Hyponatremia is a prevalent disorder marked by excess water retention and substantial morbidity, motivating interest in the CLC-Ka chloride channel as a therapeutic target. Selectively inhibiting CLC-Ka without affecting the closely related CLC-Kb is essential for preventing serious side effects. However, developing isoform-selective inhibitors has been challenging because most small molecules do not distinguish between CLC-Ka and CLC-Kb, and the basis for selectivity in the few known exceptions remains unclear. The small molecule BIM1 preferentially inhibits CLC-Ka over CLC-Kb, providing an opportunity to dissect isoform-specific pharmacology. To investigate this mechanism, we determined cryo-EM structures of BIM1 and BIM15, a related nonselective analog, bound to a CLC-K variant engineered to match the human CLC-Ka binding pocket. Structural and computational analyses reveal that inhibition and isoform selectivity are anchored by interactions with a conserved lysine, with surrounding binding-site residues subtly tuning the local electrostatic environment to promote or disfavor these contacts. These analyses further identify a dynamic extracellular loop that intermittently occludes the access pathway, indicating its role as a gate for ions and inhibitors. BIM15 engages this gating loop more extensively than BIM1, suggesting that differential loop engagement contributes to inhibitor selectivity. To probe how gating reshapes this region, we solved the structure in the presence of Ca2+, which favors channel opening, and found the gating loop ordered and withdrawn from the pathway. Together, these findings elucidate how CLC-K channels gate and how subtle binding-site differences and loop dynamics shape isoform-specific drug binding, providing a foundation for designing next-generation CLC-Ka inhibitors.

A common starting point for drug design is to find small chemical groups or "fragments" that form interactions with distinct subregions in a protein binding pocket. The subsequent challenge is to assemble these fragments into a molecule that has high affinity to the protein, by adding chemical bonds between atoms in different fragments. This "molecule assembly" task is particularly challenging because, initially, fragment positions are known only approximately. Prior methods for spatial graph completion-adding missing edges to a graph whose nodes have associated spatial coordinates-either treat node positions as fixed or adjust node positions before predicting edges. The fact that these methods treat geometry and topology prediction separately limits their ability to reconcile noisy geometries and plausible connectivities. To address this limitation, we introduce EdGr, a spatial graph diffusion model that reasons jointly over geometry and topology of molecules to simultaneously predict fragment positions and inter-fragment bonds. Importantly, predicted edge likelihoods directly influence node position updates during the diffusion denoising process, allowing connectivity cues to guide spatial movements, and vice versa. EdGr substantially outperforms previous methods on the molecule assembly task and maintains robust performance as noise levels increase. Beyond drug discovery, our approach of explicitly coupling geometry and topology prediction is broadly applicable to spatial graph completion problems, such as neural circuit reconstruction, 3D scene understanding, and sensor network design.

While generative AI is transforming the de novo design of proteins, its effectiveness for structure-based design of small molecules remains limited. Current methods, including diffusion models, often produce small molecules with difficult-to-synthesize structures, poor medicinal chemistry properties, and limited target selectivity. To address these limitations, we introduce MedSAGE, a novel generative AI framework that adapts diffusion models specifically for de novo small-molecule design. Rather than using atoms or strings, we develop a novel representation for generating molecules using fragments relevant for medicinal chemistry. Chemical and geometric information characterizing these fragments is embedded in a smooth and interpretable latent space. We also develop an algorithm to optimize the connectivity between generated fragments while preserving chemical validity and synthesizability. In a benchmark of multiple methods across 25 therapeutically relevant protein targets, MedSAGE achieved state-of-the-art performance, producing synthesizable, drug-like molecules with predicted affinity and selectivity closely matching known drugs and drug candidates. Compared to large-scale virtual screening, MedSAGE produced molecules with high predicted affinity over 100 times more efficiently. Our results demonstrate that MedSAGE is already practically useful and paves the way for next-generation tools in structure-guided drug design.

The current opioid overdose epidemic highlights the urgent need to develop safer and more effective treatments for chronic pain1. Cannabinoid receptor type 1 (CB1) is a promising non-opioid target for pain relief, but its clinical use has been limited by centrally mediated psychoactivity and tolerance. We overcame both issues by designing peripherally restricted CB1 agonists that minimize arrestin recruitment. We achieved these goals by computationally designing positively charged derivatives of the potent CB1 agonist MDMB-Fubinaca2. We designed these ligands to occupy a cryptic pocket identified through molecular dynamics simulations-an extended binding pocket that opens rarely and leads to the conserved signalling residue D2.50 (ref. 3). We used structure determination, pharmacological assays and molecular dynamics simulations to verify the binding modes of these ligands and to determine the molecular mechanism by which they achieve this dampening of arrestin recruitment. Our lead ligand, VIP36, is highly peripherally restricted and demonstrates notable efficacy in three mouse pain models, with 100-fold dose separation between analgesic efficacy and centrally mediated side effects. VIP36 exerts analgesic efficacy through peripheral CB1 receptors and shows limited analgesic tolerance. These results show how targeting a cryptic pocket in a G-protein-coupled receptor can lead to enhanced peripheral selectivity, biased signalling, desired in vivo pharmacology and reduced adverse effects. This has substantial implications for chronic pain treatment but could also revolutionize the design of drugs targeting other G-protein-coupled receptors.

CLC-2 is a voltage-gated chloride channel that contributes to electrical excitability and ion homeostasis in many different tissues. Among the nine mammalian CLC homologs, CLC-2 is uniquely activated by hyperpolarization, rather than depolarization, of the plasma membrane. The molecular basis for the divergence in polarity of voltage gating among closely related homologs has been a long-standing mystery, in part because few CLC channel structures are available. Here, we report cryoEM structures of human CLC-2 at 2.46 – 2.76 Å, in the presence and absence of the selective inhibitor AK-42. AK-42 binds within the extracellular entryway of the Cl – -permeation pathway, occupying a pocket previously proposed through computational docking studies. In the apo structure, we observed two distinct conformations involving rotation of one of the cytoplasmic C-terminal domains (CTDs). In the absence of CTD rotation, an intracellular N-terminal 15-residue hairpin peptide nestles against the TM domain to physically occlude the Cl – -permeation pathway. This peptide is highly conserved among species variants of CLC-2 but is not present in other CLC homologs. Previous studies suggested that the N-terminal domain of CLC-2 influences channel properties via a “ball-and-chain” gating mechanism, but conflicting data cast doubt on such a mechanism, and thus the structure of the N-terminal domain and its interaction with the channel has been uncertain. Through electrophysiological studies of an N-terminal deletion mutant lacking the 15-residue hairpin peptide, we support a model in which the N-terminal hairpin of CLC-2 stabilizes a closed state of the channel by blocking the cytoplasmic Cl – -permeation pathway.

The goal of designing safer, more effective drugs has led to tremendous interest in molecular mechanisms through which ligands can precisely manipulate signaling of G-protein-coupled receptors (GPCRs), the largest class of drug targets. Decades of research have led to the widely accepted view that all agonists-ligands that trigger GPCR activation-function by causing rearrangement of the GPCR's transmembrane helices, opening an intracellular pocket for binding of transducer proteins. Here we demonstrate that certain agonists instead trigger activation of free fatty acid receptor 1 by directly rearranging an intracellular loop that interacts with transducers. We validate the predictions of our atomic-level simulations by targeted mutagenesis; specific mutations which disrupt interactions with the intracellular loop convert these agonists into inverse agonists. Further analysis suggests that allosteric ligands could regulate signaling of many other GPCRs via a similar mechanism, offering rich possibilities for precise control of pharmaceutically important targets.

The mu opioid receptor (mu OR) is a target for clinically used analgesics. However, adverse effects, such as respiratory depression and physical dependence, necessitate the development of alternative treatments. Recently we reported a novel strategy to design functionally selective opioids by targeting the sodium binding allosteric site in mu OR with a supraspinally active analgesic named C6guano. Presently, to improve systemic activity of this ligand, we used structure-based design, identifying a new ligand named RO76 where the flexible alkyl linker and polar guanidine guano group is swapped with a benzyl alcohol, and the sodium site is targeted indirectly through waters. A cryoEM structure of RO76 bound to the mu OR-G(i) complex confirmed that RO76 interacts with the sodium site residues through a water molecule, unlike C6guano which engages the sodium site directly. Signaling assays coupled with APEX based proximity labeling show binding in the sodium pocket modulates receptor efficacy and trafficking. In mice, RO76 was systemically active in tail withdrawal assays and showed reduced liabilities compared to those of morphine. In summary, we show that targeting water molecules in the sodium binding pocket may be an avenue to modulate signaling properties of opioids, and which may potentially be extended to other G-protein coupled receptors where this site is conserved.

Generating ligand molecules for specific protein targets, known as structure-based drug design, is a fundamental problem in therapeutics development and biological discovery. Recently, target-aware generative models, especially diffusion models, have shown great promise in modeling protein-ligand interactions and generating candidate drugs. However, existing models primarily focus on learning the chemical distribution of all drug candidates, which lacks effective steerability on the chemical quality of model generations. In this paper, we propose a novel and general alignment framework to align pretrained target diffusion models with preferred functional properties, named AliDiff. AliDiff shifts the target-conditioned chemical distribution towards regions with higher binding affinity and structural rationality, specified by user-defined reward functions, via the preference optimization approach. To avoid the overfitting problem in common preference optimization objectives, we further develop an improved Exact Energy Preference Optimization method to yield an exact and efficient alignment of the diffusion models, and provide the closed-form expression for the converged distribution. Empirical studies on the CrossDocked2020 benchmark show that AliDiff can generate molecules with state-of-the-art binding energies with up to -7.07 Avg. Vina Score, while maintaining strong molecular properties. Code is available at https://github.com/MinkaiXu/AliDiff.

A pervasive challenge in drug design is determining how to expand a ligand-a small molecule that binds to a target biomolecule-in order to improve various properties of the ligand. Adding single chemical groups, known as fragments, is important for lead optimization tasks, and adding multiple fragments is critical for fragment-based drug design. We have developed a comprehensive framework that uses machine learning and three-dimensional protein-ligand structures to address this challenge. Our method, FRAME, iteratively determines where on a ligand to add fragments, selects fragments to add, and predicts the geometry of the added fragments. On a comprehensive benchmark, FRAME consistently improves predicted affinity and selectivity relative to the initial ligand, while generating molecules with more drug-like chemical properties than docking-based methods currently in widespread use. FRAME learns to accurately describe molecular interactions despite being given no prior information on such interactions. The resulting framework for quality molecular hypothesis generation can be easily incorporated into the workflows of medicinal chemists for diverse tasks, including lead optimization, fragment-based drug discovery, and de novo drug design.

MD simulations for "W.A. Burger, V. Pham, Z. Vuckovic, A.S. Powers, J.I. Mobbs, Y. Laloudakis, A. Glukhova, D. Wootten, A.B. Tobin, P.M. Sexton, S.M. Paul, C.C. Felder, R. Radostin, R.O. Dror, A. Christopoulos, C. Valant, and D.M. Thal. Xanomeline displays concomitant orthosteric and allosteric binding modes at the M4 mAChR. Nature Communications, 14(1): 5440 (2023)."

Generative models, especially diffusion models (DMs), have achieved promising results for generating feature-rich geometries and advancing foundational science problems such as molecule design. Inspired by the recent huge success of Stable (latent) Diffusion models, we propose a novel and principled method for 3D molecule generation named Geometric Latent Diffusion Models (GeoLDM). GeoLDM is the first latent DM model for the molecular geometry domain, composed of autoencoders encoding structures into continuous latent codes and DMs operating in the latent space. Our key innovation is that for modeling the 3D molecular geometries, we capture its critical roto-translational equivariance constraints by building a point-structured latent space with both invariant scalars and equivariant tensors. Extensive experiments demonstrate that GeoLDM can consistently achieve better performance on multiple molecule generation benchmarks, with up to 7\% improvement for the valid percentage of large biomolecules. Results also demonstrate GeoLDM's higher capacity for controllable generation thanks to the latent modeling. Code is provided at \url{https://github.com/MinkaiXu/GeoLDM}.

The M 4 muscarinic acetylcholine receptor (M 4 mAChR) has emerged as a drug target of high therapeutic interest due to its expression in regions of the brain involved in the regulation of psychosis, cognition, and addiction. The mAChR agonist, xanomeline, has provided significant improvement in the Positive and Negative Symptom Scale (PANSS) scores in a Phase II clinical trial for the treatment of patients suffering from schizophrenia. Here we report the active state cryo-EM structure of xanomeline bound to the human M 4 mAChR in complex with the heterotrimeric G i1 transducer protein. Unexpectedly, two molecules of xanomeline were found to concomitantly bind to the monomeric M 4 mAChR, with one molecule bound in the orthosteric (acetylcholine-binding) site and a second molecule in an extracellular vestibular allosteric site. Molecular dynamic simulations supports the structural findings, and pharmacological validation confirmed that xanomeline acts as a dual orthosteric and allosteric ligand at the human M 4 mAChR. These findings provide a basis for further understanding xanomeline’s complex pharmacology and highlight the myriad of ways through which clinically relevant ligands can bind to and regulate GPCRs.

Endocannabinoids (eCBs) are endogenous ligands of the cannabinoid receptor 1 (CB1), a G protein-coupled receptor that regulates a number of therapeutically relevant physiological responses. Hence, understanding the structural and functional consequences of eCB-CB1 interactions has important implications for designing effective drugs targeting this receptor. To characterize the molecular details of eCB interaction with CB1, we utilized AMG315, an analog of the eCB anandamide to determine the structure of the AMG315-bound CB1 signaling complex. Compared to previous structures, the ligand binding pocket shows some differences. Using docking, molecular dynamics simulations, and signaling assays we investigated the functional consequences of ligand interactions with the “toggle switch” residues F200 3.36 and W356 6.48 . Further, we show that ligand-TM2 interactions drive changes to residues on the intracellular side of TM2 and are a determinant of efficacy in activating G protein. These intracellular TM2 rearrangements are unique to CB1 and are exploited by a CB1-specific allosteric modulator.

A drug's selectivity for target receptors is essential to its therapeutic utility, but achieving selectivity between similar receptors is challenging. The serendipitous discovery of ligands that stimulate target receptors more strongly than closely related receptors, despite binding with similar affinities, suggests a solution. The molecular mechanism of such 'efficacy-driven selectivity' has remained unclear, however, hindering design of such ligands. Here, using atomic-level simulations, we reveal the structural basis for the efficacy-driven selectivity of a long-studied clinical drug candidate, xanomeline, between closely related muscarinic acetylcholine receptors (mAChRs). Xanomeline's binding mode is similar across mAChRs in their inactive states but differs between mAChRs in their active states, with divergent effects on active-state stability. We validate this mechanism experimentally and use it to design ligands with altered efficacy-driven selectivity. Our results suggest strategies for the rational design of ligands that achieve efficacy-driven selectivity for many pharmaceutically important G-protein-coupled receptors.

Xanomeline (formulated as KarXT) is a selective M1/M4 muscarinic acetylcholine receptor (mAChR) agonist being developed for the treatment of symptoms associated with schizophrenia and Alzheimer's disease (AD). Questions remain regarding the mechanisms of xanomeline's selectivity. Therefore, we specifically investigated molecular determinants of xanomeline binding and functional selectivity using two of the most highly homologous Gi-coupled mAChR subtypes (M2/M4).