing "ColBuilder 2.0"

The team behind this paper (Guido, Justin, Frauke, Debora, and Christoph) created an upgrade to a free software tool called ColBuilder. Think of ColBuilder as a 3D printer for collagen.

They did three main things:

1. Expanded the Library: They added the blueprints for three specific types of "Rusty Knots" (Glucosepane, Pentosidine, and MOLD) to the software.
2. The "Patch" Strategy: They figured out a clever way to insert these messy knots into the model without breaking the whole structure. It's like patching a hole in a tire; you do it in stages so the tire doesn't explode.
3. The Physics Check: They ran thousands of simulations to make sure these new digital knots actually behave like real ones. They compared their computer math against high-level quantum physics calculations to ensure accuracy.

The Experiment: Pulling on the Digital Ladder

To test if their new models worked, they built a digital collagen rope ladder and started pulling on it (simulating stress). They tested three scenarios:

• Scenario A: Only Fresh Knots (The healthy control).
• Scenario B: Only Rusty Knots (The aged/diabetic model).
• Scenario C: A mix of both.

What they found:

• The Whole Ladder Stretches the Same: Surprisingly, whether the ladder had fresh knots or rusty knots, the total length it stretched before breaking was about the same.
• The Way It Stretches is Different: This is the big discovery.
  • With Fresh Knots, the ladder stretches mostly in the "gap" sections (the loose parts between the knots), while the "overlap" sections (where the ropes are tightly packed) stay stiff.
  • With Rusty Knots, the stretching shifts! The "overlap" sections start to stretch more, and the "gap" sections stretch less.
  • The Metaphor: Imagine a rubber band with a few stiff clips on it. If you pull it, the rubber stretches between the clips. But if you cover the whole thing in sticky gum (the AGEs), the gum changes where the rubber stretches. The gum makes the stiff parts bend, changing how the whole structure handles stress.

Why Does This Matter?

This isn't just about computer games; it's about real human health.

• Aging and Diabetes: As we age, our bodies accumulate more "Rusty Knots." This makes our tissues stiffer and more likely to break (like brittle bones or stiff arteries).
• Better Medicine: By having a computer model that includes these rusty knots, scientists can finally run simulations to see exactly why aging tissues fail. They can test drugs or therapies on the computer to see if they can stop the "rust" from forming or fix the "glue."
• Open Source: The best part? They made the software free for everyone. Any scientist can now download ColBuilder, build their own collagen models with these new knots, and start their own research.

In a Nutshell

The authors built a new digital toolkit that allows scientists to simulate collagen fibers with the messy, age-related "sugar knots" that happen in real life. They proved that these knots change how collagen stretches, not just how much it stretches. This gives us a powerful new way to understand why our tissues get brittle as we age and how to potentially fix it.

Technical Summary

1. Problem Statement

Collagen fibrils are the primary load-bearing units of connective tissues, with their mechanical properties governed by molecular packing and covalent crosslinks. While enzymatic crosslinks (mediated by lysyl oxidases) are well-characterized, non-enzymatic Advanced Glycation End-product (AGE) crosslinks remain largely absent from current atomistic simulation models.

• The Gap: AGEs accumulate with age and in conditions like Type II diabetes, altering collagen mechanics (increasing stiffness, promoting brittleness) and hydration. However, existing modeling workflows (like the ColBuilder framework) lack the capability to generate simulation-ready fibril models incorporating these chemically diverse, non-enzymatic crosslinks.
• Challenge: Unlike enzymatic crosslinks which occur at specific telopeptide-helix interfaces, AGEs can form between residues distributed throughout the entire collagen sequence, requiring a more flexible approach to identifying candidate sites and optimizing geometry within the staggered D-periodic fibril structure.

2. Methodology

The authors extended the open-source ColBuilder framework to incorporate three representative AGE crosslinks: glucosepane, pentosidine, and MOLD (methylglyoxal-derived lysine dimer). The methodology involved three key technical developments:

A. Candidate Site Identification & Database Expansion

• Sequence Analysis: The team analyzed 19 vertebrate Type I collagen sequences.
• Geometric Screening: They generated periodic fibril representations and screened for candidate residue pairs based on distance constraints:
  • Lys–Arg pairs (for glucosepane and pentosidine).
  • Lys–Lys pairs (for MOLD).
  • Cutoff: A conservative $C_\alpha$–$C_\alpha$ distance of $\le 15$ Å was used to account for limited experimental data on native fibril locations.
• Result: This identified hundreds of candidate sites (e.g., ~102 Lys-Arg and ~45 Lys-Lys sites in the rat template).

B. Staged Insertion & Refinement Workflow

To handle the spatial complexity of AGEs, a new staged refinement protocol was implemented:

1. Enzymatic First: Generate a standard enzymatically crosslinked triple helix.
2. AGE Insertion: Introduce AGE markers via MODELLER patching.
3. Periodic Neighbor Selection: Unlike enzymatic crosslinks (which use a fixed interface), AGE sites can occur in "overlap" or "gap" regions with different intermolecular neighbors. The new workflow allows user-selectable unit-cell translations to refine each AGE site against its geometrically appropriate periodic neighbor.
4. Validation: Candidates are accepted if the refined geometry has no steric clashes and a maximum crosslink-forming atom separation of 5 Å.

C. Force Field Parameterization

• Target: Amber99SB*-ildnp compatibility.
• Process:
  1. Structures built and capped (acetyl/N-methyl).
  2. Geometry optimization and electrostatic potential (ESP) calculation at the B3LYP/6-31G* level (Gaussian).
  3. Partial charges derived via RESP fitting (Antechamber), constraining capping groups to match standard Amber99 charges.
  4. Parameters converted to GROMACS format.
• Validation: Parameters were benchmarked against machine-learned Grappa parameters and QM reference geometries using 100-ns MD replicas.

D. Simulation Setup

• System: Single D-period microfibril (~67 nm, ~3.5 million atoms).
• Conditions: Constant-force loading (0 to 1 nN per strand) using GROMACS 2024.
• Configurations Tested:
  1. Pure Enzymatic (PYD).
  2. Pure AGE (Glucosepane, Pentosidine, MOLD).
  3. Mixed (PYD + Glucosepane).

3. Key Contributions

1. Extended ColBuilder Framework: A modular, open-source Python package capable of generating atomistic collagen fibrils with mixed enzymatic and non-enzymatic crosslink networks.
2. New Parameter Set: Amber99-compatible force field parameters for glucosepane, pentosidine, and MOLD, validated against QM data.
3. Methodological Innovation: A "staged" refinement approach that solves the geometric complexity of placing non-enzymatic crosslinks at arbitrary axial positions within the staggered fibril lattice.
4. Comprehensive Database: A pre-screened database of candidate AGE crosslink sites across 19 vertebrate species.

4. Results

Parameter Validation

• Geometric Accuracy: The Antechamber/RESP parameters reproduced QM reference ring-core geometries with a mean absolute deviation of 2.72° (85% of angles within 5°).
• Comparison: While machine-learned Grappa parameters showed slightly better agreement (1.05° deviation), the Antechamber parameters provided a robust, stable description suitable for integration with existing enzymatic crosslink strategies.
• Stability: All-atom RMSD and RMSF values were consistent between parameter sets, indicating the ring-core geometry is preserved during simulation.

Mechanical Response (MD Simulations)

• Structural Integrity: All models (pure enzymatic, pure AGE, and mixed) maintained structural integrity under constant-force loading without unraveling.
• Deformation Patterns:
  • The fibril exhibited a two-stage response: rapid elongation followed by a plateau.
  • Total Elongation: The D-period elongated by ~24% (reaching ~83 nm), consistent with experimental AFM data.
  • Regional Differences: The gap region stretched significantly more than the overlap region (which remained relatively rigid).
• Impact of Crosslink Type:
  • Pure AGE vs. Enzymatic: Replacing terminal enzymatic crosslinks (PYD) with AGEs caused a systematic redistribution of deformation. Extension shifted slightly toward the overlap region and away from the gap region.
  • Density vs. Chemistry: Simply adding AGEs to an existing enzymatic network (Mixed system) had negligible effect on mechanics compared to the replacement of key enzymatic crosslinks with AGEs. This suggests the chemical identity and location of the crosslink are more critical than total density.
  • Specifics: AGE-only systems showed larger overlap extensions compared to the PYD-only reference, indicating AGEs alter the internal force distribution differently than enzymatic crosslinks.

5. Significance

• Bridging Biology and Simulation: This work provides the first robust, atomistic framework to simulate the specific mechanical consequences of glycation (aging/diabetes) on collagen.
• Disease Modeling: It enables researchers to investigate how AGE accumulation contributes to tissue stiffening and brittle failure, crucial for understanding diabetic complications and aging.
• Versatility: The tool supports user-defined control over crosslink density and composition, facilitating systematic in silico studies of tissue remodeling.
• Future Applications: The generated models serve as high-resolution templates for fitting cryo-EM data and can be extended to study other post-translational modifications or crosslink types as new experimental data becomes available.

In summary, the authors have successfully democratized the study of collagen aging by providing the necessary computational tools to model the complex, non-enzymatic crosslinking networks that define the mechanical fate of connective tissues in disease and aging.