Disentangling the complexity of low complexity proteins

Abstract

There are multiple definitions for low complexity regions (LCRs) in protein sequences, with all of them broadly considering LCRs as regions with fewer amino acid types compared to an average composition. Following this view, LCRs can also be defined as regions showing composition bias. In this critical review, we focus on the definition of sequence complexity of LCRs and their connection with structure. We present statistics and methodological approaches that measure low complexity (LC) and related sequence properties. Composition bias is often associated with LC and disorder, but repeats, while compositionally biased, might also induce ordered structures. We illustrate this dichotomy, and more generally the overlaps between different properties related to LCRs, using examples. We argue that statistical measures alone cannot capture all structural aspects of LCRs and recommend the combined usage of a variety of predictive tools and measurements. While the methodologies available to study LCRs are already very advanced, we foresee that a more comprehensive annotation of sequences in the databases will enable the improvement of predictions and a better understanding of the evolution and the connection between structure and function of LCRs. This will require the use of standards for the generation and exchange of data describing all aspects of LCRs.

Short abstract: There are multiple definitions for low complexity regions (LCRs) in protein sequences. In this critical review, we focus on the definition of sequence complexity of LCRs and their connection with structure. We present statistics and methodological approaches that measure low complexity (LC) and related sequence properties. Composition bias is often associated with LC and disorder, but repeats, while compositionally biased, might also induce ordered structures. We illustrate this dichotomy, plus overlaps between different properties related to LCRs, using examples.

Keywords: composition bias; disorder; low complexity regions; structure.

Figure 1

The LC diagram: sequence complexity composition versus periodicity. The diagram illustrates where several types of sequences would be placed in relation to two measures related to sequence complexity.

Figure 2

Shannon entropy value for each detected CBR against the CAST score normalized by the sequence length.

Figure 3

Motif graph based on SIMPLE analysis of CO1A1_HUMAN.

Figure 4

Comparison of positions detected to be of LC in the 21 proteins of our dataset. Methods SEG (in orange), CAST (in red), SIMPLE (in brown) and IUPred (in purple) were used. ANCHOR (in light blue), which includes structural aspects, is also compared.

Figure 5

LC diagram for various sequence datasets. The percentage of the top amino acid as a function of the percentage of mutations to perfect repeats calculated for a dataset of globular (GLOB), disordered (IUP) sequences as well as fragments of our protein dataset with LC character according to the SEG, CAST and SIMPLE methods.

Figure 6

Structural features of LC proteins. Venn diagram representing the FELLS prediction of dataset proteins, in four categories: secondary structure (SS), LCRs, disorder and aggregation. Each protein is assigned to a category if more than 30% of the residues in its sequence are predicted in that state.