Fig 1.
TGFBR2 kinase domain sequence diversity and pathogenic associations summarized along the linear sequence and our structural model.
A) The background color of the canonical sequence is shown, indicating extent of conservation across paralogs. Amino acid positions with known pathogenic mutations (n = 30) are marked by red circles and those with benign alterations (n = 4) in green. The protein secondary structure from our model is displayed above the sequence. B) Coloring the 3D structural model by sequence conservation is more informative than the linear sequence as the regions of conservation have spatial relationships. C) The kinase domain consists of two sub-domains; the N- and C-terminal lobes. The adenine binding site lies within a cleft between them. The locations of the 65 variants studied here are marked by spheres at each residue’s Cα position. Sites are colored red if the variant(s) at the site is annotated as pathogenic in ClinVar, HGMD, or UniProt. If it is annotated as benign by the same sources, or is manually chosen as a control, we color the site green. Sites with multiple annotations, or only disease phenotype associations, are colored orange. D) We validate the quality of our structural model using multiple algorithms (see Methods) including Ramachandran analysis; > 95% of residues within allowed regions. E) Overall model quality is evaluated on a per residue basis (e.g. Ramachandran outliers) by QMEAN with residues with a score of ≤ 1 colored in white and scaled linearly to red at a score of 5.8. F) Our TGFBR2 model adopts the typical kinase domain architecture. The N-lobe is primarily comprised of a sheet of 5 strands, while the C-lobe is mostly helical.
Fig 2.
Ligand binding site characteristic for TGFBR2 and paralogs.
A) Our TGFBR2 kinase domain model is superimposed on the experimental structures of 3 paralogs (TGFBR1, ACVR2A, and ACVR2B), emphasizing the consistency of this structural domain across the family. Each is colored by secondary structure elements, and the active site loop (from the DFG to the MAP sequence motifs; see Methods) in teal. The molecular surface of adenine from our TGFBR2 model is shown. B) Adenine binding site from our TGFBR2 model. Residues from both the N- and C-lobes make up the active site. Side chains closely interacting with the bound adenine are shown in detail. C) X-ray structure of TGFBR1 bound to an antitumor agent (3tzm). D) X-ray structure of ACVR2A with a different antitumor agent bound (3q4t). E) X-ray structure of ACVR2B with adenine bound. There are strong similarities to the core of the binding sites across paralogs.
Fig 3.
Canonical motions of the kinase domain architecture reveal sites important for functional motions.
A) The first mode of motion, or the least energetically taxing way that the kinase domain moves, corresponds to a twisting of the lobes relative to one another. B) The second mode corresponds to a coupled twisting and hinging of the lobes. C) The mobility of each amino acid within the structure can be summarized by Mean Square Fluctuation (MSF), computed from the same model. We plot the MSF of each residue, indicating sites of pathogenic mutations (red points) and benign (green). The inset shows the MSF on the 3D structure.
Fig 4.
Structure-based evaluations were used to evaluate benign (B) and pathogenic (P) mutations.
In these comparisons, benign simulations (n = 5; 4 benign variants and WT) act as negative controls. Variants within each group are summarized by a combined boxplot and density plot where width smoothly scales by the number of variants at each level of the score. A) The increase in folding energy upon mutation, ΔΔGfold, is greater for many pathogenic variants, compared to benign. B) Changes in the DFG structural motif tend to be larger in pathogenic variants, compared to benign and C) using the ligand binding site. D) A small number of variants lead to increased local fluctuations.
Fig 5.
Ligand binding site and active site loop conformational dynamics.
We choose representative sites on each side of the ligand-binding site. The distances between these sites are used as monitors of the conformation of each site. We analyzed the direct and allosteric effects of variants on these and other sites. A) The Cα atoms of residues around the ligand-binding site include F327 “above” the ligand, L386 below, and F255 “across from” the ligand, within the p-loop. B) We used Cα distances as summary metrics for the DFG conformation: N384, F398 in the center of the motif, and E290. C) For the active site distances, the three monitors give a point in a 3D space for each conformation. As the MD simulations progress, we generate a collection of such points, from which a 3D volume is generated that encompassed the densest region of data points, for each variant. The surfaces enclosing half of the sampled distances for our WT simulation, and an example pathogenic variant, C394W, are shown. The separation between the two indicates their conformational differences during our simulations. D) Benign variants have little effect on ligand binding site dynamics; the volumes spatially overlap each other and the WT simulation. E) Superposition of all pathogenic variants studied shows a wide range of conformational effects.
Fig 6.
Variants that are distant from the activation loop or the ligand binding site affect dynamics at these sites.
Variants that resulted in increased dynamics either the activation loop or the ligand binding site are indicated by spheres at their Cα atom position. The activation loop and ligand binding site are highlighted as in Fig 1. We defined an increase by values greater than those observed in benign simulations. Residues that when mutated alter dynamics at these sites are distributed throughout the structure.
Fig 7.
Application of structural metrics to simulations of observed variants with unknown functional consequences.
Many variants of uncertain significance, with conflicting annotations, or individual reports of disease associations, show alterations in structural features.
Fig 8.
Description of TGFBR2 variants using genomics-based and structure-based evaluations.
The same data as is presented in Table 1 is shown graphically. Genomics-based predictors provide predictions of damaging, while structure-based predictions test for specific mechanistic alterations.
Table 1.
Description of TGFBR2 variants using genomics-based and structure-based evaluations.