⚛️ biophysics

Transfer seeding reveals divergent activation ensembles across tyrosine kinases

By combining extensive molecular dynamics simulations with a novel transfer seeding strategy, this study reveals that tyrosine kinases (ABL, EGFR, and MET) follow distinct, kinase-specific activation pathways rather than a universal mechanism, highlighting unique rate-limiting steps and conformational ensembles that refine our understanding of their regulatory landscapes.

Original authors: Zhu, Y., Degiacomi, M. T., Mey, A. S. J. S., Singh, S.

Published 2026-10-15

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Zhu, Y., Degiacomi, M. T., Mey, A. S. J. S., Singh, S.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Inside every living cell, tiny molecular machines called protein kinases act as the master switches that turn biological processes on and off. These enzymes control everything from cell growth to the decision to die, functioning by attaching a small chemical tag to other proteins. When these switches malfunction, cells can grow out of control, leading to diseases like cancer. For decades, scientists have studied the shape of these machines to understand how they work, often relying on snapshots taken by powerful microscopes. However, these snapshots are static; they capture a single moment in time, much like a photograph of a runner frozen mid-stride. They miss the fluid motion and the many different shapes a kinase can take as it switches between its "off" and "on" states. Because these machines are so central to life and disease, understanding their full range of motion is critical for designing drugs that can stop them when they go wrong.

A team of researchers has now taken a deep dive into the hidden movements of three specific kinases: ABL, EGFR, and MET. These proteins are famous targets in cancer therapy, yet scientists have long wondered if they all follow the same set of rules to turn on. To answer this, the researchers did not just look at existing static images. Instead, they used a massive amount of computer power to simulate the actual movement of these molecules over time. They ran simulations that, if added together, would take over eight thousand seconds of continuous time to watch a single molecule move. This allowed them to see the molecules wiggle, twist, and shift through thousands of different shapes, revealing a dynamic landscape that static images could never show.

The study began by acknowledging a gap in what is known. While scientists have many crystal structures for ABL and EGFR, the MET kinase is poorly represented in existing databases. There are very few experimental images of MET in certain shapes, leaving a blind spot in our understanding of how it works. To fill this gap without introducing bias, the researchers developed a clever strategy they call "transfer seeding." They used a powerful artificial intelligence tool, which is excellent at predicting protein shapes, but guided it using the known shapes of similar proteins. By feeding the AI templates from ABL and EGFR, they generated new, plausible starting shapes for MET that had never been seen in a lab. This allowed them to launch their simulations from a much wider variety of starting points, ensuring they could explore the entire range of MET's movements, not just the ones we already knew about.

The results revealed that these three kinases, while built from similar parts, do not follow a universal path to activation. They all share a common set of basic shapes, but they spend different amounts of time in each one, and they move between them in unique ways. For the ABL kinase, the slowest and most difficult part of turning on involves a large rearrangement of a flexible tail called the activation loop. This tail must swing open and reorganize itself, a process that takes a significant amount of time. In contrast, the EGFR kinase is much faster to activate. Its bottleneck is not the large tail, but a tiny, local adjustment of a specific motif near the core of the machine. The MET kinase follows a third path entirely, moving through a series of intermediate shapes that were previously unknown to science. These intermediate states act as stepping stones, allowing the molecule to gradually open up its activation loop before reaching the fully active form.

Perhaps the most surprising discovery was that the "active" state is not always the best place to find the molecule ready to be switched on by chemical signals. In the case of MET and ABL, the specific spots on the activation loop that need to be tagged by the cell's signaling system are actually more exposed and accessible in the intermediate, partially open shapes. In the final, fully active state, these spots are often tucked away and hidden. This suggests that the cell might not wait for the machine to be fully active before tagging it; instead, it might catch the machine in a fleeting, intermediate moment to flip the switch. This finding challenges the simple view that a kinase is either "off" or "on" and suggests a more nuanced reality where the machine passes through specific, transient windows of opportunity.

The study also highlighted that the speed of these transitions varies wildly. Moving from an inactive state to an active one can take anywhere from a fraction of a second to several milliseconds, depending on the specific kinase. For MET, the journey from a closed, inactive shape to an open, active one takes about six milliseconds, a long time in the molecular world. For EGFR, the same transition happens much faster, in just tens of microseconds. These differences in timing and pathway mean that a drug designed to lock one kinase in an inactive shape might not work on another, even if they look similar. The researchers found that the specific shapes and the routes taken to get there are unique to each protein, creating a distinct "personality" for each kinase.

By combining massive computer simulations with a new way of generating starting shapes, the researchers have provided a systematic map of how these critical machines move. They showed that while the basic building blocks of kinases are conserved, the way they move and the paths they take to become active are highly specific to each type. This work moves beyond static pictures to a dynamic understanding of how these proteins function in real time. It reveals that the activation of a kinase is not a single, universal event but a complex, kinase-specific journey through a landscape of shapes. This level of detail offers a new way to think about how to design drugs, suggesting that the most effective inhibitors might need to target these unique, transient intermediate shapes rather than just the final active or inactive forms. The study confirms that to truly understand and control these molecular switches, we must look at the full story of their movement, not just the still frames.

Technical Summary: Transfer Seeding Reveals Divergent Activation Ensembles Across Tyrosine Kinases

Problem Statement
Protein kinase activation relies on coordinated rearrangements of conserved regulatory elements, specifically the Asp–Phe–Gly (DFG) motif and the activation loop (A-loop). While the structural states of these elements (e.g., DFG-in, DFG-out, DFG-inter) are well-defined, it remains unclear whether tyrosine kinases share a universal activation mechanism or follow kinase-specific pathways. A major barrier to resolving this is the uneven experimental structural coverage across the kinome; many kinases lack experimental structures for key intermediate states. Furthermore, conventional unbiased molecular dynamics (MD) simulations often fail to sample rare, functionally important transitions (such as DFG flips) within accessible timescales, while enhanced sampling methods can introduce biases that complicate quantitative ensemble interpretation. There is a need for a common, unbiased protocol to generate comparable conformational ensembles across diverse kinases to determine if activation pathways are conserved.

Methodology
The authors employed a systematic framework combining distributed unbiased MD simulations with Markov State Models (MSMs) to characterize the equilibrium landscapes of three apo tyrosine kinase domains (KDs): ABL, EGFR, and MET.

Simulation Scale: The study utilized over 8.3 milliseconds of aggregate unbiased MD simulation data distributed via the Folding@home platform (0.592 ms for ABL, 1.836 ms for EGFR, and 5.938 ms for MET).
Transfer Seeding Strategy: To address the lack of experimental structures for MET (which is under-represented in the PDB and lacks DFG-inter structures), the authors developed a "transfer seeding" protocol. Instead of relying solely on experimental PDB seeds or default AlphaFold2 (AF2) predictions (which failed to sample non-DFG-in states), they guided AF2 modeling using homologous templates from the KinCore database. By providing AF2 with templates annotated with specific DFG states (e.g., DFG-inter, DFG-out) from ABL and EGFR, they generated physically plausible MET structures in states absent from the experimental MET ensemble.
MSM Construction: Trajectories were processed using tICA for dimensionality reduction and k-means clustering to define microstates. Reversible maximum-likelihood MSMs were constructed to identify metastable macrostates and estimate transition timescales.
Analysis: The authors analyzed DFG geometry using the Modi–Dunbrack nomenclature (based on D1 and D2 distances and backbone dihedral angles) and characterized A-loop conformations. Activation pathways were decomposed using Transition Path Theory (TPT) to identify rate-limiting steps, and Random Forest classifiers were used to extract structural features distinguishing pre- and post-rate-limiting states.

Key Contributions

Transfer Seeding Protocol: The paper introduces a method to expand conformational sampling for structurally under-represented kinases by transferring conformational information from homologous templates to AF2 models, enabling unbiased sampling of missing intermediate states without relying on biased enhanced sampling potentials.
Kinase-Specific Landscapes: The study provides a direct comparison of ABL, EGFR, and MET, revealing that while they access similar DFG structural classes, their equilibrium populations, kinetic connectivity, and dominant activation routes differ significantly.
Resolution of Hidden Intermediates: The MSMs resolved metastable conformations not present in experimental datasets, including low-population DFG-inter intermediates along the MET activation pathway and distinct A-loop arrangements in EGFR and MET.

Key Results

Divergent Populations and Kinetics: ABL and EGFR share a subset of DFG states but exhibit different equilibrium populations. ABL is dominated by a DFG-out, A-loop closed state (State 6), while EGFR is dominated by a DFG-in, SRC-like inactive state (State 6). MET shows a distinct landscape where the active-like state is less accessible than the SRC-like ensemble.
Distinct Activation Pathways:
- ABL: The rate-limiting step involves the rearrangement of the A-loop and $\alpha$ C helix to transition from a closed A-loop DFG-out state to a SRC-like intermediate, followed by A-loop opening to reach the active state.
- EGFR: Activation is rate-limited by local rearrangements of the DFG motif (specifically the backbone dihedrals of the residue preceding the DFG motif) rather than large-scale A-loop or $\alpha$ C helix motions. The SRC-like ensemble acts as an off-pathway basin.
- MET: The pathway proceeds through low-population DFG-inter intermediates (States 1 and 3) where the A-loop is partially folded and interacts with the $\alpha$ C helix. The rate-limiting step is the initial opening of the A-loop coupled to DFG motif rearrangement.
Phosphosite Accessibility: The study found that A-loop phosphorylation sites (Tyr393 in ABL, Tyr1235 in MET) are more solvent-accessible in specific intermediate ensembles (e.g., BLBplus for ABL, BABtrans for MET) than in the final active-like BLAminus state. This suggests that phosphorylation may select or stabilize transient pre-active intermediates rather than marking the final active state. In contrast, EGFR's Tyr869 is highly exposed in the active-like state, consistent with its non-canonical regulation via dimerization.

Significance
The paper claims that kinase activation does not proceed through a conserved mechanism shared across tyrosine kinases but rather through kinase-specific transitions embedded within broader regulatory ensembles. The study establishes a systematic biophysical framework for comparative kinase ensemble characterization that integrates DFG geometry with A-loop dynamics. By resolving kinase-specific states and the atomic-resolution structural rearrangements underlying their interconversion, the work provides a foundation for understanding how conserved structural elements give rise to divergent functional dynamics. The authors suggest these findings have implications for conformation-selective inhibitor design, as inhibitor selectivity may depend on the broader conformational network and specific intermediate states unique to each kinase, rather than just local binding site interactions. The transfer seeding strategy is presented as a practical route to extend this ensemble-generation framework to kinases with incomplete experimental structural coverage.

More like this