Physicochemical classification of organisms

The hypervariable residues that compose the major part of proteins’ surfaces are generally considered outside evolutionary control. Yet, these “nonconserved” residues determine the outcome of stochastic encounters in crowded cells. It has recently become apparent that these encounters are not as random as one might imagine, but carefully orchestrated by the intracellular electrostatics to optimize protein diffusion, interactivity, and partner search. The most influential factor here is the protein surface-charge density, which takes different optimal values across organisms with different intracellular conditions. In this study, we examine how far the net-charge density and other physicochemical properties of proteomes will take us in terms of distinguishing organisms in general. The results show that these global proteome properties not only follow the established taxonomical hierarchy, but also provide clues to functional adaptation. In many cases, the proteome–property divergence is even resolved at species level. Accordingly, the variable parts of the genes are not as free to drift as they seem in sequence alignment, but present a complementary tool for functional, taxonomic, and evolutionary assignment.

Recent studies of live cells reveal that cytosolic crowding imposes some unique functional challenges that have previously been unconsidered. Essentially, the cytosolic proteins are not just sterically obstructive, but also interact electrostatically with one another through repulsive and attractive forces (1). These diffusive interactions are commonly referred to as “quinary interactions” (2), and their effect on the proteins largely exceeds that predicted from simplistic hard-sphere crowding models (3). The most dramatic effect of altering the quinary interactions is observed for the protein motions. A protein that normally diffuses relatively freely in the cytosolic compartment can be changed to get stuck to the intracellular surrounding by a single surface mutation (4). The principal determinant behind this effect is the protein-surface charge (4, 5). To maintain the cytosolic components suitably fluid, most biological macromolecules, like proteins, nucleic acids, and membranes, carry repulsive net-negative charge, and complete loss of this repulsion will naturally promote aggregation and functional arrest (1). However, the role of the protein charge has turned out to be more subtle than that: It modulates in detail the functional protein–protein encounters (1, 5–8). Because the strength and duration of these dynamic encounters need to be kept within certain limits for the cell to function optimally, the protein-charge decoration itself has been suggested to be under biological control (1, 4, 9–11). This very idea challenges the notion that the composition of the variable protein surfaces drifts freely and adds another dimension to protein evolution and organism fitness (12). Attention then shifts from the relatively small and highly conserved binding interfaces and active sites visible in crystal structures to the least-conserved parts of the protein surfaces exposed to the cytosolic surrounding. Proteome-wide studies of Escherichia coli confirm that there is indeed a systematic bias toward negative charge density and show also that not any negative charge density is acceptable: Proteins distribute around a moderately negative value, away from which few deviations are observed (1, 13) (Fig. 1). Similar results are obtained from measurements of isoelectric points, leading to the conclusion that the majority of soluble proteins are acidic and that the degree of this acidity varies across organisms (14–17). Classical examples are the proteomes of some halophilic archaea, with net-charge densities 10 times more negative than observed for most other organisms (18–23). Together, these findings show that the variable protein surfaces contain previously unrecognized evolutionary cues, which can be captured in terms of specific sets of physicochemical properties. The question is then whether organism identity can be deduced from physicochemical observables alone. To explore this possibility, we map here the divergence of proteome properties across organisms against the established taxonomic classification and demonstrate that the resolving power is indeed remarkably high. The results show that distinct clustering and separations of proteome properties not only follow taxonomic divisions, but also reflect their adaptation to various biotopes and functional specializations. Given that the data cover ∼18,000 organisms in all kingdoms of life, we focus below on a few representative examples of divergent optimizations and refer the specialist readers to our proteome explorer website for more specific analysis (https://proteome-explorer.herokuapp.com/).Open in a separate windowFig. 1.Outline of approach. (A) The UniProt Proteomes of Archaea, Bacteria, and Eukaryota were sampled and the physicochemical properties of each of their proteins calculated (Methods). From this dataset, containing ∼10,000 proteomes, any species can be further analyzed with respect to its detailed protein properties. (B) The distributions of protein MW and protein NCD for the proteomes of H. salinarum, E. coli, and H. sapiens. (C) Plot of the average MW and NCD values, derived from the protein distributions in B. This two-dimensional representation shows clear separation of the three species. Corresponding plots can be obtained for all UniProt Proteomes and, similarly, for all NCBI Assemblies (Methods; SI Appendix, Fig. S1; and Fig. 2). (D–F) Generally, the protein-surface charge of individual proteins follows closely that of the proteome average (SI Appendix, Fig. S19). Shown are surface charge potentials in vacuum (red for negative and blue for positive) of ribonucleotide reductase orthologs in H. salinarum (UniProt {"type":"entrez-protein","attrs":{"text":"Q9HMU3","term_id":"74568776","term_text":"Q9HMU3"}}Q9HMU3 modeled on PDB 5im3), E. coli (PDB 2xap), and H. sapiens (PDB 3hnc).