Improved chemistry restraints for crystallographic refinement by integrating the Amber force field into Phenix

AQuaRef: Machine learning accelerated quantum refinement of protein structures

Cryo-EM and X-ray crystallography provide crucial experimental data for obtaining atomic-detail models of biomacromolecules. Refining these models relies on library- based stereochemical restraints, which, in addition to being limited to known chemical entities, do not include meaningful noncovalent interactions relying solely on nonbonded repulsions. Quantum mechanical (QM) calculations could alleviate these issues but are too expensive for large molecules. We present a novel AI-enabled Quantum Refinement (AQuaRef) based on AIMNet2 neural network potential mimicking QM at substantially lower computational costs. By refining 41 cryo-EM and 30 X-ray structures, we show that this approach yields atomic models with superior geometric quality compared to standard techniques, while maintaining an equal or better fit to experimental data.

X-ray Crystallography Module in MD Simulation Program Amber 2023. Refining the Models of Protein Crystals

The MD simulation package Amber offers an attractive platform to refine crystallographic structures of proteins: (i) state-of-the-art force fields help to regularize protein coordinates and reconstruct the poorly diffracting elements of the structure, such as flexible loops; (ii) MD simulations restrained by the experimental diffraction data provide an effective strategy to optimize structural models of protein crystals, including explicitly modeled interstitial solvent as well as crystal contacts. Here, we present the new crystallography module xray, released as a part of the Amber 2023 package. This module contains functions to calculate and scale structure factors (including the contributions from bulk solvent), evaluate the maximum-likelihood-type crystallographic potential, and compute its derivative forces. The X-ray functionality of Amber no longer relies on external dependencies so that the full advantage of GPU acceleration can be taken. This makes it possible to refine in a short time hundreds of crystal models, including supercell models comprised of multiple unit cells. The new automated Amber-based refinement procedure leads to an appreciable improvement in Rfree (in some cases, by as much as 0.067) as well as MolProbity scores.

The bad and the good of trends in model building and refinement for sparse-data regions: pernicious forms of overfitting versus good new tools and predictions

Model building and refinement, and the validation of their correctness, are very effective and reliable at local resolutions better than about 2.5 Å for both crystallography and cryo-EM. However, at local resolutions worse than 2.5 Å both the procedures and their validation break down and do not ensure reliably correct models. This is because in the broad density at lower resolution, critical features such as protein backbone carbonyl O atoms are not just less accurate but are not seen at all, and so peptide orientations are frequently wrongly fitted by 90-180°. This puts both backbone and side chains into the wrong local energy minimum, and they are then worsened rather than improved by further refinement into a valid but incorrect rotamer or Ramachandran region. On the positive side, new tools are being developed to locate this type of pernicious error in PDB depositions, such as CaBLAM, EMRinger, Pperp diagnosis of ribose puckers, and peptide flips in PDB-REDO, while interactive modeling in Coot or ISOLDE can help to fix many of them. Another positive trend is that artificial intelligence predictions such as those made by AlphaFold2 contribute additional evidence from large multiple sequence alignments, and in high-confidence parts they provide quite good starting models for loops, termini or whole domains with otherwise ambiguous density.

Chemical features and machine learning assisted predictions of protein-ligand short hydrogen bonds

There are continuous efforts to elucidate the structure and biological functions of short hydrogen bonds (SHBs), whose donor and acceptor heteroatoms reside more than 0.3 Å closer than the sum of their van der Waals radii. In this work, we evaluate 1070 atomic-resolution protein structures and characterize the common chemical features of SHBs formed between the side chains of amino acids and small molecule ligands. We then develop a machine learning assisted prediction of protein-ligand SHBs (MAPSHB-Ligand) model and reveal that the types of amino acids and ligand functional groups as well as the sequence of neighboring residues are essential factors that determine the class of protein-ligand hydrogen bonds. The MAPSHB-Ligand model and its implementation on our web server enable the effective identification of protein-ligand SHBs in proteins, which will facilitate the design of biomolecules and ligands that exploit these close contacts for enhanced functions.

MD simulations of macromolecular crystals: Implications for the analysis of Bragg and diffuse scattering

Some of our most detailed information about structure and dynamics of macromolecules comes from X-ray-diffraction studies in crystalline environments. More than 170,000 atomic models have been deposited in the Protein Data Bank, and the number of observations (typically of intensities of Bragg diffraction peaks) is generally quite large, when compared to other experimental methods. Nevertheless, the general agreement between calculated and observed intensities is far outside the experimental precision, and the majority of scattered photons fall between the sharp Bragg peaks, and are rarely taken into account. This chapter considers how molecular dynamics simulations can be used to explore the connections between microscopic behavior in a crystalline lattice and observed scattering intensities, and point the way to new atomic models that could more faithfully recapitulate Bragg intensities and extract useful information from the diffuse scattering that lies between those peaks.

Chromogranin B (CHGB) is dimorphic and responsible for dominant anion channels delivered to cell surface via regulated secretion

Regulated secretion is conserved in all eukaryotes. In vertebrates granin family proteins function in all key steps of regulated secretion. Phase separation and amyloid-based storage of proteins and small molecules in secretory granules require ion homeostasis to maintain their steady states, and thus need ion conductances in granule membranes. But granular ion channels are still elusive. Here we show that granule exocytosis in neuroendocrine cells delivers to cell surface dominant anion channels, to which chromogranin B (CHGB) is critical. Biochemical fractionation shows that native CHGB distributes nearly equally in soluble and membrane-bound forms, and both reconstitute highly selective anion channels in membrane. Confocal imaging resolves granular membrane components including proton pumps and CHGB in puncta on the cell surface after stimulated exocytosis. High pressure freezing immuno-EM reveals a major fraction of CHGB at granule membranes in rat pancreatic β-cells. A cryo-EM structure of bCHGB dimer of a nominal 3.5 Å resolution delineates a central pore with end openings, physically sufficient for membrane-spanning and large single channel conductance. Together our data support that CHGB-containing (CHGB+) channels are characteristic of regulated secretion, and function in granule ion homeostasis near the plasma membrane or possibly in other intracellular processes.

Chemical Features and Machine Learning Assisted Predictions of Protein-Ligand Short Hydrogen Bonds

There are continuous efforts to elucidate the structure and biological functions of short hydrogen bonds (SHBs), whose donor and acceptor heteroatoms reside more than 0.3 A closer than the sum of their van der Waals radii. In this work, we evaluate 1070 atomic-resolution protein structures and characterize the common chemical features of SHBs formed between the side chains of amino acids and small molecule ligands. We then develop a machine learning assisted prediction of protein-ligand SHBs (MAPSHB-Ligand) model and reveal that the types of amino acids and ligand functional groups as well as the sequence of neighboring residues are essential factors that determine the class of protein-ligand hydrogen bonds. The MAPSHB-Ligand model and its implementation on our web server enable the effective identification of protein-ligand SHBs in proteins, which will facilitate the design of biomolecules and ligands that exploit these close contacts for enhanced functions.

Comparison of Empirical Zn2+ Models in Protein–DNA Complexes

Zinc ions are the second most abundant ions found in humans. Their role in proteins can be merely structural but also catalytic, owing to their transition metal character. Modelling their geometric–coordination versatility by empirical force fields is, thus, a challenging task. In this work, we evaluated three popular models, specifically designed to represent zinc ions with regard to their capability of preserving structural integrity. To this end, we performed molecular dynamics simulations of two zinc-containing protein–DNA complexes, which differed in their zinc coordination, i.e., four cysteines or two cysteines and two histidines. The most flexible non-bonded 12-6-4 Lennard–Jones-type model shows a preference for six-fold coordination of the Zn2+-ions in contradiction to the crystal structure. The cationic dummy atom model favours tetrahedral geometry, whereas the bonded extended zinc AMBER force field model, by construction, best preserves the initial geometry of a regular or slightly distorted tetrahedron. Our data renders the extended zinc AMBER force field the best model for structural zinc ions in a given geometry. In more complicated cases, though, more flexible models may be advantageous.

In situ ligand restraints from quantum-mechanical methods

In macromolecular crystallographic structure refinement, ligands present challenges for the generation of geometric restraints due to their large chemical variability, their possible novel nature and their specific interaction with the binding pocket of the protein. Quantum-mechanical approaches are useful for providing accurate ligand geometries, but can be plagued by the number of minima in flexible molecules. In an effort to avoid these issues, the Quantum Mechanical Restraints (QMR) procedure optimizes the ligand geometry in situ, thus accounting for the influence of the macromolecule on the local energy minima of the ligand. The optimized ligand geometry is used to generate target values for geometric restraints during the crystallographic refinement. As demonstrated using a sample of >2330 ligand instances in >1700 protein-ligand models, QMR restraints generally result in lower deviations from the target stereochemistry compared with conventionally generated restraints. In particular, the QMR approach provides accurate torsion restraints for ligands and other entities.

Cryo-EM structure of ex vivo fibrils associated with extreme AA amyloidosis prevalence in a cat shelter

AA amyloidosis is a systemic disease characterized by deposition of misfolded serum amyloid A protein (SAA) into cross-β amyloid in multiple organs in humans and animals. AA amyloidosis occurs at high SAA serum levels during chronic inflammation. Prion-like transmission was reported as possible cause of extreme AA amyloidosis prevalence in captive animals, e.g. 70% in cheetah and 57-73% in domestic short hair (DSH) cats kept in zoos and shelters, respectively. Herein, we present the 3.3 Å cryo-EM structure of AA amyloid extracted post-mortem from the kidney of a DSH cat with renal failure, deceased in a shelter with extreme disease prevalence. The structure reveals a cross-β architecture assembled from two 76-residue long proto-filaments. Despite >70% sequence homology to mouse and human SAA, the cat SAA variant adopts a distinct amyloid fold. Inclusion of an eight-residue insert unique to feline SAA contributes to increased amyloid stability. The presented feline AA amyloid structure is fully compatible with the 99% identical amino acid sequence of amyloid fragments of captive cheetah.

The temperature-dependent conformational ensemble of SARS-CoV-2 main protease (Mpro)

The COVID-19 pandemic, instigated by the SARS-CoV-2 coronavirus, continues to plague the globe. The SARS-CoV-2 main protease, or Mpro, is a promising target for the development of novel antiviral therapeutics. Previous X-ray crystal structures of Mpro were obtained at cryogenic tem-per-ature or room tem-per-ature only. Here we report a series of high-resolution crystal structures of unliganded Mpro across multiple tem-per-atures from cryogenic to physiological, and another at high humidity. We inter-rogate these data sets with parsimonious multiconformer models, multi-copy ensemble models, and isomorphous difference density maps. Our analysis reveals a perturbation-dependent conformational landscape for Mpro, including a mobile zinc ion inter-leaved between the catalytic dyad, mercurial conformational heterogeneity at various sites including a key substrate-binding loop, and a far-reaching intra-molecular network bridging the active site and dimer inter-face. Our results may inspire new strategies for antiviral drug development to aid preparation for future coronavirus pandemics.

Uptake of N2O5 by aqueous aerosol unveiled using chemically accurate many-body potentials

The reactive uptake of N₂O₅ to aqueous aerosol is a major loss channel for nitrogen oxides in the troposphere. Despite its importance, a quantitative picture of the uptake mechanism is missing. Here we use molecular dynamics simulations with a data-driven many-body model of coupled-cluster accuracy to quantify thermodynamics and kinetics of solvation and adsorption of N₂O₅ in water. The free energy profile highlights that N₂O₅ is selectively adsorbed to the liquid–vapor interface and weakly solvated. Accommodation into bulk water occurs slowly, competing with evaporation upon adsorption from gas phase. Leveraging the quantitative accuracy of the model, we parameterize and solve a reaction–diffusion equation to determine hydrolysis rates consistent with experimental observations. We find a short reaction–diffusion length, indicating that the uptake is dominated by interfacial features. The parameters deduced here, including solubility, accommodation coefficient, and hydrolysis rate, afford a foundation for which to consider the reactive loss of N₂O₅ in more complex solutions.

The reactive uptake of N₂O₅ to aqueous aerosol is a major loss channel for nitrogen oxides in the troposphere. Here authors report a theoretical investigation on the N₂O₅ uptake into aqueous aerosol and determine the hydrolysis rates by numerically solving a molecularly detailed reaction–diffusion equation.

Effective prediction of short hydrogen bonds in proteins via machine learning method

Short hydrogen bonds (SHBs), whose donor and acceptor heteroatoms lie within 2.7 Å, exhibit prominent quantum mechanical characters and are connected to a wide range of essential biomolecular processes. However, exact determination of the geometry and functional roles of SHBs requires a protein to be at atomic resolution. In this work, we analyze 1260 high-resolution peptide and protein structures from the Protein Data Bank and develop a boosting based machine learning model to predict the formation of SHBs between amino acids. This model, which we name as machine learning assisted prediction of short hydrogen bonds (MAPSHB), takes into account 21 structural, chemical and sequence features and their interaction effects and effectively categorizes each hydrogen bond in a protein to a short or normal hydrogen bond. The MAPSHB model reveals that the type of the donor amino acid plays a major role in determining the class of a hydrogen bond and that the side chain Tyr-Asp pair demonstrates a significant probability of forming a SHB. Combining electronic structure calculations and energy decomposition analysis, we elucidate how the interplay of competing intermolecular interactions stabilizes the Tyr-Asp SHBs more than other commonly observed combinations of amino acid side chains. The MAPSHB model, which is freely available on our web server, allows one to accurately and efficiently predict the presence of SHBs given a protein structure with moderate or low resolution and will facilitate the experimental and computational refinement of protein structures.

Integral equation models for solvent in macromolecular crystals

The solvent can occupy up to ∼70% of macromolecular crystals, and hence, having models that predict solvent distributions in periodic systems could improve the interpretation of crystallographic data. Yet, there are few implicit solvent models applicable to periodic solutes, and crystallographic structures are commonly solved assuming a flat solvent model. Here, we present a newly developed periodic version of the 3D-reference interaction site model (RISM) integral equation method that is able to solve efficiently and describe accurately water and ion distributions in periodic systems; the code can compute accurate gradients that can be used in minimizations or molecular dynamics simulations. The new method includes an extension of the Ornstein–Zernike equation needed to yield charge neutrality for charged solutes, which requires an additional contribution to the excess chemical potential that has not been previously identified; this is an important consideration for nucleic acids or any other charged system where most or all the counter- and co-ions are part of the “disordered” solvent. We present several calculations of proteins, RNAs, and small molecule crystals to show that x-ray scattering intensities and the solvent structure predicted by the periodic 3D-RISM solvent model are in closer agreement with the experiment than are intensities computed using the default flat solvent model in the refmac5 or phenix refinement programs, with the greatest improvement in the 2 to 4 Å range. Prospects for incorporating integral equation models into crystallographic refinement are discussed.

Modeling a unit cell: crystallographic refinement procedure using the biomolecular MD simulation platform Amber

A procedure has been developed for the refinement of crystallographic protein structures based on the biomolecular simulation program Amber. The procedure constructs a model representing a crystal unit cell, which generally contains multiple protein molecules and is fully hydrated with TIP3P water. Periodic boundary conditions are applied to the cell in order to emulate the crystal lattice. The refinement is conducted in the form of a specially designed short molecular-dynamics run controlled by the Amber ff14SB force field and the maximum-likelihood potential that encodes the structure-factor-based restraints. The new Amber-based refinement procedure has been tested on a set of 84 protein structures. In most cases, the new procedure led to appreciably lower R free values compared with those reported in the original PDB depositions or obtained by means of the industry-standard phenix.refine program. In particular, the new method has the edge in refining low-accuracy scrambled models. It has also been successful in refining a number of molecular-replacement models, including one with an r.m.s.d. of 2.15 Å. In addition, Amber-refined structures consistently show superior MolProbity scores. The new approach offers a highly realistic representation of protein-protein interactions in the crystal, as well as of protein-water interactions. It also offers a realistic representation of protein crystal dynamics (akin to ensemble-refinement schemes). Importantly, the method fully utilizes the information from the available diffraction data, while relying on state-of-the-art molecular-dynamics modeling to assist with those elements of the structure that do not diffract well (for example mobile loops or side chains). Finally, it should be noted that the protocol employs no tunable parameters, and the calculations can be conducted in a matter of several hours on desktop computers equipped with graphical processing units or using a designated web service.

Refinement of RNA Structures Using Amber Force Fields

Atomic models for nucleic acids derived from X-ray diffraction data at low resolution provide much useful information, but the observed scattering intensities can be fit with models that can differ in structural detail. Tradtional geometric restraints favor models that have bond length and angle terms derived from small molecule crystal structures. Here we explore replacing these restraints with energy gradients derived from force fields, including recently developed integral equation models to account for the effects of water molecules and ions that are not part of the explicit model. We compare conventional and force-field based refinements for 22 RNA crystals, ranging in resolution from 1.1 to 3.6 Å. As expected, it can be important to account for solvent screening of charge–charge interactions, especially in the crowded environment of a nucleic acid crystal. The newly refined models can show improvements in torsion angles and hydrogen-bonding interactions, and can significantly reduce unfavorable atomic clashes, while maintaining or improving agreement with observed scattering intensities.

The temperature-dependent conformational ensemble of SARS-CoV-2 main protease (Mpro)

The COVID-19 pandemic, instigated by the SARS-CoV-2 coronavirus, continues to plague the globe. The SARS-CoV-2 main protease, or Mpro, is a promising target for development of novel antiviral therapeutics. Previous X-ray crystal structures of Mpro were obtained at cryogenic temperature or room temperature only. Here we report a series of high-resolution crystal structures of unliganded Mpro across multiple temperatures from cryogenic to physiological, and another at high humidity. We interrogate these datasets with parsimonious multiconformer models, multi-copy ensemble models, and isomorphous difference density maps. Our analysis reveals a temperature-dependent conformational landscape for Mpro, including mobile solvent interleaved between the catalytic dyad, mercurial conformational heterogeneity in a key substrate-binding loop, and a far-reaching intramolecular network bridging the active site and dimer interface. Our results may inspire new strategies for antiviral drug development to counter-punch COVID-19 and combat future coronavirus pandemics.

Improving sampling of crystallographic disorder in ensemble refinement

Ensemble refinement, the application of molecular dynamics to crystallographic refinement, explicitly models the disorder inherent in macromolecular structures. These ensemble models have been shown to produce more accurate structures than traditional single-model structures. However, suboptimal sampling of the molecular-dynamics simulation and modelling of crystallographic disorder has limited the utility of the method, and can lead to unphysical and strained models. Here, two improvements to the ensemble refinement method implemented within Phenix are presented: DEN restraints, which guide the local sampling of conformations and allow a more robust exploration of local conformational landscapes, and ECHT disorder models, which allow the selection of more physically meaningful and effective disorder models for parameterizing the continuous disorder components within a crystal. These improvements lead to more consistent and physically interpretable simulations of macromolecules in crystals, and allow structural heterogeneity and disorder to be systematically explored on different scales. The new approach is demonstrated on several case studies and the SARS-CoV-2 main protease, and demonstrates how the choice of disorder model affects the type of disorder that is sampled by the restrained molecular-dynamics simulation.

Repurposing clinically approved drugs for COVID-19 treatment targeting SARS-CoV-2 papain-like protease

COVID-19 is a disease caused by SARS-CoV-2, which has led to more than 4 million deaths worldwide. As a result, there is a worldwide effort to develop specific drugs for targeting COVID-19. Papain-like protease (PLpro) is an attractive drug target because it has multiple essential functions involved in processing viral proteins, including viral genome replication and removal of post-translational ubiquitination modifications. Here, we established two assays for screening PLpro inhibitors according to protease and anti-ISGylation activities, respectively. Application of the two screening techniques to the library of clinically approved drugs led to the discovery of tanshinone IIA sulfonate sodium and chloroxine with their IC50 values of lower than 10 μM. These two compounds were found to directly interact with PLpro and their molecular mechanisms of binding were illustrated by docking and molecular dynamics simulations. The results highlight the usefulness of the two developed screening techniques for locating PLpro inhibitors.

Macromolecular refinement of X-ray and cryoelectron microscopy structures with Phenix/OPLS3e for improved structure and ligand quality

With the advent of the resolution revolution in cryoelectron microscopy (cryo-EM), low-resolution refinement is common, and likewise increases the need for a reliable force field. Here, we report on the incorporation of the OPLS3e force field with the VSGB2.1 solvation model in the structure determination package Phenix. Our results show significantly improved structure quality and reduced ligand strain at lower resolution for X-ray refinement. For refinement of cryo-EM-based structures, we find comparable quality structures, goodness-of-fit, and reduced ligand strain. We also show how structure quality and ligand strain are related to the map-model cross-correlation as a function of data weight, and how that can detect overfitting. Signs of overfitting are found in over half of our cryo-EM dataset, which can be remedied by a re-refinement at a lower data weight. Finally, a start-to-end script for refining structures with Phenix/OPLS3e is available in the Schrödinger 2020-3 distribution.

Adaptive Cartesian and torsional restraints for interactive model rebuilding

When building atomic models into weak and/or low-resolution density, a common strategy is to restrain their conformation to that of a higher resolution model of the same or similar sequence. When doing so, it is important to avoid over-restraining to the reference model in the face of disagreement with the experimental data. The most common strategy for this is the use of `top-out' potentials. These act like simple harmonic restraints within a defined range, but gradually weaken when the deviation between the model and reference grows beyond that range. In each current implementation the rate at which the potential flattens at large deviations follows a fixed form, although the form chosen varies among implementations. A restraint potential with a tuneable rate of flattening would provide greater flexibility to encode the confidence in any given restraint. Here, two new such potentials are described: a Cartesian distance restraint derived from a recent generalization of common loss functions and a periodic torsion restraint based on a renormalization of the von Mises distribution. Further, their implementation as user-adjustable/switchable restraints in ISOLDE is described and their use in some real-world examples is demonstrated.

Adeno-Associated Virus (AAV-DJ)-Cryo-EM Structure at 1.56 Å Resolution

Adeno-associated virus is the leading viral vector for gene therapy. AAV-DJ is a recombinant variant developed for tropism to the liver. The AAV-DJ structure has been determined to 1.56 Å resolution through cryo-electron microscopy (cryo-EM). Only apoferritin is reported in preprints at 1.6 Å or higher resolution, and AAV-DJ nearly matches the highest resolutions ever attained through X-ray diffraction of virus crystals. However, cryo-EM has the advantage that most of the hydrogens are clear, improving the accuracy of atomic refinement, and removing ambiguity in hydrogen bond identification. Outside of secondary structures where hydrogen bonding was predictable a priori, the networks of hydrogen bonds coming from direct observation of hydrogens and acceptor atoms are quite different from those inferred even at 2.8 Å resolution. The implications for understanding viral assembly mean that cryo-EM will likely become the favored approach for high resolution structural virology.

Advances in methods for atomic resolution macromolecular structure determination

Recent technical advances have dramatically increased the power and scope of structural biology. New developments in high-resolution cryo-electron microscopy, serial X-ray crystallography, and electron diffraction have been especially transformative. Here we highlight some of the latest advances and current challenges at the frontiers of atomic resolution methods for elucidating the structures and dynamical properties of macromolecules and their complexes.

What are the current limits on determination of protonation state using neutron macromolecular crystallography?

The rate of deposition of models determined by neutron diffraction, or a hybrid approach that combines X-ray and neutron diffraction, has increased in recent years. The benefit of neutron diffraction is that hydrogen atom (H) positions are detectable, allowing for the determination of protonation state and water molecule orientation. This study analyses all neutron models deposited in the Protein Data Bank to date, focusing on protonation state and properties of H (or deuterium, D) atoms as well as the details of water molecules. In particular, clashes and hydrogen bonds involving H or D atoms are investigated. As water molecules are typically the least reproducible part of a structural model, their positions in neutron models were compared to those in homologous high-resolution X-ray structures. For models determined by joint refinement against X-ray and neutron data, the water structure comparison was also carried out for models re-refined against the X-ray data alone. The homologues have generally fewer conserved water molecules where X-ray only was used and the positions of equivalent waters vary more than in the case of the hybrid X-ray model. As neutron diffraction data are generally less complete than X-ray data, the influence of neutron data completeness on nuclear density maps was also analyzed. We observe and discuss systematic map quality deterioration as result of data incompleteness.