Modulators of a robust and efficient metabolism: Perspective and insights from the Rid superfamily of proteins

Abstract

Metabolism is an integrated network of biochemical pathways that assemble to generate the robust, responsive physiologies of microorganisms. Despite decades of fundamental studies on metabolic processes and pathways, our understanding of the nuance and complexity of metabolism remains incomplete. The ability to predict and model metabolic network structure, and its influence on cellular fitness, is complicated by the persistence of genes of unknown function, even in the best-studied model organisms. This review describes the definition and continuing study of the Rid superfamily of proteins. These studies are presented with a perspective that illustrates how metabolic complexity can complicate the assignment of function to uncharacterized genes. The Rid superfamily of proteins has been divided into eight subfamilies, including the well-studied RidA subfamily. Aside from the RidA proteins, which are present in all domains of life and prevent metabolic stress, most members of the Rid superfamily have no demonstrated physiological role. Recent progress on functional assignment supports the hypothesis that, overall, proteins in the Rid superfamily modulate metabolic processes to ensure optimal organismal fitness.

1. Introduction

Microbial metabolism is a complex system of interwoven pathways coordinated by an intricate, multilayered regulatory network. Decades of study have resulted in a solid understanding of the fundamental metabolic processes carried out by microbes. New and emerging tools promise continuing progress in our efforts toward a deeper understanding of this essential feature of all living cells. Studies of microbial metabolism and physiology have evolved from the pre-sequencing era, where biochemical genetic analyses were the primary means to identify and characterize the cellular components of metabolism, to modern times. The advent of sequencing and bioinformatic analyses, and their increasing accessibility, has furthered efforts to describe genome composition and define function based on homologies and established paradigms. Today, the generation of metabolic data is accelerated by the prevalence and availability of rapid global analyses (Cain et al., 2020; Downs, Bazurto, Gupta, Fonseca, & Voit, 2018; Mallick et al., 2019; Pearcy et al., 2021; Zhou et al., 2022). By incorporating multiple -omics technologies (transcriptomics, proteomics, metabolomics, etc.), one can rapidly generate large data sets that simultaneously describe the levels of transcripts, enzymes, and metabolites in a cell (Abdelhamid et al., 2019; Chen et al., 2021; Downs et al., 2018; Tang, 2011). These technologies, and the data they generate, facilitate hypothesis generation and productively guide direct biochemical and genetic experimentation on the structure and function of metabolic networks (Downs et al., 2018). Despite this remarkable progress and expansion of tools, critical questions in metabolism remain unanswered. Somewhat surprisingly, there are still many genes with no assigned function nor homology with any characterized genes, even in the best studied model organisms – e.g. Escherichia. coli (Merlin, McAteer, & Masters, 2002; Price et al., 2018), Salmonella. enterica (Blondel, Jimenez, Contreras, & Santiviago, 2009; Porwollik et al., 2014), and Saccharomyces. cerevisiae (Christie, Hong, & Cherry, 2009; Pena-Castillo & Hughes, 2007).

None-the-less, new functions and conserved paradigms in metabolism are continually discovered and defined. Such discoveries are oftentimes the result of an arduous, winding experimental process, guided by the methodical and open-minded approach that has been trusted to generate new metabolic knowledge for decades. Studies that defined the YjgF/YER057c/UK114 (Rid) protein superfamily, along with those that identified the biochemical and physiological function of the RidA subfamily, provide an example of such a journey. In this review, we summarize the features of metabolism that contribute to its complexity and describe studies on the Rid superfamily in the context of those elements.

2. Microbial metabolism is a complex system

Microbial metabolism is a complex system that consists of individual components and their interactions with one another (Bazurto & Downs, 2011; Enos-Berlage, Langendorf, & Downs, 1998; Ernst, Borchert, & Downs, 2018; Ernst, Christopherson, & Downs, 2018; Koenigsknecht & Downs, 2010; Voit, Newstetter, & Kemp, 2012). By maintaining appropriate levels of gene expression, enzyme concentration, and flux through various pathways, microbes ensure a dynamic and efficient metabolic network that is responsive to both internal and external perturbations. The robustness of microbial metabolic systems can obscure the need for some components, thus posing a challenge in studies to identify new components, biochemical strategies, and regulatory principles.

2.1. Components of metabolism

In a general sense, proteins and metabolites are the basic components of a metabolic system. Proteins can be predicted from genome composition based on our understanding of the Central Dogma. Many proteins have a known or predicted enzymatic activity, but filling gaps in our knowledge of protein function remains a challenge. Proteins encoded in the genome fall into one of three general classes: (i) known function, (ii) predicted function, or (iii) unknown function. The latter class is problematic for investigators, and functional characterization might only be initiated when a lesion in the locus encoding the protein elicits a detectable phenotype. As such, serendipity can play an outsized role in discovering the function of uncharacterized gene products. Metabolites are the second component of a metabolic system. These molecules are not directly encoded in the genome and, therefore, gaining evidence of their existence and role in the cell is not always straightforward. Identification of metabolites is the biggest challenge we face in achieving a complete understanding of the metabolic structure of any organism.

Global approaches (e.g. “-omics”) generate data that are helpful to catalog and quantify the components of metabolism. Proteomics approaches define the proteins present and offer general ideas of their relative abundance (Chen et al., 2021; Fortuin & Soares, 2022; Tang, 2011). Metabolomics approaches have the potential to further develop our understanding of the interplay between the enzymatic components of a metabolic system. Each -omics technique comes with caveats that temper the user’s ability to make rigorous conclusions. For example, though transcriptomics provides a snapshot of all mRNA present in a cell at a given time, proteomic studies are not yet able to detect all proteins in the cell, or their relative levels. Results of metabolomic analyses are influenced by extraction methods, metabolite quenching and instability of metabolites. Despite these caveats, global approaches bring an important perspective and their continuing evolution will expand the toolbox available to probe the composition of the complex system of metabolism (Downs et al., 2018).

2.2. Metabolic network structure

Beyond the protein and metabolite components lies the second and third dimensions of a metabolic system. These layers include biochemical pathways and multiple interactions between pathways and/or individual components, respectively. In total, the interacting components are appropriately visualized as a network. A phenotype represents the overall functional status of the metabolic network in an organism. In response to internal or external perturbations, the structure (i.e. functional capacity) of the metabolic network can be altered to respond appropriately. Responses to a change in network structure are mediated by regulation at the transcriptional, translational, and post-translational levels (Chubukov, Gerosa, Kochanowski, & Sauer, 2014; Litsios, Ortega, Wit, & Heinemann, 2018). In a simple scenario, flux through a relevant pathway is moderated by changing enzyme levels or enzymatic activity, in response to an environmental or intracellular signal. Unanticipated phenotypes caused by changes in a single pathway can lead to the definition of new metabolic connections and an increased understanding of metabolic network structure (Bazurto, Dearth, Tague, Campagna, & Downs, 2017; Dougherty & Downs, 2006; Downs, 2003; Ernst & Downs, 2016; Ernst, Borchert, et al., 2018; Ernst, Christopherson, et al., 2018; Koenigsknecht, Lambrecht, Fenlon, & Downs, 2012; Lewis, Cho, Knight, & Palsson, 2009; Paxhia & Downs, 2022; Ramos, Vivas, & Downs, 2008b).

3. Cellular metabolites influence physiology

Like proteins, metabolites serve many roles in the cell. However, unlike proteins, the levels of metabolites cannot be easily determined based on the status of transcription or translation. During steady state growth, when cells are in metabolic equilibrium, each metabolite is maintained at a level that ensures metabolic robustness and optimal fitness of the organism (Sander et al., 2019). Perturbation of this equilibrium can provide insights as to the influence individual metabolites exert on fitness as the cell responds to restore balance. For instance, compromising or eliminating a gene that encodes a biosynthetic enzyme can disrupt metabolic homeostasis by depleting a product and/or accumulating a substrate of the enzyme. Such a perturbation alters the capacity of the metabolic network and often causes a growth phenotype (Epelbaum et al., 1998; Ernst & Downs, 2016; Fulton, Irons, & Downs, 2022; Johnston & Roth, 1979; Lambrecht, Schmitz, & Downs, 2013; Schmitz & Downs, 2004; Trivedi et al., 2018).

Cellular metabolites have multiple regulatory roles in establishing and maintaining metabolic network structure. There is extensive literature describing a role for metabolites as co-effectors in modulating transcriptional regulation, or as signal molecules in regulatory cascades. Importantly, some metabolites modulate the activity of key enzymes via allosteric interactions. The critical role this regulatory strategy plays in maintaining metabolic equilibrium was elegantly shown by a global study in E. coli (Sander et al., 2019). Using a combination of classical genetics, proteomics, metabolomics, and mathematical modeling approaches, the study by Sander et al. queried the effects of allosteric regulation on metabolic robustness. The data showed that the enzymes of seven amino acid biosynthetic pathways are overabundant in wild-type cells. Despite this overprogramming, flux through these pathways was kept at an appropriate level by allosteric inhibition of the relevant enzymes (Sander et al., 2019). This study showed that multi-level regulation of enzyme abundance and allosteric inhibition maximized the efficiency and robustness of amino acid biosynthesis. With additional studies, it is likely that this conclusion will be extrapolated to other nodes of metabolism. Beyond the study by Sander and coworkers, enzymes insensitive to feedback inhibition have contributed to the understanding of metabolic connections and nutritional phenotypes in multiple systems, not the least of which is the RidA system reviewed herein (Bazurto & Downs, 2011; Burns, Hofler, & Luginbuhl, 1979; Csonka et al., 1988; Enos-Berlage et al., 1998; LaRossa & Van Dyk, 1987; Lee, Lee, Lee, & Kim, 2003; Rajagopal, DePonte, Tuchman, & Malamy, 1998; Schmitz & Downs, 2004; Sheppard, 1964).

Some cellular metabolites can have detrimental effects if they accumulate above desired concentrations due to a mutation or other perturbation of the metabolic system. The potential of metabolites to cause cellular damage is not always predicable a priori. The metabolite and mechanism(s) of damage or stress are typically defined during the process of dissecting metabolic phenotypes. Some metabolites cause direct damage to cellular components and, in turn, provoke global metabolic stress if they are allowed to accumulate or persist in the cell. Such stress can be exacerbated if strategies to cope with the stress have secondary consequences. A familiar example of such collateral damage is the SOS response, which is induced to cope with DNA damage, but often introduces mutations as a consequence (McKenzie, Harris, Lee, & Rosenberg, 2000). Cysteine can be toxic at high concentrations and, in S. enterica, the mechanism of the enzyme charged with detoxifying it (cysteine desulfhydrase) proceeds through a reactive metabolite, which is itself toxic (Ernst, Lambrecht, Schomer, & Downs, 2014). Also, consider molecular species such as glyoxals, which are generated as intermediates in the oxidative degradation of glucose, peroxidation of lipids, and DNA oxidation. These molecules can irreversibly damage proteins and nucleotides (Lee & Park, 2017; Singh, Barden, Mori, & Beilin, 2001; Thornalley, 1996; Thornalley, Langborg, & Minhas, 1999). When the mechanism used to detoxify reactive electrophilic species, like glyoxals, involves redox active cofactors, the detoxification can generate a redox imbalance that then impairs electron transport (Benov & Fridovich, 2002; Wang, Liu, & Wu, 2009). Formaldehyde is a toxic metabolite that non-specifically reacts with numerous cellular components of all living cells. In some organisms, including Methylorubrum extorquens, formaldehyde is generated during the metabolism of certain substrates. These organisms have developed strategies to rapidly and efficiently detoxify this molecule to prevent direct and indirect metabolic damage (Bazurto, Bruger, Lee, Lambert, & Marx, 2021). These few examples illustrate the complexity and integration of the metabolic network and the global consequences that local perturbations can have on cellular fitness.

Significantly, a major participant in the RidA paradigm described in this review is the metabolite 2-aminoacrylate (2AA). 2AA is a reactive enamine/imine species that irreversibly damages enzymes that use a pyridoxal 5’-phosphate (PLP) cofactor. PLP-dependent enzymes are essential in multiple metabolic pathways and reactions of central metabolism (Percudani & Peracchi, 2003). Accumulation of 2AA results in irreversible damage to the PLP cofactor within the active sites of numerous enzymes and thus causes global stress (Borchert, Ernst, & Downs, 2019). 2AA stress elicits diverse phenotypes that depend on the metabolic network structure around the target enzymes in a particular organism (Irons, Hodge-Hanson, & Downs, 2020). RidA proteins moderate the effects of 2AA stress, as described herein.

4. Key challenges in modeling metabolism

With the availability of genomes and the growing ease of annotation, modeling metabolism is becoming increasingly possible in many organisms (Garcia-Jimenez, Torres-Bacete, & Nogales, 2021; Karr et al., 2012; Yang et al., 2014; Zimmermann, Kaleta, & Waschina, 2021). While metabolic models can generate valuable knowledge, they are generally unable to describe subtleties of regulation and function in metabolism (Bazurto & Downs, 2016; Downs et al., 2018; Long & Antoniewicz, 2019; McGill et al., 2021; Yung et al., 2019). Features excluded from metabolic models are often those that are difficult to extract from genomic composition yet exert significant influence over metabolic structure and organismal fitness. A number of things impact the ability to extract the structure of a metabolic system from genomic information.

4.1. Redundancy complicates interpretation of phenotypes

Cells can encode proteins of similar annotated or actual function that differ in regulation or kinetics – i.e. isozymes. This functional redundancy can complicate characterization, as it is can be difficult to recapitulate the in vivo conditions that distinguish the need for the respective isozyme. For instance, aspartate aminotransferase activity can be provided by AspC (EC 2.6.1.1) or TyrB (EC 2.6.1.57) in S. enterica (Gelfand & Steinberg, 1977; Powell & Morrison, 1978) and up to three acetolactate synthases (EC 2.2.1.6) can catalyze the second step of branched chain amino acid biosynthesis (Epelbaum et al., 1998). Pseudomonas putida encodes three putative glyceraldehyde 3-phosphate dehydrogenase (GAPDH) enzymes (Gap-1, Gap-2. PP_3443). These GAPDH isozymes are differentially regulated at the transcriptional level and thus each contribute to growth on a different carbon source (Chavarria, Goni-Moreno, de Lorenzo, & Nikel, 2016). In E. coli, the catabolic (DadX) and anabolic (Alr) alanine racemases are regulated such that each is available under the appropriate conditions (Wild, Hennig, Lobocka, Walczak, & Klopotowski, 1985). Finally, cold adapted bacteria, like Pseudomonas mandelii, often encode two glucose 6-phosphate dehydrogenase isozymes – Zwf-1 and Zwf-2 (DangThu, Jang, & Lee, 2020). When present, isozymes are differentially regulated and/or have unique kinetic properties, such that the metabolic needs of the cell are met by one or both under the appropriate conditions. Some enzymes can compensate for the loss of one another, such as the case with IlvC (ketol-acid reductoisomerase, EC 1.1.1.86) which, in E. coli and S. enterica, is an enzyme in branched chain amino acid biosynthesis that also catalyzes the reaction performed by PanE (ketopantoate reductase, EC 1.1.1.169) in pantothenate biosynthesis (Primerano & Burns, 1983). Cellular networks use isozymes in multiple ways to support an efficient and robust metabolism. Despite this logical architecture, functional overlap or redundancy can complicate phenotypic analysis aimed at defining the physiological role of a gene or protein. Regulatory and kinetic details of the above examples have been previously characterized, thus in hindsight these strategies are logical. However, in the investigative stages of a project, it is difficult to assign physiological function to two proteins that catalyze the same reaction in vitro without a distinguishing phenotype in in vivo analysis. Studies of the Rid superfamily are confronted with these complications, as described below.

4.2. Lack of data compromises metabolic models

The most pressing issue for comprehensive modeling of metabolism is the need for more information. Numerous enzymes are encoded in microbial genomes, and we still lack critical information that would allow facile, accurate prediction of properties of proteins such as expression levels, kinetic parameters, substrate specificity, etc. Each of these features can significantly impact the function and robustness of a network (Link, Christodoulou, & Sauer, 2014). The complete set of instructions for configuring a dynamic metabolic network is technically present in the genome. Still, with our current understanding and predictive tools, genomic composition can be mined to suggest metabolic potential, but not the three-dimensional configuration of a metabolic network (Bazurto & Downs, 2016; Bazurto, Farley, & Downs, 2016; Borchert & Downs, 2017b). This assessment is supported by studies showing that metabolic network architecture can differ dramatically even between closely related organisms like S. enterica and E. coli (Bailey et al., 2009; Bazurto et al., 2016; Borchert & Downs, 2017b; Meysman, Sanchez-Rodriguez, Fu, Marchal, & Engelen, 2013; Winfield & Groisman, 2004).

A classical genetic approach, using rigorous molecular analyses to describe and define underlying causes of a phenotype, can uncover metabolic details and structure not expected a priori. These types of analyses are time and labor intensive but, at present, they provide the most effective means to generate new knowledge and insights into metabolic network structure (Downs et al., 2018). The unique power of this approach is illustrated in the case studies described in subsequent sections, which identified the function of the large, highly conserved Rid protein superfamily, and provided the foundation for inquiry into uncharted areas of metabolism.

5. Description of the Rid (YjgF/YER057c/UK114) superfamily of proteins

The Rid (Reactive intermediate deaminase) superfamily, originally designated YjgF/YER057c/UK114, is a large family of sequence diverse, small proteins organized into eight subfamilies based on sequence analysis and gene synteny (Niehaus et al., 2015) (Fig. 1). The wide-ranging study by Neihaus et al. identified conserved sequence motifs, both within each subfamily and across the Rid superfamily. Key residues for Rid proteins include Tyr17, Ser30, Asn88, Arg105, and Glu120 (E. coli numbering) and variation at these positions was used as evidence to differentiate between the eight subfamilies. The Glu120 residue is the only residue that is strictly conserved in all Rid proteins. The next most conserved residue is Arg105, which is present in only the RidA and Rid1, 2, and 3 subfamilies and has been shown to be essential for (and predictive of) deaminase activity in these proteins. The residue at the 105 position varies in Rid4, 5 and 6 subfamily proteins but is consistently a tryptophan in the Rid7 proteins. The Asn88 residue is only present in the RidA, Rid1 and Rid2 proteins, and varies considerably in the remaining subfamilies. The Tyr17 residue is conserved in proteins from all subfamilies except for Rid5 proteins, which have a phenylalanine residue at this position, and the Rid2 and Rid3 proteins, which have no specific amino acid at residue 17. Finally, Ser30 is conserved in all subfamilies except Rid2, while some members of Rid6 and Rid7 subfamilies have an alanine at this position. Only the Arg105 residue has been demonstrated to have implications for function in these proteins. Establishing the significance of each of these residues will be beneficial in establishing a consistent model to predict Rid protein function and to rigorously define the properties of each Rid subfamily.

Fig. 1.

Distribution, conserved residues, and structure of Rid proteins. (A) Distribution of Rid subfamily proteins in the three domains of life. RidA proteins (dark blue) are present in each domain, while the remaining Rid1–7 subfamilies (light blue, gray) are present only in bacteria. Subfamilies Rid1–3 (light blue) retain the arginine critical for deaminase activity. Subfamilies Rid4‒7 (gray) lack the active site arginine and do not have deaminase activity. (B) Conservation of key residues assigned to the active site across Rid subfamilies. Filled boxes indicate conservation of the relevant residue (columns), empty boxes indicate a residue is not conserved in the listed Rid subfamily (rows), gray boxes with text indicate the residue is consistently replaced with the given amino acid in that subfamily (Niehaus et al., 2015). (C) Structures of RidA homotrimer (PDB: 1QU9) and monomer (PDB: 3QUW), typical of the RidA subfamily. Key residues are shown in the color indicated (Gly31, blue; Asn88, yellow; Arg105, red; Glu120, orange). Structures were generated in PyMOL. Each color in the homotrimer structure represents an individual monomer. Each color in the monomer structure represents an element of the protein secondary structure: sheets (purple), helices (cyan) and loops (pink).

Early reports referencing proteins of the Rid superfamily predate the definition of the family and are now recognized to refer to the archetypal RidA subfamily, which is the only subfamily found in eukaryotes (Irons et al., 2020; Niehaus et al., 2015). The initial reports on members of this protein family date back to the mid-1990s, and continue with numerous reports published prior to functional information that resulted in our understanding of these proteins (Burman, Stevenson, Sawers, & Lawson, 2007; Ceciliani et al., 1996; Deaconescu et al., 2002; Deriu, Briand, Mistiniene, Naktinis, & Grutter, 2003; Knapik et al., 2012; Levy-Favatier et al., 1993; Manjasetty et al., 2004; Mistiniene, 2003; Mistiniene, Pozdniakovaite, Popendikyte, & Naktinis, 2005; Parsons et al., 2003; Sinha et al., 1999; Su et al., 2015; Thakur, Praveena, & Gopal, 2009; Volz, 1999; Zhang, Gao, Li, & Chang, 2010).

The first reference to a Rid protein in the literature was an acid-soluble protein isolated from the liver and kidney of rats (Levy-Favatier et al., 1993). In their studies, these authors noticed a novel, 23 kDa protein that co-extracted with the high-mobility group chromatin components that were the focus of the work. When liver and kidney extracts from rat were treated with perchloric acid, this small protein was isolated. Alignments using the amino acid sequence, deduced from cDNA, revealed this protein had similarity to the highly conserved heat-shock protein and molecular chaperone, Hsp. Based on the sequence alignment, the authors suggested a chaperone role for the small acid-soluble protein (Levy-Favatier et al., 1993). This protein was subsequently purified in a later study by another group, who noted that the small protein inhibited translation in cell free extracts. This observation led the authors to describe the protein as a translation initiation inhibitor (Oka et al., 1995). Thus, the first proposed activities for Rid proteins were as a chaperone and a translation inhibitor, designations that persist in the annotations of many genomes today, despite the lack of rigorous confirmation supporting these functional assignments.

The earliest definition of the Rid protein family appears in a study of E. coli YjgF (now RidA) as part of a structural genomics initiative (Volz, 1999). Volz identified 24 high-identity homologs of YjgF, that were collectively termed the “YjgF family.” This study predated functional characterization or definition of an active site of RidA. Based on topology, location, and sequence conservation among other members of the protein family, Volz suggested three possible active sites. Of note, one of these possible active sites included the Arg105 residue that was later shown to be critical in the active site of RidA proteins (Lambrecht, Flynn, & Downs, 2012). The Volz study found YjgF to be a homotrimer, an oligomeric configuration that has been validated by dozens of subsequent studies of RidA proteins from all domains of life and is presumed to be the active form (Irons et al., 2020). The structure of YjgF had significant similarity to the structure of Bacillus subtilis chorismate mutase, as the two proteins have the same quaternary structure and the monomers have a single domain folded into a β-sheet that is closely situated with two helices (Volz, 1999) (Fig. 1C). Based on the structure of YjgF, Volz suggested that topological similarities between YjgF and chorismate mutase were the result of convergent evolution. This author considered that YjgF (RidA) and chorismate mutase had different functions. Further, Volz suggested the function of YjgF involved amino acids and active site features not used by the chorismate mutase mechanism. Each of these assumptions were partially validated when the function of RidA was defined (Lambrecht et al., 2012).

Prior to the functional characterization of Rid protein family members, comparative analyses of multiple sequences and structures by several groups suggested critical residues for the family (Burman et al., 2007; Ceciliani et al., 1996; Deaconescu et al., 2002; Knapik et al., 2012; Manjasetty et al., 2004; Mistiniene et al., 2005; Mistiniene, Luksa, Sereikaite, & Naktinis, 2003; Parsons et al., 2003; Pu et al., 2011; Sinha et al., 1999; Thakur et al., 2009; Volz, 1999; Zhang, Gao, et al., 2010). As summarized above, alignments of amino acid sequences from RidA homologs across the domains of life identified four invariant residues presumed to be critical for function – Gly31, Asn88, Arg105, and Glu120 (E. coli numbering) (Burman et al., 2007; Deaconescu et al., 2002; Deriu et al., 2003; Knapik et al., 2012; Mistiniene et al., 2005; Parsons et al., 2003; Pu et al., 2011; Sinha et al., 1999; Thakur et al., 2009; Volz, 1999; Zhang, Gao, et al., 2010). The invariant residues are situated within the predicted active site of the protein. Only these four residues were 100% conserved in RidA proteins, but other residues comprising the active site had similar characteristics (i.e. hydrophobicity, charge, steric properties) across all homologs (Zhang, El-Hajj, & Newman, 2010).

Prompted by the lack of biochemical information on RidA proteins, several studies co-crystallized different family members with molecules considered to be potential substrates. The hope was that functional insights could be gained from these studies. The RidA proteins and their co-crystalized metabolites include Hp14.5 from Homo sapiens (benzoate), TdcF from E. coli (ethylene glycol, propionate, serine, 2-ketobutyrate), and PSTPO-PSP from Pseudomonas syringae (glucose) (Burman et al., 2007; Knapik et al., 2012; Mistiniene et al., 2005; Zhang, El-Hajj, et al., 2010). In each of the co-crystallization studies, the relevant metabolite was visibly coordinated by conserved residues that were situated in the presumptive active site. Despite widely documented structural similarity, conserved active site residues, and co-crystal structures, no function nor activity could be assigned to this large protein family in any organism from these structural data. Over time, multiple unrelated roles for RidA proteins were proposed from disparate studies with various organisms, primarily eukaryotes. The listed functions included ribonuclease (Dhawan et al., 2012; Morishita et al., 1999), translation inhibitor (Oka et al., 1995; Schmiedeknecht et al., 1996), heat-shock protein (Volz, 1999), molecular chaperone (Farkas, Nardai, Csermely, Tompa, & Friedrich, 2004; Volz, 1999), protease activator (Melloni, Michetti, Salamino, & Pontremoli, 1998). RidA homologs were also suggested to participate in purine and fatty acid metabolism (Rappu, Shin, Zalkin, & Mantsala, 1999; Sasagawa et al., 1999). In a testament to the utility of a biochemical genetic approach, only after a serendipitous observation of an in vivo activity in S. enterica was the molecular activity of a RidA protein convincingly elucidated (Browne, Ramos, & Downs, 2006; Enos-Berlage et al., 1998; Schmitz & Downs, 2004). The studies subsequent to that observation provided rigorous biochemical analysis of the RidA protein in S. enterica and benefited from the results of the prior structural studies (Lambrecht et al., 2012; Lambrecht et al., 2013).

6. The RidA protein subfamily: From unknown function to metabolic understanding

Although references to RidA (formerly YjgF) paralogs date back to the early 1990s, work leading to the functional characterization of this protein family began with an observation in S. enterica in 1998 (Enos-Berlage et al., 1998). Studies over the course of 14 years led to the demonstration that S. enterica RidA is an imine/enamine deaminase with physiologically relevant activity on 2-aminoacrylate (2AA) and 2-amino-crotonate (2AC) (Lambrecht & Downs, 2013; Lambrecht et al., 2012; Lambrecht et al., 2013). When considered in total, investigations into this protein family over 25 years present a case study on metabolic analyses that represents a triumph of in vivo genetics and metabolic detective work. The path taken to generate new metabolic knowledge and highlight key questions about the Rid superfamily has been serendipitous and meandering, as described in the sections below. Similar winding paths through metabolism have been taken, and will continue to be critical, in efforts to define the biochemical function and physiological role of unknown proteins. Without knowledge or functional predictions, the complexities of metabolism challenge our ability to define the activities and understand the roles of proteins of unknown function, some of which are conserved across many or all microorganisms.

6.1. Loss of RidA can be beneficial or detrimental

The first reported mutation in the ridA gene of a bacterium was an insertion mutation that restored thiamine synthesis to a purF mutant of S. enterica (Enos-Berlage et al., 1998) (Fig. 2A). Thiamine synthesis was used over the years to probe metabolic integration, and numerous mutations that similarly restored thiamine synthesis to a purF mutant had been isolated (Koenigsknecht et al., 2012). Each of the previous mutations acted by diverting metabolic flux from tryptophan, histidine or phosphoribosyl pyrophosphate (PRPP) biosynthetic pathways to allow PurF-independent synthesis of phosphoribosylamine (PRA) (Downs & Roth, 1991; Koenigsknecht, Fenlon, & Downs, 2010; Ramos et al., 2008a, 2008b). The lesions in ridA stood out from the other suppressors because they were, (i) more frequent and (ii) null alleles. When a ridA insertion mutation was transduced into a wild-type genetic background, routine nutritional analysis found no growth defect on minimal medium, but a sensitivity to the addition of serine (Enos-Berlage et al., 1998) (Fig. 2B). The data showed that loss of RidA had both a positive (i.e. restored thiamine synthesis) and a negative effect (i.e. sensitivity to serine) on the growth of S. enterica. Thus, to be acceptable, any biochemical function proposed for this protein had to explain both effects.

Fig. 2.

Loss of RidA can have positive or negative effects. (A) Loss of ridA alleviates thiamine requirement in a purF background. A purF mutant (empty circles) and a ridA purF double mutant (filled circles) grown in minimal medium with adenine to satisfy the purine requirement. Thiamine synthesis in the ridA background is allowed by PurF-independent synthesis of phosphoribosylamine, as described in the text. The purine/thiamine biosynthetic pathways are schematically represented below the growth data for context. A strain lacking purF requires supplementation with both thiamine and purines under the conditions used. (B) ridA mutants exhibit sensitivity to serine that is alleviated by isoleucine. Growth is shown for wild type (closed triangles) and a ridA mutant in minimal medium supplemented with serine (closed squares), as well as a ridA mutant in minimal medium supplemented with serine and isoleucine (open circles) (Enos-Berlage et al., 1998; Schmitz & Downs, 2004). Abreviations: PRA, phosphoribosylamine; PRPP, phosphoribosyl pyrophosphate; Gln, glutamate; AIR, aminoimidazole ribotide.

A key finding was that the addition of isoleucine to the medium reversed the two effects of the ridA mutation – growth in the presence of serine was restored, and growth of the purF mutant in the absence of thiamine was eliminated. Prompted by two reports describing defects in isoleucine synthesis in S. cerevisiae strains lacking the RidA homolog Mmf1 (Kim, Hiroshi, & Shirahige, 2001; Oxelmark et al., 2000), transaminase B (IlvE, EC: 2.6.1.27) activity was assayed in S. enterica ridA mutants and found to be impaired (Schmitz & Downs, 2004). Importantly, the specific activity of IlvE was reduced via a post-translational mechanism in the ridA mutant. Exogenous isoleucine restored the activity of IlvE, just as it had ameliorated the other effects of a ridA mutation. Genetic analysis using a characterized variant of IlvA (serine/threonine dehydratase, EC: 4.3.1.19) determined that isoleucine exerted its effect in ridA mutants via allosteric inhibition of IlvA (Chen, Fonseca, Hannun, & Voit, 2013; LaRossa & Van Dyk, 1987; Schmitz & Downs, 2004). In total, data from these disparate studies were incorporated into the general model in which an IlvA reaction product, but not the accepted 2-ketobutyrate (2KB), mediated ridA phenotypes and the RidA protein neutralized its effect (Schmitz & Downs, 2004). This model proved to be prophetic, despite a lack of understanding of the metabolites involved or mechanism by which the presumed metabolite acted.

The results and working model described above were the foundation for multiple sequential studies, each of which resulted in valid, but incomplete, conclusions about the role of the RidA protein. A study that connected tryptophan biosynthesis and PRA generation gave the first hint of how PRA was being synthesized in a ridA mutant (Ramos & Downs, 2003). These studies found that, in the absence of RidA, the tryptophan biosynthetic enzyme complex anthranilate synthase-phosphoribosyl transferase (TrpDE, EC: 4.1.3.27) catalyzed PRA formation (Browne et al., 2006). Subsequent efforts confirmed synthesis of PRA could be achieved by IlvA, TrpD, PRPP and threonine in vitro. These experiments recapitulated what occurred in vivo in a ridA mutant and further, suggested there was an endogenous metabolic difference between ridA and wild-type strains. Importantly, when RidA protein was added to the in vitro reaction mixture, synthesis of PRA by TrpD was prevented (Lambrecht, Browne, & Downs, 2010). These observations were extended by in vitro experiments that defined the TrpD/2AC-dependent mechanism of PRA synthesis and the role RidA played in preventing it (Fig. 3). Simply, the 2-aminocro-tonate product generated by IlvA reacts with PRPP in the active site of TrpD, resulting in a predicted phosphoribosyl-enamine adduct that can be hydrolyzed to PRA (Lambrecht & Downs, 2013). With these data, the molecular mechanism of thiamine synthesis that caused the initial phenotype of a ridA mutant was defined. The question arose as to whether the TrpD-dependent pathway for PRA formation was a metabolic artifact in a S. enterica ridA mutant, or if this pathway might be found in another organism as part of its native metabolic structure. Unexpectedly, evidence showed this pathway exists in E. coli and operates in the presence of a functional RidA. Up to 50% of the PRA used for thiamine synthesis in E. coli required the activity of threonine dehydratase (IlvA) and anthranilate synthase component II (TrpD) (Bazurto et al., 2016). This example illustrates differences in metabolic network structure. E. coli and S. enterica share all cellular components required for the TrpD-dependent PRA pathway, yet it functions under different conditions in these two organisms. It remains to be determined what modulates these differences – potentially amino acid pool size variation, subtle differences in enzyme kinetics, etc. The conservation of RidA across the domains of life implied that a role in this non-conserved biosynthetic pathway was not the primary function for which this protein had been maintained.

Fig. 3.

PurF-independent synthesis of phosphoribosylamine requires TrpD, IlvA, threonine and PRPP. IlvA generates an enamine/imine intermediate from threonine, which is then combined with PRPP by TrpD to generate PRA. If present, RidA removes the enamine/imine species and prevents IlvA and TrpD-mediated PRA synthesis (Lambrecht & Downs, 2013). Abreviations: PRA, phosphoribosylamine; PRPP, phosphoribosyl pyrophosphate; IlvA; serine threonine dehydratase; TrpD; anthranilate phosphoribosyl transferase; RidA, reactive intermediate deaminase; Thiamine-PP, thiamine pyrophosphate.

6.2. RidA deaminates reactive enamines/imines

Results from the studies outlined above suggested that RidA acted, by an undefined mechanism, on a product of IlvA before said product: (i) was captured by TrpD to generate PRA (Lambrecht & Downs, 2013), (ii) contributed to the reduced activity of IlvE (Schmitz, 2006), and (iii) prevented growth in the presence of serine. The decreased IlvE activity was not responsible for the growth defect of a ridA mutant in the presence of serine, consistent with the demonstration that biosynthetic pathways are often overprogrammed – Section 3 (Sander et al., 2019). A closer look at the mechanism of IlvA shed light on the process occurring in a ridA mutant. In the first step of isoleucine biosynthesis, IlvA catalyzes the dehydration of threonine to the 2-aminocrotonate enamine (2AC), which tautomerizes to an imine before being hydrolyzed by solvent water, forming 2KB and free ammonia (Chargaff & Sprinson, 1943; Phillips & Wood, 1965). IlvA similarly dehydrates serine to the respective enamine 2-aminoacrylate (2AA), which likewise tautomerizes to an imine with the final hydrolysis yielding pyruvate and ammonia. Prior in vitro studies showed that enamine/imine species were obligatory intermediates in these reactions, and others catalyzed by some PLP-dependent enzymes (Walsh, 1984). Notably, the in vitro studies did not suggest, nor expect, the enamine would leave the active site of the enzyme. The instability of these molecules results in a short half-life in water, leading to the presumed role of free water in the final deamination of enamine/imine species. Lambrecht et al. (2012) made the astute suggestion that RidA acted on enamine/imine metabolites and proposed these molecules could accumulate in vivo in the absence of this protein. This scenario required that the cellular milieu be distinct from the aqueous solutions used for in vitro biochemical studies. This clever insight transformed our understanding of the Rid superfamily and emphasized the benefit of integrating results from both in vitro and in vivo experimentation. Data from experiments using a purified system determined that RidA increased the rate of IlvA-mediated 2KB formation from threonine. More specifically, RidA increased the rate of product formation by acting on the enamine/imine intermediate generated by IlvA (Lambrecht et al., 2012). This result allowed the conclusion that enamine/imine products were released from IlvA and were free in solution. Finally, the data showed that RidA was more efficient than solvent water in deaminating the enamine/imine intermediates to the ketoacid product.

In total, these data established, for the first time, a biochemical activity associated with the RidA protein from S. enterica and, as such, laid the foundation for continuing studies on this protein family. An assay was implemented to quantify the rate of deamination IlvA-derived enamine/imine products from threonine and serine – 2AC and 2AA, respectively. PLP-dependent cysteine desulfhydrase, CdsH, can also be used to generate the 2AA metabolite substrate of RidA in situ (Ernst et al., 2014). Analysis of RidA variants in these assays showed that the conserved arginine residue (Arg105 in S. enterica) is essential for deaminase activity. Consistent with this notion, a RidA^R105A variant was unable to complement a ridA mutant when provided in trans (Lambrecht et al., 2012). Models of the active site and a proposed catalytic mechanism of RidA have been reviewed (Borchert, Ernst, et al., 2019). RidA proteins from across domains had enamine/imine deaminase activity in these assays (Lambrecht et al., 2012), which continue to be used to characterize Rid proteins.

Following the identification of Rid protein deaminase activity, other assays were implemented to facilitate efforts to identify potential Rid substrates. PLP-dependent dehydratases generate a limited number of enamines, but NAD-/FAD-dependent amino acid oxidases generate imine intermediates from a wide array of amino acid substrates, which are then deaminated by water to the corresponding ketoacid (Hafner & Wellner, 1979; Niehaus et al., 2014). Analysis of Rid protein deaminase activity is facilitated by the broad substrate specificity of such oxidases, since it allows commercially available l-amino acid oxidase (LOX) to be used (Digiovanni, Degani, Popolo, & Vanoni, 2021; Niehaus et al., 2015) (Fig. 4). The addition of semicarbazide to the reaction mixtures is key to the LOX-based deaminase assay. Semicarbazide reacts with imines to generate a semicarbazone compound that can be quantified spectrophotometrically at a wavelength of ~248 nm. In general, these assays do not monitor the ketoacid product of deamination but rather, measure the ability of a Rid protein to compete with semicarbazide for the relevant imine substrate. In this way, activity of a Rid protein is detected as a decrease in the rate of semicarbazone formation. The ability of LOX to generate imines from a wide variety of amino acids has been used in efforts to define the substrate specificity of different Rid proteins (Buckner, Lato, Campagna, & Downs, 2021; Fulton et al., 2022; Hodge-Hanson & Downs, 2017). The LOX assay will be useful in studies to define active site residues that determine said specificity. At present, a limited number of studies have analyzed substrate specificity with Rid proteins. Data from those few studies suggest these in vitro experiments could provide the means to validate and improve the current subfamily classification of Rid proteins moving forward (Fulton et al., 2022; Hodge-Hanson & Downs, 2017).

Fig. 4.

Imine/enamine deaminase assays. (A) Enamine/imine intermediates are generated in the β-elimination of amino acids by PLP-dependent enzymes. The enamine/imine metabolites are deaminated by solvent water, or a capable Rid protein. Deaminase activity of a Rid protein is determined by the rate of ketoacid formation in the presence of absence of said Rid protein. (B) Imines are generated from amino acids via oxidation about the α-carbon by FAD- or NAD-dependent enzymes (e.g. commercially available l-amino acid oxidase, LOX). Imine metabolites will react with added semicarbazide to form a semicarbazone compound (green) that can be detected spectrophotometrically at a wavelength of 248 nm. The imines can also be deaminated by solvent water, or by a capable Rid protein. In this assay, deaminase activity of a Rid protein is measured as the decrease in the rate of semicarbazone formation in the presence of said protien. A decrease would be due to competition between the Rid protein and the semicarbazide for the imine substrate.

In one or more of the above assays, RidA paralogs from at least, E. coli, B. subtilis, Y. pestis, P. aeruginosa, Campylobacter jejuni, Pyrococcus furiosus, Saccharomyces cerevisiae, Arabidopsis thaliana, Zea mays, Cucimus sativus, Homo sapiens, Capra hircus, Dermatophagoides farina, and Salmo salar have deaminase activity in vitro (Degani, Barbiroli, Regazzoni, Popolo, & Vanoni, 2018; ElRamlawy et al., 2016; Ernst & Downs, 2018; Fulton et al., 2022; Irons, Hodge-Hanson, & Downs, 2018; Irons, Sacher, Szymanski, & Downs, 2019; Lambrecht et al., 2012; Martínez-Chavarría et al., 2020; Whitaker, Ernst, & Downs, 2021). To our knowledge, every Rid protein that contains the critical arginine residue has enamine/imine deaminase activity on one or more substrates in vitro. Based on these data, genomic annotations would be improved if they included the prediction of deaminase activity for those Rid family members containing the relevant arginine (Arg105). It is equally important that Rid family members lacking the active site arginine are annotated to indicate they do not have imine/enamine deaminase activity.

7. 2-Aminoacrylate is an endogenously generated metabolite stressor

In the physiological context of its discovery, RidA deaminated 2-aminocrotonate (2AC) generated from threonine by IlvA to prevent it from allowing PurF-independent PRA formation (Fig. 3) (Enos-Berlage et al., 1998; Lambrecht et al., 2012). The conservation of RidA hinted that deamination of 2AC was not the function of the protein that drove its retention in so many genomes across all domains of life. The sensitivity of a ridA mutant of S. enterica to exogenous serine led to a proposed role that would be more broadly relevant. Genetic data determined that activity of the PLP-dependent serine/threonine dehydratase, IlvA, was required for the serine-sensitive phenotype. Together these data led to a hypothesis that formation of the serine-derived enamine, 2AA, would be increased by the exogenous serine and that the reactive intermediate would accumulate in a ridA mutant. Finally, accumulation of 2AA was suggested to be the root cause of the serine induced lack of growth.

7.1. Enzymatic generation and biochemical consequences of 2-aminoacrylate

The suggestion that 2AA accumulation could negatively impact growth was prompted by previous in vitro studies. Historically, 2AA was used as a suicide inhibitor to probe the catalytic mechanism of some PLP-dependent enzymes (Bisswanger, 1981; Kishore, 1984; Tate, Relyea, & Meister, 1969; Walsh, 1982; Whalen, Wang, & Berg, 1985). In these studies, amino acid substrates with a strong electronegative leaving group (e.g. -Cl) bound to the β-carbon were provided to relevant PLP-dependent enzymes in vitro. Many of the queried enzymes had promiscuous β-elimination activity. In this case, 2AA generated during the β-elimination reaction attacks the PLP cofactor, covalently inactivating the enzyme by one of two mechanisms (Eliot & Kirsch, 2004; Likos, Ueno, Feldhaus, & Metzler, 1982; Ueno, Likos, & Metzler, 1982). This type of in vitro study provided precedent for the notion that 2AA could damage PLP-dependent enzymes in the cell and generate the phenotypes characteristic of a ridA mutant. Initially, there seemed to be two obstacles to extrapolating the in vitro results to a model for in vivo behavior of 2AA. First, the in vitro studies mentioned above involved 2AA attacking the same active site in which it was generated. These studies offered no indication that 2AA could leave the active site of one enzyme and enter the active site of another. In contrast, a suitable model for the in vivo situation required that 2AA leave the active site of IlvA, where it was generated from serine, and enter the active site of target enzyme(s) elsewhere in the cell. Second, given a reported half-life of 1–3 s, it was counterintuitive that the cellular milieu would allow persistence of 2AA to a level that would be detrimental (Burns et al., 1979; Flavin & Slaughter, 1964). Each of these concerns were lessened by data generated with targeted in vitro experiments. Data from these experiments showed that 2AA could leave the active site of one enzyme and attack PLP in the active site of another. These data were critical in the definition of the cellular role of RidA (Lambrecht et al., 2013). In the key experiments, a 2AA generator (IlvA) and a presumed target enzyme (IlvE) were combined with or without serine (the precursor to 2AA) in a reaction mixture. IlvE is a PLP-dependent transaminase, and was considered a potential target for 2AA damage in vivo based on phenotypic results (Lambrecht et al., 2013; Schmitz & Downs, 2004). Transaminase activity of IlvE was assayed after incubation with the 2AA-generating components. Data from these experiments showed that activity of IlvE was decreased following exposure to both serine and IlvA. Furthermore, IlvE was shown to be targeted by 2AA based on mass analysis of the protein before and after incubation with the 2AA-generating system. Using Mass Spectrometry, a peak consistent with a modified IlvE protein was found in the samples with serine and IlvA, but not from those without serine added (Lambrecht et al., 2013; Shen, Borchert, & Downs, 2022). Results from this study successfully recapitulated the in vivo mechanism of 2AA stress suggested for a ridA mutant. Excitingly, when RidA protein was added to the reaction mixture, the in situ generation of 2AA by IlvA and serine did not decrease the activity of IlvE. Together, these data painted the picture of metabolic differences between wild type and a ridA mutant, specifically, the persistence of 2AA. The mechanism of 2AA damage is expected to apply to several PLP-dependent enzymes, meaning that multiple enzymes might be impaired when cells experience 2AA stress. In this scenario, 2AA stress would be expected to dampen a variety of metabolic reactions, perturbing the metabolic network and generating the phenotypes observed for a ridA mutant. The correlation between in situ production of 2AA and decreased activity of a target enzyme suggested that enzymatic activity of 2AA targets could be used as a proxy for in vivo 2AA damage.

7.2. 2-Aminoacrylate damages PLP-dependent enzymes in vivo

Biochemical experiments, summarized above, convincingly showed that 2AA could damage PLP-dependent enzymes in vitro. Supported by the genetic analyses of ridA mutants, the hypothesis was extended to suggest an outline of 2AA stress in vivo. In this scenario 2AA, generated by IlvA acting on serine, would persist in the cell in the absence of the enamine/imine deaminase RidA. The persistence of 2AA would result in damage to specific PLP-dependent enzymes in the cell. 2AA differs from other enamine/imine metabolites that are generated as catalytic intermediates by PLP-dependent enzymes and those generated by NAD/FAD-dependent enzymes. While enamines and imines are characterized by their instability and short half-lives, 2AA appears unique in its ability to covalently modify PLP in enzyme active sites. This unique characteristic is likely due to 2AA being more reactive than other enamines (i.e. 2AC) and its ability to tautomerize between an enamine and imine species. Based on this attribute, 2AA was considered a potential metabolite stressor.

Results from in vivo experiments supported an emerging model that was prompted by the biochemical potential unveiled with in vitro analyses. Evidence that 2AA persisted in a ridA mutant of S. enterica was needed to validate this model. Accumulation of 2AA cannot be directly measured in cell extracts due the reactivity and short half-life of the molecule. Two approaches can be used to quantify 2AA persistence in vivo in the absence of RidA. In the first instance, enzyme activity of 2AA targets are assayed in crude extracts. A connection between activity and 2AA damage is established by comparing activity of a target enzyme in a wild-type strain and a ridA mutant strain, which is presumed to be under 2AA stress. If an enzyme is damaged by 2AA, its activity will be lower in the ridA mutant strain. Control experiments assay the same enzyme from the cells grown in a medium with added isoleucine, which allosterically inhibits IlvA – i.e. reduces its activity and decreases 2AA formation. It follows that activity of a 2AA target enzyme will be restored in the ridA mutant when isoleucine is supplemented in the growth medium. Using this strategy, multiple PLP-dependent enzymes have been shown to be in vivo targets of 2AA damage. These include serine hydroxymethyl transferase (EC: 2.1.2.1) (Flynn & Downs, 2013; Shen et al., 2022), branched-chain amino acid aminotransferase (EC: 2.6.1.42) (Ernst & Downs, 2018; Schmitz & Downs, 2004), aspartate transaminase (EC: 2.6.1.1) (Borchert & Downs, 2017b), aminolevulinic synthase (EC: 2.3.1.37) (Whitaker et al., 2021), and cysteine desulfurase (EC: 2.8.1.7) (Fulton et al., 2022). To confirm that lowered enzyme activity is caused by 2AA-induced damage, the specific activity (not simply total activity) of the enzyme should be determined. While specific activity has not been determined for each enzyme mentioned above, both IlvE and alanine racemase (Alr) had lower specific activity in a ridA mutant compared to wild type (Flynn & Downs, 2013; Schmitz & Downs, 2004; Shen et al., 2022). Thus, the data support the notion that 2AA accumulation can be measured by proxy – i.e. comparing activity of a target enzyme with and without the presence of the 2AA stressor. BCAA aminotransferase (IlvE) is the enzyme of choice for these activity assays based on the simplicity of the assay, and the fact that 2AA damage to IlvE has been well established both in vitro and in vivo (Lambrecht et al., 2013; Schmitz & Downs, 2004; Shen et al., 2022). As more PLP-dependent enzymes are identified as targets of 2AA damage, it will be beneficial to establish a streamlined method to catalog all 2AA targets in an organism.

Early biochemical studies showed that after 2AA attack, a 2AA-PLP adduct was covalently bound to the active site of the enzyme (Roise, Soda, Yagi, & Walsh, 1984). This adduct provides the means to directly measure, and quantify, enzyme damage caused by 2AA. Treatment of a 2AA-damaged protein with weak base releases the 2AA-PLP adduct from the enzyme as pyruvate-PLP (Likos et al., 1982), which can be easily distinguished from PLP by high pressure liquid chromatography (HPLC) (Fig. 5) (Flynn & Downs, 2013; Shen et al., 2022). Alanine racemase (Alr) was used as a test case, and a protocol was developed to detect and quantify damage caused by 2AA in vivo (Flynn & Downs, 2013). This assay was expanded to determine if other potential target enzymes were damaged in vivo. The relevant enzymes were purified from the wild-type strain and a ridA mutant of S. enterica and treated with base to release cofactors, which were then separated by HPLC (Flynn & Downs, 2013; Schnackerz, Ehrlich, Giesemann, & Reed, 1979; Shen et al., 2022). This protocol quantifies the ratio of damaged to undamaged PLP cofactor, and can be used with any strain background or growth condition. If an enzyme is damaged by 2AA in vivo, pyruvate-PLP will be present in some percentage of the individual enzymes purified from a ridA mutant. In most cases, the same protein purified from wild-type cells will be occupied only by a PLP cofactor. In contrast, some enzymes (e.g. IlvA) do not release any pyruvate-PLP, even when purified from the ridA mutant, indicating the active site of the protein has not been attacked by 2AA (Shen et al., 2022). In the case of IlvA, this result was particularly interesting since the catalytic mechanism of serine dehydratase generates 2AA. The reason the IlvA active site is not damaged by the 2AA it generates is not understood but is of continuing interest in the context of modeling 2AA stress in vivo.

Fig. 5.

Pyruvate-PLP adduct released from protein can be used to quantify 2AA stress. 2-Aminoacrylate (2AA), generated by the serine/threonine dehydratase IlvA, attacks the PLP in an enzyme active site. This attack results in a 2AA-PLP adduct covalently attached to the enzyme. Treatment of the respective enzyme with weak base (KOH) releases a pyruvate-PLP adduct structure (highlighted in gray) (Flynn & Downs, 2013).

Several PLP-dependent enzymes are attacked by 2AA in vivo. At present, this list includes at least isoleucine aminotransferase (IlvE), alanine racemase (Alr), serine hydroxymethyltransferase (GlyA), aspartate aminotransferase (AspC), phosphoserine aminotransferase (SerC) (Flynn & Downs, 2013; Shen et al., 2022) (Shen, unpublished) from S. enterica. This strategy was also used to assess the sensitivity of enzymes from heterologous systems. Using this approach, the PLP-dependent enzymes aminolevulinic synthase (Hem1) from S. cerevisiae (Whitaker et al., 2021) and cysteine desulfurase (IscS) from P. aeruginosa (Fulton et al., 2022) were confirmed as targets of 2AA damage. This method to detect 2AA damage in vivo has the potential to facilitate the identification of active site residues that determine 2AA sensitivity, and to explore the properties of an active site that is resistant to attack. The latter features are especially intriguing when an enzyme generates 2AA in its active site, like IlvA. A variant of cysteine desulfurase (IscS) that is less sensitive to 2AA attack than the wild-type enzyme was isolated in P. aeruginosa as a suppressor of a 2AA-induced growth defect (Fulton et al., 2022). The IscS^Q183P variant had significantly reduced enzymatic activity, highlighting the balance that exists between maintaining, or evolving, a catalytically efficient PLP-dependent active site and protecting said active site from attack by 2AA. More analysis is required to understand the parameters that constrain the evolution of active sites that are resistant to 2AA attack.

Much like the activity-based assays for single enzymes described above, analysis of protein cofactor content is currently a one-at-a-time approach that is laborious and time consuming. To improve our understanding of 2AA damage on a global scale will require creative approaches. An attractive basis for future global analysis is the strategy used by Sieber and colleagues to define the PLP-dependent proteome in different organisms (Hoegl et al., 2018). These investigators developed a workflow with functionalized cofactor probes and a strategy to label PLP-containing enzymes in vivo. Derivatization of the probe allowed the proteome to be selectively mined for PLP-associated proteins (Hoegl et al., 2018; Pfanzelt, Maher, Absmeier, Schwarz, & Sieber, 2022; Wilkinson, Pfanzelt, & Sieber, 2022). It is plausible that this workflow could be modified to specifically detect enzymes with a covalently modified PLP active site (i.e. damaged by 2AA), and in this way begin to define the proteome of 2AA-damaged PLP-dependent enzymes. It is exciting to consider that this method, and/or others, could lead to the definition of a 2AA “adduct-ome” of different organisms. Such a tool would generate insights into the response(s) of different metabolic network structures to the presence of the reactive 2AA metabolite stressor. Additionally, data from such experiments would facilitate the definition of an organism-specific “fingerprint” that would facilitate efforts to define cellular components involved in the generation of, and response to, 2AA stress. Like other global approaches, this process could define underlying consequences of 2AA stress that do not rise to the level of disruption needed to elicit a phenotype and therefore can go undetected. Ultimately, a proteomic approach could be used to define the sensitivity of different enzymes to 2AA, since not every enzyme is damaged to the same extent and not every damaged enzyme results in a detectable phenotypic consequence. Most significantly, a proteomic signature would make it practical to study 2AA stress beyond the model organisms that are easily manipulated in the laboratory.

8. The 2-aminoacrylate stress paradigm

The activity and physiological role of RidA has been defined with rigorous biochemical and phenotypic analyses, primarily in S. enterica. as outlined in the above sections. Data from these studies strongly indicate that RidA is an enamine/imine deaminase, with the physiologically relevant substrate being the reactive metabolite 2AA. Results from these studies defined a paradigm of 2AA stress that is schematically shown in Fig. 6. The framework of this stress paradigm is that: (i) PLP-dependent removal of β-functional groups from α-amines generates the enamine/imine pair, 2AA/2-iminopropionate (2IP), (ii) the deamination of 2AA/2IP can occur spontaneously by solvent water or by RidA, (iii) the rate of spontaneous hydrolysis in vivo is insufficient to prevent 2AA accumulation in the absence of RidA, and (iv) persisting 2AA results in covalent damage to some portion of sensitive PLP-enzymes. This damage results in decreased activity of the respective enzyme pool(s), eliciting metabolic stress that causes diverse phenotypes and metabolic deficiencies. In bacteria, 2AA is generated by several PLP-dependent enzymes including serine/threonine dehydratases (EC 4.3.1.19), cysteine desulfhydrases (EC 2.5.1.47) and diaminopropionate ammonia-lyases (EC 4.3.1.15). Reactions catalyzed by each of these enzymes have been associated with metabolic stress when RidA is not present, consistent with the role of 2AA as a stressor that is quenched by RidA (Ernst & Downs, 2016; Ernst et al., 2014; Schmitz & Downs, 2004).

Fig. 6.

Reactive enamines have negative effects on metabolism. The generation of the reactive enamine. 2-aminoacrylate, its potential fates, and the impact it can have on the metabolic network function are shown schematically. PLP-dependent enzymes generate reactive enamines that can be either deaminated by water (dashed arrow) or RidA (black arrow) to pyruvate. If allowed to persist, the enamine can interact with cellular components to irreversibly damage PLP-enzymes (red) to alter the metabolic network and elicit phenotypes. In the text, the deleterious effects are described as metabolic stress (i.e. the 2AA stress paradigm) and the unpictured positive effects are illustrated in the recruitment of a PurF-independent synthesis of PRA for thiamine synthesis.

The model in Fig. 6 is consistent with several relevant observations. In a wild-type cell, 2AA is continually being generated and deaminated to pyruvate without any phenotypic or biochemical evidence that it persists at a level capable of causing damage. The observation that the loss of RidA does not always impair growth is consistent with the fact that deamination of 2AA can occur non-enzymatically by solvent water. This observation indicates that RidA activity is not “essential” to remove 2AA generated under many growth conditions (Borchert, Ernst, et al., 2019; Irons et al., 2020). However, even in the absence of a detectable growth defect, there is biochemical evidence of 2AA damage to PLP-enzymes in a ridA mutant of S. enterica. This finding indicates that 2AA persists in this background, though not always to a level that perturbs the metabolic network to affect growth. Tellingly, phenotypic defects can be induced in a ridA mutant with the exogenous addition of precursors to 2AA, which increases production of the reactive molecule by various 2AA-generators (Ernst et al., 2014; Ernst, Anderson, & Downs, 2016; Irons et al., 2019; Lambrecht et al., 2013). These additions lead to a level of 2AA that exceeds the capacity of water-mediated deamination to maintain the activity of target enzymes required for normal growth. Importantly, when a wild-type strain is provided with a similar excess of a 2AA precursor metabolite, there is no apparent physiological defect, confirming that RidA can efficiently quench the elevated levels of 2AA.

2AA stress elicits multiple phenotypes that can be traced to damage of a single PLP-dependent enzyme (i.e. the “critical target” of 2AA). The specific critical target depends on the growth conditions and the metabolic architecture of the organism. Detectable consequences of 2AA damage identified so far include defects in alanine catabolism, Fe-S biosynthesis, heme biosynthesis, and one-carbon metabolism (Borchert & Downs, 2017b; Ernst & Downs, 2018; Flynn & Downs, 2013; Fulton et al., 2022; Shen et al., 2022; Whitaker et al., 2021). A phenotype for which no critical target has been identified is the compromised motility detected in ridA mutants of several organisms (Borchert & Downs, 2017a; Irons et al., 2018; Irons et al., 2019) (Collier, unpublished data). Despite the impaired motility caused by the persistence of 2AA, no PLP-dependent enzyme is known to be directly involved in motility in the relevant organisms. Campylobacter jejuni is an exception, and provides an interesting scenario to study this phenomenon. In C. jejuni, PLP-dependent enzymes are indirectly involved in motility. In this organism, flagellar assembly and export requires protein O-glycosylation. Synthesis of a relevant glycan for this process, pseudaminic acid (Pse), requires the PLP-dependent transaminase PseC (EC 2.6.1.92) (Ewing, Andreishcheva, & Guerry, 2009; Schoenhofen et al., 2006). Extrapolation of the 2AA stress paradigm defined in S. enterica suggests that 2AA would target the PseC transaminase in C. jejuni, decreasing its activity and resulting in decreased motility caused by a decreased synthesis of Pse. However, even if true in C. jejuni, damaged PseC would not explain the motility defects in ridA mutants of other bacteria. The data gathered so far suggest there may be multiple mechanisms by which 2AA affects motility in bacteria.

One molecule of 2AA is generated with every catalytic turnover of IlvA, or any other PLP-dependent enzyme that generates this reactive stressor metabolite. The RidA enzyme that removes 2AA appears to be constitutively expressed, which distinguishes this paradigm from a canonical stress response system that is regulated in response to a stressor. This feature raises questions that have yet to be addressed. Is RidA a housekeeping enzyme that is needed to maintain metabolic stability? Is it appropriate to designate it a stress-response protein rather than simply a modulator of metabolic homeostasis? Considering these questions is particularly appropriate given our emerging understanding of the function(s) of members of non-RidA subfamilies and their role(s) in the metabolic networks of various organisms, as discussed in Section 9 below.

A conspicuous characteristic of the 2AA system is that PLP-dependent enzymes comprise both the generators of 2AA and known enzyme targets of 2AA. At present, there are no structural characteristics of a PLP-dependent enzyme that can be used to predict if, or how efficiently, a protein will be targeted by 2AA. Enzymes with diverse mechanisms and of distinct fold-types have been shown to be targets of 2AA damage (Fulton, Irons, & Downs, 2022; Shen, Borchert, & Downs, 2022; Shen, unpublished). It will take the analysis of many more enzymes by methods discussed above to generate data necessary to define a consensus active site structure of a targetable enzyme. Further studies on enzyme structure and mechanism will improve our understanding of how organisms have evolved to ensure a productive, efficient, and robust metabolic network, despite the presence of continuous challenges from reactive metabolites.

8.1. 2-Aminoacrylate stress perturbs metabolic network function

The model for the RidA paradigm of 2AA stress was derived from data obtained in S. enterica. Since the major components of the system, PLP-enzymes and RidA, are highly conserved across microbes, the initial assumption was that the described paradigm could be extrapolated to other organisms. However, this assumption proved to be only partially correct. The mechanics of 2AA stress are indeed conserved in that PLP-dependent enzymes from any organism are targets of attack (Fulton et al., 2022; Shen et al., 2022; Whitaker et al., 2021), RidA homologs deaminate 2AA (Irons et al., 2019), and serine/threonine dehydratase is a highly conserved enzyme (Percudani & Peracchi, 2003). Thus, most microbes have the components to support a conserved 2AA stress response. As mentioned in the introductory sections, metabolic components do not always predict the structure of the system, and this is true in the case of the RidA paradigm. Despite the conserved ability of organisms to generate 2AA stress, the metabolic consequences (i.e. phenotypic output) of this stress vary, sometimes dramatically, between even closely related organisms. Varied phenotypic outputs reflect differences in metabolic network architecture between organisms. The RidA model presumes that 2AA stress dampens multiple metabolic enzymes (i.e. the PLP-dependent enzyme targets). This overall dampening may not result in a growth defect under most conditions, since metabolic networks often possess more enzymatic capacity than needed and are thus able to absorb perturbations (Koenigsknecht et al., 2012; Ramos et al., 2008b; Sander et al., 2019). Context-dependent nutritional requirements, or other phenotypic consequences induced by 2AA stress, can be used to identify the PLP-dependent enzyme(s) critical for stability and/or function of the metabolic network under a given condition. This knowledge is valuable for defining metabolic connections and contributes to the overall understanding of the metabolic network of the relevant organism.

The response of an organism to 2AA stress cannot be predicted a priori and has been queried in the context of physiology in relatively few organisms. In S. enterica, increased 2AA stress caused by exogenous serine, cysteine, or 2,3-diaminopropionate generates a nutritional requirement that is satisfied by glycine. A series of biochemical and genetic analyses showed that inhibition of serine hydroxymethyltransferase (GlyA) by 2AA is the cause of the glycine requirement, making this enzyme the critical target when S. enterica is experiencing 2AA stress (Ernst & Downs, 2016; Flynn & Downs, 2013). Serine hydroxymethyltransferase is one of only two PLP-dependent enzymes that is encoded in the genomes of all free-living organisms, the other being aspartate transaminase (AspC) – another known target of 2AA damage (Percudani & Peracchi, 2003). The broad conservation of GlyA suggested that it might be the conserved critical target in other organisms. Analyses of ridA mutants in several organisms showed this was not the case. Even closely related species such as E. coli and S. enterica have significant differences in their response to 2AA stress, despite sharing an overwhelming majority of components in their metabolic networks (Borchert & Downs, 2017b). E. coli has two RidA homologs (RidA and TdcF) and elimination of both of these proteins does not generate an observable metabolic defect nor nutritional requirement. However, a ridA tdcF double mutant does display a 2AA-dependent motility defect and the transcriptome differs from wild type, confirming the cell is being affected by 2AA (Collier, unpublished data; Borchert, unpublished data). Significantly, a growth defect is generated when 2AA production is artificially increased in an E. coli strain lacking the RidA homologs. This growth defect is not satisfied with glycine supplementation, indicating GlyA is not the critical target when E. coli is under 2AA stress (Borchert & Downs, 2017b). Rather, in an E. coli ridA tdcF mutant, aspartate restores growth under 2AA stress conditions, suggesting the critical target enzyme in E. coli could be aspartate transaminase. A variety of explanations for the differences in metabolic structure that would cause the phenotypic differences between S. enterica and E. coli under 2AA stress can be proposed. Differences in kinetics or substate specificity of key enzymes (i.e. IlvA), differing properties of the cellular milieu such as amino acid concentration, internal pH, local microenvironments, etc. are all considerations that could alter the 2AA stress system in one organism compared to another. In total, the differences in response to 2AA stress highlight our limited ability to define metabolic network structure based on genome composition and enzyme conservation.

In contrast to E. coli, P. aeruginosa ridA mutants appear ten-fold more sensitive to 2AA than S. enterica and exhibit a growth defect in minimal medium, even without serine added to increase 2AA stress (Fulton et al., 2022; Irons et al., 2018). Thiamine restored growth to a P. aeruginosa ridA mutant, while glycine supplementation had little to no effect. This nutritional requirement, and subsequent suppressor analysis, led to the conclusion that the cysteine desulfurase required for Fe-S cluster synthesis (IscS) is the critical target of 2AA in P. aeruginosa (Fulton et al., 2022). While IscS is not directly tied to thiamine synthesis, past work in S. enterica was leveraged to conclude that compromised Fe-S synthesis indirectly affected thiamine biosynthesis. Although decreased Fe-S cluster synthesis impacts multiple enzymes in the cell, enzymes in thiamine biosynthesis, ThiC and ThiH are most sensitive to this effect (Skovran & Downs, 2000).

The physiological role of RidA has been addressed to different extents in two eukaryotic systems, Arabidopsis thaliana and S. cerevisiase. Niehaus et al. (2014) determined that A. thaliana strains lacking RidA were defective in root development, which was exacerbated by serine and eliminated by isoleucine (Niehaus et al., 2014). These characteristics confirmed the presence of a 2AA stress system with features resembling the RidA paradigm in S. enterica. Presumably, the defect in root development is the consequence of damage to one or more PLP-dependent enzymes which have not been identified. In S. cerevisiae the RidA homolog is Mmf1, named as mitochondrial matrix factor based on the finding that strains lacking this protein lose mitochondrial DNA (Oxelmark et al., 2000) and are defective in isoleucine biosynthesis (Kim et al., 2001). These two studies were carried out before the biochemical function RidA proteins was known. Subsequent studies defined the role of Mmf1 in the context of the emerging paradigm of 2AA stress in bacteria and found many similarities with the S. enterica paradigm. In the absence of Mmf1, accumulation of 2AA causes transient metabolic stress and irreversible loss of mtDNA (Ernst & Downs, 2018). The relevant 2AA is generated from serine by one of two serine/threonine dehydratases (Cha1p, Ilv1p), which are differentially regulated by transcription and allosteric inhibition, respectively (Ahmed, Bollon, Rogers, & Magee, 1976; Bornaes, Ignjatovic, Schjerling, Kielland-Brandt, & Holmberg, 1993). The PLP-enzyme targets of 2AA that cause the pleotropic effects of 2AA accumulation in S. cerevisiae are presumed to be mitochondrially located but remain unknown. Hem1 (aminolevulinic acid synthase, EC: 2.3.1.37) and Bat1 (branched chain amino acid transferase) are mitochondrially located and damaged by 2AA (Ernst & Downs, 2018; Whitaker et al., 2021), but neither adequately explains all phenotypes caused by 2AA stress in S. cerevisiae. It is interesting to note that in both eukaryotic systems tested to date, RidA is present in a cellular compartment with several key PLP-dependent enzymes, such as those involved in the synthesis of branched chain amino acids, heme, and Fe-S clusters. Continued studies in these systems will provide a better understanding of the significance of this correlation and the critical enzymes in yeast that are affected by 2AA stress.

Campylobacter jejuni stands out among the organisms in which 2AA stress has been investigated. C. jejuni lacking the RidA homolog exhibits a motility defect, aberrant flagellar structure, and a defect in autoagglutination (Irons et al., 2019). Striking among these findings was that all phenotypes manifested on nutrient medium. In other organisms, nutrients (i.e. isoleucine) present in rich medium prevent 2AA stress. The IlvA homolog in C. jejuni lacks a regulatory domain and deletion of the gene did not rescue the observed phenotypes (Irons et al., 2019). Thus, in C. jejuni, the source of 2AA and potential target enzymes are not clearly defined, and it remains possible that the role of RidA in this organism differs from other bacteria.

Physiological and biochemical genetic studies of RidA paralogs from various organisms have strengthened the RidA paradigm and expanded our understanding of the effects of reactive intermediates in metabolism. The serendipitous discovery and subsequent characterization of RidA function in S. enterica and other organisms, prompted a growing interest in studying other members of the Rid superfamily and their contributions to cellular physiology. Together these studies suggest 2AA stress could be a useful tool in probing nuances that exist in metabolic networks. The diverse manifestations of 2AA stress across organisms emphasizes the different structures metabolic networks can have, even when they contain the same components (Downs et al., 2018).

8.2. Physiological consequences of 2-aminoacrylate stress

The phenotypes caused by the loss of a RidA protein indicate that 2AA stress impacts the overall physiology of an organism. To fully describe the effects of 2AA stress on a metabolic system, two parameters must be defined: (i) the targets of 2AA stress and (ii) how they impact the overall function of the metabolic network. At this point, all known targets of 2AA stress are PLP-dependent enzymes, and intriguing questions linger on the global level. What distinguishes the proteins targeted by 2AA from the resistant ones? What are the criteria that determine the hierarchy of damage among different target enzymes? What is the ratio of each protein pool that is inactivated by 2AA? These questions are difficult to address with our current technology and approaches, but with the development of productive global approaches in the future these questions will be approachable. An attractive reductionist approach would be to initiate in vitro “competition” between target enzymes. Such an approach would involve providing a population of different target enzymes and exposing them to 2AA generated in situ, before monitoring the activity of each enzyme.

Many consequences of 2AA stress are likely to be absorbed by the robustness of the metabolic system and will therefore fail to generate a detectable phenotype. In considering this possibility, global approaches provide the opportunity to identify metabolic changes caused by 2AA stress, and better understand the structure of the relevant metabolic network and its response to perturbation. Three studies have taken a global approach to address the indirect consequences of 2AA stress in S. enterica not detected by phenotypic analysis (Borchert & Downs, 2017a; Borchert, Gouveia, Edison, & Downs, 2020b; Borchert, Walejko, et al., 2019). In the first case, the transcriptome of a ridA mutant was compared to that of wild type when the strains were grown in minimal medium – i.e. when no growth phenotype is visible for the ridA mutant. The premise of this study was that transcriptional changes would reflect the internal cellular environment and thus identify unknown areas of metabolism that were perturbed by 2AA. It was somewhat surprising that genes involved in metabolic pathways altered by increased 2AA levels (i.e. branched chain amino acids, one carbon metabolism) were not represented by changes in the transcriptional profile of ridA mutants. These results suggested that metabolic perturbations and allosteric regulation are primarily responsible for modulating the response to 2AA stress. It was thus of interest that expression level of multiple genes encoding flagellar assembly components in the ridA mutant were decreased in the ridA mutant. Pursuit of this observation was key in identifying the 2AA-induced motility defect in S. enterica (Borchert & Downs, 2017a). Compromised motility was subsequently identified as a result of the loss of RidA activity in multiple organisms, though the causative mechanism remains unclear. Data from two metabolomic studies of ridA mutants supported the model of 2AA stress that was derived from genetic analyses and consistently, identified differences in metabolite profiles between ridA mutant and wild-type cells. Significantly, these studies showed that the disruption of the metabolic profile caused by a ridA mutation was stabilized when the cells were grown with the addition of isoleucine, consistent with the role of isoleucine prevent 2AA synthesis (Borchert, Gouveia, Edison, & Downs, 2020a; Borchert, Walejko, et al., 2019). Future studies are needed to expand on these findings, and further expansion of global analyses of ridA mutants in other organisms will increase our understanding of 2AA stress.

Progress in our physiological understanding of 2AA stress is also facilitated by isolating and analyzing suppressing mutations of ridA mutant phenotypes. There are at least two mechanisms to eliminate 2AA-dependent growth defects by altering the structure of the metabolic network: (i) mutations that reduce or eliminate 2AA stress, and (ii) mutations that allow growth in the presence of 2AA stress, possibly by bypassing a critical enzyme under the relevant condition. A few of these studies have been performed in S. enterica and contributed to the details of the 2AA stress paradigm represented in Fig. 6. Suppressor mutations that restore growth of a ridA mutant in the presence of excess 2AA stress most often reduce or eliminate the generation of 2AA. Mechanisms that reduced 2AA production most commonly altered the catalytic capacity of IlvA (Christopherson, Lambrecht, Downs, & Downs, 2012) or increased the synthesis of threonine/isoleucine (Borchert & Downs, 2018; Christopherson et al., 2012; Fulton et al., 2022). Some studies identified genes that could reduce 2AA stress when provided in trans on a multi-copy plasmid. Specifically, expression of cystathionine β-lyase (MetC, EC4.4.1.8) from S. enterica or the aspartate/glutamate racemase from multiple organisms restored growth of a ridA mutant in the presence of serine (Ernst, Borchert, et al., 2018; Ernst, Christopherson, et al., 2018; Hodge-Hanson, Zoino, & Downs, 2018). In each case, the accumulation of 2AA was reduced, as judged from the restored activity of the target protein IlvE, which was used as a proxy for 2AA damage (Fulton et al., 2022; Shen et al., 2022). Biochemical models and in vitro experiments suggested the decrease in 2AA was mediated by a catalytic intermediate produced by the respective enzymes. These metabolites were suspected of reacting with 2AA to generate a stable adduct, which prevented damage to cellular targets. These data show there are multiple means by which a cell can quench 2AA and suggest that, in some organisms, 2AA stress could be managed without RidA. From another perspective, perhaps RidA is conserved because it forms a functional product from 2AA that is used in a metabolic pathway. In contrast, quenching 2AA with a metabolite would remove the reactive enamine but generate an unusable adduct product, likely an inefficient metabolic strategy.

In a second scenario, mutations that altered critical targets, or bypassed the need for them, should arise as suppressors. Experiments in P. aeruginosa yielded a mutation that altered a target enzyme, IscS, and restored growth to a ridA mutant. The resulting IscS variant had two defining features: (i) it retained cysteine desulfurase activity, although it was reduced, and (ii) the IscS variant was less sensitive to damage by 2AA than the wild-type protein (Fulton et al., 2022). The decreased desulfurase activity of the variant IscS restored growth, but not to the level of wild-type cells. Other suppressor screens have failed to identify variants of additional target enzymes that had increased resistance to 2AA. The difficulty in isolating these mutations may reflect how difficult it is to retain catalytic activity while simultaneously altering sensitivity to 2AA attack. Perhaps this observation is not surprising, as the same active site is responsible for both properties.

9. Rid family function beyond RidA and 2-aminoacrylate stress

In S. enterica RidA modulates pathway flux, blocks a moonlighting activity, and rapidly deaminates the reactive 2AA generated by multiple enzymes (i.e. IlvA, CdsH, DapL) (Browne et al., 2006; Ernst et al., 2014; Ernst et al., 2016; Lambrecht et al., 2013). Significantly, each of these roles was a result of the enamine/imine deaminase activity that unifies members of the RidA subfamily. Once it was defined, the RidA paradigm of 2AA stress provided a framework to consider the Rid protein superfamily as a whole (Hodge-Hanson & Downs, 2017; Irons et al., 2020; Niehaus et al., 2015). Unlike members of other Rid subfamilies, RidA proteins are found in all domains, leading to the conclusion that the RidA was the ancestral protein from which the remaining seven subfamilies descended (Fig. 1A) (Niehaus et al., 2015). Rid1, 2, 3, 4, 5, 6 and 7 proteins are found only in prokaryotic genomes, primarily bacteria, and are particularly abundant in the Pseudomonadota (i.e. Proteobacteria). The NCBI conserved domain database uses a position-specific scoring matrix (Marchler-Bauer et al., 2013) to define the various Rid subfamilies, but it remains difficult to classify Rid proteins by sequence alone. In contrast to the somewhat imprecise definition of the eight subfamilies by sequence characteristics, members of the Rid superfamily are easily separated into two groups based on a critical arginine residue. The relevant arginine (Arg105 in S. enterica) is essential for imine/enamine deaminase activity in all cases where it has been queried experimentally (Buckner et al., 2021; Lambrecht et al., 2012; Niehaus et al., 2015). Based on these data, and the conservation of sequence and structure within the Rid superfamily, proteins with the Arg105 residue (members of the RidA and Rid1, 2 and 3 subfamilies) are considered deaminases. Similarly, those lacking the residue (members of the Rid 4, 5, 6 and 7 subfamilies) are assumed to lack deaminase activity. No member of the latter subfamilies has been reported to have deaminase activity, making it likely these proteins have a distinct, non-deaminase function(s) that may be conserved. Thus far, no exceptions to this designation have been reported.

In the absence of a putative, physiologically relevant imine/enamine substrate, Rid family members are assayed using a generic imine generator. In the commonly used assay, commercially available FAD-dependent l-amino acid oxidase (LOX) generates imine intermediates in situ from a spectrum of amino acids (Buckner et al., 2021; Digiovanni et al., 2021; Fulton & Downs, 2022; Hodge-Hanson & Downs, 2017; Irons et al., 2018; Niehaus et al., 2015) (Fig. 4). While convenient, this assay does not give information about the physiological relevance of deaminase activity, or potential substrates. Rather, the LOX assay provides a panel of imine substrates with which various Rid proteins can be tested and compared.

Data from LOX-deaminase assays, combined with in vivo complementation experiments, support some distinction between the Rid subfamilies. For instance, among the subfamilies carrying the Arg105 residue – Rid1 proteins consistently complement growth defects of a ridA mutant of S. enterica when expressed in trans, albeit less effectively than RidA proteins. In contrast, Rid2 and Rid3 proteins must be highly expressed to restore growth to a ridA mutant under conditions of 2AA stress, and some of these proteins may not complement a ridA mutant at all (Hodge-Hanson & Downs, 2017). In general, Rid2 and Rid3 proteins deaminate substrates that RidA and Rid1 proteins have little to no activity with in vitro, specifically iminohistidine, iminophenylalanine and iminoarginine (Fulton et al., 2022; Hodge-Hanson & Downs, 2017). While trends seem to be emerging, the correlations between substrate specificity and Rid subfamily are currently based on relatively few data points. Expansion of this type of biochemical analysis will improve our understanding of structure/function relationships in the Rid superfamily and allow a better delineation of the subfamilies.

The study of Rid proteins is emerging as a field with the potential to provide exciting insights into biochemistry, bacterial physiology, and metabolism. Despite the rigorous characterization of numerous RidA paralogs, little is known about the physiological role(s) of members of other Rid subfamilies. Recent work has demonstrated a physiological role for DadY (Rid2 subfamily) in P. aeruginosa and Rid7C (Rid7 subfamily) in Nonomuraea garenzanensis (Damiano et al., 2022; Fulton & Downs, 2022) and presented strong biochemical evidence for the function of RutC (Rid1 subfamily) in E. coli (Buckner et al., 2021).

9.1. Genomic context suggests Rid protein function

Gene clustering in Prokaryotes can imply function, particularly if genes group together with varying configurations across multiple organisms. However, the physiological role of a gene product can rarely be defined by this consideration alone. Analysis of the genomic context of genes encoding Rid proteins in 981 representative genomes found they clustered with genes involved in a variety of metabolic processes (Niehaus et al., 2015). Rid proteins of the subfamilies expected to catalyze deamination tend to cluster with genes encoding PLP-, FAD-, or NAD-dependent enzymes. Enzymes in each of these classes have the ability to generate imine/enamine intermediates in their catalytic mechanisms (Fulton & Downs, 2022; Hafner & Wellner, 1979; He, Li, & Lu, 2011; Irons et al., 2020; Niehaus et al., 2015; Porter & Bright, 1972). For example, proteins DadY (PA5303) and DguB (PA5083) belong to the Rid2 subfamily and are encoded in the genomes of many Pseudomonas, as well as some Acinetobacter, species (PubSeed Database, Pseudomonas Genome Database). These genes are in loci with FAD-dependent d-amino acid dehydrogenases, which have been shown to proceed through imine intermediates in their catalytic mechanism (Fulton & Downs, 2022; Hafner & Wellner, 1979; He et al., 2011; He, Li, Yang, & Lu, 2014; Porter & Bright, 1972).

Other members of the Rid1, 2, and 3 subfamilies cluster with genes that encode FAD-, NAD-, or PLP-dependent enzymes that are predicted to use diverse substrates such as alcohols, amines, and aromatic compounds (Boulette et al., 2009; Chirino, Strahsburger, Agullo, Gonzalez, & Seeger, 2013; Chou, Kwon, Hegazy, & Lu, 2008; Davis, He, Somerville, & Spain, 1999; He & Spain, 1998; He et al., 2011; He et al., 2014; Irons et al., 2020; Muraki, Taki, Hasegawa, Iwaki, & Lau, 2003; Orii, Takenaka, Murakami, & Aoki, 2004; Park & Kim, 2001; Takenaka, Murakami, Kim, & Aoki, 2000; Zhou et al., 2013). Based largely on this genomic context, Rid proteins are implicated in the degradation of aromatic compounds in various Pseudomonas, Burkholderia, Paraburkholderia, and Bordatella species (Davis et al., 1999; He & Spain, 1998; Irons et al., 2020; Muraki et al., 2003; Niehaus et al., 2015; Orii et al., 2004; Park & Kim, 2001). Definitive classification of the Rid subfamilies, and their functions, awaits definition of clear biochemical properties that distinguish them from one another.

9.2. New functions defined for Rid family members

Considerable strides have been made in recent years to enhance our understanding of Rid proteins and their contributions to a robust metabolism. As was the case with RidA, the study of other Rid subfamilies has benefited considerably from investigation in numerous organisms with varying metabolic architectures. These initial studies, summarized below, support the importance of biochemical analysis and creative experimental design in the study of this ubiquitous family of proteins.

9.2.1. RutC is a 3-aminoacrylate deaminase

The first Rid protein beyond RidA to have a rigorously defined, physiologically relevant biochemical activity is RutC from E. coli. RutC is situated in the rut operon (rutABCDEFG) for pyrimidine utilization. The Rut pathway was reconstituted in vitro without RutC, RutD or RutE, which led investigators to propose multiple non-enzymatic reactions in the pathway (Kim et al., 2010). This initial work was carried out before deaminase activity had been associated with the Rid family. Based on knowledge of the deaminase activity of those members of the Rid family that contain the critical arginine, further characterization of RutC was pursued (Buckner et al., 2021). In a pure system, RutC had 2AA deaminase activity comparable to that of RidA, yet the protein only partially complemented growth defects of a ridA mutant when provided in trans. With the semicarbazide-based LOX assay RutC deaminated multiple imines. Significantly, RutC was more active on imines derived from histidine, arginine, and phenylalanine than RidA. Thus, RutC had a similar substrate profile as Rid2 and Rid3 proteins (Hodge-Hanson & Downs, 2017). Critically, RutC has in vitro deaminase activity on the substrate implicated in the uracil utilization pathway (Buckner et al., 2021) (Fig. 7). In an assay for RutC activity, purified RutB (peroxyureidoacrylate amidohydrolase) generated the 3-aminoacrylate (3AA) enamine species in situ from chemically synthesized ureidoacrylate. Consistent with the biochemistry of the proposed pathway for uracil utilization, RutC increased the rate of 3AA deamination over that achieved by solvent water, when quantified with a coupled assay (Buckner et al., 2021). These data suggested that RutC increases flux through the Rut pathway in vivo by removing a reactive 3AA metabolite released by RutB. Based on the genomic location and defined in vitro substrate for deamination, it was rational to predict a rutC mutation should compromise the utilization of uridine via the Rut pathway. Lesions in the other genes in the operon (i.e. rutB, rutE) had this expected phenotype (Kim et al., 2010), but loss of RutC activity did not result in a detectable defect in uridine catabolism. Despite the lack of an obvious phenotype, this work described the first activity for a non-RidA subfamily member that could be placed in a physiological context. The study on RutC provided a blueprint for using gene context and the predicted biochemistry of nearby gene products to predict a Rid protein function and furthermore, suggested the need to more closely examine metabolic phenotypes.

Fig. 7.

The Rut pathway for uracil utilization in E. coli. RutA/RutF convert uracil to uredoacrylate, which serves as the substrate of RutB. RutB releases carbamate and 3-aminoacrylate (3AA). 3AA can then be deaminated spontaneously by water (thin arrow), or more rapidly by the Rid protein RutC (bold arrow) to malonate semialdehyde (Buckner et al., 2021). RutE converts malonate semialdehyde to 3-hydroxypropionate in the final step of the pathway. FAD-dependent enzymes are shown in orange, and the Rid protein is shown in blue.

9.2.2. DadY is an iminoalanine deaminase

In many Pseudomonas species, the dad locus encodes genes that comprise a defined catabolic pathway for alanine – an AsnC-type regulator (DadR), FAD-dependent d-amino acid dehydrogenase (DadA) and a PLP-dependent alanine racemase (DadX). Some dad loci also encode a member of the Rid2 subfamily (locus tag PA5303) (Fulton et al., 2022; He et al., 2011). No functional role in the catabolism of alanine was attributed to the Rid2 protein in initial studies of the pathway (Boulette et al., 2009; He et al., 2011). Recent work in P. aeruginosa led to the designation of the Rid2 protein as DadY and determined its physiological role was to deaminate iminoalanine to pyruvate in the alanine catabolic pathway (Fig. 8A). Like many Rid proteins, DadY has 2AA deaminase activity in vitro, but does not complement a ridA mutant in vivo (Buckner et al., 2021; Fulton & Downs, 2022; Hodge-Hanson & Downs, 2017). Even if expression of the plasmid-borne dadY gene is induced, growth is not restored to the ridA mutant (Fulton & Downs, 2022). This apparent inconsistency between in vivo and in vitro assays that are presumed to measure the same activity (i.e. 2AA deamination) has arisen with multiple Rid proteins and represents an opportunity to refine the in vitro assay. Beyond activity on 2AA, DadY had significant deaminase activity with several imine substrates in the LOX-based assay. The substrate specificity of DadY distinguished it from RidA despite an overlap in 2AA deaminase activity in vitro, a feature also noted for other non-RidA proteins. A model that would incorporate the activity of DadY in alanine catabolism required that the d-amino acid dehydrogenase (DadA) released an imine product into the cellular milieu. Biochemical analyses confirmed that DadA releases an imine into solution where it could form a semicarbazone compound. The addition of DadY to the DadA reaction mixture decreased the rate of semicarbazone formation, demonstrating that DadY was competing with semicarbazide for free imine in aqueous solution (Fig. 8B) (Fulton et al., 2022). These data confirmed the biochemical potential for a catabolic pathway involving a deaminase, though a strain lacking dadY had no detectable growth defect. This work provided an opportunity to test a key hypothesis about the general contribution of Rid deaminases to a robust metabolic network, discussed below.

Fig. 8.

DadY contributes to alanine catabolism by deaminating imines released by DadA. (A) Pathway for alanine catabolism in P. aeruginosa. DadA (red) oxidizes D-alanine, releasing an imine intermediate that is then deaminated by water (thin arrow) or by the Rid protein DadY (bold arrow). (B) DadY removes imines released by DadA. Data show the reaction rate (i.e. the rate of semicarbazone formation) in reactions containing DadA alone (red) or in those containing both DadA and DadY (gray). A decrease in the rate of semicarbazone formation indicates an increased rate of imine removal in the presence of the Rid protein. (C) DadY is required for competitive fitness on alanine. Schematic representation of competition experiments between wild-type P. aeruginosa (blue circles) and a dadY mutant (red triangles). Represented is the culture composition after growth when equal numbers of wild type and the dadY mutant were inoculated into a medium with ammonia or alanine as a nitrogen source. The relative abundance of each strain after incubation in coculture is represented by the bar across the bottom of each tube (wild type, blue; dadY mutant, red) (Fulton & Downs, 2022).

9.2.3. Rid7C is an endoribonuclease

All Rid proteins of the “deaminase class” (i.e. those with the critical arginine residue) queried thus far have some manner of enamine/imine deaminase activity. The challenge to identify the relevant substrate(s) and physiological role of each Rid deaminase remains. However, members of four Rid subfamilies (Rid 4, 5, 6 and 7) do not have deaminase activity. It remains to be seen if there is single activity that unifies these remaining proteins, or if multiple activities are represented in these subfamilies. Excitingly, a study in N. garenzanensis led to the definition of the first biochemical activity of a Rid family member lacking the critical arginine residue. In a situation reminiscent of the initial RidA observations, a role for a Rid7 protein was uncovered serendipitously. While investigating potential redundancy between two paralogs of the beta subunit of RNA polymerase (RpoB) in N. garenzanensis, Damiano and colleagues determined that a Rid7 protein (Rid7C) had endoribonuclease activity (Damiano et al., 2022). Relevant to this study is the fact that this organism encodes two types of RNA polymerase, the result of two distinct beta subunits, RpoB(R) and RpoB(S) (Vigliotta et al., 2005). Significantly, the polymerase containing RpoB(R) controls expression of genes involved in generating secondary metabolites such as the antimicrobial A40926. Based on this involvement, the regulation and physiological function of RpoB(R) was of interest. Two mRNA transcripts encoding for RpoB(R) were found, presumed to be the result of two transcription start sites, and termed TSS1 and TSS2 (Tanaka, Tokuyama, & Ochi, 2009). The TSS1mRNA formed a secondary structure including a hairpin that occluded the Shine-Dalgarno sequence and therefore suppressed translation of the rpoB(R) gene. In contrast, the TSS2 mRNA lacked the hairpin structure. The second transcript (TSS2) was derived from TSS1 by RNA processing mediated with the endoribonuclease activity of Rid7C (Damiano et al., 2022). Thus, Rid7C indirectly modulated the translation of RpoB(R). Importantly, overexpressing rid7C increased A40926 antibiotic production, a phenotype consistent with identified endoribonuclease activity (Damiano et al., 2022). While details of the endonuclease activity across organisms remain to be explored, the biochemical characterization of a Rid protein lacking the Arg105 residue provides a starting point for efforts to understand the Rid4, 5, 6 and 7 subfamilies. Although Arg105 is the only residue definitive of an activity thus far, variations in other conserved residues could hold the key to the divergent functions of proteins belonging to different subfamilies. The initial characterization of a Rid7 protein from N. garenzanensis extends the understanding of Rid family members that function outside of the RidA paradigm and raise new questions about Rid family members and their role in modulating the metabolic network in diverse microbes (Downs, 2022).

9.3. Competitive fitness is an informative phenotype

The general process to define the role of an uncharacterized gene involves both in vitro and in vivo components. When a function for the gene product is demonstrated biochemically and a metabolic role can be hypothesized based on the genomic context of the gene, the definitive result is a phenotype generated by lesions in the relevant gene, the simplest of which are null mutations. Phenotypes that can be easily screened in this process include nutritional requirements, sensitivity to various supplements or antimicrobials, excretion of metabolites, changes in motility and/or biofilm formation, etc. In situations where the predicted phenotype cannot be detected, it is worth considering whether the proposed gene function is incorrect, or whether the condition queried prevents visualization of the phenotype. In some cases, contribution of a gene product to the metabolic network is subtle or context-dependent, such that it complicates or even prevents the detection of a phenotype associated with a lesion in the relevant gene (Borchert & Downs, 2017b; Buckner et al., 2021; Damiano et al., 2022; Fulton & Downs, 2022; He et al., 2014; Irons et al., 2020; Lambrecht et al., 2013; Schmitz & Downs, 2004).

When an anticipated phenotype is not detected for a genetic lesion, two general explanations can be suggested. First, if a redundant activity is encoded in the genome, the absence of the subject gene may not produce a phenotype. Considered in isolation, such a result might prompt the incorrect conclusion that the relevant gene has little or no role in the metabolic network. This caveat is particularly applicable in the cases considered herein, since genomes often encode multiple Rid family members with similar or indistinguishable activities in vitro. Second, the environment (i.e. temperature, medium, etc.) used for the phenotypic screening may be one in which the relevant gene product is not required and/or not present. This caveat is particularly problematic, as there are countless conditions in which phenotypes can be monitored. This point prompts the question of how one should interpret negative phenotypic results – i.e. does the lack of a phenotype indicate a lack of functional requirement? One way to increase the probability that a phenotype is detected is to sensitize the system of interest. This can be done by further perturbing the metabolic network, by an external stress or genetic modification, such that the system is more sensitive to a slightly compromised function. While it is difficult to design the right combination(s) of stressors in an absence of foundational knowledge of the system, careful selection of experimental conditions and genetic background, along with some amount of creativity and persistence, has proven successful in generating insights into physiological function. Probing varying environmental conditions and/or genetic backgrounds has led to the characterization of multiple genes of unknown function, not the least of which is RidA (Enos-Berlage et al., 1998). Phenotypic analysis has primarily been conducted in a laboratory setting with the relevant strains in pure culture. The contribution(s) of a gene product to the metabolic system might be too subtle to generate a detectable phenotype under standard laboratory conditions. However, if conservation is taken as an indication of significance, one would expect broadly conserved genes to contribute to the fitness of the organism under some condition.

Thus far, Rid proteins with the critical arginine residue catalyze reactions that can also occur non-enzymatically. Thus, it may not be surprising that these proteins are not essential for function of their respective pathways in all situations. In the case of S. enterica, RidA is not required for steady state growth under most conditions. However, an increase in 2AA levels, caused by serine supplementation, overburdens the rate of non-enzymatic deamination, resulting in a detectable need for RidA. Recall that this perturbation was key in the functional characterization of RidA. The need to monitor the fitness cost(s) associated with loss of a Rid protein with an unknown functional role was first brought to light by the studies on RutC. As noted above, strains lacking RutC activity showed no detectable impairment in their ability utilize uridine as a source of nitrogen (Buckner et al., 2021). The proposed fitness cost of losing RutC activity is yet to be experimentally validated. A similar situation arose with DadY (PA5303) in P. aeruginosa, where both the genomic location and biochemical analyses implicated its deaminase activity in the catabolism of alanine (Fulton et al., 2022; Irons et al., 2020). Deletion of the dadY gene had no measurable effect on the ability of P. aeruginosa to use alanine as a carbon or nitrogen source in pure culture under standard laboratory conditions (Fulton et al., 2022). Importantly, strains lacking genes encoding other enzymes in the dad locus (dadA, and dadX) were required for alanine catabolism (Fulton et al., 2022; He et al., 2011). When grown with wild-type P. aeruginosa under conditions requiring the catabolism of alanine, a dadY mutant had a clear defect in competitive fitness (Fulton et al., 2022). These data, coupled with biochemical evidence discussed above, showed the imine deaminase activity of DadY was physiologically significant in a natural, competitive environment (Fig. 8). This fitness defect described a critical metabolic role for DadY, despite the protein being considered “non-essential” based on a lack of detected phenotypes in the laboratory. The DadY study highlights the use of competition to probe the role of proteins that may have subtle contributions to fitness. Like the other ideas discussed throughout, productive use of this approach is likely to require some idea of a metabolic context for the protein of interest. The characterization of DadY represented a milestone in our understanding of Rid proteins and provided the first clear support for the emerging model that members of the Rid superfamily modulate metabolism.

10. General model for the Rid superfamily and remaining questions

Analysis of RidA led to the identification of a biochemical activity (i.e. enamine/imine deamination) that was consistent with the phenotypes generated by its absence in S. enterica. The most impactful observation from these studies was that a protein could be required in vivo despite the relevant reaction occurring spontaneously. The characterization of 2AA damage, and the subsequent construction of a stress response model, has led all consequences of a ridA mutation to be viewed through this lens. For the most part, this view has been supported by the data obtained. However, RidA may also contribute to modulating metabolic flux. For instance, by deaminating 2AC to the pathway intermediate 2KB, RidA potentially maximizes the efficiency of branched chain amino acid biosynthesis. In this scenario, ridA mutants would have a fitness defect when grown on minimal medium, despite lack of growth defect in laboratory conditions. Competition experiments, like those implemented to characterize DadY in P. aeruginosa are needed to expand our understanding of these possibilities (Fulton et al., 2022).

ridA mutants in bacteria tested thus far have a pronounced motility defect, and/or problems with the generation or assembly of flagella (Borchert & Downs, 2017a; Fulton et al., 2022; Irons et al., 2018, 2019). Studies in these bacteria (except for C. jejuni) found the motility defects were caused by 2AA accumulation by an unknown mechanism (Borchert & Downs, 2017a; Fulton et al., 2022; Collier, unpublished). The only characterized targets of 2AA damage are PLP-dependent enzymes, none of which are known to be directly involved in motility (Borchert & Downs, 2017a; Macnab, 1992). Compromised motility may be an indirect result of a damaged PLP-dependent enzyme or may indicate a novel mechanism of damage by the reactive species. In either case, the connection between motility and 2AA accumulation provides an opportunity to generate new physiological insights related to the 2AA stress paradigm.

An extended, or additional, role for RidA proteins is supported by the finding that some organisms encode multiple RidA paralogs. The significance of this apparent redundancy is unclear, as only a few systems have been rigorously investigated. P. aeruginosa encodes two RidA homologs, one of which (PA5339) was designated the housekeeping RidA, based on phenotypes paralleling the ridA mutant in S. enterica (Irons et al., 2018). The other RidA homolog (PA3123) had no role in ameliorating 2AA stress in vivo, despite biochemical and complementation studies supporting its classification as a bona fide RidA protein (Irons et al., 2018; Niehaus et al., 2015). The model eukaryote, S. cerevisiae encodes two RidA paralogs, Mmf1 and Hmf1, located in the mitochondria and cytoplasm, respectively (Kim et al., 2001; Oxelmark et al., 2000). Strains lacking the mitochondrial RidA (Mmf1) have phenotypes reminiscent of a bacterial ridA mutant, consistent with mitochondrial location of relevant 2AA-generating and target PLP-enzymes (Ernst & Downs, 2018; Whitaker et al., 2021). Somewhat surprisingly, the loss of the cytoplasmic RidA (Hmf1) elicits no detectable phenotype, even in the absence of Mmf1. The conservation of RidA homologs in eukaryotes strongly suggests they have an important physiological role.

While the RidA paradigm has been well-established and allows for facile modeling of the role(s) of RidA proteins in various organisms, identifying the role(s) of the non-RidA subfamilies remains challenging. The non-RidA subfamilies can be grouped based on the presence or absence of the arginine critical for deaminase activity (Arg105). The current model suggests that proteins containing the Arg105 residue catalyze deamination reactions that can occur non-enzymatically. The implication of this model is that the relevant Rid proteins are only required for fitness under conditions where the rate of the relevant reaction becomes limiting for growth or fitness. Defining the two Rid protein functions that support this model relied on their genetic context to identify the relevant imine substrate (Buckner et al., 2021; Fulton & Downs, 2022). Many Rid proteins across genomes are in loci that contain genes of unknown function, or that are only annotated with general catalytic functions with no defined substrate (Christie et al., 2009; Merlin et al., 2002; Pena-Castillo & Hughes, 2007; Porwollik et al., 2014; Porwollik, Wong, & McClelland, 2002; Price et al., 2018). Therefore, there is a need to characterize genes surrounding the Rid-encoding gene before a hypothesis can be made for the role of the relevant Rid protein. The general idea that Rid proteins increase fitness by facilitating reactions that are otherwise spontaneous has the potential to define roles for them in diverse metabolic processes. This notion raises an interesting question. Might there be additional reactions in cellular metabolism that have been attributed to spontaneous chemistry that can uncover the role of additional Rid proteins? This exciting possibility should be considered as metabolic studies continue to focus on this large protein superfamily. Moving forward, creative approaches will be needed to make significant progress in understanding the diverse reactions these proteins are likely to facilitate in supporting fitness of the organism.

One of the most significant recent advances in the understanding the Rid superfamily was the identification of an endonuclease activity for a Rid protein (Rid7C) that lacked the arginine critical for deaminase activity (Damiano et al., 2022). The knowledge that Rid proteins have a physiologically relevant activity not tied to imine/enamine deaminase activity raises exciting questions for future investigation. Will this broad class of Rid proteins (i.e. the Rid4, 5, 6 and 7 subfamilies) be distinguished by substrate specificity, or are there additional biochemical activities yet to be uncovered in these protein subfamilies? What is the connection between the deaminase and endonuclease activities? Does the ancestral RidA subfamily have both activities? Is it possible that some Rid proteins simultaneously lost deaminase activity while gaining a new function, or that they simply lost deaminase activity while retaining another, conserved activity? Future efforts are needed to determine if there is a general activity that ties the “non-deaminase” proteins, or the whole superfamily, together. The conservation of several key residues between deaminase and non-deaminase subfamilies suggests shared activities even between the more distantly related RidA and Rid7 subfamilies (Burman et al., 2007; Niehaus et al., 2015). Excitement about the demonstrated endonuclease activity aside, it is worth remembering that broad assumptions about the Rid family as whole may not be possible based on the activity of a single member. Although precedent set by the RidA paradigm in S. enterica appears to hold true thus far, questions raised above about the role of RidA proteins in modulating metabolic flux have not been addressed. Rigorous experimentation into the potential of all Rid proteins to perform additional, regulatory functions within their respective pathways could bring our study of this diverse family of proteins closer to establishing a unified model that allows reliable predictions of the functions of various Rid proteins.

Whether accelerating non-enzymatic reactions or regulating gene expression with endonuclease activity, the apparent lack of dramatic phenotypes supports a unifying role of these proteins as modulators of the metabolic network. While the significance of these proteins might be overlooked based on results from classical laboratory experimentation, their broad conservation indicates strong selection for their maintenance and is likely due to evolutionarily significant contributions to fitness in the natural environments of each organism.

11. Insights gained, and lessons learned from studies of the Rid superfamily

The RidA paradigm of 2AA stress, defined in S. enterica, provides a context to interpret results in other organisms. While acknowledging there is much to be learned about RidA and the larger Rid protein superfamily, it is worth considering the insights generated by the foundational studies with respect to metabolic network structure. Unraveling the perturbations caused by 2AA stress revealed new insights about metabolic connectivity (Bazurto et al., 2016; Borchert & Downs, 2017b; Browne et al., 2006; Christopherson, Schmitz, & Downs, 2008; Ernst, Borchert, et al., 2018; Ernst, Christopherson, et al., 2018; Flynn, Christopherson, & Downs, 2013; Fulton et al., 2022). These results suggest that 2AA stress can be used as a tool to study differences in metabolic network structure, much like the thiamine metabolic node has been (Koenigsknecht & Downs, 2010). Identifying mechanistic details of 2AA-induced phenotypes, defining critical target enzymes and the resulting perturbations of the networks in various organisms, will provide fundamental new knowledge.

The recruitment of 2AC for the synthesis of the thiamine precursor, phosphoribosylamine (PRA) in a TrpD-dependent pathway was defined in the absence of RidA in S. enterica (Figs. 2 and 3) (Browne et al., 2006; Lambrecht et al., 2010). Knowledge of this recruited pathway proved to be critical in showing that E. coli not only has this pathway, but it functions in the presence of RidA and generates ~50% of the PRA for thiamine synthesis (Bazurto et al., 2016). These results provide precedent for the possibility that other moonlighting activities or recruited pathways that depend on RidA, or other Rid proteins exist.

Characterization of RidA deaminase activity transformed the way we think about non-enzymatic reactions in the cell. Phenotypes resulting from the lack of RidA challenged the often-held notion that significant water was available in the cellular milieu. Despite the short half-life of 2AA in the presence of water, the genetic analysis convincingly led to the conclusion that this enamine could persist in the cellular milieu in the absence of RidA. This conclusion was substantiated by identifying the scars of 2AA damage in vivo.

Among Rid family members, those that are deaminases (i.e. have the critical arginine) showed some level of 2AA deaminase activity when tested in vitro. However, these non-RidA proteins do not complement a ridA mutant to the same degree as ridA when expressed in trans. This apparent inconsistency raises a concern about in vitro assays of Rid proteins and indicates nuances in cellular milieu that are not accounted for with in vitro conditions. As it stands, when genome annotation indicates there are multiple RidA proteins, the designation of the relevant housekeeping RidA has been based on phenotypic analysis (Irons et al., 2018, 2019). Perhaps this approach is not optimal, considering that the loss of Rid proteins often fails to result in a detectable phenotype. The inconsistencies that arise as Rid proteins are analyzed in more organisms emphasize the need to better understand the role(s) of Rid proteins, and the differences between metabolic networks of these organisms. Perhaps more consistent and extensive correlations will emerge as knowledge accumulates from studies of individual Rid proteins across the domains of life. It is worth reiterating that biochemical activity in vitro provides information about the potential of an enzyme within a given metabolic network, but does not define its function within the network. A more rigorous correlation between activity in vitro and in vivo for members of this protein superfamily will be critical to piece together the contribution of each Rid protein to the metabolic network it inhabits.

Analysis of the Rid superfamily has emphasized that competitive fitness can be a valuable tool to address the role(s) of genes of unknown function. Characterization of phenotypes displayed in pure culture can be expanded by considering the contribution of a gene product to fitness in the natural niche of an organism. Fitness can be difficult to monitor in pure culture, but it is a critical consideration from an evolutionary perspective. As we continue to refine our understanding of genomes and gene function, it is possible that many of the remaining genes of unknown function will play a modulatory, rather than mandatory, role in maintaining optimal function of the metabolic network. Studies of the RidA subfamily, emphasized how minimal dampening of multiple enzymes by 2AA can perturb the metabolic network and how the consequences vary across organisms (i.e. metabolic network architecture). Further, results discussed herein emphasize the many aspects of metabolism that cannot be predicted a priori even with well annotated genomes. These conclusions illustrate the current inability to extract sufficient knowledge from a genome to model metabolic details and define network structure.

12. Concluding remarks

Our current understanding of the RidA paradigm of 2AA stress, and the limited knowledge we have on members of other Rid subfamilies, represents only the tip of the iceberg in terms of the role of proteins in this superfamily. Though RidA has been studied in several bacterial systems, we lack an in vivo context for RidA activity in higher-order eukaryotes, most eukaryotic microbes, and archaea. Study of the RidA paradigm in the limited number of organisms tested so far has highlighted details and differences in both the paradigm itself, and the architectures of each metabolic network queried. Continuing studies on this intriguing protein superfamily across all domains of life will generate new and exciting insights into the metabolism and physiology of all living cells.