Library docking for Cannabinoid-2 Receptor ligands

This study demonstrates that structure-based docking of a 2.6 billion molecule library against the Cannabinoid-2 receptor successfully identified eight diverse, potent, and subtype-selective ligand families by targeting polar orthosteric residues, with experimental validation confirming the accuracy of the docking predictions and the efficacy of subsequent structure-based optimization.

The Gist

The Big Picture: Finding the Right Key for a Very Picky Lock

Imagine the human body has thousands of different "locks" (receptors) on the surface of its cells. These locks control everything from pain to mood to inflammation. Two of these locks, CB1 and CB2, are very similar twins. They look almost identical, but they live in different neighborhoods: CB1 hangs out in the brain, while CB2 patrols the immune system.

For decades, scientists have tried to design "keys" (drugs) that fit perfectly into the CB2 lock to treat pain or inflammation without accidentally unlocking the CB1 lock in the brain (which causes the "high" associated with marijuana).

This paper is the story of a massive, high-tech treasure hunt to find those perfect keys using a computer.

The Strategy: The Digital Fishing Net

Instead of physically testing millions of chemicals in a lab (which would take forever and cost a fortune), the researchers used a supercomputer to simulate "docking" billions of virtual molecules into the CB2 lock.

Think of it like this:

• The Lock (CB2 Receptor): A complex, 3D puzzle with a specific shape and some sticky spots (polar residues) inside.
• The Keys (Molecules): Billions of different shapes and sizes.
• The Computer: A giant fishing net that casts out into a sea of 2.6 billion virtual molecules to see which ones fit.

The Three Big Discoveries

The researchers ran three different "fishing" campaigns, and here is what they learned:

1. The "Polar" Trick: Finding Selective Keys

In the past, when scientists tried to find keys for the twin CB1 lock, they found powerful keys, but they were "sloppy"—they opened both CB1 and CB2.

• The Insight: The researchers noticed that the inside of the CB2 lock has some specific "sticky" spots (polar interactions) that the CB1 lock doesn't have as strongly.
• The Analogy: Imagine the CB2 lock has a specific Velcro patch inside. If you design a key with a fuzzy patch that sticks to that Velcro, it will snap into place. If you make the key too greasy (hydrophobic), it slides right past.
• The Result: By programming the computer to prioritize these "sticky" interactions, they found keys that fit CB2 perfectly but ignored CB1. This solved the "sloppy key" problem.

2. Size Matters: The More, The Merrier

They ran two main searches:

• Search A: A small net with 7 million molecules.
• Search B: A massive net with 2.6 billion molecules.

The Result: The bigger net caught much better fish.

• From the small net, they found keys that worked, but they were weak (like a key that turns the lock with a lot of effort).
• From the giant net, they found keys that were 14 to 100 times stronger.
• The Lesson: When you search a tiny library, you might get lucky. But when you search a library the size of a small country's population, you are almost guaranteed to find the "perfect" key that nature didn't give us yet.

3. The "On/Off" Switch Problem

Receptors have two main states: Active (ON) and Inactive (OFF). Scientists hoped that if they docked keys into the "Active" shape of the lock, they would only find "ON" switches (agonists), and if they docked into the "Inactive" shape, they would only find "OFF" switches (inverse agonists).

• The Reality: It didn't work that cleanly. Even when they targeted the "OFF" shape, they found some "ON" keys, and vice versa.
• The Analogy: It's like trying to predict if a person will be happy or sad just by looking at their face. Sometimes a smile (active shape) hides a sad person, and sometimes a frown (inactive shape) hides a happy one. The computer couldn't perfectly predict the "mood" of the drug just by the shape of the lock.

The Proof: Taking the Keys to the Lab

Finding a key on a computer is one thing; proving it works in real life is another.

1. Testing: They synthesized the best virtual keys and tested them in real cells. They worked! They found 8 completely new families of drugs that had never been seen before.
2. The "X-Ray" Check: To prove the computer wasn't lying, they took two of the best new drugs and used a high-tech microscope (Cryo-EM) to take a 3D picture of the drug sitting inside the lock.
   • The Result: The picture matched the computer prediction almost perfectly. The drug was sitting exactly where the computer said it would be.

Why This Matters

This paper is a huge win for drug discovery for three reasons:

1. Selectivity: We can now design drugs that target specific body parts (like the immune system) without messing up the brain.
2. Scale: Bigger libraries = better drugs. We don't need to stop at millions; we should be searching billions.
3. Novelty: They found 8 totally new types of chemical structures. It's like finding a new species of animal in the ocean; it opens up entirely new avenues for making medicines.

The Bottom Line

The researchers used a supercomputer to sift through 2.6 billion virtual molecules to find a few dozen perfect keys for a specific body lock. By being smart about how they looked (focusing on sticky spots) and looking at a huge number of options, they found powerful new drugs that are highly selective and work exactly as the computer predicted. It's a triumph of "digital fishing" leading to real-world medicine.

Technical Summary

1. Problem Statement

The study addresses several critical challenges in structure-based drug discovery (SBDD) for G protein-coupled receptors (GPCRs), specifically the Cannabinoid-2 (CB2) receptor:

• Subtype Selectivity: CB1 and CB2 receptors share high sequence identity (~44% overall, ~68% in the orthosteric binding site, and up to ~90% in binding site residues). Previous docking campaigns against CB1 yielded potent but non-selective ligands. The challenge is to design strategies to discover CB2-selective ligands.
• Functional Bias: It is unclear if docking against specific receptor conformations (active vs. inactive) reliably biases the discovery of agonists vs. inverse agonists, respectively.
• Library Size Impact: The relationship between library size (7 million vs. billions of molecules) and the quality of hits (affinity and hit rate) in lipid-receptor sites is not fully established.
• Enumerated vs. Implied Libraries: A comparison is needed between brute-force docking of a fully enumerated library (2.6 billion molecules) and "synthon-based" docking of an implied library (11.5 billion molecules) to determine efficiency and efficacy.
• Chemical Novelty: There is a need to discover ligands with chemotypes distinct from known cannabinoid scaffolds found in databases like ChEMBL.

2. Methodology

The researchers employed a multi-stage virtual screening and experimental validation pipeline:

• Target Structures:
  • Active State: CB2 receptor bound to agonist WIN 55,212-2 (PDB: 6PT0).
  • Inactive State: CB2 receptor bound to inverse agonist AM10257 (PDB: 5ZTY).
• Docking Strategy (DOCK 3.8):
  • Selectivity Filter: To overcome the high similarity between CB1 and CB2, the team prioritized polar interactions within the otherwise lipophilic binding pocket. They increased the local dipoles of polar residues (T3.33, S7.39, S2.60, H2.65, and L182ECL2 backbone) in electrostatic calculations and applied post-docking filters for hydrogen bonds with these residues.
  • Functional Bias Filter: For the inactive state screen, an additional filter was applied to ensure proximity to the "toggle switch" residue W6.48 to stabilize the inactive conformation.
• Library Screening Campaigns:
  1. 7 Million (In-Stock): Docked against the active state. ~34 molecules were synthesized/purchased and tested.
  2. 1.6 Billion (Make-on-Demand): Docked against the active state with reduced sampling but higher score stringency. ~151 molecules synthesized and tested.
  3. 2.6 Billion (Fully Enumerated): Docked against the inactive state with high sampling. ~290 molecules synthesized and tested.
• Experimental Validation:
  • Binding Assays: Radioligand displacement assays using [³H]-CP55,940 (for 7M/1.6B) and [³H]-WIN 55212-2 (for 2.6B) to determine $K_i$.
  • Functional Assays: BRET (Bioluminescence Resonance Energy Transfer) for G-protein dissociation and GloSensor cAMP accumulation assays to determine EC50/IC50 for agonism and inverse agonism.
  • Selectivity: Parallel testing against CB1 receptors.
  • Cryo-EM: Structures of two optimized agonists ('5249 and '1029) bound to CB2-Gi1 complexes were solved to validate docking poses.
• Optimization: Structure-Activity Relationship (SAR) studies were conducted on hit series to improve potency and selectivity.

3. Key Contributions

• Polar Interaction Strategy: Demonstrated that explicitly targeting polar residues in a lipophilic GPCR binding site is an effective strategy for achieving subtype selectivity (CB2 over CB1).
• Library Size Scaling: Provided empirical evidence that increasing library size from millions to billions significantly improves hit rates and, more importantly, the potency of discovered ligands (shifting from µM to low nM).
• Functional Selectivity Limits: Confirmed that while larger libraries yield more potent ligands, docking against specific receptor states (active/inactive) does not reliably predict functional bias (agonist vs. inverse agonist) for CB2, likely due to minimal structural differences (<1.5 Å) between states compared to other GPCRs.
• Enumerated vs. Synthon Comparison: Showed that brute-force docking of a 2.6 billion molecule library yielded similar hit rates but significantly higher potency (4-fold in binding, 44-fold in function) compared to synthon-based docking of an 11.5 billion implied library.
• Novel Chemotypes: Discovered eight distinct ligand families with high potency (<100 nM) that are topologically unrelated to known cannabinoid ligands in ChEMBL.

4. Key Results

• Selectivity: By prioritizing polar interactions, the team identified ligands with up to 115-fold selectivity for CB2 over CB1 (e.g., compound '5249).
• Potency & Hit Rates:
  • 7M Screen: Hit rate ~17%; best $K_i$ ~0.9 µM; no measurable functional activity.
  • 1.6B Screen: Hit rate ~17%; best $K_i$ ~1.5 µM; identified agonists and inverse agonists.
  • 2.6B Screen: Hit rate ~17%; best $K_i$ ~65 nM (agonist) and ~28 nM (inverse agonist). Functional activities reached low nanomolar ranges.
• Functional Diversity: The 2.6B screen against the inactive state surprisingly yielded potent agonists (e.g., Z2158902960, EC50 2.5 nM) alongside inverse agonists, challenging the assumption that inactive-state docking yields only inverse agonists.
• SAR Optimization:
  • Series derived from the 7M screen ('6138) were optimized to achieve 140-fold improvement in potency.
  • Series from the 1.6B screen ('0055) were optimized to 50-fold improvement, with stereochemical purification yielding a cis-isomer ('1029) with 5.0 nM EC50.
• Cryo-EM Validation:
  • Structures of agonists '5249 and '1029 were solved at 2.9 Å resolution.
  • Experimental poses superposed well with docking predictions (RMSD 1.8 Å and 1.1 Å).
  • Some discrepancies were noted (e.g., water-mediated interactions or isomer flips), but the core binding modes were validated.
• Novelty: 7 out of 8 discovered chemotypes had Tanimoto coefficients <0.35 compared to 2,391 known CB2 ligands in ChEMBL, representing truly novel chemical space.

5. Significance

• Therapeutic Potential: The discovery of potent, selective, and novel CB2 agonists and inverse agonists offers new tool molecules for studying immune modulation, inflammation, and pain, potentially bypassing the psychoactive side effects associated with CB1 activation.
• Methodological Benchmark: This study serves as a benchmark for ultra-large library screening (ULLS), demonstrating that brute-force enumeration of billions of molecules can outperform synthon-based approaches in terms of hit quality, despite the computational cost.
• GPCR Docking Insights: It highlights the limitations of current structure-based methods in predicting functional bias for receptors with subtle conformational differences (like CB2), suggesting that ligand-specific features rather than receptor state alone may dictate function.
• Future Directions: The success of prioritizing polar interactions in lipid receptors suggests a generalizable strategy for improving selectivity in other GPCR families. The findings support the continued expansion of virtual libraries into the trillions of molecules.