Protein Variants Form a System of Networks: Microdiversity of IMP Metallo-Beta-Lactamases

Abstract

Genome and metagenome sequencing projects support the view that only a tiny portion of the total protein microdiversity in the biosphere has been sequenced yet, while the vast majority of existing protein variants is still unknown. By using a network approach, the microdiversity of 42 metallo-β-lactamases of the IMP family was investigated. In the networks, the nodes are formed by the variants, while the edges correspond to single mutations between pairs of variants. The 42 variants were assigned to 7 separate networks. By analyzing the networks and their relationships, the structure of sequence space was studied and existing, but still unknown, functional variants were predicted. The largest network consists of 10 variants with IMP-1 in its center and includes two ubiquitous mutations, V67F and S262G. By relating the corresponding pairs of variants, the networks were integrated into a single system of networks. The largest network also included a quartet of variants: IMP-1, two single mutants, and the respective double mutant. The existence of quartets indicates that if two mutations resulted in functional enzymes, the double mutant may also be active and stable. Therefore, quartet construction from triplets was applied to predict 15 functional variants. Further functional mutants were predicted by applying the two ubiquitous mutations in all networks. In addition, since the networks are separated from each other by 10–15 mutations on average, it is expected that a subset of the theoretical intermediates are functional, and therefore are supposed to exist in the biosphere. Finally, the network analysis helps to distinguish between epistatic and additive effects of mutations; while the presence of correlated mutations indicates a strong interdependency between the respective positions, the mutations V67F and S262G are ubiquitous and therefore background independent.

Metallo-β-lactamases in the Age of Multidrug Resistance: From Structure and Mechanism to Evolution, Dissemination, and Inhibitor Design

“…On the other side, if IMP-1 is an ancestor of IMP-6, the effect of mutation S262G would have been to improve the activity toward newer carbapenems at the expense of the penicillinase activity. These hypotheses heavily rely on assumptions of the evolutionary timeline of these enzymes that cannot be verified . Finally, position 262 seems to act independently of the presence of other substitutions, as similar effects were observed when comparing IMP-7 vs IMP-51 (IMP-7 S262G) and IMP-11 vs IMP-68 (IMP-11 S262G), i.e., a drop in MIC values of imipenem, ceftazidime, and cephaloridine and an increase in MIC values of newer carbapenems.…”

“…Along the IMP family, position 262 is occupied by a Ser (80%, including IMP-1) or Gly (19%, including IMP-3 and IMP-6) residue, with the exception of IMP-80 that harbors an Ala. The effect of ubiquitous substitutions, such as S262G or V67F (see below), inside the IMP family are more easily tracked when analyzing IMP variants by a network approach instead of classical phylogenetic trees …”

The scale-free nature of protein sequence space

“…In the absence of knowledge about the extant sequence space, relationships between known sequences can be measured by a metric based on global sequence identity or by neighborhood relationships in a protein sequence network where sequences form nodes that are connected by edges [ 15 , 16 ]. While pairwise sequence identity can be determined for all protein families, extended protein sequence networks only exist for families with high microdiversity such as TEM β-lactamases, which form a single connected network of more than 260 single point variants.…”

Network Analysis of Sequence-Function Relationships and Exploration of Sequence Space of TEM β-Lactamases

“…The interpretation of phylogenetic trees is highly intuitive, and they are an appropriate model for assigning sequences to families, subtypes, or clades. However, phylogenetic tree analysis will fail if all members have similar distances from each other (20). This becomes especially apparent for variants forming multiple connected quartets which cannot be represented in a binary tree assuming additivity of distances (see Fig.…”