RCrane: semi-automated RNA model building

Abstract

RNA crystals typically diffract to much lower resolutions than protein crystals. This low-resolution diffraction results in unclear density maps, which cause considerable difficulties during the model-building process. These difficulties are exacerbated by the lack of computational tools for RNA modeling. Here, RCrane, a tool for the partially automated building of RNA into electron-density maps of low or intermediate resolution, is presented. This tool works within Coot, a common program for macromolecular model building. RCrane helps crystallographers to place phosphates and bases into electron density and then automatically predicts and builds the detailed all-atom structure of the traced nucleotides. RCrane then allows the crystallographer to review the newly built structure and select alternative backbone conformations where desired. This tool can also be used to automatically correct the backbone structure of previously built nucleotides. These automated corrections can fix incorrect sugar puckers, steric clashes and other structural problems.

Figure 1

The RNA backbone, with suite and nucleotide divisions indicated. The backbone torsions belonging to suite i are also shown.

Figure 2

The RCrane interface. (a) RCrane assists the crystallographer in tracing phosphates and bases in the electron-density map. Here, the traced nucleotides are shown in green and orange, with the nucleotide that is currently being traced shown entirely in orange. The RCrane window on the right allows the user to select alternate phosphate and base locations. (b) After the backbone has been traced, RCrane automatically builds an all-atom model of the traced nucleotides. The user may then review the newly built structure and select alternate conformers where desired.

Figure 3

Coordinates built using RCrane are highly accurate. A comparison of models built using RCrane (magenta) and published coordinates (green) is shown. Suite numbers are as indicated. Note that the structures built using RCrane have not yet undergone crystallographic refinement. (a) The GANC tetraloop from the group II intron (Toor et al., 2010 ▶). This structure was built into the 3.1 Å experimentally phased map (Toor et al., 2008 ▶) shown contoured at 3.0σ. (b) An S-turn motif from the lysine riboswitch (Garst et al., 2008 ▶) built into a 2.8 Å density map shown contoured at 1.8σ.

Figure 4

Two dimensions of the sugar-center location can be accurately predicted using only the coordinates of the 3′ and 5′ phosphates. (a) The interphosphate distance d is used to predict the radial and vertical components of the sugar-center location (r and z, respectively). (b) The radial component. Each point represents a nucleotide in the RNA05 data set, with sugar puckers as indicated. The quadratic regression (equation 1) is shown. For this regression, r ² = 0.74. (c) The vertical component. The linear regression (equation 3) is shown. For this regression, r ² = 0.78. Note that the regressions in (b) and (c) were calculated using all data points, regardless of sugar pucker.

Figure 5

The coordinates of the C1′ atom can be accurately determined after locating the sugar center. (a) The Cartesian coordinate system defined using the locations of the sugar center, 3′ phosphate and 5′ phosphate (see §3). (b) Plots showing all C1′ atoms from the RNA05 data set relative to their respective sugar center, using the axes shown in (a) with sugar puckers as indicated. The top plot shows the C1′ coordinates in the x and y axes, while the bottom plot shows the x and z axes. The mean C1′ location is shown as a yellow diamond in both plots. During backbone tracing, RCrane places new C1′ atoms using this mean location.

Figure 6

Positioning bases in density. (a) When placing bases into density, the nucleoside base is first approximated as a 5 Å vector, shown here in transparent gray. This vector is anchored at the C1′ candidate, which is shown as a cyan sphere. The vector is then replaced by a pyrimidine base, which may be computationally mutated to a purine if desired. To align the purine with the pyrimidine, the midpoint of the purine C4—C5 bond (yellow sphere) is aligned with the 5 Å vector. (b) When flipping a pyrimidine between anti and syn configurations, the base is rotated about the C1′ atom (cyan sphere) and the midpoint of the C4—C5 bond (yellow sphere). Flipping the base in this manner ensures that it is not moved out of the electron density.