TL;DR
This paper provides rigorous statistical evidence that the conformational preferences of amino acids in a polypeptide chain are not solely determined by their immediate neighbors independently, but rather by a coupled effect of both left and right neighbors. This finding necessitates considering at least three-residue fragments as the minimal building blocks for understanding sequence-structure relationships in proteins.
What Problem Does This Paper Address?
Proteins and peptides are fundamental to life, and their functions are intimately linked to their three-dimensional structures. The backbone dihedral angles, denoted as $\phi$ and $\psi$, are key descriptors of these structures. For decades, researchers have studied how the allowed values of these angles and their statistical distributions are influenced by the surrounding amino acid sequence. While it's generally accepted that neighboring residues play a role (known as nearest-neighbor effects), there's been ongoing debate about whether the influence of the left neighbor and the right neighbor can be considered independently. This paper aims to settle this debate by applying robust statistical methods to definitively test the interdependence of these nearest-neighbor effects.
Methods
The authors employ statistical hypothesis testing methods, specifically goodness-of-fit (GoF) tests based on the Wasserstein distance, to evaluate two main hypotheses:
- Rejection of Flory's isolated-pair hypothesis: This hypothesis suggests that the conformational preferences of an amino acid depend only on its own identity, independent of its neighbors. The authors test this by comparing the $\phi/\psi$ distributions of a central residue when its neighbors are considered versus when they are ignored.
- Assessment of interdependence between left and right neighbors: This is the core contribution. The authors use a $\chi^2$ independence test to determine if the identities of the left and right neighbors can be treated independently, given the central residue and its conformation. This involves discretizing the Ramachandran space ($\phi/\psi$ angles) and performing tests for each discretized region.
To handle the complexities of the data and the statistical tests, they developed and applied three discretization methods for the $\phi/\psi$ space. They also introduced an AUC (Area Under the Curve) score to quantify the strength of interdependence based on the p-values obtained from the independence tests.
Datasets
The study utilizes two datasets derived from experimentally determined high-resolution protein structures:
- All dataset: This dataset includes all three-residue fragments (tripeptides) from a non-redundant set of protein structures (95% sequence-identity filtered).
- Coil dataset: This dataset is a subset of the "All" dataset, specifically excluding tripeptides located within $\alpha$-helices or $\beta$-strands, focusing on coil regions.
These datasets were built from the SCOPe 2.07 release, with structures determined by X-ray crystallography and NMR experiments.
Key Insights and Ideas
- The "coupled" nature of neighbor effects: The most significant finding is the unambiguous statistical proof that the influence of left and right neighbors on a central amino acid's conformation is interdependent. This means you can't simply add up the effects of each neighbor independently.
- Three-residue fragments are minimal: This interdependence implies that three-residue fragments (tripeptides) are the smallest meaningful units for understanding sequence-structure relationships.
- Flory's hypothesis is rejected: The study provides strong statistical evidence against Flory's isolated-pair hypothesis, even in disordered (coil) regions, reinforcing the importance of sequence context.
- Physicochemical properties matter: The strength of this interdependence is influenced by the physicochemical properties (polarity and size) of the neighboring amino acids, with some properties leading to stronger interdependence than others.
- Coil regions show stronger interdependence: The analysis suggests that neighbor interdependence is stronger in coil regions compared to the overall dataset, which has implications for modeling intrinsically disordered proteins.
Key Results
- The authors found highly significant p-values (often < $10^{-10}$) when testing for the interdependence of left and right neighbors across all amino acid types and both datasets (All and Coil). This strongly rejects the hypothesis of independent neighbor effects.
- The study provides statistically strong evidence to reject Flory's isolated-pair hypothesis, with p-values consistently below the significance level of $\alpha = 0.05$.
- Analysis of physicochemical properties revealed that interdependence is generally stronger when neighbors are polar and weaker when they are large.
- Comparisons between the "All" and "Coil" datasets indicated that neighbor interdependence is stronger in coil regions.
Limitations and Future Work
- Limited to nearest neighbors: The current study focuses exclusively on the effects of residues at positions $i \pm 1$. The authors suggest that extending the analysis to more distant neighbors ($i \pm n$, where $n = 2, 3, 4, \dots$) would be a valuable future direction. However, they note that the current amount of experimental data is insufficient for such an analysis.
- Mathematical challenges: Analyzing interactions beyond nearest neighbors would require new methodological developments to address non-trivial mathematical challenges.
- Data availability: The feasibility of analyzing more distant neighbors is contingent on the increase of available experimental data and high-quality structural models.
- Implications for modeling: The findings have direct implications for molecular modeling and protein design, suggesting that algorithms should explicitly consider triplets of amino acids for more accurate conformational ensemble models and structure prediction.
Summary updated