Accelerating crystal structure determination with iterative AlphaFold prediction

Hidden Structural States of Proteins Revealed by Conformer Selection with AlphaFold-NMR

Recent advances in molecular modeling using deep learning can revolutionize our understanding of dynamic protein structures. NMR is particularly well-suited for determining dynamic features of biomolecular structures. The conventional process for determining biomolecular structures from experimental NMR data involves its representation as conformation-dependent restraints, followed by generation of structural models guided by these spatial restraints. Here we describe an alternative approach: generating a distribution of realistic protein conformational models using artificial intelligence-(AI-) based methods and then selecting the sets of conformers that best explain the experimental data. We applied this conformational selection approach to redetermine the solution NMR structure of the enzyme Gaussia luciferase. First, we generated a diverse set of conformer models using AlphaFold2 (AF2) with an enhanced sampling protocol. The models that best-fit NOESY and chemical shift data were then selected with a Bayesian scoring metric. The resulting models include features of both the published NMR structure and the standard AF2 model generated without enhanced sampling. This "AlphaFold-NMR" protocol also generated an alternative "open" conformational state that fits nearly as well to the overall NMR data but accounts for some NOESY data that is not consistent with first "closed" conformational state; while other NOESY data consistent with this second state are not consistent with the first conformational state. The structure of this "open" structural state differs from that of the "closed" state primarily by the position of a thumb-shaped loop between α-helices H5 and H6, revealing a cryptic surface pocket. These alternative conformational states of Gluc are supported by "double recall" analysis of NOESY data and AF2 models. Additional structural states are also indicated by backbone chemical shift data indicating partially-disordered conformations for the C-terminal segment. Considered as a multistate ensemble, these multiple states of Gluc together fit the NOESY and chemical shift data better than the "restraint-based" NMR structure and provide novel insights into its structure-dynamic-function relationships. This study demonstrates the potential of AI-based modeling with enhanced sampling to generate conformational ensembles followed by conformer selection with experimental data as an alternative to conventional restraint satisfaction protocols for protein NMR structure determination.

The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins

Artificial intelligence-driven advances in protein structure prediction in recent years have raised the question: has the protein structure-prediction problem been solved? Here, with a focus on nonglobular proteins, we highlight the many strengths and potential weaknesses of DeepMind's AlphaFold2 in the context of its biological and therapeutic applications. We summarize the subtleties associated with evaluation of AlphaFold2 model quality and reliability using the predicted local distance difference test (pLDDT) and predicted aligned error (PAE) values. We highlight various classes of proteins that AlphaFold2 can be applied to and the caveats involved. Concrete examples of how AlphaFold2 models can be integrated with experimental data in the form of small-angle X-ray scattering (SAXS), solution NMR, cryo-electron microscopy (cryo-EM) and X-ray diffraction are discussed. Finally, we highlight the need to move beyond structure prediction of rigid, static structural snapshots toward conformational ensembles and alternate biologically relevant states. The overarching theme is that careful consideration is due when using AlphaFold2-generated models to generate testable hypotheses and structural models, rather than treating predicted models as de facto ground truth structures.

Structure of a phosphodiesterase from Streptomyces sanglieri with a novel C-terminal domain

A glycerophosphoethanolamine ethanolaminephosphodiesterase (GPE-EP) from Streptomyces sanglieri hydrolyzes glycerophosphoethanolamine to phosphoethanolamine and glycerol. The structure of GPE-EP was determined by the molecular replacement method using a search model generated with AlphaFold2. This structure includes the entire length of the mature protein and it is composed of an N-terminal domain and a novel C-terminal domain connected to a flexible linker. The N-terminal domain is the catalytic domain containing calcium ions at the catalytic site. Coordination bonds were observed between five amino acid residues and glycerol. Although the function of the C-terminal domain is currently unknown, inter-domain interactions between the N- and C-terminal domains may contribute to its relatively high thermostability.

AlphaFold-assisted structure determination of a bacterial protein of unknown function using X-ray and electron crystallography

Macromolecular crystallography generally requires the recovery of missing phase information from diffraction data to reconstruct an electron-density map of the crystallized molecule. Most recent structures have been solved using molecular replacement as a phasing method, requiring an a priori structure that is closely related to the target protein to serve as a search model; when no such search model exists, molecular replacement is not possible. New advances in computational machine-learning methods, however, have resulted in major advances in protein structure predictions from sequence information. Methods that generate predicted structural models of sufficient accuracy provide a powerful approach to molecular replacement. Taking advantage of these advances, AlphaFold predictions were applied to enable structure determination of a bacterial protein of unknown function (UniProtKB Q63NT7, NCBI locus BPSS0212) based on diffraction data that had evaded phasing attempts using MIR and anomalous scattering methods. Using both X-ray and micro-electron (microED) diffraction data, it was possible to solve the structure of the main fragment of the protein using a predicted model of that domain as a starting point. The use of predicted structural models importantly expands the promise of electron diffraction, where structure determination relies critically on molecular replacement.

ZP2 cleavage blocks polyspermy by modulating the architecture of the egg coat

Following the fertilization of an egg by a single sperm, the egg coat or zona pellucida (ZP) hardens and polyspermy is irreversibly blocked. These events are associated with the cleavage of the N-terminal region (NTR) of glycoprotein ZP2, a major subunit of ZP filaments. ZP2 processing is thought to inactivate sperm binding to the ZP, but its molecular consequences and connection with ZP hardening are unknown. Biochemical and structural studies show that cleavage of ZP2 triggers its oligomerization. Moreover, the structure of a native vertebrate egg coat filament, combined with AlphaFold predictions of human ZP polymers, reveals that two protofilaments consisting of type I (ZP3) and type II (ZP1/ZP2/ZP4) components interlock into a left-handed double helix from which the NTRs of type II subunits protrude. Together, these data suggest that oligomerization of cleaved ZP2 NTRs extensively cross-links ZP filaments, rigidifying the egg coat and making it physically impenetrable to sperm.

Systematic enhancement of protein crystallization efficiency by bulk lysine-to-arginine (KR) substitution

Structural genomics consortia established that protein crystallization is the primary obstacle to structure determination using x-ray crystallography. We previously demonstrated that crystallization propensity is systematically related to primary sequence, and we subsequently performed computational analyses showing that arginine is the most overrepresented amino acid in crystal-packing interfaces in the Protein Data Bank. Given the similar physicochemical characteristics of arginine and lysine, we hypothesized that multiple lysine-to-arginine (KR) substitutions should improve crystallization. To test this hypothesis, we developed software that ranks lysine sites in a target protein based on the redundancy-corrected KR substitution frequency in homologs. This software can be run interactively on the worldwide web at https://www.pxengineering.org/. We demonstrate that three unrelated single-domain proteins can tolerate 5-11 KR substitutions with at most minor destabilization, and, for two of these three proteins, the construct with the largest number of KR substitutions exhibits significantly enhanced crystallization propensity. This approach rapidly produced a 1.9 Å crystal structure of a human protein domain refractory to crystallization with its native sequence. Structures from Bulk KR-substituted domains show the engineered arginine residues frequently make hydrogen-bonds across crystal-packing interfaces. We thus demonstrate that Bulk KR substitution represents a rational and efficient method for probabilistic engineering of protein surface properties to improve crystallization.

Community recommendations on cryoEM data archiving and validation

In January 2020, a workshop was held at EMBL-EBI (Hinxton, UK) to discuss data requirements for the deposition and validation of cryoEM structures, with a focus on single-particle analysis. The meeting was attended by 47 experts in data processing, model building and refinement, validation, and archiving of such structures. This report describes the workshop's motivation and history, the topics discussed, and the resulting consensus recommendations. Some challenges for future methods-development efforts in this area are also highlighted, as is the implementation to date of some of the recommendations.

Predictive modeling and cryo-EM: A synergistic approach to modeling macromolecular structure

Over the last 15 years, structural biology has seen unprecedented development and improvement in two areas: electron cryo-microscopy (cryo-EM) and predictive modeling. Once relegated to low resolutions, single-particle cryo-EM is now capable of achieving near-atomic resolutions of a wide variety of macromolecular complexes. Ushered in by AlphaFold, machine learning has powered the current generation of predictive modeling tools, which can accurately and reliably predict models for proteins and some complexes directly from the sequence alone. Although they offer new opportunities individually, there is an inherent synergy between these techniques, allowing for the construction of large, complex macromolecular models. Here, we give a brief overview of these approaches in addition to illustrating works that combine these techniques for model building. These examples provide insight into model building, assessment, and limitations when integrating predictive modeling with cryo-EM density maps. Together, these approaches offer the potential to greatly accelerate the generation of macromolecular structural insights, particularly when coupled with experimental data.

AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination

Artificial intelligence-based protein structure prediction methods such as AlphaFold have revolutionized structural biology. The accuracies of these predictions vary, however, and they do not take into account ligands, covalent modifications or other environmental factors. Here, we evaluate how well AlphaFold predictions can be expected to describe the structure of a protein by comparing predictions directly with experimental crystallographic maps. In many cases, AlphaFold predictions matched experimental maps remarkably closely. In other cases, even very high-confidence predictions differed from experimental maps on a global scale through distortion and domain orientation, and on a local scale in backbone and side-chain conformation. We suggest considering AlphaFold predictions as exceptionally useful hypotheses. We further suggest that it is important to consider the confidence in prediction when interpreting AlphaFold predictions and to carry out experimental structure determination to verify structural details, particularly those that involve interactions not included in the prediction.

Enzymes in 3D: Synthesis, remodelling, and hydrolysis of cell wall (1,3;1,4)-β-glucans

Recent breakthroughs in structural biology have provided valuable new insights into enzymes involved in plant cell wall metabolism. More specifically, the molecular mechanism of synthesis of (1,3;1,4)-β-glucans, which are widespread in cell walls of commercially important cereals and grasses, has been the topic of debate and intense research activity for decades. However, an inability to purify these integral membrane enzymes or apply transgenic approaches without interpretative problems associated with pleiotropic effects has presented barriers to attempts to define their synthetic mechanisms. Following the demonstration that some members of the CslF sub-family of GT2 family enzymes mediate (1,3;1,4)-β-glucan synthesis, the expression of the corresponding genes in a heterologous system that is free of background complications has now been achieved. Biochemical analyses of the (1,3;1,4)-β-glucan synthesized in vitro, combined with 3-dimensional (3D) cryogenic-electron microscopy and AlphaFold protein structure predictions, have demonstrated how a single CslF6 enzyme, without exogenous primers, can incorporate both (1,3)- and (1,4)-β-linkages into the nascent polysaccharide chain. Similarly, 3D structures of xyloglucan endo-transglycosylases and (1,3;1,4)-β-glucan endo- and exohydrolases have allowed the mechanisms of (1,3;1,4)-β-glucan modification and degradation to be defined. X-ray crystallography and multi-scale modeling of a broad specificity GH3 β-glucan exohydrolase recently revealed a previously unknown and remarkable molecular mechanism with reactant trajectories through which a polysaccharide exohydrolase can act with a processive action pattern. The availability of high-quality protein 3D structural predictions should prove invaluable for defining structures, dynamics, and functions of other enzymes involved in plant cell wall metabolism in the immediate future.

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.

Crystal structure of adenylosuccinate lyase from the thermophilic bacterium Thermus thermophilus HB8

Adenylosuccinate lyase (PurB) catalyzes two distinct reactions in the purine nucleotide biosynthetic pathway using the same active site. The ability to recognize two different sets of substrates is of structural and evolutionary interest. In the present study, the crystal structure of PurB from the thermophilic bacterium Thermus thermophilus HB8 (TtPurB) was determined at a resolution of 2.38 Å by molecular replacement using a structure predicted by AlphaFold2 as a template. The asymmetric unit of the TtPurB crystal contained two TtPurB molecules, and some regions were disordered in the crystal structure. The disordered regions were the substrate-binding site and domain 3. TtPurB forms a homotetramer and the monomer is composed of three domains (domains 1, 2 and 3), which is a typical structure for the aspartase/fumarase superfamily. Molecular dynamics simulations with and without substrate/product were performed using a full-length model of TtPurB which was obtained before deletion of the disordered regions. The substrates and products were bound to the model structures during the MD simulations. The fluctuations of amino-acid residues were greater in the disordered regions and became smaller upon the binding of substrate or product. These results demonstrate that the full-length model obtained using AlphaFold2 can be used to generate the coordinates of disordered regions within the crystal structure.

A structural classification of the variant surface glycoproteins of the African trypanosome

Long-term immune evasion by the African trypanosome is achieved through repetitive cycles of surface protein replacement with antigenically distinct versions of the dense Variant Surface Glycoprotein (VSG) coat. Thousands of VSG genes and pseudo-genes exist in the parasite genome that, together with genetic recombination mechanisms, allow for essentially unlimited immune escape from the adaptive immune system of the host. The diversity space of the "VSGnome" at the protein level was thought to be limited to a few related folds whose structures were determined more than 30 years ago. However, recent progress has shown that the VSGs possess significantly more architectural variation than had been appreciated. Here we combine experimental X-ray crystallography (presenting structures of N-terminal domains of coat proteins VSG11, VSG21, VSG545, VSG558, and VSG615) with deep-learning prediction using Alphafold to produce models of hundreds of VSG proteins. We classify the VSGnome into groups based on protein architecture and oligomerization state, contextualize recent bioinformatics clustering schemes, and extensively map VSG-diversity space. We demonstrate that in addition to the structural variability and post-translational modifications observed thus far, VSGs are also characterized by variations in oligomerization state and possess inherent flexibility and alternative conformations, lending additional variability to what is exposed to the immune system. Finally, these additional experimental structures and the hundreds of Alphafold predictions confirm that the molecular surfaces of the VSGs remain distinct from variant to variant, supporting the hypothesis that protein surface diversity is central to the process of antigenic variation used by this organism during infection.

Predicted models and CCP4

In late 2020, the results of CASP14, the 14th event in a series of competitions to assess the latest developments in computational protein structure-prediction methodology, revealed the giant leap forward that had been made by Google's Deepmind in tackling the prediction problem. The level of accuracy in their predictions was the first instance of a competitor achieving a global distance test score of better than 90 across all categories of difficulty. This achievement represents both a challenge and an opportunity for the field of experimental structural biology. For structure determination by macromolecular X-ray crystallography, access to highly accurate structure predictions is of great benefit, particularly when it comes to solving the phase problem. Here, details of new utilities and enhanced applications in the CCP4 suite, designed to allow users to exploit predicted models in determining macromolecular structures from X-ray diffraction data, are presented. The focus is mainly on applications that can be used to solve the phase problem through molecular replacement.

Facing the phase problem

The marvel of X-ray crystallography is the beauty and precision of the atomic structures deduced from diffraction patterns. Since these patterns record only amplitudes, phases for the diffracted waves must also be evaluated for systematic structure determination. Thus, we have the phase problem as a central complication, both intellectually for the field and practically so for many analyses. Here, I discuss how we - myself, my laboratory and the diffraction community - have faced the phase problem, considering the evolution of methods for phase evaluation as structural biology developed to the present day. During the explosive growth of macromolecular crystallography, practice in diffraction analysis evolved from a universal reliance on isomorphous replacement to the eventual domination of anomalous diffraction for de novo structure determination. As the Protein Data Bank (PDB) grew and familial relationships among proteins became clear, molecular replacement overtook all other phasing methods; however, experimental phasing remained essential for molecules without obvious precedents, with multi- and single-wavelength anomalous diffraction (MAD and SAD) predominating. While the mathematics-based direct methods had proved to be inadequate for typical macromolecules, they returned to crack substantial selenium substructures in SAD analyses of selenomethionyl proteins. Native SAD, exploiting the intrinsic S and P atoms of biomolecules, has become routine. Selenomethionyl SAD and MAD were the mainstays of structural genomics efforts to populate the PDB with novel proteins. A recent dividend has been paid in the success of PDB-trained artificial intelligence approaches for protein structure prediction. Currently, molecular replacement with AlphaFold models often obviates the need for experimental phase evaluation. For multiple reasons, we are now unfazed by the phase problem. Cryo-EM analysis is an attractive alternative to crystallography for many applications faced by today's structural biologists. It simply finesses the phase problem; however, the principles and procedures of diffraction analysis remain pertinent and are adopted in single-particle cryo-EM studies of biomolecules.

Visualizing everything, everywhere, all at once: Cryo-EM and the new field of structureomics

Twenty years ago, the release of the first draft of the human genome sequence instigated a paradigm shift in genomics and molecular biology. Arguably, structural biology is entering an analogous era, with availability of an experimentally determined or predicted molecular model for almost every protein-coding gene from many genomes-producing a reference "structureome". Structural predictions require experimental validation and not all proteins conform to a single structure, making any reference structureome necessarily incomplete. Despite these limitations, a reference structureome can be used to characterize cell state in more detail than by quantifying sequence or expression levels alone. Cryogenic electron microscopy (cryo-EM) is a method that can generate atomic resolution views of molecules and cells frozen in place. In this perspective I consider how emerging cryo-EM methods are contributing to the new field of structureomics.

The current role and evolution of X-ray crystallography in drug discovery and development

INTRODUCTION

Macromolecular X-ray crystallography and cryo-EM are currently the primary techniques used to determine the three-dimensional structures of proteins, nucleic acids, and viruses. Structural information has been critical to drug discovery and structural bioinformatics. The integration of artificial intelligence (AI) into X-ray crystallography has shown great promise in automating and accelerating the analysis of complex structural data, further improving the efficiency and accuracy of structure determination.

AREAS COVERED

This review explores the relationship between X-ray crystallography and other modern structural determination methods. It examines the integration of data acquired from diverse biochemical and biophysical techniques with those derived from structural biology. Additionally, the paper offers insights into the influence of AI on X-ray crystallography, emphasizing how integrating AI with experimental approaches can revolutionize our comprehension of biological processes and interactions.

EXPERT OPINION

Investing in science is crucially emphasized due to its significant role in drug discovery and advancements in healthcare. X-ray crystallography remains an essential source of structural biology data for drug discovery. Recent advances in biochemical, spectroscopic, and bioinformatic methods, along with the integration of AI techniques, hold the potential to revolutionize drug discovery when effectively combined with robust data management practices.