T-RMSD: A Fine-grained, Structure-based Classification Method and its Application to the Functional Characterization of TNF Receptors

Cedrik Magis, François Stricher, Almer M. van der Sloot, Luís Serrano, Cédric NotredameJournal of Molecular Biology 2010 · DOI

TL;DR

This paper introduces T-RMSD, a novel method for classifying protein structures based on fine-grained structural differences using distance RMSD (dRMSD). T-RMSD not only recovers existing classifications but also refines them, revealing subtle structural variations that correlate with distinct functional roles, as demonstrated on TNF receptor domains and small GTPases.

What Problem Does This Paper Address?

Accurately classifying protein structures is crucial for understanding their function. While existing methods often focus on identifying broad structural "folds," they tend to overlook subtle structural variations within families that can have significant functional implications. This gap makes it difficult to assign proteins to precise functional categories, especially when experimental data is limited. The authors aim to bridge this gap by developing a method that can identify these fine-grained structural differences and link them to specific functional roles, thereby improving protein annotation.

Methods

The core of the T-RMSD method lies in its novel approach to structural clustering. Instead of relying on global structural comparisons, T-RMSD leverages the concept of distance RMSD (dRMSD). For each position in a multiple sequence alignment (MSA) of protein structures, T-RMSD calculates the differences in distances between equivalent Cα atoms for pairs of residues. These dRMSD values are then used to build a distance matrix for each alignment column.

Crucially, T-RMSD generates a separate tree for each alignment column, effectively creating a collection of "alternative clusters." These trees are then combined into a consensus tree. A key innovation is the bootstrap value assigned to each node in the consensus tree, indicating the fraction of alignment columns that support a particular split. This provides a measure of robustness for each cluster, akin to bootstrap support in phylogenetic analysis. T-RMSD is implemented as an open-source freeware within the T-Coffee package.

Datasets

The method was validated on two well-characterized protein domain families:

  • Small GTPase RAS superfamily: A dataset of 43 non-redundant protein domains with known 3D structures was used. This dataset included domains from RAB, RAS, ARF, RHO, RAN, and RGK subfamilies.
  • Cysteine-rich domains (CRDs) associated with the Tumor Necrosis Factor Receptors (TNFRs) family: A dataset of 22 non-redundant CRD structures was analyzed.

Key Insights and Ideas

  • Fine-grained structural differences matter: T-RMSD demonstrates that subtle structural variations, often missed by broader classification methods, can be directly linked to specific functional differences.
  • Bootstrap support for structural clusters: The ability to assign bootstrap values to structural clusters provides a crucial measure of reliability, allowing researchers to assess the robustness of identified subgroups.
  • dRMSD as a powerful metric: Using distance RMSD (dRMSD) allows for a structure-based comparison that is independent of superposition, making it suitable for identifying subtle variations within homologous sequences.
  • Functional implications of structural subgroups: The identified structural subgroups for both RAS GTPases and TNFR CRDs showed clear correlations with known functional distinctions, validating the biological relevance of the T-RMSD classification.
  • Refining existing classifications: T-RMSD not only recovers established classifications but also refines them, identifying new subgroups that were not apparent through sequence-based methods or broader structural analyses.

Key Results

  • RAS GTPases: T-RMSD successfully distinguished between GTP-bound and GDP-bound states of the ARF subfamily, a feat not achievable by sequence-based methods alone. This distinction was supported by a bootstrap value of 38%.
  • TNFR CRDs: T-RMSD identified three well-defined structural groups within the TNFR CRD family, with high bootstrap support (e.g., 100% support for one major split). These groups were found to be more functionally homogeneous than existing classifications.
  • HMM-based functional annotation: Hidden Markov Models (HMMs) built from the T-RMSD-derived CRD groups were highly effective in classifying unannotated CRDs, demonstrating the practical utility of the method for genome annotation. Specifically, the HMMs could distinguish between TNFL binders (Type II) and PLAD CRDs (Type I).
  • High specificity of HMMs: A leave-one-out validation showed that 100% of CRD sequences achieved a better score when aligned to the HMM corresponding to their original T-RMSD group, indicating high specificity.

Limitations and Future Work

  • Data dependency: The T-RMSD method, like any structure-based approach, is dependent on the availability of sufficient 3D structural data for the protein family of interest. The authors note that further expansion of classification might be needed for families with limited structural information.
  • Functional validation: While the study demonstrates biological relevance for two well-studied families, expanding the validation to other protein families would be beneficial. The authors acknowledge that a lack of functional evidence can be a bottleneck in assessing the biological meaning of new clusters.
  • Prioritizing structure determination: The identification of unassigned CRDs suggests potential new categories that could be explored through structure determination, which T-RMSD itself cannot directly facilitate due to its reliance on existing structures.
  • Future applications: The authors plan to apply this methodology to other important protein families, such as metabolic enzymes, protein-binding domains involved in signaling pathways, and DNA-binding proteins, to further explore its broad applicability. They also suggest that the reliability values and column-specific information could be useful for homology modeling and drug design.

Summary updated

BibTeX
@article{magis2010t,
  author = {Cedrik Magis and François Stricher and Almer M. van der Sloot and Luís Serrano and Cédric Notredame},
  title = {{T-RMSD: A Fine-grained, Structure-based Classification Method and its Application to the Functional Characterization of TNF Receptors}},
  journal = {Journal of Molecular Biology},
  year = {2010},
  doi = {10.1016/j.jmb.2010.05.012},
}
Figure