A study of the structural properties of sites modified by the O-linked 6-N-acetylglucosamine transferase

Abstract

Protein O-GlcNAcylation (O-GlcNAc) is an essential post-translational modification (PTM) in higher eukaryotes. The O-linked β-N-acetylglucosamine transferase (OGT), targets specific Serines and Threonines (S/T) in intracellular proteins. However, unlike phosphorylation, fewer than 25% of known O-GlcNAc sites match a clear sequence pattern. Accordingly, the three-dimensional structures of O-GlcNAc sites were characterised to investigate the role of structure in molecular recognition. From 1,584 O-GlcNAc sites in 620 proteins, 143 were mapped to protein structures determined by X-ray crystallography. The modified S/T were 1.7 times more likely to be annotated in the REM465 field which defines missing residues in a protein structure, while 7 O-GlcNAc sites were solvent inaccessible and unlikely to be targeted by OGT. 132 sites with complete backbone atoms clustered into 10 groups, but these were indistinguishable from clusters from unmodified S/T. This suggests there is no prevalent three-dimensional motif for OGT recognition. Predicted features from the 620 proteins were compared to unmodified S/T in O-GlcNAcylated proteins and globular proteins. The Jpred4 predicted secondary structure shows that modified S/T were more likely to be coils. 5/6 methods to predict intrinsic disorder indicated O-GlcNAcylated S/T to be significantly more disordered than unmodified S/T. Although the analysis did not find a pattern in the site three-dimensional structure, it revealed the residues around the modification site are likely to be disordered and suggests a potential role of secondary structure elements in OGT site recognition.

Fig 1. Sequence relative entropy of sites (+/- 7 residues) from 4 posttranslational modifications.

Three kinases with most sites in PhosphoSitePlus database [14] protein kinase A (PKA with 1285 sites), protein kinase C (PKC with 930 sites) and casein kinase 2 (CK2 with 742 sites). 1530 OGT sites were compiled from the same database. The sequence relative entropy was calculated with the WebLogo library [25]. Lines show mean relative entropy and the semi-transparent area represents 95% confidence intervals.

Fig 2. Diagram of the relationships of among the 5 datasets used in this work.

See Methods for details.

Fig 3. RSA of modified S/T in the SS143 dataset and unmodified S/T in same proteins.

DSSP calculated solvent accessibility was normalised by the residue theoretical maximum accessibility and the derived scores were reduced to three levels: buried (RSA ≤ 0.05), partially buried (0.05 < RSA ≤ 0.25) and exposed (RSA > 0.25) levels. The y-axis and x-axis carry the RSA levels and the RSA distribution for each level, respectively. The mean RSA is equivalent between modified and unmodified residues, at all three levels.

Fig 4. Structural superimpositions for the 10 clusters comprising 96 sites in the SS132 dataset.

Pairs of sites were superimposed on their 7 Cα atoms and the Cβ of the central S/T. Their pairwise RMSD were clustered with complete linkage and Euclidean distance. Clusters were defined by a 3 Å threshold. Green, yellow and grey represent residues in H, E, C secondary structures respectively.

Fig 5. Predicted disorder around O-GlcNAc-sites in the MSS compared to randomly selected S/T in the GS-dataset.

The y-axis shows the log₁₀ odds ratio of the between the proportion of disordered residues in the MSS dataset and the proportion of disordered residues in the GS dataset. The semi-transparent area represents 95% confidence intervals. A residue was defined as disordered according to each method’s threshold. The x-axis represents the distance in residues to the central residue which is always a S/T. DisEMBL-REM465, IUpred-short predict protein structural disorder specifically around the modification site, while the other methods predict intrinsic disorder over O-GlcNAcylated proteins. DisEMBL-REM465 shows a less pronounced increase in predicted disorder compared to the other methods.