- Research article
- Open Access
A three dimensional visualisation approach to protein heavy-atom structure reconstruction
BMC Structural Biologyvolume 14, Article number: 27 (2014)
A commonly recurring problem in structural protein studies, is the determination of all heavy atom positions from the knowledge of the central α-carbon coordinates.
We employ advances in virtual reality to address the problem. The outcome is a 3D visualisation based technique where all the heavy backbone and side chain atoms are treated on equal footing, in terms of the Cα coordinates. Each heavy atom is visualised on the surfaces of a different two-sphere, that is centered at another heavy backbone and side chain atoms. In particular, the rotamers are visible as clusters, that display a clear and strong dependence on the underlying backbone secondary structure.
We demonstrate that there is a clear interdependence between rotameric states and secondary structure. Our method easily detects those atoms in a crystallographic protein structure which are either outliers or have been likely misplaced, possibly due to radiation damage. Our approach forms a basis for the development of a new generation, visualization based side chain construction, validation and refinement tools. The heavy atom positions are identified in a manner which accounts for the secondary structure environment, leading to improved accuracy.
Protein structure validation methods like MolProbity  and Procheck  help crystallographers to find and fix potential problems that are incurred during fitting and refinement. These methods are commonly based on a priori chemical knowledge and utilise various well tested and broadly accepted stereochemical paradigms. Likewise, template based structure prediction and analysis packages  and molecular dynamics force fields  are customarily built on such paradigms. Among these, the Ramachandran map , has a central role. It is widely deployed both to various analyses of the protein structures, and as a tool in protein visualisation. The Ramachandran map describes the statistical distribution of the two dihedral angles φ and ψ that are adjacent to the Cα carbons along the protein backbone. A comparison between the observed values of the individual dihedrals in a given protein with the statistical distribution of the Ramachandran map is an appraised method to validate the backbone geometry.
In the case of side chain atoms, visual analysis methods like the Ramachandran map have been introduced. For example, the Janin map  can be used to compare observed side chain dihedrals such as χ1 and χ2 in a given protein, against their statistical distribution, in a manner which is analogous to the Ramachandran map.
Crystallographic refinement and validation programs like Phenix , Refmac  and others, often utilize the statistical data obtained from the Engh and Huber library ,. This library is built using small molecular structures that have been determined with a very high resolution. At the level of entire proteins, side chain restraints are commonly derived from analysis of high resolution crystallographic structures , in Protein Data Bank (PDB) . A backbone independent rotamer library  makes no reference to backbone conformation. But the possibility that the side-chain rotamer population depends on the local protein backbone conformation, was considered already by Chandrasekaran and Ramachandran . Subsequently both secondary structure dependent , see also  and , and backbone dependent rotamer libraries , have been developed. We note that the subject remains under active investigation -.
The information content in the secondary structure dependent libraries and the backbone independent libraries essentially coincide . Both kinds of libraries are used extensively during crystallographic protein structure model building and refinement. But for the prediction of side-chain conformations, for example in the case of homology modeling and protein design, there can be an advantage to use the more revealing backbone dependent rotamer libraries.
In x-ray crystallographical protein structure experiments, the skeletonisation of the electron density map is a common technique to interpret the data and to build the initial model . The Cα atoms are located at the branch points between the backbone and the side chain. As such they are subject to relatively stringent stereochemical constraints; this is the reason why model building often starts with the initial identification of the skeletal Cα trace. The central role of the Cα atoms is widely exploited in structural classification schemes such as CATH  and SCOP , in various threading modeling techniques such as I-Tasser  and homology base approaches including SWISS-MODEL  and other related methods , in de novo approaches , and in the development of coarse grained energy functions for folding prediction . As a consequence the so-called Cα-trace problem has become the subject of extensive investigations -. The resolution of the problem would consist of an accurate main chain and/or all-atom model of the folded protein, based on the knowledge of the positions of the central Cα atoms only. Both knowledge-based approaches such and MAXSPROUT  and de novo methods including PULCHRA  and REMO  have been developed, to try and resolve the Cα- trace problem. In the case of the backbone atoms, the geometric algorithm introduced by Purisima and Scheraga , or some variant thereof, is commonly utilized in these approaches. For the side chain atoms, most approaches to the Cα trace problem rely either on a statistical or on a conformer rotamer library in combination with steric constraints, complemented by an analysis which is based on diverse scoring functions. For the final fine-tuning of the model, all-atom molecular dynamics simulations can also be utilised.
In the present article we introduce and develop new generation visualisation techniques that we hope will become a beneficial complement to existing methods for protein structure analysis, refinement and validation. We use the Cα Frenet frames , to visualise the side chain. The output we aim at, is a 3D “what-you-see-is-what-you-have” type visual map of the statistically preferred all-atom model, calculable in terms of the Cα coordinates. As such, our approach should have value for example during the construction and validation of the initial backbone and all-atom models of a crystallographic protein structure.
Our approach is based on developments in three dimensional visualisation and virtual reality, that have taken place after the Ramachandran map was introduced. In lieu of the backbone dihedral angles that appear as coordinates in the Ramachandran map and correspond to a toroidal topology, we employ the geometry of virtual spheres that surround each heavy atom. We visually describe all the higher level heavy backbone and side chain atoms on the surface of a sphere, level-by-level along the backbone and side chains, exactly in the manner how they are seen by an imaginary, geometrically determined and Cα based miniature observer who roller-coasts along the backbone and climbs up the side chains, proceeding from one Cα atom to the next. At the location of each Cα our virtual observer orients herself consistently according to the purely geometrically determined Cα based discrete Frenet frames ,. Thus the visualisation depends only on the Cα coordinates, and there is no reference to the other atoms in the initialisation of the construction. The other atoms - including subsequent Cα atoms along the backbone chain - are all mapped on the surface of a sphere that surrounds the observer, as if these atoms were stars in the sky.
At each Cα atom, the construction proceeds along the ensuing side chain, until the position of all heavy atoms have been determined. As such our maps provide a purely geometric and equitable, direct visual information on the statistically expected all- atom structure in a given protein.
The method we describe in this article, can form a basis for the future development of a novel approach to the Cα trace problem. As a complement to the existing approaches such as MAXSPROUT , PULCHRA  and REMO , the method we envision accounts for the secondary structure dependence in the heavy atom positions, which we here reveal. A secondary-structure dependent method to resolve the Cα trace problem should lead to an improved accuracy in the heavy atom positions, in terms of the Cα coordinates. In particular, since rotameric states do display clear secondary structure dependence, a fact that is sometimes overlooked in the development of rotamer libraries. The present article serves as a proof-of-concept.
Method and results
Cα based frenet frames
Let ri (i = 1,…, N) be the coordinates of the Cα atoms. The counting starts from the N terminus. At each ri we introduce the orthonormal, right-handed, discrete Frenet frame (ti, ni, bi) . As shown in Figure 1 the tangent vector t points from the center of the ith central carbon towards the center of the (i + 1)st central carbon,
The binormal vector is
The normal vector is
We also introduce the virtual Cα backbone bond (κ) and torsion (τ) angles, as follows (see in Additional file 1: Figure S1),
We identify the bond angle κ ∈ [0, π] with the latitude angle of a sphere which is centered at the Cα carbon. We orient the sphere so that the north-pole where κ = 0 is in the direction of t. The torsion angle τ ∈ [−π, π] is the longitudinal angle. It is defined so that τ = 0 on the great circle that passes both through the north-pole and through the tip of the normal vector n. The longitude angle increases towards the counter-clockwise direction around the vector t. Additional visual gain can be obtained, by stereographic projection of the sphere onto the plane. The standard stereographic projection from the south-pole of the sphere to the plane with coordinates (x, y) is given by
This maps the north-pole where κ = 0 to the origin (x, y) = (0, 0). The south-pole where κ = π is sent to infinity; see Figure 2. The visual effects can be further enhanced by sending
where f(κ) is a properly chosen function of the latitude angle κ. Various different choices of f(κ) will be considered in the sequel.
The Cα map
We first describe, how to visually characterize the Cα trace in terms of the Cα based Frenet frames (1)-(3). We introduce the concept of a virtual miniature observer who roller-coasts the backbone by moving between the Cα atoms. At the location of each Cα the observer has an orientation that is determined by the Frenet frames (1)-(3). The base of the ith tangent vector ti is at the position ri. The tip of ti is a point on the surface of the sphere (κ, τ) that surrounds the observer; it points towards the north-pole. The vectors ni and bi determine the orientation of the sphere, these vectors define a frame on the normal plane to the backbone trajectory, as shown in Figure 1. The observer uses the sphere to construct a map of the various atoms in the protein chain. She identifies them as points on the surface of the sphere that surrounds her, as if the atoms were stars in the sky.
The observer constructs the Cα backbone map as follows . She first translates the center of the sphere from the location of the ith Cα, all the way to the location of the (i + 1)st Cα, without introducing any rotation of the sphere, with respect to the ith Frenet frames. She then identifies the direction of ti+1, i.e. the direction towards the site ri+2 to which she proceeds from the next Cα carbon, as a point on the surface of the sphere. This determines the corresponding coordinates (κi, τi). After this, she redefines her orientation to match the Frenet framing at the (i + 1)st central carbon, and proceeds in the same manner. The ensuing map, over the entire backbone, gives an instruction to the observer at each point ri, how to turn at site ri+1, to reach the (i + 2)nd Cα carbon at the point ri+2.
In Figure 3 (top) we show the Cα Frenet frame backbone map. It describes the statistical distribution that we obtain when we plot all PDB structures which have been measured with better than 1.5 Å resolution, using the stereographic projection (6); in the sequel we then consider a subset with resolution better than 1.0 Å. There are presently 7548 entries measured with better than 1.5 Å resolution in PDB, and 488 entries with resolution better than 1.0 Å.
For our observer, who always fixes her gaze position towards the north-pole of the surrounding sphere at each Cαi.e. towards the red dot at the center of the annulus, the color intensity in this map reveals the probability of the direction at position ri, where the observer will turn at next Cα carbon, when she moves from ri+1 to ri+2. In this way, the map is in a direct visual correspondence with the way how the Frenet frame observer perceives the backbone geometry. We note that the probability distribution concentrates within an annulus, roughly between the latitude angle values κ ~ 1 and κ ~ 3/2. The exterior of the annulus is a sterically excluded region while the entire interior is in principle sterically allowed but not occupied in the case of folded proteins. In the figure we identify four major secondary structure regions, according to the PDB classification. These are α-helices, β-strands, left-handed α-helices and loops. In this article we will use this rudimentary level PDB classification thorough.
We imagine surrounding Cα,i with sphere, with Cα,i at the origin, we may choose the radius of the sphere to coincide with the (average) virtual covalent bond length value which is 3.8 Å in the case of Cα atoms, excluding the cis-proline. See  for a recent statistical analysis of various virtual and non-virtual variables in protein structures. The variations in the covalent bond lengths are in general minor, and in this article we do not account for deviations in covalent bond lengths from their ideal values.
We note that the visualisation in Figure 3 (top) resembles the Newman projection of stereochemistry: The vector ti which is denoted by the red dot at the center of the figure, points along the backbone from the promixal Cα at ri towards the distal Cα at ri+1. This convention will be used thorough the present article.
For comparison, we also show in Figure 3 (bottom) the standard Ramachandran map. The sterically allowed and excluded regions are now intertwined, while the allowed regions are more localized than in Figure 3 (top). We point out that the map in Figure 3 (top) provides non-local information on the backbone geometry, it extends over several peptide units, and tells the miniature observer where the backbone turns at the next Cα. As such it goes beyond the regime of the Ramachandran map, which is localized to a single Cα carbon and does not provide direct information how the backbone proceeds: The two Ramachandran angles φ and ψ are dihedrals for a given Cα, around the N - Cα and Cα - C covalent bonds. These angles do not furnish information about neighboring peptide groups.
Backbone heavy atoms
Consider our imaginary miniature observer, located at the position of a Cα atom and oriented according to the discrete Frenet frames. She observes and records the backbone heavy atoms N, C and the side-chain Cβ that are covalently bonded to a given Cα, and the O atom that is located in the peptide plane which is located after the given Cα along the backbone. In Figure 4a) - d) we show the ensuing density distributions, on the surface of the Cα centered sphere. These figures are constructed from all the PDB entries that have been measured using diffraction data with better than 1.0 Å resolution.
We note clear rotamer structures: The Cβ, C, N and O atoms are each localised, in a manner that depends on the underlying secondary structure . Both in the case of Cβ and N, the left-handed α-region (L-α) is a distinct rotamer which is detached from the rest. In the case of C and O, the L-α region is more connected with the other regions. But for C and O, the region for residues before cis-prolines becomes detached from the rest. In the case of C and Cβ we do not observe any similar isolated and localised cis-proline rotamer.
The C and O rotamers concentrate on a circular region, with essentially constant latitude angle with respect to the Frenet frame tangent vector; for the O distribution, the latitude is larger. The N rotamers form a narrow strip in the longitudinal direction, while the map for Cβ rotamers form a shape that resembles a horse shoe.
For comparison, in Figure 5 we visualise the Cβ and N distributions in the coordinate system that is utilised in REMO . In these frames, the secondary structures can be identified. But the rotamers are clearly much more delocalised than in the case of the Frenet frame map, shown in Figure 4a) and c). This delocalisation persists in the case of backbone C and O atoms (not shown). Similarly, we have found that in the case of the coordinate system of PULCHRA , the rotamers are similarly clearly more delocalised than in the Frenet frames (not shown).
One may argue that the stronger the localisation of rotamers, the more precise will structure analysis, prediction and validation become: Strong localisation enables a more precise identification of both outliers and misplaced atoms. From this perspective, the Frenet frames used here, appear to have a definite advantage over the frames used e.g. in PULCHRA and REMO.
Apparently, the secondary structure dependence of the distribution of the N, C and Cβ atoms is mainly due to the Discrete Frenet Frame. However, we have to emphasize the secondary structures also deform the very local sp3-hybridized tetrahedron structure centered on Cα with the N, C and Cβ atoms at corners. We consider the three bond angles
The ϑNC angle relates to the backbone only, while the definition of the other two involves the side chain Cβ. In Figure 6 we show the distribution of the three tetrahedral bond angles (8)-(10) in our PDB data set. We find that in the case of the two side chain Cβ related angles ϑNβ and ϑβC, the distribution has a single peak which is compatible with ideal values; the isolated small peak in Figure 6b) is due to cis-prolines. But in the case of the backbone-only specific angle ϑNC we find that in our data set this is not the case. The PDB data set we use and display in Figure 6a) shows, that there is a correlation between the ϑNC distribution and the backbone secondary structure. See also Table 1.
For example, the deviation of the Cβ atom from its ideal position is among the validation criteria in MolProbity , that uses it to identify potential backbone distortions around Cα. But several authors , have pointed out that certain variation in the values of the ϑNC can be expected, and is in fact present in PDB data. Accordingly, the protein backbone geometry does not appear to obey the single ideal value paradigm ,; we refer to ,, for extended analysis.
We remind that ϑNC pertains to the two peptides planes that are connected by the Cα. The Ramachandran angles (φ, ψ) are the adjacent dihedrals, but unlike ϑNC they are specific to a single peptide plane; the Ramachandran angles describe the twisting of the ensuing peptide plane. If the internal structure of the peptide planes is assumed to be rigid, the flexibility in the bond angle ϑNC remains the only coordinate that can contribute to the bending of the backbone. Consequently a systematic secondary structure dependence, as displayed in Figure 6, is to be expected. It could be that the lack of any observable secondary structure dependence in ϑNβ and ϑβC suggests that existing validation methods distribute all refinement tension on ϑNC.
The side chains are connected to the Cα backbone by the covalent bond between Cα and Cβ. Consequently the precision, and high level of localisation in the Cβ map as shown in Figure 4a) becomes pivotal for the construction of accurate higher level side chain maps.
Cβ at termini
We have analysed those Cβ atoms that are located in the immediate proximity of the N and the C termini in the PDB data. For this, we have considered the first two Cβ atoms starting from the N terminus, and the last two Cβ atoms that are before the C terminus. Note that in the data that describes a crystallographic PDB structure, these do not need to correspond to the actual biological termini of the biological protein. In case the termini of the biological protein can not be crystallised, the PDB data describes the first two residues after the N terminus resp. the last two residues prior to the C terminus that can be crystallised. Here we consider the termini, as they appear in the PDB data.
Recall, that the termini are commonly located on the surface of the protein. As such, they are accessible to solvent and quite often oppositely charged. It is frequently presumed that the termini are unstructured and highly flexible. They are normally not given any regular secondary structure assignment in PDB. But the Figure 7 shows that in the Cα Frenet frames the orientations of the two terminal Cβ atoms are highly regular. Their positions on the surface of the Cα centered sphere are fully in line with that of all the other Cβ atoms, as shown in Figure 4a). In particular, there are very few outliers. Moreover, the few outliers are (mainly) concentrated in a small region which is located towards the left from the β-stranded structures.
Cβ and proline
Proline is different from the other amino acids, as its side chain connects to the backbone nitrogen atom N. There is an increased propensity to form trans- peptide planes. Thus, we analyze the distribution of proline and those amino acids that are its nearest neighbors separately, in detail.
In Figure 8 we compare the individual proline contributions in our data set with the Cβ background in Figure 4a). In Figure 8a) we show the trans-proline, and in Figure 8b) we show the cis-proline. The trans-proline has a very good match with the background. There are very few outliers. These outliers are predominantly located in the same region as in Figure 7, towards the left from the main distribution i.e. towards increasing longitude. We observe that all the cis-prolines are located outside of the main Cβ distribution, towards the increasing longitude from the main distribution.
In Figure 9a)-d) we display the Cβ carbons that are located either immediately after or right before a proline. We observe the following:
In Figure 9a) we have the Cβ that are immediately after the trans-proline. The distribution matches the background, with very few outliers that are located mostly in the same region as in Figures 7, 8 i.e. towards increasing longitude. But there is a very high density peak in the figure, that overlaps with the α-helical region: We remind that proline is commonly found right before the first residue in a helix.
In Figure 9b) we display those Cβ atoms which are immediately after the cis- prolines. There is again a good match with the background. The cis-proline is relatively rare. Nevertheless, we observe an apparent increase in the number of points located in the β-stranded region. There are very few outliers, again mainly towards increasing longitude.
In Figure 9c) we have those Cβ that are right before a trans-proline. There is a clear match with the background distribution. But there are relatively few entries in the α-helical position: It is known that helices rarely end in a proline. The intensity is very large in the loop region overlapping the β-strand region (see in Additional file 1: Figure S2); we always use the classification of the secondary structure of an entry, following PDB. There are also a few outliers. Again, the outliers are mainly located in the region towards increasing longitude.
In Figure 9d) we show the Cβ distribution for residues that are right before a cis-proline. There are no entries in the background region of Figure 4a). The distribution is almost fully located in the previously observed outlier region, towards the left of the background in the figure. In addition, we observe an extension of this region towards increasing latitude, reaching all the way to the south-pole.
Finally, we demonstrate the effect of proline on the covalent tetrahedron which is centered on Cα atom. For this we recall that in Figure 4b) the region that corresponds to the effect of cis-prolines in the preceding C rotamer, is clearly visible. But in the case of Cβ and N atoms, we do not observe any similar high density isolated cis-region.
In Figure 10 we show the distribution of the three angles; see also Table 3. We observe a small deviation in the angle N - Cα - C. In comparison to proline values in Table 4, the value we find in our data set is smaller.
Cβ and histidine
Histidine has a side chain with pKa around physiological PH. But we find that its Cβ distribution is not affected by this property (see in Additional file 1: Figure S3).
We proceed upwards along the side-chain, to the level-γ heavy atoms that are covalently bonded to Cβ. Conventionally, these atoms are described by the side-chain dihedral angle χ1. This angle is determined by the three covalently bonded heavy atoms Cα, Cβ and N. The angle χ1 determines the dihedral orientation of the level-γ carbon atom, in terms of these three atoms.
We remind that ALA and GLY do not contain any level-γ atoms. In the case of ILE and VAL we have two Cγ while in the case of CYS there is an Sγ atom.
We first define a χ1-framing, where the rotamer angle χ1 appears as a dihedral coordinate. For this we introduce the following Cα based orthonormal triplet
with rα, rβ and rN the coordinates of the pertinent Cα, Cβ and N atoms, respectively. This constitutes our χ1-framing, with Cα at the origin. We introduce a sphere around Cα, oriented so that the north-pole is in the direction of tx 1. Now the dihedral χ1 coincides with the ensuing longitude angle.
In Figure 11 we show the distribution of level-γ carbon atoms. The Figure 11a) shows the distribution on the surface of the Cα centered two-sphere. In Figure 11b) we use the stereographic projection (6) with the choice
in equation (7). The three rotamers gauche ± (g±) and trans (t) have been identified in this figure. The prolines are also visible, as rotamers. In addition, in Figure 11b) we have a circle that shows the average distance of the data points from the north-pole (origin) on the stereographic plane. A number of apparent outliers are visible in Figure 11b).
We note that the underlying secondary structure of the backbone is not visible in Figure 11. This is a difference between Figures 4 and 11, in the former the underlying backbone secondary structure is visible in the density profile.
In Figure 12 we show how the Cγ atoms are seen by the observer who is located at the Cα atom, and oriented according to the backbone Frenet frames; these are the frames used in Figure 4. Now both the rotamer structure and the various backbone secondary structures are clearly seen.
Secondary structure dependent level-γ rotamers:
In the Cα Frenet frame Figure 12 the secondary structure dependence is visible. But unlike Figure 11a) the Cα Frenet frame Figure 12 lacks an apparent symmetry. This complicates the implementation of the stereographic projection, such as the one shown in Figure 11b). We proceed to introduce a new set of frames, that enables us to analyse the secondary structure dependence of the γ-level atoms in terms of the stereographic projection:
We want this frame construction method to also remain valid for higher levels of side chains. For this we introduce the following notation. Suppose our observer is located at a generic atom X. She inquires about the distribution of another atom Y. She introduces an X centered frame as follows: With Z the atom where the observer made her previous observations, we set
where rZ, rX are the coordinates of atom Z and X and tα is the tangent vector in the discrete Frenet frame.
In the case of Cγ level side chain, the atoms Z, X take Cα and Cβ, respectively. We may choose either Cα or Cβ to coincide with the origin; the Cα centered coordinate system is the original roller coasting observer while the Cβ centered coordinate system corresponds to an observer who has climbed “one-step-up” along the side chain. We map the level-γ atoms on the surface of the pertinent, surrounding two- spheres. We note that the difference between the Cα and Cβ centered distributions appears mainly in the latitude i.e. in the distance from the north-pole (see in Additional file 1: Figure S4).
In Figure 13 we have stereographically projected the distribution on the sphere, in combination with the map (14). The distribution displays clear localization, both in secondary structure and rotamer structure. The individual distributions for α-helices, β-strands and prolines are shown in Additional file 1: Figure S5, where a few outlying prolines are highlighted as examples. There are also outliers that are outside of the range of the stereographic projection in Figure 13. The projection - to the extent it has been plotted - covers a disk-like region around the north-pole i.e. around the tip of vector t in the figure. The far-away outliers can be visualised by properly rotating the sphere (see in Additional file 1: Figure S6).
Finally, we notice the fact that starting from γ-level atoms, the non-carbon heavy atoms appear in the side chain for some amino acids. However, it seems that these non-carbon heavy atoms obey the similar distributions as carbon atoms (see in Additional file 1: Figure S7).
Standard dihedral angle
We proceed upwards along the side-chain, to describe level-δ atoms. We start with a coordinate frame which is centered at the Cγ atom. We note that in the case of ILE, two alternatives exist and we choose the Cγ carbon which is covalently bonded to the Cδ atom. We start with the standard way to describe the distribution of Cδ atom. It uses the dihedral angle χ2 defined in terms of the atoms Cα, Cβ, Cγ and Cδ. Correspondingly, a frame can be defined as
In Figure 14 we show the distribution of heavy atoms in level-δ, after stereographic projection (6). The longitude in these figures coincides with the standard χ2 dihedral angle, modulo a global π/2 rotation around the center. In addition, we introduce the following version of (7)
In Figure 14, we have separately displayed the distribution of the aromatic (a) and the non-aromatic (b) amino acids; we find that starting at level-δ this is a convenient bisection. We observe that the distributions in the case of aromatic and non-aromatic side chains are different. A clear trimodal rotamer structure is present in Figure 14b). Some outliers have been highlighted with circles, as generic examples. The individual distributions for PRO and O atoms at δ-level are shown in Additional file 1: Figure S8.
However, as in Figure 12, in the backbone Frenet frames where the Cα is located at the center of the sphere, the secondary structure dependence becomes visible in the level-δ rotamers. As an example, we show in Figure 15 how some of the regions in Figure 12 are seen on the surface of the ensuing Cα centered sphere, by the roller coasting observer. The examples we have displayed are the overlap of the α-helical structures with the g − rotamer (marked α-g − in the figure) and t rotamer (α-t), and the overlap of the β-stranded structures with the g − rotamer (β-g−) and t rotamer (β-t). A secondary structure dependent trimodal rotamer structure is clearly present, in each of the distributions.
Secondary structure dependent level-δ rotamer angles
Following (15)-(17) and Figure 13 we proceed to visually inspect secondary structure dependence in the level-δ rotamers. In equations (15)-(17), the Z, X now correspond to the Cβ and Cγ atoms, respectively.
We start with the non-aromatic amino acids. In Figure 16 we show the distribution of all the Cδ non-aromatic atoms in our data set. In this figure we have also identified those apparent rotamers that are classified either as α-helical or β-stranded in PDB. The figure shows that there is clear secondary structure dependence in these rotamers. The three corresponding level-γ rotamer subsets are also labeled in Figure 16a). To see it more clearly we draw the individual distributions for the three level-γ rotamer subsets and prolines (see in Additional file 1: Figure S9). Far-away outliers also exist (not shown), these can be located and visualised by rotating the original sphere as in Additional file 1: Figure S6. For the aromatic amino acids, we show all level-δ aromatic carbons (CD1 and CD2 in PDB) in Additional file 1: Figures S10 and S11. Again, the distributions of the secondary structures are localized well.
Levels ε, ζ and η
Finally, we proceed to the ε, ζ and η levels. Following the analysis of Cγ and Cδ distributions, we use the frame (15)-(17) to visualise the secondary structure dependent distribution of these levels. For this, we introduce the analogous Cδ, Cε and Cζ frames to display the distributions of atoms at the corresponding levels.
For the Cδ frame, we choose the atoms Z, X in equation (15)-(17) to coincide with the Cγ and Cδ, respectively. Note that in the case of PHE and TYR two essentially identical choices can be made. In the case of TRP there are also two choices, and we choose the one denoted CD2 in PDB, it is covalently bonded to the higher level C atoms. In the case of HIS a framing could also be based on the level-δ N atom, but here we select the level-δ C atoms that are denoted CD2 in PDB.
In Figure 17a) - f) we show various examples of level-ε atoms. We observe that in addition of rotamers in the longitude, there are also rotamer-like variations in the latitude angle, as shown in black circles in each figure.
Similarly, we observe ζ-level atoms in Cε frame, where Z and X atoms in equations (15)-(17) take atoms Cδ and Cε, respectively. As an example, in Figure 18 we identify one rotamer. In the case of β-stranded structures we observe three rotamers. We observe that the β-stranded rotamers are not distributed evenly. The rotamers are not related to each other by (regular) 120° rotations.
Finally, we use Cζ frame to observe η-level atoms with Z and X taken as Cε and Cζ in the definition (15)-(17). As an example, the Nη2 distribution in ARG is shown in Figure 19. Now there is a very strong two-fold localisation of the distribution, shown in Figure 19a). Some of the outliers are encircled, as examples, in a).
We have utilised modern 3D visualisation techniques and advances in virtual reality to describe how to construct an entirely Cα geometry based visual library of the backbone and side chain atoms: There has been substantial progress in visualisation techniques, since the inception of the Ramachandran map. In lieu of a torus, our approach engages the geometry of a sphere and as such it has a direct “what-you- see-is-what-you-have” visual correspondence to the protein structure. In particular, we utilise the geometrically determined discrete Frenet frames of . We propose the concept of an imaginary observer, chosen so that the discrete Frenet frames determine the orientation of the observer when she roller-coasts along the backbone and climbs up the side chains. She maps the directions of all the heavy atoms on the surface of a two-sphere that surrounds her, exactly as these atoms are seen in her local frame like stars in the sky.
Since the discrete Frenet frames can be unambiguously determined in terms of the Cα trace only, we can analyse both the backbone atoms and the side chain atoms on equal footing, in a single geometric framework. This is not possible in the conventional Ramachandran approach, that assumes a priori knowledge of the peptide planes, to define the dihedral angles.
As examples of the approach, we have analysed the orientation of various heavy atoms that are located both along the backbone and in the side chains. Our approach also enables a direct, visual identification of outliers.
In particular, we have found that in terms of the discrete Frenet frames, the secondary structure dependence becomes clearly visible in the rotamer structure, both in the case of the backbone atoms and in the case of the side chain atoms. Apparently this is not always the case, in conventional approaches such as ,,:
According to  conventional secondary structure dependent rotamer libraries do not provide much more information than backbone-independent rotamer libraries. But by using the Frenet frame coordinate system chosen here, we observe that there is a clear correlation between secondary structures and rotamer positions. Thus the approach we have presented, can form a basis for the future development of a novel approach to the Cα trace problem. As a complement to existing approaches ,, the one we envision accounts for the secondary structure dependence in the heavy atom positions that we have revealed, which should lead to an improved accuracy in determining the heavy atom positions.
In this paper, we introduced a new method to visualise the heavy atom structure of a protein. In particular, our method easily detects those atoms in a crystallographic protein structure which are either outliers, or have been likely misplaced. Our approach can form a basis for the development of a new generation, visualisation based side chain construction, validation and refinement tool. Since the heavy atom positions are identified in a manner which correlates strongly with the secondary structure environment, this could lead to an improved accuracy, in particular when used in combination with existing methods.
Chen VB, Arendall WB III, Headd JJ, Keedv DA, Immormino RM, Kapral GJ, Murray LW, Richardson JS, Richardson DC: MolProbity: all-atom structure validation for macromolecular crystallography. Acta Cryst D 2010, 66: 12–21. 10.1107/S0907444909042073
Laskowski RA, MacArthur MW, Moss DS, Thornton JM: PROCHECK: a program to check the stereochemical quality of protein structures. J App Cryst 1993, 26: 283–291. 10.1107/S0021889892009944
Qu X, Swanson R, Day R, Tsai J: A guide to template based structure prediction. Curr Protein Pept Sci 2009, 10: 270–285. 10.2174/138920309788452182
Freddolino PL, Harrison CB, Liu Y, Schulten Y: Challenges in protein-folding simulations. Nature Phys 2010, 6: 751–758. 10.1038/nphys1713
Ramachandran GN, Ramakrishnan C, Sasisekharan V: Stereochemistry of polypeptide chain configurations. J. Mol. Biol. 1963, 7: 95–99. 10.1016/S0022-2836(63)80023-6
Carugo O, Carugo KD: Half a century of Ramachandran plots. Acta Cryst D 2013, 69: 1333–1341. 10.1107/S090744491301158X
Janin J, Wodak S, Levitt M, Maigret B: Conformation of amino acid side-chains in proteins. J. Mol. Biol. 1978, 125: 357–386. 10.1016/0022-2836(78)90408-4
Adams PD, Afonine PV, Bunkoczi G, Chen VB, Davis IW, Echols N, Headd JJ, Hung LW, Kapral GJ, Grosse-Kunstleve RW, McCoy AJ, Moriarty NW, Oeffner R, Read RJ, Richardson DC, Richardson JS, Terwilliger TC, Zwart PH: PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Cryst D 2010, 66: 213–221. 10.1107/S0907444909052925
Murshudov GN, Vagin AA, Dodson EJ: Refinement of macromolecular structures by the maximum-likelihood method. Acta Cryst. D. 1997, 53: 240–255. 10.1107/S0907444996012255
Engh RA, Huber R: Accurate bond and angle parameters for X-ray protein structure refinement. Acta Cryst A 1991, 47: 392–400. 10.1107/S0108767391001071
Engh RA, Huber R: Structure quality and target parameters. In: International Tables for Crystallography. Vol. F. Edited by Rossmann MG and Arnold E. Dordrecht, Kluwer Academic Publishers 2001: 382–392
Ponder JW, Richards FM: Tertiary templates for proteins: use of packing criteria in the enumeration of allowed sequences for different structural classes. J. Mol. Biol. 1987, 193: 775–791. 10.1016/0022-2836(87)90358-5
Dunbrack RL Jr: Rotamer Libraries in the 21st Century. Curr. Op. Struc. Biol. 2002, 12: 431–440. 10.1016/S0959-440X(02)00344-5
Berman HM, Westbrookm J, Feng Z, Gilliland G, Bhat TH, Weissig H, Shindyalov IN, Bourne PE: The protein data bank. Nucl. Acids Res. 2000, 28: 235–242. 10.1093/nar/28.1.235
Lovell SC, Word J, Richardson JS, Richardson DC: The penultimate rotamer library. Proteins 2000, 40: 389–408. 10.1002/1097-0134(20000815)40:3<389::AID-PROT50>3.0.CO;2-2
Chandrasekaran R, Ramachandran GN: Studies on the conformation of amino acids: XI. Analysis of the observed side group conformations in proteins. Int J Protein Res 1970, 2: 223–233. 10.1111/j.1399-3011.1970.tb01679.x
Schrauber H, Eisenhaber F, Argos P: Rotamers: to be or not to be?: an analysis of amino acid side-chain conformations in globular. J Mol Biol 1993, 230: 592–612. 10.1006/jmbi.1993.1172
Dunbrack RL Jr, Karplus M: Backbone-dependent Rotamer library for proteins application to side-chain prediction. J. Mol. Biol. 1993, 230: 543–574. 10.1006/jmbi.1993.1170
Shapovalov MS, Dunbrack RL Jr: A smoothed backbone-dependent Rotamer library for proteins derived from adaptive kernel density estimates and regressions. Structure 2011, 19: 844–858. 10.1016/j.str.2011.03.019
Islam SM, Stein R, Mchaourab H, Roux B: Rotamer library of spin labeled cysteines attached to T4 lysozyme deduced from molecular dynamics simulations constrained by double electron–electron resonance (Deer) experiments. Biophys J 2013, 104: 335A. 10.1016/j.bpj.2012.11.1862
Alexander NS, Stein RA, Koteiche HA, Kaufmann KW, McHaourab HS, Meiler J: RosettaEPR: rotamer library for spin label structure and dynamics. PloS One 2013, 8: e72851. 10.1371/journal.pone.0072851
Subramaniam S, Senes A: An energy-based conformer library for side chain optimization: improved prediction and adjustable sampling. Proteins: Struct., Funct., Bioinf 2012, 80: 2218–2234. 10.1002/prot.24111
Kirys T, Ruvinsky AM, Tuzikov AV, Vakser IA: Rotamer libraries and probabilities of transition between rotamers for the side chains in protein-protein binding. Proteins: Struct, Funct, Bioinf 2012, 80: 2089–2098.
Subramaniam S, Senes A: Backbone dependency further improves side chain prediction efficiency in the Energy-based Conformer Library (bEBL). Proteins: Struct., Funct., Bioinf 2014, 82: 3177–3187. 10.1002/prot.24685
Peterson LX, Kang X, Kihara D: Assessment of protein side-chain conformation prediction methods in different residue environments. Proteins: Struct, Funct, Bioinf 2014, 82: 1971–1984. 10.1002/prot.24552
Jones TA, Zou JY, Cowan SW, Kjeldgaard M: Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Cryst A 1991, 47: 110–119. 10.1107/S0108767390010224
Sillitoe I, Cuff AL, Dessailly BH, Dawson NL, Furnham N, Lee D, Lees JG, Lewis TE, Studer RA, Rentzsch R, Yeats C, Thornton JM, Orengo CA: New functional families (FunFams) in CATH to improve the mapping of conserved functional sites to 3D structures. Nucleic Acids Res 2013, 41(D1):D490-D498. 10.1093/nar/gks1211
Murzin AG, Brenner SE, Hubbard T, Chothia C: SCOP: A structural classification of proteins database for the investigation of sequences and structures. J. Mol. Biol. 1995, 247: 536–540.
Roy A, Kucukural A, Zhang Y: I-TASSER: a unified platform for automated protein structure and function prediction. Nature Protocols 2010, 5: 725–738. 10.1038/nprot.2010.5
Schwede T, Kopp J, Guex N, Peitsch MC: SWISS-MODEL: an automated protein homology-modeling server. Nucleic Acids Res. 2003, 31: 3381–3385. 10.1093/nar/gkg520
Zhang Y: Protein structure prediction: when is it useful? Curr Opin Struct Biol 2009, 19: 145–155. 10.1016/j.sbi.2009.02.005
Dill K, Ozkan SB, Weikl TR, Chodera JD, Voelz VA: The protein folding problem: when will it be solved? Curr Op Struct Biol 2007, 17: 342–346. 10.1016/j.sbi.2007.06.001
Scheraga HA, Khalili M, Liwo A: Protein-folding dynamics: overview of molecular simulation techniques. Ann Rev Phys Chem 2007, 58: 57–83. 10.1146/annurev.physchem.58.032806.104614
Holm L, Sander C: Database algorithm for generating protein backbone and side-chain coordinates from a Cα trace: Application to model building and detection of co-ordinate errors. Journ Mol Biol 1991, 218: 183–194. 10.1016/0022-2836(91)90883-8
DePristo MA, Bakker PIW, Shetty RP, Blundell TL: Discrete restraint-based protein modeling and the Cα-trace problem. Prot. Sci. 2003, 12: 2032–2046. 10.1110/ps.0386903
Lovell SC, Davis IW, Arendall WB III, Bakker PIW, Word JM, Prisant MG, Richardson JS, Richardson DC: Structure validation by Cα geometry: ψ, φ and Cβ deviation. Proteins 2003, 50: 437–450. 10.1002/prot.10286
Rotkiewicz P, Skolnick J: Fast procedure for reconstruction of full-atom protein models from reduced representations. Journ Comp Chem 2008, 29: 1460–1465. 10.1002/jcc.20906
Li Y, Zhang Y: REMO: A new protocol to refine full atomic protein models from C-alpha traces by optimizing hydrogen-bonding networks. Proteins 2009, 76: 665–676. 10.1002/prot.22380
Purisima EO, Scheraga HA: Conversion from a virtual-bond chain to a complete polypeptide backbone chain. Biopolymers 1984, 23: 1207–1224. 10.1002/bip.360230706
Hu S, Lundgren M, Niemi AJ: Discrete Frenet frame, inflection point solitons, and curve visualisation with applications to folded proteins. Phys. Rev. E 2011, 83: 061908. 10.1103/PhysRevE.83.061908
Lundgren M, Niemi AJ, Sha F: Protein loops, solitons, and side-chain visualization with applications to the left-handed helix region. Phys Rev E 2012, 85: 061909. 10.1103/PhysRevE.85.061909
Hinsen K, Hu S, Kneller GR, Niemi AJ: A comparison of reduced coordinate sets for describing protein structure. J Chem Phys 2013, 139: 124115. 10.1063/1.4821598
Lundgren M, Niemi AJ: Correlation between protein secondary structure, backbone bond angles, and side-chain orientations. Phys Rev E 2012, 85: 021904. 10.1103/PhysRevE.86.021904
Touw WG, Vriend G: On the complexity of Engh and Huber refinement restraints: the angle τ as example. Acta Cryst D 2010, 66: 1341–1350. 10.1107/S0907444910040928
AJ Niemi thanks A Elofsson, J Lee and A Liwo for a discussion. This research hs been supported by a CNRS PEPS Grant, Region Centre Recherche d’Initiative Academique grant, Cai Yuanpei Exchange Program, Qian Ren Grant at BIT, Carl Trygger’s Stiftelse för vetenskaplig forskning, and Vetenskapsrådet.
The authors declare that they have no competing interests.
XP and AN conceived and designed the study. XP and SH developed the analysis method. XP, AC, YZ and AN analysed the PDB data. XP and AN wrote the article. All authors read and approved the final manuscript.
Electronic supplementary material
Authors’ original submitted files for images
Below are the links to the authors’ original submitted files for images.