The network of sequence flow between protein structures

Leonid Meyerguz, Jon Kleinberg, Ron ElberProceedings of the National Academy of Sciences 2007 · DOI

TL;DR

This paper introduces a novel computational model that maps the "sequence flow" between protein structures, revealing a surprisingly interconnected protein space. They demonstrate that while many sequences can fold into a single structure, the transitions between different protein folds are less common but significant, suggesting a dynamic evolutionary landscape.

What Problem Does This Paper Address?

Proteins are the workhorses of biology, and their function is intimately tied to their three-dimensional structure. However, the relationship between a protein's amino acid sequence and its final folded structure is not a simple one-to-one mapping. Instead, many different sequences can fold into the same structure, a phenomenon known as "sequence capacity" or "designability." Conversely, a single sequence typically folds into a specific, stable structure. This asymmetry raises fundamental questions: How many sequences can adopt a particular fold? Can a protein's structure change through small, point mutations (sequence migration)? Understanding these dynamics is crucial for protein design, evolutionary studies, and deciphering the complex interplay between sequence, structure, and function. Previous studies often focused on individual proteins or simplified models, leaving a gap in understanding the global network of these relationships across the vast landscape of known protein structures.

Methods

The authors developed a computational model to construct a directed graph representing the flow of protein sequences between different structural folds. Each node in this graph represents a distinct protein structure (fold), and a directed edge signifies a "sequence flow" from one fold to another. This flow is defined by the probability that a sequence energetically compatible with one fold becomes more compatible with another fold after a single point mutation.

Key methodological components include:

  • Protein Data Bank (PDB) Dataset: They utilized a curated dataset of 2,060 experimentally determined protein structures from the PDB, ensuring minimal sequence identity (less than 70%) and structural redundancy (TM-score > 0.8) to represent a diverse set of protein folds.
  • THOM2 Energy Function: A simplified energy function, THOM2, was employed to estimate the energetic compatibility of sequences with protein structures. This function assigns scores based on amino acid type and its local environment.
  • Sequence Capacity (N(E)) and Competitive Sequence Capacity (C(E)) Calculation: They computed N(E), the number of sequences with energy below a certain threshold E for a given fold, and C(E), which accounts for sequences that not only have low energy for a target fold but also do not have better energies for any other competing fold. This competitive capacity is crucial for understanding sequence migration.
  • Stochastic Sampling and Markov Chains: The calculations relied on stochastic sampling of sequences and the analysis of Markov chains to estimate these capacities. The authors developed heuristics to manage the computational cost of evaluating millions of sequences.
  • Edge Definition: An edge was established between two protein folds if a significant fraction (a defined cutoff 'c') of sequences that are energetically compatible with the source fold migrate to the target fold.

Datasets

  • Protein Data Bank (PDB): The primary dataset comprised 2,060 experimentally determined protein structures. This set was carefully curated to represent a broad coverage of SCOP database classes and to minimize redundancy.
  • NR Database (for validation): The authors also used the non-redundant (NR) protein sequence database to identify experimentally observed sequences that match specific folds, providing a biological validation for their computational findings.

Key Insights and Ideas

  • A Densely Interconnected Protein Space: Contrary to a view of isolated protein folds, the study reveals a highly connected network where sequences can "flow" between structures. This suggests that protein evolution might explore a more fluid landscape than previously assumed.
  • "Sinks" Attract Sequences: The graph exhibits "sinks" – protein structures that attract a disproportionately large number of sequences from other folds. These sinks are often rich in beta-sheet structures and are correlated with protein length.
  • Migration is Less Common Than Retention: While sequence migration between folds is significant, the number of sequences that transition to a new fold is considerably smaller than the number of sequences that remain within their native fold. This explains the observed asymmetry in sequence-structure relationships.
  • Protein Length and Beta Sheets as Key Factors: The study highlights that protein length and beta-sheet content are strongly correlated with a fold's "in-degree" (its ability to attract sequences). This provides a structural basis for understanding why certain folds act as attractors.
  • Implications for Protein Evolution and Design: The findings suggest that protein evolution might leverage this interconnectedness for structural flexibility and adaptation. The ability of sequences to migrate could allow proteins to adjust their shapes in response to environmental pressures, offering a new perspective on evolutionary mechanisms.

Key Results

  • The constructed network consists of 2,060 nodes (protein folds).
  • The largest strongly connected component of the graph contains 320 structures, with the second largest having 90.
  • Proteins in the largest component are significantly longer on average (516 amino acids) compared to the overall average (260 amino acids).
  • A strong correlation (Spearman's ρ = 0.215) was found between beta-sheet content and in-degree, indicating that beta-sheet-rich proteins are better at attracting sequences.
  • The number of experimentally observed sequences matching a particular fold correlates with the number of in-edges (Spearman's ρ = 0.223), suggesting that evolutionary pressures align with the network's structure.
  • The study found that the density of sequence capacity (C(E)) shows an intriguing maximum at a contact density of approximately 1.5, suggesting an optimal structural feature for designability.

Limitations and Future Work

  • Incomplete Set of Structures: The authors acknowledge that their set of alternative structures is incomplete, as they used only gapless alignments. The inclusion of gaps could significantly increase the number of alternate structures and thus the network's density.
  • Simplified Energy Function: The THOM2 energy function is a one-body potential and is less accurate than more sophisticated models. This simplification might lead to a less dense network.
  • Computational Cost: The computational demands of evaluating millions of sequences for each fold are substantial. The authors employed heuristics to manage this, but a more comprehensive analysis with more complex models would be computationally prohibitive.

Future work could involve:

  • Exploring the impact of more accurate energy functions and the inclusion of gaps on the network structure.
  • Investigating the role of these sequence flows in specific evolutionary scenarios and protein design applications.
  • Extending the analysis to other biomolecules, such as RNA, which may exhibit similar sequence-structure dynamics.
  • Guiding experimental searches for rare sequence migration events by leveraging the insights from the computational model.

Summary updated

Abstract

Sequence-structure relationships in proteins are highly asymmetric because many sequences fold into relatively few structures. What is the number of sequences that fold into a particular protein structure? Is it possible to switch between stable protein folds by point mutations? To address these questions, we compute a directed graph of sequences and structures of proteins, which is based on 2,060 experimentally determined protein shapes from the Protein Data Bank. The directed graph is highly connected at native energies with "sinks" that attract many sequences from other folds. The sinks are rich in beta-sheets. The number of sequences that transition between folds is significantly smaller than the number of sequences retained by their fold. The sequence flow into a particular protein shape from other proteins correlates with the number of sequences that matches this shape in empirically determined genomes. Properties of strongly connected components of the graph are correlated with protein length and secondary structure.
BibTeX
@inproceedings{meyerguz2007network,
  author = {Leonid Meyerguz and Jon Kleinberg and Ron Elber},
  title = {{The network of sequence flow between protein structures}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2007},
  doi = {10.1073/pnas.0701393104},
}
Figure