Case 1: Domain level; overall fold [≥novice]

The overall fold is often the first figure in a structural report, usually depicted as a cartoon ribbon. Figure 1 shows an example of this type of graphic. This figure aims to depict and clearly label the domains and 2° structural elements (SSEs) comprising the protein. As the key to understanding the relationship between structural elements and protein function, it will be a figure of repeated reference for the reader. There are four critical goals in constructing such a figure: to optimally orient the molecule, choose between stereo or mono representation, settle on a meaningful coloring scheme, and clearly label relevant structural entities.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. The overall fold of a carbonic anhydrase.

The three domains of CsoS3 (PDB 2G13; [21]) are distinguished by separate coloring (blue, yellow, red). The orientation was chosen to feature the location of the active site (outlined by side chains and zinc ion), and to show the two-fold symmetry relationship between the active domain (yellow) and homologous but defunct domain (red). Structural elements are labeled directly on the individual SSEs. Domain labels are colored to correspond to the domains they are labeling. Labels in the active site are given an “outer glow” to make them legible in a region of the figure that is dense in detail. Depth is conveyed by use of fog, veiling less important structural features toward the back of the enzyme.

https://doi.org/10.1371/journal.pcbi.1000918.g001

The molecule should be oriented so as to minimize unnecessary overlap of SSEs. Neglecting to do so will make it difficult to perceive the relative depth of the overlapping features and the overall fold. Ideally, the viewer should be able to visually trace the path of a protein chain from N′ to C′ terminus without ambiguity. Finding this “canonical” orientation can be difficult, especially for large proteins or complexes, so survey many options. As if these are not enough constraints, one must also consider the layout of later figures. Subsequent figures may involve close-up views of the active site, protein interaction interfaces, or other areas of special interest. Therefore, if at all possible, orient the molecule in this initial figure so that spatial relationships among these important sites are clearly illustrated too. If chosen carefully, it will be possible to minimize the total number of different orientations of the molecule illustrated in the paper, thereby minimizing the chance of spatial disorientation and confusion of the viewer. In many cases, no more than two orthogonal orientations of the molecule will be required for the whole paper. The relationship between the two orientations should be clearly labeled by a curved arrow that shows the direction and magnitude of rotation (Figure S2G), or else the relationship should be explicitly stated in the legend. If terms such as “front view,” “top view,” “side view” are included in the text, then also include these labels in the figure so that the reader can become properly oriented with just a quick glance.

Split stereo views greatly aid depth perception (Section 2.7 of Text S1). It is simple to make a stereo figure in PyMOL: chose an orientation, activate stereo (using the “stereo” command), and render the image. Unfortunately, not everyone is able to perceive stereo, and it is sometimes inconvenient to find stereo glasses. Instead, a pair of orthogonal views can convey the perception of depth without the need for glasses or the training in how to view stereo. Whether or not stereo is used, shadowing and fog supply valuable depth-cues. (These features are activated by default in PyMOL.) The effect of fog is enhanced by adjusting the front and rear clipping planes (Box 1) so they touch the front and rear of the molecule, respectively. Features in the rear of the molecule will be lightly veiled in fog, while features in the front will be sharp, crisp, and bright.

As a general rule, figures showing the overall fold (Figures 1, S2, and S3A) should label each SSE, domain, or any other structural element of special interest (such as features referred to in later figures or text). Labels should be large enough to be clearly visible, but not so large that the figure looks cluttered. Labeling can be done in PyMOL, or via graphical editing software (e.g., Photoshop or Illustrator) for more fine-grained control of label placement, font, color, and size. If possible, label helices and strands directly on the elements themselves, as in Figure 1. In some cases it might be necessary to place the label a short distance away from the SSE, but this should generally be avoided since it can become ambiguous to the reader to which element a label refers; it may be necessary to increase the thickness of the helix or strand to accommodate direct labeling. For large (e.g., multi-domain) structures, consider simplifying the figure by portraying helices as cylinders instead of ribbons (the “cartoon_cylindrical_helices” setting in PyMOL). If the shading or lines of the SSE obscure the legibility of the label, it is possible to impart an “outer glow”; this creates a small halo of white color around the label, thereby better contrasting it against the background. For stereo figures, take special care to place the labels so they appear at the same depth as the feature being labeled. While PyMOL renders labels in stereo at proper depths, one does this manually in Photoshop: First, place labels in one of the panels, either left or right. Copy the labels to the other panel so they are at the exact same height as the corresponding labels in the original panel. Put on stereo glasses. Then, simply adjust the horizontal position of the labels in the left panel until they appear to be at the same depth as the feature being labeled.

Strive to limit the number of colors used in the figure of the overall fold. A simple grayscale can be sufficient to convey depth, avoid journal color charges, and reproduce accurately on black/white photocopiers (see also Box S1 in Text S1). For multi-domain proteins, a separate color is often assigned to each domain (Figures 1 and S3A). If a color scheme is chosen for this figure, adhere to this scheme in later figures so that the reader can easily recognize how structural elements relate to one another, in space and in position along the amino acid sequence. If the structure described is related to a previously described structure in the literature, consider adopting the existing color scheme; this will minimize confusion, and readers familiar with that literature will appreciate the continuity of convention. If the illustrated structure is a new fold, the chain is often color-ramped so that successive structural elements (domains or SSEs) are graded from blue (N′)→red (C′). Such a scheme is helpful in distinguishing, at a glance, whether a particular structural feature arises near the beginning, middle, or end of the polypeptide chain. In depicting multi-subunit assemblies, consider separately coloring each subunit, employing particular coloring schemes for homo-/heteromeric complexes, etc.; doing so is especially helpful for intricate quaternary structures, such as domain-swapped oligomers [15]. An example of a ribbon cartoon for a moderately complex dimer, alongside its 2D topology diagram, can be found in Figures 2 and 3 of ref. [16].

Case 2: Ligand-binding sites [≥novice / intermediate]

Evolution has molded ligand-binding sites to perform specific and unique functions, giving a protein its identity, and often its name. It follows that figures portraying ligand-binding or active sites (Box 3) are often the primary focus of the results and discussion sections, and therefore should be as clear as possible. However, lucidly illustrating an active site can be more difficult than showing the overall fold, because optimal renditions of binding pockets often include both low-resolution (cartoon) and high-resolution (wireframe/stick) representations (“reps”), as exemplified in Figure 2. In addition, many text labels are often required to describe a ligand-binding site in terms of particular atoms, residues, SSEs, dashed lines, and associated distances (hydrogen bonds, ionic interactions, etc.). Beyond the examples provided in this guide (Figures 2 and S3B and S3C), literature examples include (i) a combination of wireframe, stick, sphere, cartoon, and surface reps for the ligand-binding site of a hexamer (Figure 7 in [17]); and (ii) ion-binding sites in DNA portrayed via a mixture of cartoon tubes, spheres, sticks, and isocontour surfaces (Figure 4 in [18]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. The active site of Mycobacterium tuberculosis dUTPase.

The orientation of the active site of this dUTPase (PDB 1SIX; [35]) was chosen to feature the geometry of the reaction it catalyzes, specifically the in-line nucleophilic attack of water 212 on the alpha-phosphate of dUTP. The orientation was fine-tuned to eliminate overlap of side chains and to make all hydrogen bonds (dashed lines) visible. Only side chains directly involved in catalysis are depicted. Carbons are colored according to five conserved motifs of the dUTPase family. Non-conserved residues are given a less-distracting gray color and are veiled in fog. Label colors match the side chains being labeled.

https://doi.org/10.1371/journal.pcbi.1000918.g002

As a first step in binding-site figures, take care to eliminate structural features that are irrelevant to ligand binding. Begin constructing this figure by displaying only the ligand and the residues directly involved in binding (e.g., restrict atom selections to a ≈5–10 Å neighborhood about the ligand). Typically, these atoms are rendered as sticks or ball-and-sticks (Figure 2). Residues that connect the ligand binding residues can be shown as cartoon, or not at all. To reduce clutter, the four main-chain atoms (N, C_α, C, O) should not be drawn as sticks, unless they interact with the ligand too. It requires time and patience to specify which residues are shown and which are not, and this choice may be optimized more efficiently using a script rather than interactively—a script is easily edited and re-run, and makes a figure easy to reproduce in the distant future (Box 2, Rule 9). Binary files such as PyMOL sessions (“.pse” suffix) also can be saved for reuse, but are not as amenable to future modification as are (plaintext) scripts.

As with the overall fold, it is important to find an orientation of the ligand-binding site and its proximal environment that avoids overlapping side chains, ligand atoms, or other relevant structural features. It is also helpful if the orientation of the active site is similar, or at least explicitly relatable, to the canonical orientation used to portray the overall fold (see, e.g., the green arrow “attached” to a β-hairpin to help orient the viewer in Figure 7 of [19]). By default, standard colors (oxygen, red; nitrogen, blue; etc.) should be used in these atomistic illustrations; note that carbon is often colored white, gray, or green. Because of the general crowdedness of binding sites, judicious use of depth-fog or stereoviews can greatly help in depicting these structures. If shadows cast distracting patterns on important active site side chains, then consider disabling or modifying the shadowing effect.

Case 3: Structural comparison [≥novice / intermediate]

Structures are often compared by superimposing C_α atoms of two or more molecules, as illustrated in Figure 3. Properly constructed, these versatile figures can be used to convey bioinformatic information for a protein family (evolutionary distances) or structural and dynamical results for a single protein, such as conformational heterogeneity in an NMR bundle, structural changes inferred from different crystal forms, or thermal motion computed from MD simulations. The procedure occurs in two stages: establish equivalencies between positions in the two structures (A, B), and then compute the alignment so as to optimize some function (e.g., minimize sum of least-squares differences in positions of paired atoms). Note that both the calculation and graphical illustration tasks are greatly simplified if A and B are merely different conformations of the same molecule (or simple point mutants, as in Figure 3). Also, pairwise comparisons are more easily illustrated than multiple structure alignments. For multiple structures superimposed to show a progression (e.g., hinge motion), it may be helpful to assign colors in a stepwise gradient; for instance, the most closed hinge is black, the most open hinge is white, and intermediate states are gray. This scheme can also work with rainbow colors (Figure 5 in [20]). In large structures, individual domains should be colored separately. Only when the structures are highly divergent should one consider using cylinders to represent helices, as cylinder orientations are highly dependent on exact residues included in the helix definition. (Though often assigned via pre-processing in a 2° structure calculation program, note that residue structure classifications can be manually altered in PyMOL.)

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. An overlay of five simvastatin synthetase crystal structures, illustrating varying degrees of hinge closure imparted by ligand binding.

Hinge motion in this two-domain enzyme (PDB 3HLB, 3HLC, 3HLE, 3HLF, and 3HLG) is highlighted by superimposing only atoms in one of the domains (colored grey in this figure). The range of motion is highlighted by the rainbow colors assigned to the upper domain. The orientation of the molecule is chosen to make the range of motion evident (hinge axis normal to the plane of the page). Each of the structures is labeled explicitly in the figure, rather than burying the information in the figure legend. Color coding the labels makes it easier to comprehend how each ligand affects the hinge motion. The structures are represented as C_α traces rather than cartoon ribbons because the motion is relatively small, and the C_α trace allows a more exact representation of the position of the atoms.

https://doi.org/10.1371/journal.pcbi.1000918.g003

The clearest way to represent superimpositions depends on the degree of 3D similarity and the size of the structural unit. Single domains with small coordinate RMSD (<2 Å) should be superimposed as C_α backbone traces rather than cartoons, because a trace will be interpretable at this high level of similarity and is more precise than cartoon renderings. Cartoons are more helpful than backbone traces when superimposing entities (single- or multi-domain) with large RMSDs, exceeding ≈2–3 Å. The smooth cartoon eliminates some of the structural details that would otherwise obscure or distract from the relevant, larger-scale differences. In this respect, it is preferable for multi-structure alignments to use thin, spaghetti-like cartoons—these are well-suited to showing 3D alignments of homologous protein structures (Figure 1B in [21]), frames periodically sampled from MD simulations (Figure 3 in [18]), and regularly spaced (interpolated) structures generated by projection along the principal components of a dynamics trajectory (Figure 2 in [22]). Note also the strategic use of color in some of these illustrations—structural regions that are irrelevant or virtually identical can be de-emphasized by coloring them uniformly in a subdued hue (e.g., light grey domain on a white background).

Case 4: Volumetric data, surface mapping [≥intermediate]

Volumetric properties vary as a function of 3D position (r), in and around a molecule. Depending on the physical property, volumetric data may be scalar-valued (e.g., electron density, ρ(r)) or take the form of a vector field (e.g., electrostatic field). This is one of the basic difficulties of rendering volumetric data: It is continuous through space, unlike discrete molecular entities such as atoms or bonds, and is therefore less amenable to representation via geometric primitives. The other intrinsic difficulty lies in representing higher-rank tensors in 2D formats, such as a computer display or illustration. To address this well-known [4], [23] problem, volumetric data are usually represented as isovalue meshes or surfaces/contours. For instance, isocontour “chicken-wire” maps are indispensible for building a structure into electron density, and illustrating the agreement between final 3D model and crystallographic data. (Though not strictly volumetric, many “derived” physicochemical or bioinformatic properties [surface curvature, polarity, residue conservation, diffusional accessibility, etc.] can be mapped onto similar surfaces or meshes.)

Electron density is typically illustrated as a mesh surface (Figure S4A). A mesh is used rather than a solid surface so that one can evaluate the fit of the density to the underlying atoms. Appropriate contour levels are ≈1.0σ for 2F_o–F_c maps and ≈±3.0σ for F_o–F_c maps; those values should be specified in the legend. Positive contours are conventionally shown in shades of green, blue, or purple, and negative contour levels are generally colored red. When showing both positive (P) and negative (N) contours, avoid misleading the viewer by choosing values such that P = −N.

Biophysical properties are most often rendered as a gradient of colors mapped onto a molecular surface (Figures 4 and S4). The value of a given property at each point on the surface is encoded by a color that can be interpreted by a key given in the figure. A surface is used rather than a mesh, because it is continuous and so better suited to illustrate the gradient of colors, uninterrupted by the holes of the mesh. The use of a surface does, however, conceal the atoms beneath, making it difficult to know what residue lies below a given point on the surface. It is tempting to make the surface semitransparent in order to see the hidden 2° structure cartoons or stick representations. However, overlaying many objects and surfaces can be difficult for the viewer to interpret. Thus, it may be best to forego transparency and label the positions of important residues at the surface using arrows, or show a 2° structure ribbon diagram side-by-side (same scale and orientation). As with electron density, positive (P) and negative (N) isocontours of physicochemical values generally should be scaled so that P = −N (e.g., −10↔+10 k_BT/e for electrostatic potentials).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. An intermolecular interface from the M. tuberculosis PE-PPE system (PDB 2G38).

The complex between PE and PPE is shown as a ribbon cartoon (A); the two proteins are colored separately to make the interface evident. Hydrophobic side chains involved in the interface are labeled. The complementary surfaces are illustrated more clearly in (B), by splitting the complex apart like a clamshell (triangular wedge in A). Labels identify the PE and PPE proteins. Hydrophobicity is indicated by grading of cool→warm colors, as shown. Both panels are labeled and depicted on the same scale, so readers can easily see how 3D residue positions at the interface (A) correspond to apolar surface patches in (B).

https://doi.org/10.1371/journal.pcbi.1000918.g004

Case 5: Interfaces [≥intermediate]

Like ligand-binding sites, molecular interfaces are challenging to illustrate because of their inherent structural complexity. Examples are shown in Figures 4 and S2F. Difficulty arises from the need to show two or more molecular surfaces as well as stick or wireframe reps of the atoms in contact across the interface. If the interface is small or mainly one-dimensional, then a single panel is sufficient (Fig. 4D in [24]). If the interface is large or two-dimensional, multiple panels may be necessary, each illustrating a different slice through the interface (Figure 5A in [24]). To help distinguish the two molecules, the surface of one molecule should be colored differently than the other. Another alternative is to split the interface apart like a clamshell, exposing both sides of the interface to the viewer; this technique is employed in Figure 4B. Although it does not reveal surface complementarity, another approach is to show one molecule with an overlaid surface and the other as a bare cartoon in front of the first (Figure 3D [25]).

Case 6: Higher-order structures [≥intermediate/advanced]

Higher-order structures consist of many subunits assembled into a large-scale complex, possibly spanning hundreds to thousands of ångströms. Common applications include illustrations of crystal packing, large complexes (ribosomes, viruses) or multi-domain proteins (antibodies, Dscam), and biological assemblies such as cytoskeletal filaments and lipid bilayers (Figure 5). A common purpose of these images is to show the relationships between domains or molecular subunits, not atomic interactions (Figure S1). For these reasons, the molecules are typically rendered more schematically, with far less detail shown. Only for polymers of very small molecules should stick representations be used (Figure S3A in [26]; Figure S3 in [27]). Cartoon representations can be used, but care should be taken to eliminate complex shadowing; not doing so may render the ray-tracing step computationally infeasible, and may degrade the appearance of the final image by producing visual artifacts (particularly in periodic structures such as 1D polymers). Simplification should be introduced whenever possible. For example, smoothed loops and cylindrical helices may help simplify large proteins. At length scales exceeding ≈100–200 Å, entire molecules can be shown as molecular envelopes (e.g., the middle layer in Figure 3 of [28]). These envelopes can be calculated from atomic coordinates as the ASA, using an inflated probe radius so that the molecule is effectively viewed, for instance, at 20 Å resolution. Even more aggressive methods may be necessary in the ≈500–1,000 Å range, including, for instance, schematizing entire oligomers as polygonal plates and using color gradients as additional depth cues (Figure 3, [28]). Note that lighting and outlining effects can be tuned to further clarify the highly schematic renderings often necessary for higher-order structures. The sheer visual and computational complexity of higher-order structures drives the development of efficient multi-scale/multi-resolution approaches as a major current area of MolVis research. Though it lies beyond the scope and space limitations of the present tutorial, we note that this area is rapidly advancing in terms of basic algorithms and methodologies (e.g., [29], [30]) as well as practical tools (e.g., the Situs software package for multi-resolution structural work [31]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Best-fit planes.

This figure illustrates the results from computing and rendering the best-fit plane to a set of atoms. As described in the text, the plane was calculated via SVD on the phosphate atoms (orange spheres) of one leaflet of this POPC bilayer, and is rendered as a semi-transparent orange surface. The two leaflets are shown as wireframes, with a single lipid shown as CPK spheres; carbons are colored wheat in one leaflet and light grey in the other. The computed bilayer normal is drawn as an arrow, as are the two basis vectors which span the subspace defining the planar membrane.

https://doi.org/10.1371/journal.pcbi.1000918.g005

Case 7: Animations [≥intermediate]

If a picture is worth a thousand words, then a movie is worth a million. Though animations are more costly in terms of user effort and computing time (hundreds to thousands of individual images must be rendered as movie frames), nothing makes complex mechanisms clearer—see, for example, Video S1 or the helicase animations in ref. [32]. Movies fundamentally differ depending on whether the molecule is static or dynamic (Table S1)—one class shows only rigid-body motion (static coordinates, rotating camera), while the other types animate changes in atomic coordinates over frames (i.e., conformational motion). Tools like iPyMOL simplify movie-making by automating the process of interpolation between conformational states, and recent PyMOL versions provide enhanced functionality for constructing movie scenes, keyframes, and transitions. In some cases, improved 3D depth perception can be achieved by rocking back and forth 15°, while for other scenes a 360° rotation about the vertical axis may suffice. If the exact aim of the animation is quite intricate, complex modes may become necessary (rocking about multiple axes, successively or simultaneously). In general, 40 frames per revolution is sufficient; for conformational changes, 20 frames might be enough. It is helpful to have the movie end with the same frame as it starts, so that there is a smooth transition when the movie cycles to the beginning. Conformational dynamics (e.g., MD trajectories) can be combined with camera motion in the most complex type of movies, but simply making additional movies from alternative perspectives might be preferred in order to avoid viewer motion sickness. A variation on this theme is to construct the movie using scenes containing duplicate molecules that differ only in orientation (e.g., perpendicular views). The molecular motion is synchronized between the copies so that the trajectories progress identically; this is an effective means by which to visualize simultaneous events in the molecular dynamics of distant regions of a molecule (see, e.g., animations accompanying ref. [18]).

Case 8: Compute and display a best-fit plane [≥advanced]

Advanced visualization projects straddle the line between molecular modeling and molecular graphics. For example, consider the task of computing and illustrating the best-fit plane to a set of atoms. Such a problem arises in many contexts, including membrane proteins and the lipid bilayers in which they are embedded. We address this problem (Figure 5) by starting with the PDB file for a bilayer composed of 200 POPC lipids (100/leaflet); our starting structure is the bilayer slab after equilibration via MD simulations [33]. The mathematical approach of singular value decomposition (SVD) was then used to transform the atomic coordinates into a new reference frame, defined by the three basis vectors that capture the underlying geometry of the bilayer. Most importantly, these three singular vectors correspond to the bilayer normal (z-direction) and the two vectors which span the 2D plane of best fit (“best” in the sense of linear least-squares minimization of the deviation of z-coordinates of all atoms from the plane). To represent this plane, we chose to display the minimal-area rectangle that contains all the planar projections of the selected atoms (Figure 5). Computationally, this method was implemented as two stages: (i) a lower-level Python module (“svdPLoS.py”) to perform SVD on an arbitrary selection of atoms (and related manipulations, such as coordinate transformations, computing planes as linear combinations of basis vectors, drawing PyMOL compiled graphics objects, etc.); and (ii) a higher-level PyMOL macro (“svdPLoS_fig.pml”) that “wraps” the Python code to create the actual image file—from PDB input, to SVD, to final ray-tracing. The scripts used to produce the final result (Figure 5) are available from the PyMOL wiki site.