Hydrogen-Bond-Basicity Predictions for Scaffold Hopping in PDE2A Inhibitors

by Corin Wagen · Feb 14, 2025

In a 2018 study (J. Med. Chem. 2018, 61, 1001–1018), Chris Helal and co-workers at Pfizer reported the development of phosphodiesterase 2A (PDE2A) inhibitors as potential treatments for cognitive disorders in schizophrenia. Their core scaffold—pyrazolopyrimidine 1—exhibited good potency but remained relatively lipophilic, leading to high human-liver-microsome clearance and thus an excessively high estimated dose.

The authors investigated peripheral modifications to decrease lipophilicity, but found that the perturbations they tried either compromised activity or led to increased values of MDR BA/AB. (MDR BA/AB is the ratio of basolateral-to-apical flux to apical-to-basolateral flux of a compound in MDCK cells engineered to express P-glycoprotein (MDR1), reflecting P-gp–mediated efflux). Accordingly, the Pfizer team next investigated re-engineering the central heterocyclic core to preserve potency while improving lipophilicity and efflux.

A few different peripheral modifications.

Peripheral modifications to 1 were unsuccessful.

LMP2 Calculations Identify a More Potent Scaffold

Inspired by results from Bayer showing that different heterocyclic isomers could lead to 50x potency improvements for PDE5A, the team explored an imidazotriazine ring to replace the pyrazolopyrimidine scaffold. Counterpoise-corrected LMP2/cc-pVTZ//X3LYP/6-31G** calculations in Jaguar indicated that key hydrogen-bond interactions in the enzyme's active site would be strengthened with the imidazotriazine core. This prediction was borne out experimentally: after extensive optimization, the new scaffold led to the clinical candidate PF-05180999, which demonstrated higher PDE2A affinity and improved brain penetration.

Illustration of the PDE2A binding site.

Illustration of the inhibitor class bound to PDE2A and the key hydrogen bonds made with the protein (PDB 5UZL).

pKBHX Workflow as a Black-Box Alternative

Unfortunately, the LMP2-based calculations run by Helal and co-workers are challenging for non-expert practitioners to perform accurately: proper setup of these calculations requires properly aligning new ligands with experimental or predicted protein–ligand complexes, deleting all but the relevant regions of the protein, and then hoping that the generated structure is both small enough to enable practical MP2 calculations and large enough to capture all relevant interactions.

Rowan's hydrogen-bond-basicity-prediction workflow (pKBHX) offers a more accessible way to estimate site-specific hydrogen-bond acceptor strengths. Whereas LMP2-based workflows require specialized resources and expertise, the pKBHX approach uses only a single density-functional-theory calculation per molecule, automatically handles different conformers, and has been calibrated and validated against a large experimental database. This method identifies the most strongly hydrogen-bonding heteroatoms, enabling medicinal chemists to prioritize promising scaffold hops without advanced quantum-chemistry infrastructure.

Agreement with High-Level QM Calculations

In an independent analysis of the same scaffold hop, pKBHX calculations confirmed that the imidazotriazine ring would generally strengthen the critical hydrogen bond to the five-membered ring (N2 when using the pyrazolopyrimidine numbering) compared to the original pyrazolopyrimidine core: Rowan predicts an increase in pKBHX of 0.88 units (almost an order of magnitude), while Helal and co-workers predict that the hydrogen bond will be 1.4 kcal/mol stronger.

Both the LMP2-based computations and the pKBHX results also agree qualitatively on the overall increase in hydrogen-bond strengths upon scaffold modification. The two methods disagree on the effect of the scaffold hop on N5, suggesting that in this case side-chain and water-mediated interactions can slightly alter outcomes for ligand-only predictions. Nonetheless, the overall conclusion is the same: substituting imidazotriazine yields a net improvement in hydrogen bonding and potency.

ChemDraw comparison of compounds 10 and 11

Comparison of the hydrogen-bond-acceptor strength of the pyrazolipyrimidine 10 and the imidazotriazine 11. More negative LMP2 values imply stronger hydrogen bonding; higher pKBHX values indicate stronger hydrogen-bond acceptors.

This case study demonstrates how modeling and simulation tools like the pKBHX workflow can help make complex decisions like scaffold hops more data-driven, accelerating project timelines and decreasing the role of luck in hit-to-lead optimization. While top pharmaceutical companies like Pfizer have long relied on sophisticated computational workflows to help make these high-stakes decisions (as shown by this paper), it's now possible for even small teams to take advantage of these insights.

Rowan is working to democratize computational chemistry and bring advanced design & simulation tools to every scientist, not just computational experts. If you're interested in working with us, reach out to our team at contact@rowansci.com or start using our web-based computational chemistry platform today!

Banner background image

What to Read Next

Efficient Black-Box Prediction of Hydrogen-Bond-Donor and Acceptor Strength

Efficient Black-Box Prediction of Hydrogen-Bond-Donor and Acceptor Strength

Here, we report a robust black-box workflow for predicting site-specific hydrogen-bond basicity and acidity in organic molecules with minimal computational cost.
Jul 1, 2025 · Corin C. Wagen
Tracking Boltz-2 Benchmarks

Tracking Boltz-2 Benchmarks

Tracking the community's response to the new Boltz-2 model, plus some notes about Chai-2.
Jul 1, 2025 · Corin Wagen
g-xTB, Credit Usage, & More

g-xTB, Credit Usage, & More

the new g-xTB model from Grimme and co-workers; an easy visual overview of credit usage; better credit handling for organizations; bulk PDB download; a new collapsible JSON viewer
Jun 27, 2025 · Jonathon Vandezande, Ari Wagen, Spencer Schneider, and Corin Wagen
Representing Local Protein Environments With Atomistic Foundation Models

Representing Local Protein Environments With Atomistic Foundation Models

A guest post about how to use NNP embeddings for other prediction tasks.
Jun 20, 2025 · Meital Bojan and Sanketh Vedula
Co-Folding Updates

Co-Folding Updates

Boltz-2 FAQ and launch event recap; new visuals for co-folding workflows; new submission options; PDB bugfixes; new credit-management tools
Jun 12, 2025 · Ari Wagen, Spencer Schneider, and Corin Wagen
The Boltz-2 FAQ

The Boltz-2 FAQ

Questions and answers about the Boltz-2 biomolecular foundation model.
Jun 9, 2025 · Corin Wagen and Ari Wagen
Cleaning the Tap Room

Cleaning the Tap Room

beer and bezos; terms-of-service and privacy-policy updates; more deployment options; compliance requirements and country restrictions; a blog post about transition states
Jun 6, 2025 · Ari Wagen and Corin Wagen
BREAKING: Boltz-2 Now Live On Rowan

BREAKING: Boltz-2 Now Live On Rowan

This morning, a team of researchers from MIT and Recursion released Boltz-2, an open-source protein–ligand co-folding model.
Jun 6, 2025 · Corin Wagen, Spencer Schneider, and Ari Wagen
How to Run Boltz-2

How to Run Boltz-2

Step-by-step guides on how to run the Boltz-2 model locally and through Rowan's computational_chemistry platform.
Jun 6, 2025 · Corin Wagen
Guessing Transition States

Guessing Transition States

Methods for generating guess transition states for reaction modeling.
Jun 5, 2025 · Jonathon Vandezande