Name: modeling ligands into maps derived from electron cryomicroscopy
Uploaded: 2024-07-19T13:00:00.000+00:00
Duration: 9 min 30 s
Description: Deciphering the protein-ligand interactions in a macromolecular complex is crucial for understanding the molecular mechanism, underlying biological ...

SJ is a recipient of the PhD studentship from DAE-TIFR, and the funding is acknowledged. KRV acknowledges DBT B-Life grant DBT/PR12422/MED/31/287/2014 and the support of the Department of Atomic Energy, Government of India, under Project Identification No. RTI4006.

Modeling Small Molecule Ligands in CryoEM Maps of Macromolecules

The authors have nothing to disclose.

The improvements in microscope hardware and software have resulted in an increase in the number of cryoEM structures in recent years. Although the highest resolution achieved at the moment in single particle cryoEM is 1.2 &#197;57,58,59, the majority of the structures are being determined around 3-4 &#197; resolution. Modeling ligands in medium to low-resolution maps can be tricky and often fraught with ambiguity. Given the widespread use of cryoEM in both academia and the pharma industry for translational research and drug discovery, ensuring that ligands are modeled correctly and without ambiguity is essential. Thus, it is prudent to quantify the resolvability of ligand atoms by calculating Q-scores60, which is now available in the EMDB as a metric to evaluate the quality of the maps and the fit of the model as well as in Chimera.
In the first example, ChimeraX was used to compute the difference map in real space between the apo and the ligand-bound maps in the M. tuberculosis enolase enzyme. The additional density at a high threshold suggests the presence of the ligand, phosphoenolpyruvate, in the active site and, Mg2+ bound to the ligand. It is important to note that, in this case, the map is of medium resolution (3.2 &#197;), and water molecules cannot be modeled confidently (Figure 1). The limitation associated with this method is that it can only be applied when ligand binding does not induce significant conformational changes in the protein. Map normalization was not performed in this case since both the apo and ligand-bound datasets were acquired at the same pixel size and processed with identical parameters in Relion. However, it is worth noting that when comparing maps generated by different reconstruction programs, such as Relion46,47&#160;and CryoSparc61, or with different quality, normalization of the maps becomes essential before meaningful comparisons can be made.
The next example is mGlu5, which undergoes large molecular reorganization upon agonist binding, as evident from the cryoEM structures51,62 (Figure 2). In this scenario, it is not feasible to compute a simple difference map between the unbound (apo) and ligand-bound receptors because of substantial differences between the maps. Here, Servalcat, which uses unsharpened and unweighted half-maps as input for refinement in reciprocal space and subsequently computes a difference map between the experimental map and the map derived from the model, was utilized. At a high threshold, one can visualize differences and act as a guide for correction and improvement of the model. Several unmodeled blobs in the extracellular and near the transmembrane domain of mGlu5 were observed and used as a guide to model ligands (Figure 2).
The third example shows how resolution (2.3 &#197;) plays a crucial role in map interpretation and modeling of a fragment-sized inhibitor in &#946;-galactosidase. Here, the challenge was to identify a very small ligand (&#60;200 Da) in the Servalcat difference map amidst the inherent noise in the data and model it accurately. In addition to the high global resolution determined using Fourier Shell Correlation (FSC), the local resolution specific to the ligand was also sufficiently high to ensure accurate placement of the chiral centres of the ligands (Figure 3). Density for solvent molecules was observed in the difference map throughout the enzyme and especially surrounding the ligand. The effect of resolution on modeling ligands and solvent atoms was also demonstrated (Figure 4). It is important to exercise caution when modeling water or solvent molecules because, on occasion, noise at a low threshold may resemble water or solvent molecules, leading to potential misinterpretations.
Another important consideration is that the cryoEM maps alone may not be sufficient to accurately identify a metal ion on its own. Additional biophysical methods like Extended X-ray Absorption Fine Structure (EXAFS) or Energy-Dispersive X-ray Spectroscopy (EDX) are often necessary to confirm the presence and identity of the metal ion. In both enolase and &#946;-galactosidase enzymes, Mg2+ was modeled due to the wealth of information already available on these proteins, confirming the identity of the metal ion. Also, the coordination of the metal ions in these cases, exemplified by the classic octahedral geometry and the near-ideal coordination distances of Mg2+, provided substantial proof of their identity.
In general, a few important considerations should be taken into account while modeling ligands in cryoEM maps. To begin with, choosing the correct map for ligand identification and modeling is important. In all cases, an unsharpened and unweighted map or half-map is advisable over a sharpened and weighted map to visualize the ligand density, as sharpening and weighting could result in under-sharpened or over-sharpened (noise due to series termination) regions in the map. This may result in a suboptimal ligand density, and the use of different B-factor sharpening can be employed to evaluate the density during model building in Coot. Although there is a risk in using the sharpened map to identify ligands in a cryoEM map, the unsharpened map may not show the full detail of the ligand density but can be used for demonstration purposes, as shown in Figure 1B,C.
The pose of the modeled ligand should be validated, especially in cases where the data is weak. As shown here for mGlu5, the local resolution varies across the cryoEM map, and unbiased modeling of the ligand can be challenging. Servalcat can be used as a valuable tool for detecting potential inaccuracies in protein and ligand modeling22.
Compositional heterogeneity can be present in a protein-ligand complex where only a specific population may have the ligand present (if the ligand is of low molecular weight, then the classification step might not remove the heterogeneity). Nevertheless, it is important to perform the 3D classification63 step iteratively during image processing prior to ligand modeling and verify if the density of the ligand improves. If multiple copies of protein are present, caution should be exercised when applying symmetry across the map during initial model generation and refinement, as this may average out the ligand density in all the symmetry-related molecules. Symmetry should only be imposed after the map has been thoroughly inspected to confirm the presence of ligand density in all the protein chains.
Depending on the state (crystal or solution) and the location (buried or surface), the atoms can be dynamic, and in model refinement, this is referred to as the atomic displacement parameter (ADP). Along with the difference map, which provides visual clues to possible inaccuracies in the model, ADP values can be used to evaluate the accuracy of the ligands after refinement64,65,66,67. Typically, the ligand should have similar ADP values as the surrounding residues, i.e., if they are stably bound and modeled accurately. However, ligands at the periphery or atoms of a ligand (like lipids) that are far from macromolecules can have higher ADP values. In addition to refining the coordinates, both Refmac33,34 and Phenix allow refinement of ADP values28,&#160;68. In Refmac, the Mott-Bethe approximation is used to calculate the electron scattering factor of individual atoms during map calculation. In the recent version of Phenix, individual B-factor refinement, similar to crystallography, has been introduced to account for atom disorder. Very often, in the refined models derived from cryoEM, a wide range of ADP values are observed (sometimes values close to zero), and even the Q-score that is used in EMDB for evaluating the fit of the model with the map depends on the primary map deposited and the nature of B-factor sharpening60. In cryoEM model building, multiple maps are often used, and thus, the maps used in modeling and refinement should clearly be mentioned in the methods, as due to anisotropic resolution in many macromolecules, one single map may not be sufficient to explain all the details.
In modeling of ligands in macromolecules, one of the main limitations of cryoEM maps (like in crystallography) is that if the ligand binding site is of low resolution or if the bound ligand is dynamic, then ascertaining the correct conformation may prove difficult. In addition, most cryoEM structures have resolutions below 3 &#197;, and the representation of water molecules in the maps is limited, making it difficult to assess the role of hydration in ligand or drug binding (as shown here with the examples of enolase and mGluR). Computational methods can be used in combination with cryoEM data to address these limitations69. In contrast to crystallography, only the model undergoes refinement, not the map. Presently, the only method to indicate the presence of a ligand or ensure accurate modeling is by generating omit maps (using tools like Servalcat). Thus, there are a number of tools to help researchers build and evaluate the model, but there are several areas in model refinement where newer approaches or modifications of current approaches can be expected in the near future.
In this article, we have focused on the current approaches for modeling ligands, which include manual inspection of the ligand density, generating the ligand geometry file, followed by modeling the ligand in the cryoEM map. This is an exciting period in structural biology and drug discovery, as direct electron detectors with faster frame rates and the use of faster data acquisition70 have resulted in the obtaining of high-resolution maps (&#60;2.8 &#197;) of several macromolecules often bound with small molecule ligands in a relatively short time. The recent introduction of automated ligand modeling tools such as GEMspot69 in the Schrodinger suite and EMERALD17 in the Rosetta suite, which attempts to find the most probable bound pose of the ligand while factoring in the experimental cryoEM data, holds promise to streamline and automate this process. Similar to X-ray crystallography, it is envisioned that identifying the binding modes of small-molecule ligands by cryoEM, perhaps two or more in a single day, will become a realistic possibility.

Cells accomplish their functions by carrying out innumerable chemical reactions simultaneously and independently, each meticulously regulated to ensure their survival and adaptability in response to environmental cues. This is achieved by molecular recognition, which enables biomolecules, especially proteins, to form transient or stable complexes with other macromolecules as well as small molecules or ligands1. Thus, protein-ligand interactions are fundamental to all processes in biology, which include the regulation of protein expression and activity, the recognition of substrates and cofactors by enzymes, as well as how cells perceive and relay signals1,2. A better understanding of the kinetic, thermodynamic, and structural properties of the protein-ligand complex reveals the molecular basis of ligand interaction and also facilitates rational drug design by optimizing drug interaction and specificity. An economical and faster approach to studying protein-ligand interaction is to use molecular docking, which is a computational method that virtually screens a diverse range of small molecules and predicts the binding mode and affinity of these ligands to target proteins3. However, experimental evidence from high-resolution structures determined by X-ray Diffraction (XRD), Nuclear Magnetic Resonance (NMR), or electron cryomicroscopy (cryoEM) provides the essential proof for such predictions and aids in the development of newer and more effective activators or inhibitors for a given target. This article uses the abbreviation &#39;cryoEM&#39; as the technique is commonly referred to. However, there is an ongoing debate on choosing the correct nomenclature, and recently, the term cryogenic-sample Electron Microscopy (cryoEM) has been proposed to indicate that the sample is at cryogenic temperature and imaged with electrons4. Similarly, the maps derived from cryoEM have been called electron potential, electrostatic potential, or Coulomb potential, and for simplicity, here we use cryoEM maps5,6,7,8,9,10.
Although XRD has been the gold standard technique in high-resolution structure determination of protein-ligand complexes, post resolution-revolution11 cryoEM has gained momentum, as indicated&#160;by the surge of Coulomb potential maps or cryoEM maps deposited in the Electron Microscopy Database (EMDB)12,13 over the last few years14. Owing to advancements in sample preparation, imaging, and data processing methods, the number of Protein Data Bank (PDB)14 depositions employing cryoEM increased from 0.7% to 17% between 2010 and 2020, with approximately 50% of reported structures in 2020 being determined at 3.5&#8197;&#197; resolution or better15,16. CryoEM has been rapidly adopted by the structural biology community, including the pharmaceutical industry, as it allows the study of flexible and non-crystalline biological macromolecules, especially membrane proteins and multi-protein complexes, at near-atomic resolution, overcoming the process of crystallization and obtaining well-diffracting crystals required for high-resolution structure determination by XRD.
Accurate modeling of the ligand in the cryoEM map is paramount, as it serves as a blueprint of the protein-ligand complex at a molecular level. There are several automated ligand-building tools used in X-ray crystallography&#160;that depend on the shape and topology of the ligand density in order to fit or build the ligand into the electron density17,18,19,20. Nevertheless, if the resolution is lower than 3 &#197;, these approaches tend to produce less desirable outcomes because the topological features on which they depend for recognition and building become less defined. In many instances, these methods have proven ineffective in accurately modeling ligands into cryoEM maps, as these maps have been determined in the low-to-medium resolution range, typically between 3.5 &#197;-5 &#197;17.
The first step in 3D structure determination of a protein-ligand complex by cryoEM involves either co-purifying the ligand with the protein (when the ligand has a high binding affinity to the protein) or incubating the protein solution with the ligand for a specific duration before grid preparation. Subsequently, a small sample volume is placed onto a plasma-cleaned holey TEM grid, followed by flash freezing in liquid ethane and ultimately imaging with a cryo-TEM. The 2D projection images from hundreds of thousands to millions of individual particles are averaged to reconstruct a 3-dimensional (3D) Coulomb potential map of the macromolecule. Identifying and modeling ligands and solvent molecules in these maps pose significant challenges in many cases due to the anisotropic resolution across the map (i.e., the resolution is not uniform across the macromolecule), flexibility in the region where the ligand is bound, and the noise in the data. Many of the modeling, refinement, and visualization tools that were developed for XRD are now being adapted for use in cryoEM for the same purposes18, 19, 20, 21. In this article, an overview of various methods and software currently used to identify ligands, build models, and refine the coordinates derived from cryoEM is presented. A step-by-step protocol has been provided to illustrate the processes involved in modeling ligands using specific protein-ligand complexes with varying resolution and complexity.
The first step in modeling ligands in cryoEM maps includes the identification of the ligand density (non-protein) in the map. If ligand binding does not induce any conformational change in the protein, then calculating a simple difference map between the protein-ligand complex and the apo-protein essentially highlights the regions of extra density, suggesting the presence of the ligand. Such differences can be observed immediately, as it just requires two maps, and even intermediate maps during the process of 3D refinement can be used to check if the ligand is present. Additionally, if the resolution is high enough (&#60;3.0 &#197;), then the difference map can also provide insights into the location of water molecules as well as ions interacting with the ligand and the protein residues.
In the absence of the apo-protein map, it is now possible to use Servalcat22, which is available as a standalone tool and has also been integrated into the CCP-EM software suite23,24 as part of the Refmac refinement and in CCP4 8.0 release25,26. Servalcat allows the calculation of an FSC weighted difference (Fo-Fc) map using the unsharpened half-maps and the apoprotein model as input. The Fo-Fc omit map represents the disparity between the experimental map (Fo) and the map derived from the model (Fc). In the absence of a ligand in the model, a positive density in a Fo-Fc map that overlaps with the experimental EM map typically suggests the presence of the ligand. The assumption here is that the protein chain is well-fitted in the map, and the remaining positive density indicates the location of the ligand. However, it is important to meticulously examine whether the positive density stems from modeling inaccuracies, such as the wrong rotamer of a protein side-chain.
The second step involves obtaining or creating a cartesian coordinate file of the ligand with well-defined geometry from the available chemical information. Standard ligands (for example, ATP and NADP+) that are already available in the CCP4 monomer library can be used for refinement by retrieving the coordinate and geometry files via their monomer accession code. However, for unknown or non-standard ligands, various tools are available to create the geometry files. Some of the examples include the eLBOW27 - (electronic ligand builder and optimization workbench) in Phenix28, Lidia - an in-built tool in Coot29, JLigand/ACEDRG30,31, CCP-EM23,24, Ligprep32-a module of Glide within the Schrodinger suite. The ligand coordinate file is then fitted in the density, guided by both the experimental cryoEM map and the difference map in Coot. This is followed by real-space refinement in Phenix28 or reciprocal refinement in Refmac33. A Linux workstation or a laptop equipped with a good graphics card and the above-mentioned software is required. Most of these programs are included in various suites. CCP-EM24&#160;and Phenix28 are freely available to academic users and include a variety of tools that are used in this article, including Coot, Refmac533,34,35,36, Servalcat, phenix.real_space_refine, etc. Similarly,&#160;Chimera37, and ChimeraX38 provide free licenses to academic users.

<table><tbody><tr><td>CCP4-8.0</td><td>Consortium of several institutes</td><td>https://www.ccp4.ac.uk</td><td>Free for academic users and includes Coot and list of tools developed for X-ray crystallography</td></tr><tr><td>CCP-EM</td><td>Consortium of several institutes</td><td>https://www.ccpem.ac.uk/download.php</td><td>Free for academic users and includes Coot, Relion and many others</td></tr><tr><td>Coot&amp;nbsp;</td><td>Paul Emsley, LMB, Cambridge</td><td>https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/</td><td>General software for model building but also available with other suites described above</td></tr><tr><td>DockinMap (Phenix)</td><td>Consortium of several institutes</td><td>https://phenix-online.org/documentation/reference/dock_in_map.html</td><td>Software inside the Phenix suite for docking model into cryoEM maps</td></tr><tr><td>Electron Microscopy Data Bank&amp;nbsp;</td><td>Consortium of several institutes</td><td>https://www.ebi.ac.uk/emdb/</td><td>Public Repository for Electron Microscopy maps</td></tr><tr><td>Falcon</td><td>Thermo Fisher Scientific&amp;nbsp;</td><td>https://assets.thermofisher.com/TFS-Assets/MSD/Technical-Notes/Falcon-3EC-Datasheet.pdf</td><td>Commercial, camera from Thermo Fisher&amp;nbsp;</td></tr><tr><td>Phenix&amp;nbsp;</td><td>Consortium of several institutes</td><td>https://phenix-online.org/download</td><td>Free for academic users and includes Coot</td></tr><tr><td>Protein Data Bank</td><td>Consortium of several institutes</td><td>https://rcsb.org</td><td>Public database of macromolecular structures</td></tr><tr><td>Pymol</td><td>Schrodinger</td><td>https://pymol.org/2/</td><td>Molecular viusalization tool. Educational version is free but comes with limitation. The full version can be obtained with a small fee.</td></tr><tr><td>Relion</td><td>MRC-LMB, Cambridge</td><td>https://relion.readthedocs.io/en/release-4.0/Installation.html</td><td>Software for cryoEM image processing, also available with CCP-EM</td></tr><tr><td>Titan Krios</td><td>Thermo Fisher Scientific&amp;nbsp;</td><td>https://www.thermofisher.com/in/en/home/electron-microscopy/products/transmission-electron-microscopes/krios-g4-cryo-tem.html?cid=msd_ls_xbu_xmkt_tem-krios_285811_gl_pso_gaw_tpne1c&amp;amp;&lt;br/&gt; gad_source=1&amp;amp;gclid=CjwKCAiA-P-rBhBEEiwAQEXhHyw5c8MKThmdA&lt;br/&gt; AkZesWC4FYQSwIQRk&lt;br/&gt; ZApkj08MfYG040DtiiuL8&lt;br/&gt; RihoCebEQAvD_BwE</td><td>Commercial, cryoTEM from Thermo Fisher</td></tr><tr><td>UCSF Chimera</td><td>UCSF, USA</td><td>https://www.cgl.ucsf.edu/chimera/download.html</td><td>General purpose software for display, analysis and more</td></tr><tr><td>UCSF Chimera X</td><td>UCSF, USA</td><td>https://www.cgl.ucsf.edu/chimerax/</td><td>General purpose software for display, analysis and more</td></tr></tbody></table>

modeling ligands into maps derived from electron cryomicroscopy

Deciphering the protein-ligand interactions in a macromolecular complex is crucial for understanding the molecular mechanism, underlying biological processes, and drug development. In recent years, cryogenic sample electron microscopy (cryoEM) has emerged as a powerful technique to determine the structures of macromolecules and to investigate the mode of ligand binding at near-atomic resolution. Identifying and modeling non-protein molecules in cryoEM maps is often challenging due to anisotropic resolution across the molecule of interest and inherent noise in the data. In this article, the readers are introduced to various software and methods currently used for ligand identification, model building, and refinement of atomic coordinates using selected macromolecules. One of the simplest ways to identify the presence of a ligand, as illustrated with the enolase enzyme, is to subtract the two maps obtained with and without the ligand. The extra density of the ligand is likely to stand out in the difference map even at a higher threshold. There are instances, as shown in the case of metabotropic Glutamate receptor mGlu5, when such simple difference maps cannot be generated. The recently introduced method of deriving the Fo-Fc omit map can serve as a tool for validating and demonstrating the presence of the ligand. Finally, using the well-studied &#946;-galactosidase as an example, the effect of resolution on modeling the ligands and solvent molecules in cryoEM maps is analyzed, and an outlook on how cryoEM can be used in drug discovery is presented.

Example 1 
The enzyme enolase from M. tuberculosis catalyzes the penultimate step of glycolysis and converts 2-phosphoglycerate to phosphoenolpyruvate (PEP), which is an essential intermediate for several metabolic pathways44,45. CryoEM data for the apo-enolase and PEP-bound enolase samples were collected at the same pixel size of 1.07 &#197;, and image processing was performed with Relion 3.146,47. The apo-enolase and the PEP-enolase structures were determined at 3.1 &#197; and 3.2 &#197;, respectively48. The maps and models were deposited in EMDB and PDB49,50 (EMD-30988, EMD-30989, PDB-7e4x, and PDB-7e51). The cryoEM map of the enzyme shows that it is an octamer in the solution (Figure 1A). In order to identify the ligand density in the PEP-enolase map, the unsharpened maps of the apoenzyme and PEP-bound enzyme were selected, and a difference map was calculated in ChimeraX, by subtracting the PEP-enolase map from the apoenzyme map. A distinct density (green) at a high threshold was observed, which suggested the presence of the ligand (Figure 1B). Modeling the protein chain in the unsharpened map clearly indicated that the extra density is present in the active site of the protein (Figure 1C). The ligand, PEP, was then modeled in the B-factor sharpened map using Coot, and the protein+ligand model was refined in real space with Phenix. Two Mg2+&#160;ions were modeled in the density observed in the vicinity of the ligand (Figure 1D). The ligand, PEP, adopts a similar orientation as observed in other enolase homologs, and several active site residues, such as Lys-386, Arg-364, form hydrogen bond&#160;interactions with the ligand PEP. The Mg2+ ions form metal coordination bonds with Asp-241, Glu-283, Asp-310, and the phosphate of PEP (Figure 1D).
Example 2 
In the absence of an available apo-protein structure or if the protein undergoes a large conformational change, calculating the difference maps as described above is not possible. In 2021, Garib Murshudov&#39;s group at the Laboratory of Molecular Biology, Cambridge, introduced Servalcat22, which implements a refinement workflow using Refmac and also calculates a Fo-Fc difference map after refinement. A positive Fo-Fc difference density suggests the presence of molecules/ligands that were not included in the model during refinement, essentially an omit&#160;map. However, it is recommended to first assess the fit of the model to the map in general and then evaluate the difference density map.
To illustrate the use of Servalcat/Refmac, mGlu5, a dimeric G-protein coupled receptor that binds to the neurotransmitter, L-Glutamate was chosen. Upon binding of the agonist, L-quisqualate, the extracellular domain reorients, which triggers the rotation of the 7TM, bringing them closer to stabilize the activated state. Thus, there is a large conformational change observed between the apo/antagonist vs. agonist bound states51&#160;(Figure 2A and Figure 2E). The two half-maps for both the agonist (EMD-31536) and the antagonist (EMD-31537) bound complexes were obtained from EMDB, and the cryoEM maps show varied resolution across the molecule and better-resolved extracellular domain. Subsequently, these were utilized as inputs in Servalcat along with the apo-protein as the model to compute the difference or Fo-Fc map for each dataset. This map distinctly showed the presence of various ligand molecules (non-protein). The resolution estimated by FSC (Fourier Shell correlation) for the agonist and antagonist bound complexes were 3.8 &#197; and 4.0 &#197;, respectively. In the case of the agonist-bound mGlu5, the Servalcat Fo-Fc difference map showed the presence of both the agonist (L-quisqualate) (Figure 2B) and N-AcetylGlucosamine (NAG) (Figure 2C) in the ECD of the receptor (due to lower resolution of the TMD, here we focus only in the ECD and the top of the TMD). Protein residues, including Tyr-64, Trp-100, Ser-151, and Thr-175, are seen to interact with the agonist. Density near the Asn-210 residue suggested the presence of N-acetyl glucosamine (Figure 2C). A density consistent with cholesteryl hemisuccinate, which was added during the purification of mGlu5, was observed near the top of transmembrane helix 1 (Figure 2D). As the resolution is moderate and the ligand, L-quisqualate, can be placed in different orientations, prior&#160;structure of the extracellular domain with the ligand (PDB-6N50) was used as a guide to model the ligand. Antagonist binding stabilizes the open or resting state of the receptor (Figure 2E). A density consistent with the antagonist LY341495 was observed at the hinge of lobe I and lobe II of the venus fly trap domain in the ECD. The antagonist interacts with residues similar to those of the agonist. Stacking interaction between Tyr-223 in lobe II with the antagonist stabilizes the receptor in an open state (Figure 2F). Similar to the agonist structure, glycosylation or the presence of N-acetylglucosamine moiety was observed near Asn-210 (Figure 2G).
Example 3 
The third example elucidates the protocol to model fragment-sized or small-sized ligands and solvent molecules in high-resolution CryoEM maps. Fragment-based Drug Discovery (FBDD) has emerged as a powerful and innovative method in the development of target-based novel therapeutics in various disease areas, making it a promising avenue in pharma R&#38;D52,53. FBDD begins with screening and careful selection of small, highly soluble, low-molecular-weight fragments of molecules that bind to specific target proteins or biomolecules of interest. Determining the structures of these protein-fragment complexes reveals the mode of binding of these fragments, which serves as a guide to designing larger and more complex drug-like molecules with increasing affinity and specificity towards the target protein54. However, this method demands a high-resolution ligand density to&#160;accurately determine the pose&#160;and correctly place the functional groups of the ligand15.
&#946;-galactosidase, one of the first high-resolution structures to be determined from the advancements in cryoEM technology, is a well-studied 450kDa homotetrameric enzyme that catalyzes the hydrolysis of lactose to glucose and galactose55. To showcase the use of cryoEM in FBDD, Astex, UK, determined the structure of &#946;-galactosidase with a fragment-sized inhibitor, deoxygalacto-nojirimycin (DGN) bound in the active site (EMDB-10563, PDB:6tsh)56. This dataset is used to illustrate the protocol to model ligands and solvents unambiguously in high-resolution maps. To showcase the effect of resolution on ligand modeling and visualization, the maps were filtered to 3.0 &#197; and 3.5 &#197; in the post-processing step in Relion. This highlights the quality of the map density at different resolutions and underscores the need for higher resolution for modeling ligands and solvents.
The enzyme is a tetramer with D2 symmetry in solution (Figure 3A). Difference map (between map and model), as computed by Servalcat, suggested the presence of DGN and several solvent molecules in the active site of the enzyme (Figure 3B). At an estimated resolution of 2.3 &#197;, the density showed high-resolution features, which helped in the accurate modeling of the inhibitor in the protein active site. Interactions between DGN and Tyr-503 and His-540 were observed (Figure 3C). The difference&#160;density also suggested the presence of solvent molecules that interact with DGN as well as the protein residues. Mg2+ and several water molecules were modeled in the density (Figure 3D). Metal coordination bonds between Mg2+ and Glu-416, Glu-461, and several water molecules are observed (Figure 3D). Mg2+ was seen to interact with DGN via a water molecule.
At a lower resolution of 3.5 &#197; and 3.0 &#197;, the ligand density resembles a blob and lacks high-resolution features crucial for accurate modeling of the ligand (Figure 4A,B). Density for water molecules was almost non-existent at these resolutions. In summary, with increasing resolution, especially higher than 3.0 &#197;, the density enabled the modeling of a greater number of water molecules (Figure 4C,D). The correct positioning of the ligand&#39;s chiral centre became achievable at ~2.3 &#197; (Figure 4C,D) due to the presence of distinct features in the map that guided the placement as well as modeling of water molecules and Mg2+. In comparison, the density for Mg2+ remained discernible across the resolution range (Figure 4A-C).
<img alt="Figure 1" class="xfigimg" src="/files/ftp_upload/66310/66310fig01v3.jpg" /> 
Figure 1: Modeling the ligand phosphoenolpyruvate in M. tuberculosis enolase. (A)&#160;shows the B-factor sharpened map of the PEP-bound enolase enzyme. The map suggests that enolase is octameric in solution, and each monomer in the map is colored differently. (B)&#160;displays an unsharpened cryoEM map of apo-Enolase enzyme in grey with the difference map (between PEP-bound and apo-Enolase unsharpened maps) overlaid in green, suggesting the presence of ligand, PEP. This is a demonstration map to show the presence of ligands. (C) displays the fit of the enolase model in the unsharpened cryoEM map, highlighting the position of the difference density relative to the protein. The protein model is shown in cartoon representation and colored in Chainbow. This figure shows that the extra density (green) is present in the active site of each monomer. (D) shows the ligand, PEP, encased in the B-factor sharpened map, colored blue. Additionally, density for two Mg2+ ions was also observed, which forms metal coordination bonds with several active site residues, including Ser-42, Asp 241, Glu-283, Asp-310, and ligand atoms. The ligand makes hydrogen bond interactions with Lys-386, Lys-335, and Arg-364. The protein residues are shown in stick representation, and Mg2+&#160;ions are shown as&#160;purple spheres. The figures in panels (A-D) were generated with Pymol. <a href="https://www-jove-com.remotexs.ntu.edu.sg/files/ftp_upload/66310/66310fig01largev3.jpg" target="_blank">Please click here to view a larger version of this figure.</a>
<img alt="Figure 2" class="xfigimg" src="/files/ftp_upload/66310/66310fig02.jpg" /> 
Figure 2: Identification, modeling, and visualization of various ligands in mGlu5 receptor. The Fo-Fc omit maps were obtained using Servalcat.&#160;(A) shows mGlu5 receptor dimer structure (PDB-7fd8) in cartoon representation, with each monomer colored in teal and wheat, respectively, and bound to the agonist L-quisqualate. All the ligands that were identified in the extracellular and on the top of the transmembrane domain of the receptor are encased in the Fo-Fc omit map, colored green, and contoured at 6&#963;.(B) highlights the fit of the agonist, L-quisqualate encased in the difference map. L-quisqualate interacts with several mGlu5 residues including Tyr-64, Trp-100, Ser-151, Thr-175 and Gly-280. The H-bond interactions are represented by red dashes. In (C), an additional density is evident near Asn-210, which is present in the receptor&#39;s extracellular domain, and N-acetylglucosamine molecule (NAG) was modeled in this density. For clarity, the NAG is not linked to Asn in the current figure. (D)&#160;shows the difference in density for cholesterol hemisuccinate (CHS) in green near the lipid-exposed surface of the receptor. The CHS molecule, represented in the sticks, was modeled in this density. (E) shows mGlu5 receptor structure (PDB-7fd9) bound to the antagonist LY341495 in cartoon representation. In (F), an extra density in the Fo-Fc difference map located at the hinge between lobe I and lobe II of the extracellular domain indicates the presence of the antagonist. Key residues (Tyr-64, Trp-100, Ser-152, Ser-173, Thr-175, and Tyr-223) around the antagonist are shown in stick representations, and potential hydrogen bond interactions with the antagonist are shown in red dashes. (G) shows the difference in density suggestive of the presence of NAG molecule near Asn-210 in the antagonist-bound structure (note, the NAG is not linked with Asn for clarity). The figures were generated with Pymol. <a href="https://www-jove-com.remotexs.ntu.edu.sg/files/ftp_upload/66310/66310fig02large.jpg" target="_blank">Please click here to view a larger version of this figure.</a>
<img alt="Figure 3" class="xfigimg" src="/files/ftp_upload/66310/66310fig03.jpg" /> 
Figure 3:&#160;Identification, modeling, and refinement of a small inhibitor and solvent molecules in the high-resolution map of &#946;-galactosidase (EMD-10563). (A) shows the &#946;-galactosidase model resolved to 2.3 &#197; (PDB: 6tsh) in cartoon representation, where each monomer is colored distinctly. The grey box highlights the ligand binding site. (B) displays the Fo-Fc difference density (from Servalcat) in green mesh in the active site of the enzyme. The difference in density suggests the presence of the inhibitor (DGN) and several solvent molecules in the active site. (C) demonstrates that guided by the Fo-Fc map, the inhibitor deoxygalacto-nojirimycin (DGN) is modeled in the active site. This modeled ligand is depicted in stick format and encased in the Fo density (by Servalcat), which is colored in blue_mesh. Hydrogen bond interactions between DGN and several protein residues, including Tyr-503 and His-540, are observed. The additional Fo-Fc difference density around the ligand (green) is indicative of the solvent molecules. (D) The map shows several solvent molecules, including water and Mg2+, represented as red and purple spheres, respectively,&#160;are modeled in the active site after ensuring that each solvent molecule is bonded to either protein (Glu-416, His-418, and Glu-461) or ligand residues. The water molecules and Mg2+ are encased in the Fo density (blue mesh). Mg2+ is seen to interact with the ligand, DGN via a water molecule. The Fo-Fc map (green mesh) in panels (B,C) is contoured at 6&#963;, whereas the Fo density - blue mesh in C and D (from the Servalcat refinement after modeling) is contoured at 3&#963;. The figures in panels (A-D) were generated with Pymol. <a href="https://www-jove-com.remotexs.ntu.edu.sg/files/ftp_upload/66310/66310fig03large.jpg" target="_blank">Please click here to view a larger version of this figure.</a>
<img alt="Figure 4" class="xfigimg" src="/files/ftp_upload/66310/66310fig04.jpg" /> 
Figure 4: Effect of resolution on ligand modeling in &#946;-galactosidase. Half-maps from EMD-10563 were used as input in the Relion post-process step, and the combined post-process maps were filtered to resolutions 2.3 &#197;, 3.0 &#197;, and 3.5 &#197; with different B-factors. The map shown in all the panels is contoured at 6&#963;. For clarity, only the protein backbone is shown with no side chains, ligand or solvent molecules in panels A, B, and C.&#160;(A) The map is filtered to 3.5 &#197;, and the active site of &#946;-galactosidase is shown.&#160;A blob resembling the ligand, DGN, is seen at this resolution, accompanied by a few smaller blobs in the vicinity. Modeling the ligand in the correct orientation proves challenging due to the lack of distinct features in the map. (B) The map filtered to 3.0 &#197; resolution is shown. Here, the ligand blob becomes slightly more defined but still lacks features in general. A few more small blobs suggestive of solvent molecules are also observed. (C) The map filtered to 2.3 &#197; resolution reveals the ligand density with distinct features, notably revealing the chair conformation of the iminosugar. A significant number of small blobs corresponding to water molecules are observed at this resolution. The automatic B-factor estimation/sharpening in Relion post-process gives a value of -18 &#197;2 for the maps filtered to 3 &#197; and 3.5 &#197;, while the value is -52 &#197;2 for the map filtered to 2.3 &#197;. Different B-factor sharpening of EM maps can also be performed with Coot and useful in model building. (D) The panel illustrates that the ligand DGN (stick representation), Mg2+, and water molecules&#160;(as spheres)&#160;as modeled in the active site and the sharpened map at 2.3 &#197;, shown in blue mesh surrounding these atoms. The figures in panels (A-D) were generated with Pymol. <a href="https://www-jove-com.remotexs.ntu.edu.sg/files/ftp_upload/66310/66310fig04large.jpg" target="_blank">Please click here to view a larger version of this figure.</a>

Author Spotlight: Exploring Cellular Processes by Modeling Ligands in Cryo-EM Maps

This protocol introduces the tools available for modeling small-molecule ligands in cryoEM maps of macromolecules.

Modeling Ligands into Maps Derived from Electron Cryomicroscopy

Deciphering the protein-ligand interactions in a macromolecular complex is crucial for understanding the molecular mechanism, underlying biological ...

modeling-ligands-into-maps-derived-from-electron-cryomicroscopy

Nanyang Technological University

Research

JoVE Journal

Biochemistry

1.2K Views.  Tata Institute of Fundamental Research. This protocol introduces the tools available for modeling small-molecule ligands in cryoEM maps of macromolecules.

Video: Author Spotlight: Exploring Cellular Processes by Modeling Ligands in Cryo-EM Maps

High-resolution Single Particle Analysis from Electron Cryo-microscopy Images Using SPHIRE

SPHIRE (SPARX for High-Resolution Electron Microscopy) is a novel open-source, user-friendly software suite for the semi-automated processing of single particle electron cryo-microscopy (cryo-EM) data. The protocol presented here describes in detail how to obtain a near-atomic resolution structure starting from cryo-EM micrograph movies by guiding users through all steps of the single particle structure determination pipeline. These steps are controlled from the new SPHIRE graphical user interface and require minimum user intervention. Using this protocol, a 3.5 &#197; structure of TcdA1, a Tc toxin complex from Photorhabdus luminescens, was derived from only 9500 single particles. This streamlined approach will help novice users without extensive processing experience and a priori structural information, to obtain noise-free and unbiased atomic models of their purified macromolecular complexes in their native state.

This paper presents a protocol for processing cryo-EM images using the software suite SPHIRE. The present protocol can be applied for nearly all single particle EM projects that target near-atomic resolution.

SPHIRE (SPARX for High-Resolution Electron Microscopy) is a novel open-source, user-friendly software suite for the semi-automated processing of single ...

Measuring Biomolecular DSC Profiles with Thermolabile Ligands to Rapidly Characterize Folding and Binding Interactions

Differential scanning calorimetry (DSC) is a powerful technique for quantifying thermodynamic parameters governing biomolecular folding and binding interactions. This information is critical in the design of new pharmaceutical compounds. However, many pharmaceutically relevant ligands are chemically unstable at the high temperatures used in DSC analyses. Thus, measuring binding interactions is challenging because the concentrations of ligands and thermally-converted products are constantly changing within the calorimeter cell. Here, we present a protocol using thermolabile ligands and DSC for rapidly obtaining thermodynamic and kinetic information on the folding, binding, and ligand conversion processes. We have applied our method to the DNA aptamer MN4 that binds to the thermolabile ligand cocaine. Using a new global fitting analysis that accounts for thermolabile ligand conversion, the complete set of folding and binding parameters are obtained from a pair of DSC experiments. In addition, we show that the rate constant for thermolabile ligand conversion may be obtained with only one supplementary DSC dataset. The guidelines for identifying and analyzing data from several more complicated scenarios are presented, including irreversible aggregation of the biomolecule, slow folding, slow binding, and rapid depletion of the thermolabile ligand.

We present a protocol for rapid characterization of biomolecular folding and binding interactions with thermolabile ligands using differential scanning calorimetry.

Differential scanning calorimetry (DSC) is a powerful technique for quantifying thermodynamic parameters governing biomolecular folding and binding ...

Semi-automated Biopanning of Bacterial Display Libraries for Peptide Affinity Reagent Discovery and Analysis of Resulting Isolates

Biopanning bacterial display libraries is a proven technique for peptide affinity reagent discovery for recognition of both biotic and abiotic targets. Peptide affinity reagents can be used for similar applications to antibodies, including sensing and therapeutics, but are more robust and able to perform in more extreme environments. Specific enrichment of peptide capture agents to a protein target of interest is enhanced using semi-automated sorting methods which improve binding and wash steps and therefore decrease the occurrence of false positive binders. A semi-automated sorting method is described herein for use with a commercial automated magnetic-activated cell sorting device with an unconstrained bacterial display sorting library expressing random 15-mer peptides. With slight modifications, these methods are extendable to other automated devices, other sorting libraries, and other organisms. A primary goal of this work is to provide a comprehensive methodology and expound the thought process applied in analyzing and minimizing the resulting pool of candidates. These techniques include analysis of on-cell binding using fluorescence-activated cell sorting (FACS), to assess affinity and specificity during sorting and in comparing individual candidates, and the analysis of peptide sequences to identify trends and consensus sequences for understanding and potentially improving the affinity to and specificity for the target of interest.

Biopanning bacterial display libraries is a proven technique for discovery of peptide affinity reagents, a robust alternative to antibodies. The semi-automated sorting method herein has streamlined biopanning to decrease the occurrence of false positives. Here we illustrate the thought process and techniques applied in evaluating candidates and minimizing downstream analysis.

Biopanning bacterial display libraries is a proven technique for peptide affinity reagent discovery for recognition of both biotic and abiotic targets. ...

Analyzing Dynamic Protein Complexes Assembled On and Released From Biolayer Interferometry Biosensor Using Mass Spectrometry and Electron Microscopy

In vivo, proteins are often part of large macromolecular complexes where binding specificity and dynamics ultimately dictate functional outputs. In this work, the pre-endosomal anthrax toxin is assembled and transitioned into the endosomal complex. First, the N-terminal domain of a cysteine mutant lethal factor (LFN) is attached to a biolayer interferometry (BLI) biosensor through disulfide coupling in an optimal orientation, allowing protective antigen (PA) prepore to bind (Kd 1 nM). The optimally oriented LFN-PAprepore complex then binds to soluble capillary morphogenic gene-2 (CMG2) cell surface receptor (Kd 170 pM), resulting in a representative anthrax pre-endosomal complex, stable at pH 7.5. This assembled complex is then subjected to acidification (pH 5.0) representative of the late endosome environment to transition the PAprepore into the membrane inserted pore state. This PApore state results in a weakened binding between the CMG2 receptor and the LFN-PApore and a substantial dissociation of CMG2 from the transition pore. The thio-attachment of LFN to the biosensor surface is easily reversed by dithiothreitol. Reduction on the BLI biosensor surface releases the LFN-PAprepore-CMG2 ternary complex or the acid transitioned LFN-PApore complexes into microliter volumes. Released complexes are then visualized and identified using electron microscopy and mass spectrometry. These experiments demonstrate how to monitor the kinetic assembly/disassembly of specific protein complexes using label-free BLI methodologies and evaluate the structure and identity of these BLI assembled complexes by electron microscopy and mass spectrometry, respectively, using easy-to-replicate sequential procedures.

Here we present a protocol to monitor the assembly and disassembly of the anthrax toxin using biolayer interferometry (BLI). Following assembly/disassembly on the biosensor surface, the large protein complexes are released from the surface for visualization and identification of components of the complexes using electron microscopy and mass spectrometry, respectively.

In vivo, proteins are often part of large macromolecular complexes where binding specificity and dynamics ultimately dictate functional outputs. In this ...

Analysis of &#946;-Amyloid-induced Abnormalities on Fibrin Clot Structure by Spectroscopy and Scanning Electron Microscopy

This article presents methods for generating in vitro fibrin clots and analyzing the effect of beta-amyloid (A&#946;) protein on clot formation and structure by spectrometry and scanning electron microscopy (SEM). A&#946;, which forms neurotoxic amyloid aggregates in Alzheimer&#39;s disease (AD), has been shown to interact with fibrinogen. This A&#946;-fibrinogen interaction makes the fibrin clot structurally abnormal and resistant to fibrinolysis. A&#946;-induced abnormalities in fibrin clotting may also contribute to cerebrovascular aspects of the AD pathology such as microinfarcts, inflammation, as well as, cerebral amyloid angiopathy (CAA). Given the potentially critical role of neurovascular deficits in AD pathology, developing compounds which can inhibit or lessen the A&#946;-fibrinogen interaction has promising therapeutic value. In vitro methods by which fibrin clot formation can be easily and systematically assessed are potentially useful tools for developing therapeutic compounds. Presented here is an optimized protocol for in vitro generation of the fibrin clot, as well as analysis of the effect of A&#946; and A&#946;-fibrinogen interaction inhibitors. The clot turbidity assay is rapid, highly reproducible and can be used to test multiple conditions simultaneously, allowing for the screening of large numbers of A&#946;-fibrinogen inhibitors. Hit compounds from this screening can be further evaluated for their ability to ameliorate A&#946;-induced structural abnormalities of the fibrin clot architecture using SEM. The effectiveness of these optimized protocols is demonstrated here using TDI-2760, a recently identified A&#946;-fibrinogen interaction inhibitor.

Presented here are two methods that can be used individually or in combination to analyze the effect of beta-amyloid on fibrin clot structure. Included is a protocol for creating an in vitro fibrin clot, followed by clot turbidity and scanning electron microscopy methods.

This article presents methods for generating in vitro fibrin clots and analyzing the effect of beta-amyloid (A&#946;) protein on clot formation and ...

Expression and Purification of the Human Lipid-sensitive Cation Channel TRPC3 for Structural Determination by Single-particle Cryo-electron Microscopy

Transient receptor potential channels (TRPCs) of the canonical TRP subfamily are nonselective cation channels that play an essential role in calcium homeostasis, particularly store-operated calcium entry, which is critical to maintaining proper function of synaptic vesicle release and intracellular signaling pathways. Accordingly, TRPC channels have been implicated in a variety of human diseases including cardiovascular disorders such as cardiac hypertrophy, neurodegenerative disorders such as Parkinson&#39;s disease, and neurologic disorders such as spinocerebellar ataxia. Therefore, TRPC channels represent a potential pharmacologic target in human diseases. However, the molecular mechanisms of gating in these channels are still unclear. The difficulty in obtaining large quantities of stable, homogeneous, and purified protein has been a limiting factor in structure determination studies, particularly for mammalian membrane proteins such as the TRPC ion channels. Here, we present a protocol for the large-scale expression of mammalian ion channel membrane proteins using a modified baculovirus gene transfer system and the purification of these proteins by affinity and size-exclusion chromatography. We further present a protocol to collect single-particle cryo-electron microscopy images from purified protein and to use these images to determine the protein structure. Structure determination is a powerful method for understanding the mechanisms of gating and function in ion channels.

This protocol describes techniques used to determine ion channel structures by cryo-electron microscopy, including a baculovirus system used to efficiently express genes in mammalian cells with minimum effort and toxicity, protein extraction, purification, and quality checking, sample grid preparation and screening, as well as data collection and processing.

Transient receptor potential channels (TRPCs) of the canonical TRP subfamily are nonselective cation channels that play an essential role in calcium ...

Single Particle Cryo-Electron Microscopy: From Sample to Structure

Cryo-electron microscopy (cryoEM) is a powerful technique for structure determination of macromolecular complexes, via single particle analysis (SPA). The overall process involves i) vitrifying the specimen in a thin film supported on a cryoEM grid; ii) screening the specimen to assess particle distribution and ice quality; iii) if the grid is suitable, collecting a single particle dataset for analysis; and iv) image processing to yield an EM density map. In this protocol, an overview for each of these steps is provided, with a focus on the variables which a user can modify during the workflow and the troubleshooting of common issues. With remote microscope operation becoming standard in many facilities, variations on imaging protocols to assist users in efficient operation and imaging when physical access to the microscope is limited will be described.

Structure determination of macromolecular complexes using cryoEM has become routine for certain classes of proteins and complexes. Here, this pipeline is summarized (sample preparation, screening, data acquisition and processing) and readers are directed towards further detailed resources and variables that may be altered in the case of more challenging specimens.

Cryo-electron microscopy (cryoEM) is a powerful technique for structure determination of macromolecular complexes, via single particle analysis (SPA). The ...

Modeling an Enzyme Active Site using Molecular Visualization Freeware

Biomolecular visualization skills are paramount to understanding key concepts in the biological sciences, such as structure-function relationships and molecular interactions. Various programs allow a learner to manipulate 3D structures, and biomolecular modeling promotes active learning, builds computational skills, and bridges the gap between two dimensional textbook images and the three dimensions of life. A critical skill in this area is to model a protein active site, displaying parts of the macromolecule that can interact with a small molecule, or ligand, in a way that shows binding interactions. In this protocol, we describe this process using four freely available macromolecular modeling programs: iCn3D, Jmol/JSmol, PyMOL, and UCSF ChimeraX. This guide is intended for students seeking to learn the basics of a specific program, as well as instructors incorporating biomolecular modeling into their curriculum. The protocol enables the user to model an active site using a specific visualization program, or to sample several of the free programs available. The model chosen for this protocol is human glucokinase, an isoform of the enzyme hexokinase, which catalyzes the first step of glycolysis. The enzyme is bound to one of its substrates, as well as a non-reactive substrate analog, which allows the user to analyze interactions in the catalytic complex.

A key skill in biomolecular modeling is displaying and annotating active sites in proteins. This technique is demonstrated using four popular free programs for macromolecular visualization: iCn3D, Jmol, PyMOL, and UCSF ChimeraX.

Biomolecular visualization skills are paramount to understanding key concepts in the biological sciences, such as structure-function relationships and ...

Simplified Volumetric Models as an Effective Strategy for Segmenting Actin Networks in Cryo-Electron Tomograms

Efficient methods for the extraction of features of interest remain one of the biggest challenges for the interpretation of cryo-electron tomograms. Various automated approaches have been proposed, many of which work well for high-contrast datasets where the features of interest can be easily detected and are clearly separated from one another. Our inner ear stereocilia cryo-electron tomographic datasets are characterized by a dense array of hexagonally packed actin filaments that are frequently cross-connected. These features make automated segmentation very challenging, further aggravated by the high-noise environment of cryo-electron tomograms and the high complexity of the densely packed features. Using prior knowledge about the actin bundle organization, we have placed layers of a highly simplified ball-and-stick actin model to first obtain a global fit to the density map, followed by regional and local adjustments of the model. We show that volumetric model building not only allows us to deal with the high complexity, but also provides precise measurements and statistics about the actin bundle. Volumetric models also serve as anchoring points for local segmentation, such as in the case of the actin-actin cross connectors. Volumetric model building, particularly when further augmented by computer-based automated fitting approaches, can be a powerful alternative when conventional automated segmentation approaches are not successful.

Here, we present a protocol to place simplified volumetric models into noisy, complex, tomographic 3D volumes. This allows the fast segmentation of actin filament densities, detection of systematic filament bending and of gaps in hair bundle filaments, as well as convenient quantification of volumetric model properties, such as distances.

Efficient methods for the extraction of features of interest remain one of the biggest challenges for the interpretation of cryo-electron tomograms. ...

Modeling Ligands into Maps Derived from Electron Cryomicroscopy

Summary

Explore More Videos

Chapters in this video