Chemical reactivity drives spatiotemporal organisation of bacterial metabolism

In this review, we examine how bacterial metabolism is shaped by chemical constraints acting on the material and dynamic layout of enzymatic networks and beyond. These are moulded not only for optimisation of given metabolic objectives (e.g. synthesis of a particular amino acid or nucleotide) but also for curbing the detrimental reactivity of chemical intermediates. Besides substrate channelling, toxicity is avoided by barriers to free diffusion (i.e. compartments) that separate otherwise incompatible reactions, along with ways for distinguishing damaging vs. harmless molecules. On the other hand, enzymes age and their operating lifetime must be tuned to upstream and downstream reactions. This time dependence of metabolic pathways creates time-linked information, learning and memory. These features suggest that the physical structure of existing biosystems, from operon assemblies to multicellular development may ultimately stem from the need to restrain chemical damage and limit the waste inherent to basic metabolic functions. This provides a new twist of our comprehension of fundamental biological processes in live systems as well as practical take-home lessons for the forward DNA-based engineering of novel biological objects.

INTRODUCTION: DEALING WITH METABOLIC FRUSTRATION

The traditional textbook depiction of prokaryotic cells often sketches bacteria as sort of stand-alone containers lacking organelles and where the gene expression machinery executes the instructions that emanate from DNA, following the central dogma of Molecular Biology. Enzymes then handle metabolites through the action of a large network of biochemical reactions that appear to co-occur in a single vessel, the cytoplasm. The implicit message behind this representation is that one could simply change the instructions encoded in the DNA or parts of them and make cells to carry thereby a different physiological or biochemical agenda. This view is not only simplistic but also profoundly misleading, as it conceals that the extant configuration of metabolic systems – from single reactions to whole, multitiered pathways – are run in time and space by physical objects (enzymes) that handle chemically reactive molecules which operate in a nonhomogeneous and very crowded milieu (Weiss 2014). It is now clear that artificially placing genes in and out a heterologous host is not enough for robust metabolic engineering or any other genetic engineering (see e.g. Jarboe et al., 2013). This is because the pre-existing molecular network or particular chemical constraints often reject artificially made implants and eventually favour overgrowth of those members of the population that have lost the man-made construct (Silva et al., 2012). Why? Knocked-in biological functionalities are embodied in actual material objects (e.g. proteins) which do not operate in the vacuum, but in specific physical and (bio)chemical scenarios where the new biological object (e.g. the enzymes of an engineered pathway) has to nest. One extreme challenge is posed by the central claim of contemporary Synthetic Biology that cells can be programmed at will within a biological chassis employing re-usable, standardised components (e.g. BioBricks; Stephanopoulos 2012). This often leads to the opening up of novel unexpected pathways and conflicts related to chemical interference and intractable spatial (physical) puzzles. Physicists have explored the way such mutually conflicting set-ups with similar probabilities could organise in space within the confinement of a common vessel. This results in compartmentalised states, in an overall process named frustration (Schiffer, 2002). The term expresses how co-occurrent but mutually incompati-ble processes must find a way to get together in space, resulting in a nontrivial order. In a similar way, metabolic frustration is expected to select local molecular architectures allowing cells to deploy conflicting metabolic pathways and explore new ones. Alternatively, frustration can be alleviated by genetic patches that allow co-existence of otherwise incompatible activities (Conrado et al., 2008) or alternative expression of the contentious processes (Kitao et al., 2006; Danchin 2009b; Silva-Rocha et al., 2011).

This review attempts to make a provisional catalogue of the evolutionary challenges that the robust, complex metabolism that we see today in bacteria had to overcome and what solutions were found to get to the extant biochemical networks found in the natural world (see Concept Box). In particular, much experimental data reveal that many bacteria have created specific organelles where metabolism is compartmentalised. In addition, management of time and regulation of gene expression help preventing reactive molecules to meet. The pursued outcome of this exercise is to identify design principles that can guide future endeavours in Synthetic biology, in particular metabolic engineering. Let us thus get started by reasoning as chemical engineers would.

METABOLISM IS CHEMISTRY STRUCTURED IN SPACE AND TIME

The overall chemical setting of cells is driven by a central necessity: generating building blocks able to link to each other through stable covalent links for synthesising macromolecules. This limits biological chemistry essentially to the atoms of the first and second rows of the Mendeleieff's table, with carbon-derived chemistry at its core. Carbon provides four possible stable covalent bonds that can be shared with one or several of the three bonds of nitrogen, two bonds of oxygen and one bond of hydrogen. Boron, which would be able to fit the picture, does not belong to the significant atoms of life simply because it is a rare element in the Universe (Bernas et al., 1967), but no-thing would prevent it from belonging to the xenobiology of the future (Schmidt 2010). Fluorine is also extremely rare (Chan and O'Hagan 2012), mainly because carbon–fluorine bonds are all too stable at the temperature of the Earth to allow metabolic transactions (Furuya et al., 2008). Single bonds are usually fairly stable, and their reactivity landscape is shaped by highly specific enzymes, often associated with co-enzymes or prosthetic groups of great specificity. Asymmetry is an efficient way to separate molecules without physical borders: once l-amino acids were selected to make proteins, d-sugars (based on d-glycerate) followed as an efficient way to prevent interference with translation (Danchin and Sekowska 2014). Constructing the crucial components of cells, essentially macromolecules, requires that chemical compounds are extracted from the environment, modified or entirely rebuilt, polymerised, repaired, degraded and finally exported as waste. This general process also requires an efficient management of energy. Overall, this constitutes the intermediary metabolism of the cell, where chemical bonds are continuously created, modified and disrupted. Chemical reactivity is essential in the process. As a consequence, and this has been much discussed in the past (see e.g. Srere 1987; Welch and Easterby 1994; Huang et al., 2001), there is a continuous need for proper channelling of metabolites to make them meet with the right chemical partners while avoiding chemical conflicts (Iturrate et al., 2009). The principal way to bring this about is by avoiding free diffusion, and the way to make it happen is by having a more or less stringent physical separation of other-wise incompatible reactions. This led cells, very early in evolution, to a growing need to orchestrate the different domains of metabolism (and other biological processes) and thus avoid a chaotic chemical mishmash. Note that we employ the term ‘compartmentalisation’ throughout this article as any physical constraint that limits free diffusion. This includes not only more or less permeable barriers (e.g. membranes) but also any other molecular architectures or physical barrier that prevents phy-sical and/or chemical transactions between co-occurrent partners.

Compartmentalisation of metabolism has three major drivers based on their role in the sustained development of life. First, compartments limit the blowout of reactive compounds. Chemical processes may generate unsuitable intermediates, either because they spontaneously follow a route that deviates from the expected path (paralogous metabolism; Danchin and Sekowska 2014) or because they are inherently harmful (i.e. they bear chemical groups that react with important components of the cell and inactivate them). Often such unwanted effects are due to reactive oxygen (ROS), nitrogen (RNS) and sulphur (RSS) species originated as side products of imperfect metabolic reactions (Giles and Jacob 2002; Antelmann and Helmann, 2011; Perez-Pantoja et al., 2013). But in other cases, production of the same reactive species reflects an adaptive trait, for example for defence against pathogens and predators or stress signalling (Coppola et al., 2013). Chemicals can also cause noxious effects indirectly when, for example, they build up intracellularly and change osmotic pressure, alter regulatory and metabolic networks or interfere with promiscuous enzymes (Khersonsky and Tawfik 2010). Second, by counteracting diffusion, compartmentalisation helps metabolites to react together closer and for longer – and this is important for some reactions involving unstable or volatile intermediates such as ROS and other ra-dicals, for example the ubiquitous 5′-deoxyadenosyl radical derived from S-adenosyl-methionine (Shisler and Broderick 2012). Finally, physical separation of opposing metabolic fluxes prevents the rise of futile cycles. It is a rule of thumb that anabolism and catabolism cannot act simultaneously on the very same molecules, and the reactions at stake should therefore be either physically detached or must not occur at the same time in the same location (Sauer and Eikmanns 2005). A large variety of processes have evolved to cope with these metabolic challenges. They include physical solutions implemented as properties of the cellular milieu, chemical solutions to metabolic hurdles and regulatory solutions that act as traffic lights that sort out conflicting processes.

While each metabolite constitutes a particular instance of reactivity, having a complete evolutionary profile of their chemical conflicts would be an unrealistic endeavour. We thus focus here on a limited number of examples where solution of a chemical problem involves the set-up of specialised compartments. In turn, biochemical channelling and physical barriers paved the way for selective processes that allowed metabolism to build learning and memorisation capacities, that is enabled a mechanism for capturing information, using molecular devices that are reminiscent of Maxwell's demons (Maxwell, 1871; Binder and Danchin, 2011).

HAZARDOUS COMPOUNDS ARE UBIQUITOUS IN METABOLISM

Although we often visualise bacterial metabolism as a whole of many reactions occurring in a single microreactor (the cell cytoplasm), chemical engineering has known for a long time that assembly of a multitask system requires a specific tri-dimensional (3D) positioning of its constituents (Zhao et al., 2012). Under this design principle, the role of a compartment is to cluster together components that must interact, while separating those which should not interfere with one another. Compartments can also function for storage of items not being used at a given time (Chandra et al., 2011). Man-made reactors compartmentalise compounds to maintain and control desired transformations for an optimal balance of energy consumption vs. synthesis of products of interest. The same design challenges have to be met by cells. Construction of new covalent bonds ultimately implies reacting metabolites, the role of enzymes (and their cognate co-enzymes) being just the lowering of the activation energy for a specific stereochemical configuration. For this to happen, biological compounds must have an inherent reactive potential, which in some cases can be traced to their ability to take or deliver energy-rich phosphate bonds (Westheimer 1987).

One can classify metabolism and metabolites through different and somewhat arbitrary criteria. One customary division is between primary metabolism (all the compounds required to construct and maintain the cell) and secondary metabolism that helps the cell (or the organism) to cope with its environment, for example by fighting competitors (Demain and Fang, 2000). This scenario immediately suggests a requirement for compartmentalisation: primary metabolites should remain inside the cell, while secondary counterparts must get out to reach their targets. Antibiotics, the most conspicuous molecules of this type, are frequently released into the medium through specific transporters (Huang et al., 2006; Caballero et al., 1991). But the split primary/secondary metabolism is however very problematic. Not only do the two overlap (Craney et al., 2012) but often intermediary compounds of a metabolic pathway are released into the external medium. This is a typical situation in commensal environmental bacteria that degrade complex chemicals: the product of the first part of a given route is secreted and then captured by other bacteria that do the rest until complete mineralisation of the substrate (Pazos and Valencia, 2008). But bacteria possess also a nondiffusible metabolome that – unless cells are lysed – is never secreted. Other metabolites (typically amino acids and organic acids) can diffuse out the cells under certain circumstances, although the ease of their consumption often makes its concentration insignificant. Finally, compounds that are metabolised slowly have a chance to diffuse out and become available to members of the community other than those that produce them. This is well documented under the name of exometabolome (Paczia et al., 2012; Romano et al., 2014) or epi-metabolome (de Lorenzo 2008), the latter meaning the freely diffusible pool of chemicals available to the members of a microbial consortium. But one can also classify metabolites using other criteria, for example mapping them along the history of life before or after the explosion of chemical structures allowed by the emergence of dioxygen (Wang et al., 2011; Jiang et al., 2012). In this case, both O₂ as a gas and ROS determined the physico-chemical and regulatory solutions evolved to cope with high reactivity. Some of exemplary metabolites in this regard are exa-mined below.

DEALING WITH DEAD-END OR UNSAFE HOUSEKEEPING METABOLITES

Despite their apparently inconspicuous chemical formulas, housekeeping metabolites are necessarily reactive for playing their biological role. It is noteworthy that the catalogue of macromolecule-building blocks is limited to a narrow list of compounds out of the many more that could have been selected early in evolution to the same end. We focus below on the chemical logic behind the evolutionary choice of the canonical proteinogenic amino acids and the components of carbohydrates. Alas, we cannot cover the entire repertoire of macromolecular precursors (e.g. the origin of nucleosides). Still, these examples provide good study cases for the issue under examination.

Nonproteinogenic amino acids in the primary metabolism

Some amino acids are ubiquitous in the primary metabolism, but they do not form part of proteins. Why 20 amino acids were retained to this end while others did not? Many bacteria make homoserine, homocysteine (a neurotoxic for humans; Tchantchou 2006) and ornithine, but these amino acids are not found in ribosome-mediated protein synthesis. Yet, they are found in nonribosomal polypeptide synthesis (Cheung et al., 2009). This seems to reflect a conflict between the chemical properties of some amino acids and their loading in cognate tRNAs for translation. In fact, homoserine, homocysteine and ornithine are prone to cyclise when activated by ATP to make the aminoacyl adenylate, the form required for reacting with free tRNAs (Jakubowski and Fersht, 1981; Sekowska et al., 2000), thereby becoming unavailable for protein synthesis and crea-ting dead-end and energy-consuming intermediates. Citrulline is another conspicuous example. This amino acid is often present in proteins as the result of a post-translational modification either programmed or due to chemical ageing (see below), but it is not incorporated during translation. Why? Citrulline has a tendency to form isopeptides, thereby leading to branched or cyclic peptides (a property co-opted for nonribosomal peptide synthesis; Yin and Zabriskie 2006). As a consequence, a citrulline-specific tRNA has never made it to the protein synthesis machinery. Cells have thus evolved mechanism for tRNA editing to isolate nonproteinogenic from proteinogenic amino acids as well as noncognate proteinogenic amino acids (Yadavalli and Ibba, 2012).

Many other amino acids are found in primary metabolic pathways. In some cases, the same compound (e.g. lysine) has a canonical form (l-Lys) and a nonproteinogenic isomer (d-Lys). The issue here is that both forms are essential, but d-Lys (as do other d-amino acids) interferes with translation (Banik and Nandi 2012). Interestingly, bacteria produce l-lysine out of meso-diaminopimelate (one key component of murein), while Archaea and some Bacteria synthesise the same from α-aminoadipate, a nonproteinogenic homologue of glutamate (Busch et al., 2003; Nishida and Nishiyama 2012). But, one way or the other, murein requires d-amino acids (d-alanine and d-glutamate), which are produced through the action of l-amino acid racemases. The way to avoid mixing the pools of d- and l-amino acids is to have some of the machinery for the turnover of envelope components (e.g. d-amino acids) outside the membrane (Mattei et al., 2010; Alvarez et al., 2014). In addition, tRNA editing is frequently able to put d-amino acids aside. As a consequence, inactivation of such editing has clearly harmful consequences (Wydau et al., 2009).

Still, the most important case of toxicity associated to the synthesis of an amino acid is that of serine. This is an often overlooked matter that provides an archetypal example of the interplay between the chemical constraints that frame a given pathway and its evolutionary solution. As examined below, dealing with serine toxicity poses a serious practical problem for metabolic engineering which might be ultimately solved only through the construction of ad hoc compartments. This is clearly evident in plants (Christensen and MacKenzie, 2006), where se-rine biosynthesis follows different routes in each of three plant compartments (Ros et al., 2014).

The serine puzzle

It may come as a surprise that core-cell-building blocks may be harmful. However, as early as 1970, Cosloy and McFall (1970) identified mutants of E. coli highly sensitive to serine. Subsequently, while trying to identify the enzyme l-serine deaminase (dehydratase), Rasko and Alfoldi (1971) noticed that an excess of l-serine transiently inhibited bacterial growth. The effect was competitively reversed by threonine and noncompetitively by isoleucine or its precursor, 2-ketobutyrate. With the onset of the concept of allostery (established on threonine deaminase; Changeux 1961), the growth inhibition effect of serine was interpreted as the result of the competition with the chemically related threonine, which could be alleviated by the allosteric re-gulator isoleucine. However, it was found a few years later in a strain auxotrophic for threonine (and therefore in the presence of that amino acid in the growth medium) that addition of the one-carbon pool amino acids (serine, methionine and glycine) inhibited growth of a relA strain (Uzan and Danchin 1976). This mutation uncouples translation from rRNA and tRNA biosynthesis. Further work established that serine was the key agent of such inhibition and that the same amino acid also transiently inhibited growth in the wild type, not only a relA mutant. Note that RelA produces ppGpp as the result of accumulating uncharged tRNAs when translation is interrupted because of the dearth of amino acids. Further research found out that these phenomena were not related to allosteric regulation of threonine deaminase (as first thought, see above), but it reflected a more intricate scenario. In the absence of relA (and thus low ppGpp levels), serine prevents the derepression (i.e. expression) of the isoleucine–valine biosynthesis pathway, and thus, cells become phenotypic isoleucine auxotrophs (Uzan and Danchin 1978). In addition, E. coli cannot grow on serine as sole carbon source, despite its central position in metabolism (Newman and Walker 1982). All these unexpected connections make serine to be one of the most difficult amino acids to produce by fermentation at an industrially cost-effective level in E. coli.

Although most observations on serine toxicity have been made in E. coli growing on minimal medium (when the whole metabolism is in full gear), the biochemical basis of the negative effects of this amino acid is likely to be shared by many other organisms. Fortunately, there are alternatives to the serine pathway present in most bacteria: methylotrophic bacteria have a route to assimilate methanol that goes via serine (Peyraud et al., 2011). Yet, the production levels are still lower than that of other amino acids (Hagishita et al., 1996; Shen et al., 2010). The very high concentrations of serine released to the medium by some industrial Corynebacteria (Stolz et al., 2007; Lai et al., 2012) are probably enabled by the specific composition of the growth media and also by the very secretion of serine, which is in itself a detoxification mechanism (see Concept Box). In fact, the Corynebacteria could be a special case of mitigated serine toxi-city that makes this type of microorganisms appealing for industrial processes. Biotechnological issues notwithstanding, what is behind the surprising detrimental action of high concentrations of this amino acid?

The harmful effect of serine stems from two different che-mical faults at the interface between anabolism and catabolism (Fig. 1). Biosynthesis of most amino acids begins with formation of a α-ketoacid, which is subsequently trans-aminated by a variety of aminotransferases, the synthesis and activity of which is finely tuned for stoichiometric nitrogen assimilation (Hutson 2001; Muro-Pastor et al., 2005). Going in the other direction, catabolism of amino acids frequently involves the action of aminotransferases that yield α-ketoacids, which are subsequently decarboxylated. Amino acid synthesis and consumption are often physically separated (i.e. compartmentalised) in plants (Mooney et al., 2002). But in bacteria, the two processes are co-occurrent, and once a given amino acid reaches a certain concentration limit, its amino group is lost through the action of trans/de-aminases and then decarboxylated.

Under this general frame, serine seems to be dealt differently, as bacterial genomes persistently have a specific enzyme (annotated as l-serine deaminase-dehydratase: serine ammonia-lyase, EC 4.3.1.17; Fang et al., 2005; Acevedo-Rocha et al., 2013) with dehydratase activity that, following its action on serine, produces pyruvate. This is in contrast with the other housekeeping, ubiquitous aminotransferases (e.g. IlvE or TyrB in E. coli) which produce β-hydroxypyruvate out of the same substrate (Blatt et al., 1966). And, here comes the chemical angle: although similarly to other α-ketoacids β-hydroxypyruvate can be disposed of by decarboxylating enzymes (Hedrick and Sallach 1961b), the molecule is very reactive chemically, triggering unwanted effects on thiamine via formation of nonfunctional or suicide products (Thomas et al., 1990; Fiedler et al., 2002; Duggleby 2005; Karsten and Cook 2009) and taking over the role of pyruvate in other reactions involving α-ketoacids (Baykal et al., 2006). Furthermore, in the presence of O₂, hydroxypyruvate undergoes spontaneous auto-oxidation and triggers ROS formation (Hedrick and Sallach 1961a). Finally, hydroxypyruvate might also react with ammonium, giving the enamine hydroxyaminoacrylate, that is toxic via formation of dead-end products with aminotransferases (Karsten and Cook 2009). In this context, it makes sense that one extra serine dehydratase activity had to be persistently patched in the genomic complement of bacteria for preventing the build up of hydroxypyruvate out of serine.

But this is not the whole story, because serine dehydratase does not completely detoxify the side effects of se-rine metabolism. The same enzyme can also generate the reactive α-aminoacrylic acid (also known as 2-aminopropenoic acid or dehydroalanine; Seebeck and Szostak 2006), which can tautomerise to iminopropionic acid (Fig. 2). This metabolite is prone to react with nucleophilic groups (such as the ɛ-amino group of Lys or the -SH of Cys) resulting in formation of lysino-alaninyl and lanthionyl residues (Reid, 2000). Aminoacrylate is also produced by other enzymes that use serine itself or its variants (e.g. O-acetylserine and P-serine) as substrates. The same compound can be generated by enzymes of cysteine or tryptophan biosynthesis (Levy and Danchin 1988; Woehl et al., 1996). Other reactions leading to counterparts of aminoacrylate include transamination of threonine, which results in the toxic enamine intermediate vinylglycine. In reality, all enzymes using pyridoxal phosphate as cofactor produce enamine intermediates (Zheng et al., 1994; Sharif et al., 2006) that become toxic if they leave the catalytic site instead of reacting with their canonical substrates.

Cells are thus bound to live with a degree of stress stemming from essential chemical reactions and for which evolution has only found partial solutions. The biochemical fate of serine, as summarised above, exemplifies how major chemical hurdles plague the functioning of the cells. The complex mechanism by which bacteria deal with serine interference is examined later in this review. But the question remains as why this amino acid has not been entirely erased from the core metabolism? In fact, despite negative metabolic crosstalk, serine is hard to counter-select for, as it is both a structural component of proteins and the precursor of one-carbon metabolites. But this is not the only case: many other metabolites are surely prone to yield trouble-making reactions, although most of them remain to be discovered. One approach to explore them is the adoption of bio-orthogonal chemistry methods, in which reaction components must react rapidly and selectively with each other under physiological conditions in the presence of the plethora of functionality necessary to sustain life (Sletten and Bertozzi, 2009). In the meantime, we can examine how bacteria have created cognate detoxifying systems and compartments as a way to remediate – if not altogether solve, specific problems of the sort.

α-dicarbonyls and other reactive species

Harmless as they may look, many carbohydrates count among the most toxic metabolites owing to their qualification as α-dicarbonyl compounds and their ensuing ability to react with sulfhydryl and free amino groups (Jakas et al., 2008). It is well known that deoxyglucosone, glyoxal and methylglyoxal are potent protein-glycating agents (Saraiva et al., 2006) and that, more generally, α-dicarbonyls adducts are markers of aged proteins (Yim et al., 2001). Surprisingly, some of such reactive carbohydrate variants are not produced accidentally, but generated through dedicated pathways (e.g. synthesis of methylglyoxal from dihydroxy acetylphosphate in E. coli; Ferguson 1999), an issue that has caused much speculation. This may be an example of the controversy surrounding the role of reactive oxygens species (ROS), which are generally assumed to be deleterious, while they may be helpful under some circumstances (Watson, 2014). In any case, upon glycation, proteins tend to aggregate spontaneously forming microstructures that often go to the poles of the bacterial cell (Lloyd-Price et al., 2012). This is an interesting hint that cells organise their physical layout as a function of their age and suggests the existence of molecular devices that can be compared with a Maxwell's demon – as long as they distinguish the old from the new (see below). These devices should allow collecting aged proteins and move them to specific places. One extreme case of the kind in the microbial world is the different fates of budding yeasts where the mother cell is playing the role of a garbage bin (Erjavec and Nystrom 2007). α-dicarbonyl compounds thus rank along with archetypal reactive species (ROS, NRS, SSC, reactive chlorine species) as toxicants that cell must keep in check. And, as discussed below, compartmentalisation plays an important role in curbing their negative effects (Jacob 2006; Forrester and Foster 2012; Mishra and Imlay 2012; Gray et al., 2013).

Side reaction products

It is common to consider a metabolic pathway as a direct route going from a starting input (substrate) and ending up with a well-identified output (product). This oversimplified view misses common side products that do not appear to be immediately relevant to the pathway. H₂O or carbon dioxide are often taken for granted despite the fact that intracellular water is not simply a bathing medium of the cell: 70% of the water molecules in a bacterium result from metabolic transformations (Kreuzer-Martin et al., 2006). Reactions involving oxidoreductases provide another illustration of our sketchy description of pathways, where it is not unusual to do as if NADH and NADPH were equivalent (e.g. for many reactions the symbol NAD(P)H is used for describing the pathway of interest). Yet, under physiological conditions, NAD and NADP differ in their reduced states. Although the standard redox potentials of NAD and NADP are identical (E°′ = −320 mV), all organisms keep NADPH/NADP = 100 (E′= −380 mV) and NADH/NAD = 0.05 (E′= −280 mV), see Bennett et al. (2009). A similar ambiguity persists for pathways involving aminotransferases. These enzymes transfer an amino group from a donor molecule to relevant substrates, but, ty-pically, the origin of the NH₂ donor is given little importance. Yet, it does make a difference in terms of the fate (and possible toxicity) of the corresponding reactions. There are, for instance, cases where the α-NH₂ donor comes from glutamine or asparagine instead of glutamate or aspartate (Berger et al., 2003). In these cases, the products of the transamination are 2-ketoglutaramate or succinaramate, two compounds somewhat alien to the customary cell metabolism and which need to be hydrolysed by an omega-amidase to prevent their toxic effects or entry into further unwanted pathways (Belda et al., 2013; Cooper and Kuhara, 2013). The same compounds can be cycled to produce, for example, reactive 2-hydroxy-5-oxoproline and succinimide. While such glutamine-dependent reactions are used in fungi to drive ammonium acquisition (Mora 1990), it is intri-guing why some bacteria have also kept them in face of such downstream chemical consequences. It is possible that some sort of compartmentalisation relieves the potential damage in these cases (Ogawa et al., 1996). However, a compelling reason for using these otherwise problematic reactions is to push a pathway in a particular direction using hydrolysis as the dri-ving force. This is illustrated by the methionine salvage pathway, which must prevent its catabolism via its cognate keto acid and uses glutamine, not glutamate, as the alpha-amino group donor (Belda et al., 2013).

Metabolic imbalance

The ratio between the various pools of metabolites has no mechanisms for being constant in time. Changes in the composition of the environment will often affect the level of some compounds within the cell, and in turn, this may affect the re-lative concentrations of other metabolites, thereby causing useless perturbations. One notable case is that of the pools of phosphorylated metabolites. Because many sugars are transported in bacteria through the phosphoenolpyruvate:phosphotransferase system (PTS), intake of the corresponding carbohydrates typically depletes the pool of phophoenolpyruvate and increases intracellular levels of phosphosugars, thereby causing a distinct type of metabolic stress (Richards et al., 2013). Similarly, reactive phosphorylated compounds such as acetylphosphate or carbamoylphosphate need to be somehow compartmentalised, as they are prone to acetylate (Kuhn et al., 2014), carbamoylate (Shomura and Higuchi, 2012) or phosphorylate (McCleary and Stock 1994; Wolfe 2010) important enzymes, structural components or regulators with considerable consequences on the overall behaviour of the cell.

Compounds that permeate or disrupt membranes

If the cell is the ‘atom’ of life (i.e. the minimum live entity endowed of autonomous existence), the key feature of a cell is being surrounded by a lipid bilayer impermeable to water and to charged molecules. This allows bacteria to build up the electrochemical gradient used as the driving force to select imported and exported molecules. Compounds that can permeate the membrane (whether via solubility in the lipid bilayer or through the protein complement of the membrane; Kell, 2013) tend to diffuse in or out of the cell where they interplay with the biochemical network. However, some metabolites, for example dipolar compounds acting like detergents, can destroy the bilayer structure of the membrane. Also, uncharged short chain fatty acids (e.g. acetate) and aromatic molecules produce deleterious effects: at low concentration they equilibrate between the inside and the outside, but at higher doses they can disrupt the membrane's structure. These circumstances become important in some cell locations. Because of respiration (that releases protons to the outside of the cytoplasmic membrane), the periplasm behaves as an acidic compartment. Consistently with this, the optimum reaction pH of typical periplasmic enzymes (e.g. the acidic phosphatase/phytase) is very low (c. 2.5; Greiner et al., 1993). This makes acetic acid not to be charged and thus able to cross the membrane without interference with the membrane potential. Although protons in the periplasm tend to equilibrate with the external pH (Martinez et al., 2012) E. coli often goes through low-pH environments (e.g., the mammalian stomach) that maintain all the milieu outside the cytoplasmic membrane acidic. In these cases, bacteria must not only react to a general stress condition (low pH) but also to counteract the undue traffic of uncharged fatty acids (Kanjee and Houry, 2013) and aromatic molecules. Not in vain benzoic acid is used as a food preservative that enhances killing of bacteria in acid medium (Friedman et al., 2003).

In fact, aromatic compounds are particularly prone to be membrane permeable (see e.g. Corre et al., 1990), and this imposes considerable constraints on the way they can be metabolised. The behaviour of aromatics in water is still not fully understood and sometimes counterintuitive. For example, des-pite the presence of an extra hydroxyl group, tyrosine is much less soluble than phenylalanine (Hernandez et al., 2010). Aromatics are usually assumed to aggregate by stacking onto one another. Yet, in proteins, they make stable interactions that are usually perpendicular to one another (edge to face; Hunter et al., 1991). Such connections possibly enable proteins rich in aromatic residues to make gluons that stabilise multisubunit complexes (Pascal et al., 2005). The molecular mechanisms by which aromatics interfere with both the lipidic and the protein components of the membranes (including the protein–lipid interface) have been studied in detail using phenethyl alcohol as a model damaging compound (Silver and Wendt 1967; Killian et al., 1992; Anbazhagan et al., 2010). It seems to be the rule that hydrophilic and charged groups make aromatics to remain on either side of the lipid bilayer. How this physically fits with the known biochemical routes for production, consumption and transport of aromatic compounds remains an open question, although logic would anticipate a degree of compartmentalisation to avoid biochemical traffic jams.

Finally, gases can interact with membranes as well. Although some textbook views argue that biological membranes are highly permeable to volatile molecules (Gutknecht et al., 1977; Windrem and Plachy 1980; Wang et al., 2010), this may not be entirely true, as dedicated membrane proteins act as gas channels (Itel et al., 2012; Kell 2013). Contemporary metabolic engineering often sees gas release as a way to alleviate toxicity of a given product using diffusion and external traps to maintain a low constant pressure in the planned factories (van Leeuwen et al., 2012). Yet, in the real bacterial world, the problem is not releasing gases so much but preventing their leakage when needed for cell metabolism. Carbonic anhydrase, for example, is essential to preserve the level of intracellular CO₂ necessary for many reactions of intermediate metabolism (Lotlikar et al., 2013). Finally, preventing escape of intracellular gases can be evolutio-narily co-opted to change the buoyancy of the cells (Tuval et al., 2005).

PARALOGOUS METABOLISM

Chemical challenges do not stem only from canonical reactions in central metabolic networks, as shown in the examples above. It happens also that a large number of physico-chemical processes at the temperature of life – including mere stochastic variations – yield a significant proportion of errors in otherwise well-accredited enzymatic reactions. While standard metabolism operates, variants of the correct reaction products are synthesised accidentally at a low level (Van Schaftingen et al., 2013; Chan et al., 2014), forming chemical homologues of the basic building blocks of the cell: nucleotides and amino acids. Furthermore, proteins age as a result of their inherent chemistry and metabolic accidents (Robinson and Robinson, 2004; Gonidakis and Longo 2013). Aged proteins, which typically bear modified residues, are subsequently degraded into variants of the standard proteinogenic amino acids (Helou et al., 2014). Although the pool of such misfit metabolites is relatively minor, it cannot be ignored, as it constitutes a sort of shadow of the core biochemical network that we have named paralogous metabolism (Chan et al., 2014). That such parallel metabolism is not to be neglected is indicated by a large number of observations on the unexpected presence of nonproteinogenic amino acids that end up incorporated in proteins upon entering the metabolic pool as side products of branched-chain amino acids biosynthesis. Typically, noncanonical amino acids such as norvaline, norleucine, homoisoleucine and β-methylnorleucine are found in recombinant proteins overproduced in E. coli (Apostol et al., 1997; Muramatsu et al., 2002). Norvaline production is related to an imbalance of the synthesis of the branched-chain amino acids under conditions were pyruvate level is high (Kisumi et al., 1976, 1977). Consistently with this, a downshift in O₂ availability is sufficient to cause a considerable norvaline accumulation (up to mM levels) as a consequence of the build up of pyruvate and its chain elongation over α-ketobutyrate and α-ketovalerate. Pyruvate and α-ketovalerate then yield norleucine (Soini et al., 2008), which is an excellent analogue of methionine (Budisa et al., 1995) that can substitute for the bulk of methionine residues in proteins under conditions of amino acid limitation in E. coli (Gilles et al., 1988; our unpublished experiments). Even very small amounts of unwanted metabolites may have considerable consequences in terms of fitness of the organism. Cells must therefore have evolved a variety of ways to counteract the effects of paralogous metabolism, although the matter remains largely unexplored.

SECONDARY METABOLISM

Occupying a specific niche is essential for all organisms, and niche destruction is often equivalent to dooming the species to extinction. Microbes do not escape this constraint, and a large part of their genome, what has been called the cenome (Acevedo-Rocha et al., 2013), is earmarked to coping with specific environments. Some functions encoded in the cenome deal with degradation of the products of the paralogous metabolism, most of which are toxic for the cell (Chan et al., 2014). But other functions found in this variable share of genome contents typically correspond to synthesis of a large number of compounds that deli-neate a secondary metabolism that allows the organism to thrive in its niche. One archetypal example is the synthesis of compounds with antibiotic activity that act both as quorum-sensing signal and as deterrents for unwanted competitors (Davies and Davies 2010). A major family is that of polyketides (Hertweck et al., 2007; Jenke-Kodama and Dittmann 2009; Zhan, 2009) and other peptides synthesised by nonribosomal machinery (Condurso and Bruner 2012). These compounds are often toxic, and the organism must produce them under special circumstances or in special cell types (e.g. aerial mycelia in Streptomycetes (Vetsigian et al., 2011). These antibiotics often contain ci-trulline, dehydroalanine or ornithine (see above) as well as other nonproteinogenic amino acids such as β-hydroxyasparagine (Worthington and Burkart 2007) or chloro-β-hydroxytyrosine (Puk et al., 2004).

The machinery that allows synthesis of many secondary metabolites with the help of polyketide synthases (Van Lanen and Shen 2008) and nonribosomal peptide synthetases (Condurso and Bruner 2012) is yet another example of the way a cellular device is able to limit diffusion of intermediates, preventing their harmful action through construction of scaffolds that resemble veritable assembly lines (Danchin, 2012). In these enzymes, a sequence of modules channelling intermediates accommodates the sequence of reactions that leads to a final, generally complex, compound, using a swinging arm, 4-phosphopantetheine (Zettler and Mootz 2010).

The number of secondary metabolites originating from either dedicated pathways or side products of canonical enzymatic activities keeps on expanding. One remarkable case is that of the α-ketoglutarate-dependent l-isoleucine dioxygenase of Bacillus thuringiensis which, by default, generates hydroxyisoleucine out of its canonical substrate. However, depending on the amino acid entered in the reaction, the same enzyme catalyses three different types of oxidations: hydroxylation, dehydrogenation and sulfoxidation. For example, this enzyme is able to sulfoxide sulphur-containing l-amino acids, thereby converting l-methionine in the corresponding methionine sulfoxide (Hibi et al., 2011). Analogues of nucleosides are also ubiquitous as the result of secondary reactions and paralogous metabolism (Budovsky et al., 2010).

ONE-POT SOLUTIONS FOR CURBING TOXICITY OF METABOLIC INTERMEDIATES

Whether from essential metabolic reactions, from environmental conditions (e.g. low pH) or from secondary/paralogous enzymatic reactions, the state of affairs is that cells do produce reactive chemicals that are prone to divert pathways from their expected course. Note that from an evolutionary perspective, even a small concentration of a highly reactive intermediate may reduce fitness. This situation asks for evolutionary strategies that have made bacteria to be generally immune to their own harmful compounds. The positive outcome may happen without involvement of compartmentalisation (i.e. one-pot solutions) or with the establishment of actual barriers to free diffusion. The first type of scenarios includes (bio)chemical stratagems as well as regulatory tinkering. Some archetypal cases are examined below.

Preventing accumulation

The simplest and most direct way to dispose of harmful intermediates is to degrade them as soon as they reach a detrimental level. However, this obvious tenet can be extremely difficult to implement biochemically. The issue of deleterious serine effects enunciated above and how bacteria manage to overcome is a good example in this regard. As mentioned before, serine prospective noxiousness is the consequence of two processes. First, serine transamination yields a reactive product, hydroxypyruvate, the accumulation of which is prevented by diverting serine catabolism towards pyruvate using serine dehydratase. However, the very detoxification reaction produces another hazardous intermediate, 2-aminoacrylate, albeit at a low level. Cells have two different problems to solve when they have much serine: dealing with excess hydroxypyruvate and dealing with excess aminoacrylate. Bacteria thus need to strike a fine balance between the two equally knotty fates of serine. For this, we have to consider the outcome of each compound.

First, let us deal with 2-aminoacrylate. When analysing the gene encoding serine dehydratase in a suite of bacterial genomes, the presence of a persistently associated gene (called yjgF in E. coli; Fang et al., 2005) became evident. This gene encoded an enigmatic enzyme belonging to the ubiquitous YjgF/YabJ/Yer057p/UK114 family of proteins, which is functionally related to various effects of serine (Enos-Berlage et al., 1998) despite the lack of clear known mechanisms. The YjgF protein was suspected of acting on analogues of ketoacids, because analysis of the 3D structure of the E. coli's homologous variant (TcdF) was predicted to interact with 2-ketobutyrate and/or its unstable enamine precursor (Parsons et al., 2003; Burman et al., 2007). The role of YjgF was recently proven in Salmonella enterica as an enzyme able to deaminate reactive enamine/imine intermediates of pyridoxal 5′-phosphate (PLP)-dependent enzyme reactions and thus able to detoxify 2-aminoacrylate coming from serine dehydration (Lambrecht and Downs 2013). The multiple targets of aminoacrylate account for the complex phenotype of mutants of S. enterica unable to get rid of this reactive compound (Flynn et al., 2013). The YjgF protein has been renamed RidA (Lambrecht et al., 2012), and it provides a mechanistic rationale of how bacteria deal with one of two issues involved in serine toxicity. The other issue is, as mentioned above, elimination of hydroxypyruvate. This requires a more complex pathway that, interestingly, is compartmentalised in plants (Timm et al., 2011; Ros et al., 2014). The most widespread solution is isomerisation of hydroxypyruvate into tartronate semialdehyde (Njau et al., 2000), which is subsequently oxidised into d-glycerate (Osipiuk et al., 2009), although other alternative routes may exist.

Serine is just one case of intermediate metabolites that are prone to generate reactive chemicals. Aminoacrylate is also produced during cysteine biosynthesis (Fig. 3). The curbing of such compounds is likely to be found among those encoding persistent proteins such as E. coli YggS (B. subtilis YlmE) and other thus far unexplored activities encoded in the bacterial genome.

Toxicity sinks

Hazardous molecular species can also be rendered ineffective by making them react immediately after their synthesis with chemical sinks that remove them from the medium. This is a widespread mechanism for dealing with reactive oxygen species (ROS), which most often involves the sulphydryl groups deli-vered by methionine either as the free amino acid or already inserted in proteins. That methionine is highly reactive towards ROS is illustrated by the hypertoxicity of cigarette smoke on individuals with alleles of the lung α-antitrypsin lacking a methionine residue (Gazzani et al., 1989). Upon contact with ROS, Met residues of polypeptide chains are modified to either of two sulfoxide stereoisomers in the protein backbone. Subsequently, methionine sulfoxide reductases repair this modification back to the original methionine residue (Zhang and Weissbach, 2008), thereby completing the ROS detoxification cycle. These methionine sulfoxide reductases use NADPH as the reducing cofactor (not NADH), highlighting importance of this pathway during anabolism. This mechanism allows methionine residues in proteins to behave as chemical sponges which, if located on the protein surface, shield polypeptides against ROS damage (Ezraty et al., 2004; Bender et al., 2008; Luo and Levine 2009). Cys residues may also act as ROS buffers through the action of peroxiredoxins (Dietz, 2011). However, cysteine is prone to be further oxidised into forms that can no longer be repaired. Additional protein repair systems are instrumental also for reverting the damage made by other reactive species, for example the undue binding of α-dicarbonyls to -NH₂ and -SH groups (see above) using de-glycating enzymes (Deppe et al., 2011).

Protection/deprotection of reactive groups

Production of amino acids, nucleotides and other macromolecule-building blocks often involve long step-wise and frequently branched routes that are reminiscent of strategies that chemists draw for synthesis of organic compounds. A frequent challenge in organic chemistry is the preservation of given reactive groups in the final molecule through the intermediate steps of the process. One recurrent strategy to this end is to make such groups nonreactive via protective chemical modifications and removing the protection at the end of the synthesis scheme, thereby delivering the chemical of interest at the end. The same problem (reactivity of intermediates in a stepwise pathway) repeatedly happens in the cell's metabolism, and, amazing, as it may look, the solution is also similar: protection and deprotection of problematic functional groups of the molecules at stake in different stages of the biosynthetic route. Again, the case of serine provides a remarkable example of how metabolic challenges recurrently lead to surprising solutions. How does it work in this case? Some textbooks state that serine originates in direct transamination of hydroxypyruvate, but this does not always reflect the actual biosynthetic pathway (Snell 1986). This solution appears to be used only in one particular compartment of plant cells (Ros et al., 2014). Indeed, we have discussed already that hydroxypyruvate may be harmful and therefore such a direct transformation is hardly viable in the cell's context. In reality, serine is synthesised from phosphohydroxypyruvate, not from hydroxypyruvate. This apparently neglectable detail is in fact highly meaningful. Why synthesis of each serine molecule requires wasting a phosphate bond that required energy to be built up if there were a direct manner to generate the amino acid? The rationale for this odd occurrence is that the phosphate group alleviates the reactivity of hydroxypyruvate. In this way, transamination operates not on hydroxypyruvate (to yield serine) but on phosphohydroxypyruvate to produce phosphoserine. Hydrolysis of the phosphate group of the phospho-amino acid then yields the final product, serine.

The protection/deprotection evolutionary strategy for hand-ling toxicity of metabolic intermediates appears to be quite general. Modifications of reactive groups involve not only phosphorylation (as in the case of serine biosynthesis), but also methylation and acylation (e.g. succinylation or acetylation). In particular, acetylation reactions have been recruited to this end in many different contexts, from arginine biosynthesis via N-acetyl-glutamate and N-acetyl-ornithine to lysine biosynthesis via N-acetyl-diaminopimelate (Dairi et al., 2011). Succinylation is used as well in many organisms, sometimes overlapping acetylation (Ledwidge and Blanchard 1999; Lal et al., 2014). These modi-fications are often instrumental to protect producing strains from antibiotic action (as a sort of immunity against its effects; Benveniste and Davies 1973; Davies and Benveniste, 1974). Management of the acylation process can even be a way to sepa-rate anabolic and catabolic reactions: in Pseudomonas aeruginosa arginine catabolism proceeds mostly through succinylated rather than acetylated compounds (Jann et al., 1986). Acetylation also curbs unwanted reactivity of intermediates in degradation pathways. A dramatic example is observed in the disposal of S-alkyl-cysteine species generated by the paralogous metabolism (see above), the first step of which is an N-acetylation reaction that prevents the compound from interfering with the translation machinery. This is then followed by de-acetylation when catabolic enzymes (in this case, monooxygenases) have modified the molecule back to cysteine (Chan et al., 2014).

Regulatory solutions to metabolic traffic jams

One intriguing mechanism that bacteria have evolved for avoiding co-occurrence of conflicting metabolic routes and toxic intermediates is to regulate the production of both the enzymes that create the engagement and others that help to alleviate the problem. The co-habitation of two different catabolic pathways for degradation of benzoate and 3-methylbenzoate in the soil bacterium Pseudomonas putida mt-2 is a good case example of this (Fig. 4). This strain is the carrier of the so-called TOL plasmid pWW0, which encodes one of the best characterised routes for degradation of aromatics (Ramos et al., 1997). This system allows mineralisation of toluene and m-xylene. It is composed of two operons encoding an upper and a lower pathway. One CH₃ group of the substrates is sequentially oxidised through the enzymes of the upper route, to render benzoate (Bz) or 3-methylbenzoate (3MBz), respectively. Bz is then converted to catechol by either the lower route enzymes (encoded by xylXYZL) or by the products of the highly similar genes benABCD encoded in the chromosome, both of which being induced by Bz. The resulting catechol can then enter either a meta-cleavage pathway led by catechol 2,3-dioxygenase (encoded by the xylE gene) or an ortho-fission route initiated by the chromosomally encoded catechol 1,2-dioxygenase. In contrast, the generation of 3MBz (formed from m-xylene catabolism) into the corresponding 3-methylcatechol creates a metabolic conflict, because the subsequent action of catechol 1,2-dioxygenase generates 2-methyl-2-enelactone, a dead-end toxic product (Timmis et al., 1994). The solution to this conundrum is regulatory. Benzoate activates expression of both the plasmid-encoded and chromosome-encoded routes for its biodegradation, while 3MBz activates only the plasmid-encoded route. The consequence is that metabolism does not have a chance to build the toxic product (Perez-Pantoja et al., 2014) as 3MBz never sees the enzymes that drive it to a dead-end. But, on top of this, a second regulatory device (which we have called metabolic amplifier, Fig. 5) ensures that the actual levels of intracellular 3MBz are kept to a minimum (Silva-Rocha et al., 2011). How? The transcriptional control circuit of the TOL plasmid is such that when cells are exposed to m-xylene (the precursor of 3MBz), this aromatic substrate activates simultaneously expression of the upper operon (for conversion m-xylene → 3MBz) and the lower operon (3MBz → TCA cycle intermediates). As a consequence, the enzymes for degradation of 3MBz are in place before the cells have actually generated this compound. This is a fascinating case of metabolic memory and anticipation in which the entire regulatory network seems to be shaped by the need to avoid any significant accumulation of a potentially conflicting metabolite: 3MBz (Silva-Rocha et al., 2011).

Finally, because of the same genetic and metabolic arrangement, growth of P. putida mt-2 on benzoate only creates an additional problem, as the simultaneous action of the plasmid-encoded toluate dioxygenase (xylXYZL) and the chromosomal benzoate dioxygenase (benABCD) generates a high intracellular level of catechol, which is toxic beyond a threshold level. In this case, the solution is a chemical/genetic patch involving a second catechol 1,2 dioxygenase with a high K_M for the substrate, so it becomes operative only when catechol has reached a high intracellular level. This extra dioxygenase thereby provides a metabolic safety valve for excess catechol that alleviates the metabolic conflict generated by simultaneous expression of the meta and ortho pathways (Jimenez et al., 2013). Regulatory tinkering thus seems to ease the coexistence of pathways that run nonfully compatible metabolic routes. This happens in the case of the TOL plasmid by creating a sort of traffic control that ensures conflicting biochemical actors never to occur at the same time in the same place.

PHYSICAL MECHANISMS FOR CURBING TOXICITY IN ISOLATED CELLS

Sophisticated as they are, one-pot solutions to chemical reactivity or toxicity may not cover the entire repertoire of biological needs, in particular when cells may need to employ reactive intermediates for given functions or when metabolically significant gases need to be prevented from escaping into the environment. The evolutionary answer to these challenges is generally more physical than (bio)chemical, although both merge in the final outcome as exposed in the cases discussed below.

Coping with intracellular build up of unfavourable chemicals

Deleterious effects of metabolites may stem not only from their inherent chemical reactivity but also from their intracellular accumulation beyond a certain osmotic tolerance. While electrochemical potential is necessary for the cell to interact efficiently with its environment, aquaporins and glyceroporins-related permeases allow facilitated diffusion of water, glycerol and other polar, but generally uncharged compounds, without significant effect on osmotic pressure (Rosen and Tamas 2010). In contrast, active permeases allow metabolites to climb up concentration gradients up to a thousand-fold. As a consequence, when given metabolites increase their cellular concentration (either because of endogenous synthesis or environmental import), osmotic pressure can easily build up, thus putting cells at the risk of breakup. To face this challenge, two strategies have evolved, both to be found in extant bacteria. First, the metabolite at stake may be polymerised, thereby creating a sort of local domain with defined boundaries that considerably lowers osmotic pressure while keeping the polymer as a reserve material. Typically, this is what happens with glucose, which can form a variety of polymers, commonly glycogen. Interestingly, the reversible conversion glucose |${\leftrightarrows}$| glycogen acts as an osmotic buffer in some bacteria (Seibold et al., 2007; Xu et al., 2013). Excess of intermediate carbon compounds can lead also to reversible formation of intracellular polyesters (polyhydroxybutyrate and other polyhydroxyalkanoates) which can depolymerise when needed for supplying carbon back to the central metabolism (Escapa et al., 2012; Khosravi-Darani et al., 2013). A second solution to prevent osmotic pressure is mediated by membrane-bound proteins the function of which is reminiscent of the safety valves of pressure cookers in that they export metabolites when their concentration reaches an unsafe level. To distinguish the in-form and the out-form of the same metabolite, the corresponding molecules might be earmarked for export by means of a covalent modification (e.g. acetylation) to avoid the set-up of a futile permeation cycle. This could be one role of some of the many acetyltransferases and ABC transporters encoded in bacterial genomes whose role is uncertain. The case of the lactose acetyltransferase encoded by the lacA gene (the last gene of the lac operon) is quite revealing in this respect. LacA acetylates lactose with a K_M indicative of a low affinity for the substrate, so the reaction may occur only when the intracellular concentration of sugar is high (Danchin 2009c). It is thus possible that each type of efficient permease is matched by a cognate device for controlling osmotic pressure, whether through polymerisation (if the compound at stake is amenable to it) or by straight expulsion after the excess of the metabolite is pretagged through a covalent reaction. This is the simplest case of compartmentalisation for avoiding toxicity.

Specific export also tackles the problem of disposal of leftovers (toxic or not) of biochemical pathways as well as transient concentration peaks or ordinary metabolites. This is a central problem in chemical engineering that has been surprisingly overlooked when studying the fate of macromolecule biosynthesis despite the fact that bioengineering has been widely exploited for the production of small molecules. Every consolidated chemical process needs to streamline the balance of input and output atoms of the synthetic pathway to the inflow and outflow of feedstocks and intermediate products in the core reactor. By the same token, cells need to fit the relative amount of nitrogen and sulphur atoms needed for amino acid biosynthesis to the influx of carbon and this may require adjustment of import/export of the corresponding metabolic intermediates (Venditti et al., 2013; You et al., 2013). In some cases (e.g. polyamine metabolism in E. coli), valuable intermediate molecules like methylthioribose are exported as such (i.e. the are not degraded further) for the sake of maintaining the balance of the cognate pathway (Hughes 2006). Even amino acids (e.g. cysteine, lysine) are exported when their concentrations reach a threshold level (Vrljic et al., 1996; Ohtsu et al., 2010). The active agents of such a control often include exporter systems of the MDR family that are expected to keep fluctuations of metabolic pools within limits for optimal performance of the chemical network (Jack et al., 2001).

Separation between anabolic and catabolic reactions

The frequent sharing of the same metabolic intermediate in both anabolic and catabolic reactions poses an interesting case of chemical decision making, as cells have to resolve whether the same molecule will go upwards for constructing biomass or downwards to waste after extraction of useful chemical groups and energy. Proline catabolism in Salmonella enterica serovar Typhimurium is a case in point. In the presence of proline, PutA (a peripheral membrane protein that exhibits activity both of a proline dehydrogenase and a repressor of the put operon) associates with the cytoplasmic membrane. When proline levels decrease, PutA accumulates in the cytoplasm where it is free to interact with put-specific operator sites and repress transcription. In order for PutA to accomplish these multiple tasks, it shuttles between the membrane and cytoplasm in response to the intracellular concentration of proline. Such alternative localisation of PutA allows to earmark cytoplasmic pyrroline-5-carboxylate for either biosynthesis of proline or as an intermediate of its catabolism (Surber and Maloy 1998).

Metabolism of quinate provides a further illustration of the problem as well as its compartmentalised solution. Quinic acid is a cyclitol present in many plants and thus a nutrient-to-be for plant-associated bacteria. Catabolism of the molecule goes via 3-dehydroshikimate, which is also an intermediate in anabolism of aromatic metabolites. How do bacteria deal then with this compound? The solution is the physical separation of the downward and the upward reactions. The 3-dehydroshikimate anabolic pathway proceeds through cytoplasmic enzymes that convert it in shikimate and chorismate, the precursors of virtually all aromatics of the cell. In contrast, the first steps of quinate degradation and the ensuing catabolism of 3-dehydroshikimate occur in the periplasm at least in Acinetobacter, Pseudomonas and related species. The periplasmic dehydrogenase producing 3-dehydroshikimate out of quinate is a pyrroloquinoline quinone (PQQ) enzyme (Vangnai et al., 2004). PQQ is the cofactor of many dehydrogenases acting on periplasmic substrates (including glucose; Snell 1986), in contrast with equivalent anabolic cytoplasmic dehydrogenases, which use NADP (Adachi et al., 2010). Although details remain elusive at the time of writing this review, it seems that the subsequent periplasmic catabolism of 3-dehydroshikimate yields products that are transported into the cell and feed to the TCA cycle (Tresguerres et al., 1972). In view of all this, it is tempting to generalise that in–out division afforded by the cytoplasmic membrane and the periplasm allows separation of a number of processes that would be incompatible otherwise if they were in the same reaction vessel. Going back to the chemical engineering rationale, it makes sense that enzymes dealing with general degradation processes (e.g. proteases or nucleases) are kept physically disconnected from the inner biochemical network. And that the products of their reactions are subsequently transported within the cell where they can be used to construct the biomass. One extreme manifestation of such a cellular architecture is the presence of feeding complexes on the surface of some bacterial pathogens. Capnocytophaga canimorsus feeds on mammalian cells by harvesting the glycan moiety of glycoproteins, that is shaving the glycans away from the glycosylated proteins of the cell surface and using them as nutrients (Mally et al., 2008; Renzi et al., 2011). The correspon-ding feeding machinery, which is displayed on the bacterial envelope, is reminiscent of the celullosomes that some environmental bacteria use for degrading cellulose into metabolisable sugars (Bayer et al., 2004).

THE MAKING OF COMPARTMENTS PROPER

Excepting the separation of anabolic/catabolic reactions by the physical action of the cell membrane, we have taken thus far the term compartment in the noncommittal sense of physical spaces bounded by hindrances to free diffusion. This includes, for instance macromolecular complexes, localised synthesis of proteins according to their chromosomal location and membrane ancho-ring of related enzyme partners (often at the poles; Rokney et al., 2009). In general, large complexes have less room for free diffusion. It has been argued, for instance, that transcription and translation happen in defined intracellular sites and that the whole-gene expression flow is implemented through a highly organised superstructure that uses the chromosome layout as a template (Norris et al., 2007). But even in these cases, the physical boundaries at stake limit diffusion rather than impeding it altogether. However, bacteria deal also with toxicity and separation of incompatible processes by means of self-aggregating structures that cause a sharp split between the corresponding actors located inside or outside them. Most of such structures generate a variety of specialised microcompartments. These include the acidocalcisomes, membrane-bound organelles containing polyphosphates and calcium which are present from Bacteria to higher Eukarya (Docampo et al., 2005). The volutin granules of polyphosphate may be a particularly simple form of these organelles, meant to store an ultimate form of energy in the cell which seems to be used for managing ageing and senescence (Danchin, 2009b). Bona fide acidocalcisomes are surrounded by a membrane and resemble lysosomes (Docampo and Moreno, 2011). Anammoxosomes of ammonium-oxidising bacteria play the important role of detoxifying toxic NO₂- and NH₃ into dinitrogen N₂ under anaerobic conditions. Such structures retain NH₃ inside the cells, as this chemical species is a volatile gas at neutral/basic pH and thus tends to be little available for bacterial growth. The anammoxosome performs the task while recovering carbon from CO₂ and energy (Agrawal et al., 2011; Kartal et al., 2013). Carboxysomes have similarly evolved to keep a gas (CO₂) inside the bacterial cell and make it available for a suite of metabolic pathways. They are large, polyhedral, cytosolic microcompartments that capture ribulose 1,5-bisphosphate carboxylase/oxygenase (RuBisCO) and carbonic anhydrase for carbon fixation in Cyanobacteria and other carbon-fixing microorganisms (Bonacci et al., 2012; Roberts et al., 2012). Note, however, that bicarbonate is formed by carbonic anhydrase both in the cell cytoplasm and in the carboxysomes where it follows different metabolic fates. A different type of functional compartment is the chlorosome, the major light-harvesting antenna complex in green sulphur bacteria, filamentous anoxygenic phototrophs and phototrophic acidobacteria (Oostergetel et al., 2010; Adams et al., 2013). They contain self-assembling bacteriochlorophyll c, d and e as their principal light-harvesting pigments along with variable amounts of carotenoids. For an entirely different role, ethanolaminosomes of Salmonella and Clostridium enable bacteria to use ethanolamine as carbon and nitrogen source (Tanaka et al., 2010; Pitts et al., 2012). Their layout is intriguing, as they consist of structures surrounded by hexameric tiles with a pore for exchange with the outside. This structure contains the enzymes for ethanolamine catabolism in an arrangement where the B12-dependent metabolism occurs at the outside of the organelle, acetaldehyde being formed and degraded by enzymes present in its lumen so as to prevent its toxic action (Huseby and Roth, 2013). Also, for a very different and still uncertain function, magnetosomes are single-domain iron-mineral crystals of either magnetite (Fe₃O₄) or greigite (Fe₃S₄) confined within a lipid-bilayer membrane (Bazylinski and Frankel, 2004). They are widely spread in Bacteria, and they are believed to be engaged in geo-metabolic processes that involve iron. The genes coding for the synthesis and maintenance of magnetosomes are split in genomic islands, comprising the mam, mtx and mms clusters (Komeili 2012). They may have been present very early in evolution, as it is likely that iron sulphur particles played a seminal role in the origin of life (Wachtershauser 2007). Other intracellular structures have a more explicit metabolic role. For instance, propanediol organelles degrade 1,2-propanediol in a coenzyme B12-dependent manner in certain enteric bacteria. The role of the microcompartment is to protect the cell against the deleterious action of the toxic intermediate propionaldehyde that is produced during propanediol degradation (Sampson and Bobik 2008). In a related fashion, Clostridium phytofermentans possesses three loci encoding polyhedral microcompartments which contain the fermentation of fucose and rhamnose via 1,2-propanediol (Petit et al., 2013), their role being presumably the same to degrade propanediol as in, for example, Salmonella. Finally, microcompartments found in Bacteria and Archaea can also capture gases, form intracellular vesicles and function thereby as flotation devices to maintain cells at a suitable depth in water columns. The wall of these vesicles is permeable to gas molecules and is composed of a small hydrophobic protein, GvpA, which forms a single-layer wall. The structure is assembled with the help of accessory proteins (Pfeifer 2012).

The repertoire of microcompartments just enumerated is by no means complete. Comparative genomic studies has revealed the presence of numerous gene clusters that are similar to others coding for identified structures. Alas, many of them are automatically annotated as coding for unknown metabolic functions (Jorda et al., 2013). The structure of many of these microcompartments is reminiscent of viral capsids (Bobik 2006; Yeates et al., 2007; Rapaport 2010). They look often as polyedral organelles that may have been recruited from prophages and co-opted to compartmentalise metabolism (Bobik 2006). There could even be an evolutionary connection between viral capsids and virus-encoded metabolic reactions: it might useful for a virus to carry over some of the functions that could allow its own multiplication by helping its host organism or by creating pathways that are virus specific. Although this notion is difficult to generalise with available data, a gene cluster borne by a cyanophage has been reported to code for synthesis of the alternative DNA base 2,6 diaminopurine (Kirnos et al., 1977). Furthermore, an entire photosynthesis system I has been identified in the genome of marine bacteriophages (Sharon et al., 2009). Viral in origin or not, intracellular organelles and microcompartments are currently considered as good assets for engineering cells: their shell could in principle be redesigned for encasing a specific metabolic process, thereby preventing the leak of toxic intermediates (Frank et al., 2013).

METABOLISM FRAMES EMERGENCE OF MULTISCALE COMPLEXITY

Beyond its impressive technical achievement, the genome transplantation experiment by (Lartigue et al., 2007) that changed one bacterial species into another is conceptually of major importance for reasons that may not be evident in a first sight. This work not only accredited that the genetic program (the DNA) of a live system can be separated from the machine that reads it, as in a Turing Machine. Also, it showed that the program (i.e. the DNA sequence) remained the same throughout the transplantation process, while the biological machine at the end of the experiment (Mycoplasma capricolum) differed from the initial host (M. mycoides). The programme thus made an identical copy of itself (replication), but the machine made a similar, nonidentical duplicate or the original, that is, it reproduced. As a matter of fact, reproduction of metabolism must have predated the emergence of replication at the origin of life (Dyson 1985), when nucleic acids that had substituted for the mineral support of metabolism became templates for replication (Danchin 1989).

Symbionts and metabolic division of labour

One plausible scenario during this early time is that protocells kept fusing and splitting, mixing up metabolic pathways and creating novel ones by trials and errors. The ancestral population of what is contemporary life was likely made of phagocytes, cells that were prone to engulf other cells, either keeping them functional as a whole or in part, or digesting them (Kurland et al., 2007; Wang et al., 2007). The next step was either to escape being phagocyted (via the synthesis of a protecting envelope) or to colonise hostile environments where phagocytosis was difficult. This could have made the ancestral population to split in Eukarya (still amenable to phagocytal promiscuity), the phagocytosis-resistant Bacteria, and Archaea, thriving in hostile environments. This hypothetical road map emphasises the role of compartmentalisation in the origin of metabolism and vice versa.

A first consequence of phagocytosis is that cells may develop within other cells and end up as subsidiary organs. This type of symbiosis is ubiquitous in life (de Bary 1879; Portier 1918; Buchner 1953), for example in nitrogen fixation by plants and synthesis of other essential components (McBurney et al., 1935). Because natural selection tends to streamline costly constructs, it may be expected that symbiosis is metabolically advantageous. The metabolism of symbionts is thus informative to reveal hidden constraints. Perusal of insect symbionts show these organisms have a compact genome along with plasmids that retained specific biosynthetic pathways: sulphate assimilation and cysteine synthesis, branched-chain amino acids and tryptophan synthesis (van Ham et al., 1997; Van Ham et al., 1999; Gil et al., 2003; Gosalbes et al., 2008). Remarkably, these compact genomes generally code for RidA (YjgF, see above) showing that degradation of 2-aminoacrylate (which is a toxic side product of cysteine, tryptophan and serine synthesis) is an important requirement, possibly because it interferes with branched-chain amino acids synthesis (Schmitz and Downs 2004) in the same fashion as vinylglycine does (Gehring et al., 1977). This substantiates the chemical incompatibility between serine and branched-chain amino acids: the host makes serine while the symbiont makes isoleucine, leucine and valine. The symbiont–host interplay also brings about new metabolic reactions: enamines can be substrates of novel metabolic pathways, as demonstrated by the moonlighting role of anthranilate phosphoribosyl transferase (TrpD) in the synthesis of phosphoribosylamine. This compound can be used for a minor pathway of thiamine synthesis in the absence of the canonical enzyme RidA.

More generally, the chemical conflict between the different amino acid biosynthetic pathways explains why many animals require essential amino acids: histidine, isoleucine, leucine, lysine, methionine, phenylalanine, threonine, tryptophan and valine in humans. Remarkably, the genome of symbionts of aphids (which are animals quite distant from humans) like Buchnera sp. code for the complete histidine biosynthesis pathway, for pheA and aroH, for the complete threonine biosynthesis pathway, for the complete diaminopimelate-dependent lysine pathway, for the branched-chain amino acids and tryptophan pathway (including components determined by a plasmid). MetE is also present, together with sulphate reducing genes and CysE and CysK allowing the cell to make cysteine (Shigenobu et al., 2000). In short, symbiosis (i.e. compartmentalisation) provides the ultimate solution to the chemical conflicts involved in biosynthesis of amino acids. There are other benefits as well. Mitochondria are likely derived from primitive Alphaproteobacteria that progressively lost their genes to the nucleus of the host cell to retain only the ability to couple respiration to ATP synthesis (Andersson and Kurland 1999; Andersson et al., 2003; Tamames et al., 2007). In this case, compartmentailsation is required by the need to use vectorial (proton) transport to move the rotor of ATP synthase, together with the need for synthesis of highly hydrophobic proteins that would be impossible to transport efficiently from the cytoplasm to the inside of the other cell organelles (nucleus, Golgi).

In either case, symbiosis provides an important take-home lesson for metabolic engineering: it may be particularly difficult to reconcile chemical pathways within the same cell compartment. Combining microbial communities that would share complementary pathways should be considered as an efficient outcome (Brenner et al., 2008; de Lorenzo 2008). Plants have exploited this engineering solution with their many compartments (mitochondria and chloroplasts in particular; Agrawal et al., 2011), and this likely explains why plants are so efficient in developing a rich secondary metabolism as compared to the animal counterparts (Jenkins et al., 2011).

Monoderms and diderms

Bacteria are usually split between two types of cells, based on Gram staining. Gram-positive bacteria and Gram-negative bacteria differ by the nature of their envelope. As a rule of thumb, Gram-positive bacteria generally have one cytoplasmic membrane and a thick layer of murein, made of peptidoglycan, teichoic acids and many other protective polymers. Gram-negative bacteria have two membranes and murein in between. However, as the details of the structure of more and more Bacteria have been identified, this split is no longer relevant. A Gram-negative bacterium such as Thermoanaerobacter tengcongensis is in fact a Firmicute, with only one membrane (Bao et al., 2002), while a Gram-positive bacterium such as Mycobacterium tuberculosis has two membranes (Niederweis 2003). We thus advocate the nomenclature based on compartmentalisation proposed by R.S. Gupta: monoderms have one membrane and diderms two membranes (Gupta 1998). Intriguingly, antibiotic selection pressure (rather than endosymbiosis) has been proposed as the likely evolutionary driver of bacterial cells with two membranes (Gupta 2011). Diderms are particularly relevant for the sake of this review as they hold a distinct close space: the periplasm, where a specific part of metabolism can be developed separately from the cytoplasm (see above). But can bacteria have other authentic compartments? For some time, it was thought that Planctomycetes had a primitive nucleus. Recent work has in fact demonstrated that these bacteria are diderms with the cytoplasmic membrane deeply folded within the cell volume, so that it may look as if it were a nuclear membrane (Santarella-Mellwig et al., 2013). Cyanobacteria have a set of thylakoid membranes and phycobilisomes meant to harvest photons and fix carbon dioxide while producing dioxygen. Compartmentalisation in this case is meant both to concentrate carbon dioxide and to prevent the deleterious actions of dioxygen.

FILLING-IN AND USING COMPARTMENTS: MAXWELL'S DEMONS IN THE INTRACELLULAR MILIEU

The rise of different types of compartments at various stages of biological evolution created also the necessity to manage positional and compositional information connected mostly to the proteins involved. These have to move from their site of synthesis (the cytoplasmic ribosomes) to generate the various components that make the compartment. The mini-organelle itself may simply be self-assembled by spontaneous association of particular proteins (often building blocks that originate polyedral structures). In the cases where these structures are delimi-ted by membranes, compartments may just be built from some sites at the cell's lipid bilayers. But then, specific proteins have to be allocated to remain in the cytoplasm or to be part of the nondiffusible microscaffold. The process or earmarking proteins for a distinct position of the cellular body is illustrated by the way aged polypeptides are disposed of in the cell, either at its poles (typically in bacteria) or using the mother cell as a sort of garbage bin for aged or damaged proteins, for example in Saccharomyces cerevisiae (Erjavec and Nystrom 2007). Also, enzymes that form part of organelles must find their place in the architecture of the compartment. To this end, cells must have biomolecular versions of what in physics is known as Maxwell's demons, that is agents able to convert information into distribution of distinct molecules at the sides of a divided space (Maxwell 1871). The role of such demons is to operate as a sort of information ratchet mechanism, in which knowledge of a particle's initial position or other characteristics is used to guide its transport to another place (i.e. away from thermal equili-brium). Such biodevices have been engineered with the tools of Synthetic Biology (Serreli et al., 2007) and are likely to exist at the basis of different protein-targeting phenomena. Furthermore, information-based choices very frequently become manifest during the cell's lifetime. One revealing case is the fate of aged proteins. Apart from typical stress-related damage (e.g. carbonylation by ROS, sulphoxidation), proteins spontaneously undergo chemical modifications due to the reaction of the side chain of asparagine or aspartic acid residues with the adjacent peptide group, forming a symmetric succinimide intermediate (isoaspartate; Robinson and Robinson, 2004). Some of such modifications can be fixed. This is the role of the S-adenosyl-L-methionine (AdoMet)-dependent isoaspartate methylase by ring opening (Danchin 2012). But other damages cannot be reverted, and the aged proteins are thus signalled for energy-dependent degradation. How do cells distinguish between aged and fresh proteins and how are they directed to different cell locations? Some type of Maxwell's demon must decide the proteins to be degraded and those to be conserved and fixed (Danchin 2009a). The scenario creates connections between chemistry (reactive groups on the protein surface), information (ways to divide the new from the old) and thermodynamics (degrading proteins is more costly than fixing them) that have hardly been tackled thus far. In any case, note that information-based earmaking of protein fate and location can create a partition in metabolism that does not require a physical division between impermeable compartments.

CONCLUSION: CHEMISTRY SHAPES THE MULTISCALE ARCHITECTURE OF BIOLOGICAL SYSTEMS

As exposed by the limited number of cases (mostly from the bacterial world) that we have examined above, metabolism is not just the final manifestation of a chain of command that starts in selfish DNA molecules placed on top of the biomolecular hierarchy (de Lorenzo 2014). Quite on the contrary, metabolism is the ultimate raison d'être of biological systems and the main drive for the emergence of multiscale biological complexity. Chemical engineering provides a good frame to pose the central question in this respect: how continuous matter fluxes are dynami-cally maintained between the spatially and chemically segregated zones of the prokaryotic cell but in the absence of any membrane or predetermined material structure (Le Saux et al., 2014). Resolution of metabolic problems (what we have called metabolic frustration) seems to be at the basis of the evolution of regulatory networks (one-pot solutions) and compartmen-talisation, processes that require information management and memory for fixing outcomes compatible with thermodynamic constraints. We can still find many indications of these challenges when looking at cases (e.g., serine toxicity) where cells have not found a perfect solution to the problem and they fix one aspect of it at the cost of creating another (see above). It is possible that evolutionary emergence of multicelullarity and developmental programs was also triggered by the need to overcome chemical toxicity, as intracellular compartments and division metabolic labour in biofilms and structured syntrophic communities (e.g. microbial mats) could suggest. These are not just academic concerns, but very practical issues to take into account for modern metabolic engineering, in which the biochemical network of a platform microorganism (a chassis in the jargon of Synthetic Biology; Andrianantoandro et al., 2006) is genetically edited for execution of new chemical reactions. While the problem of toxicity of intermediates in engineered pathways has been frequently recognised, others such as disposal of reaction leftovers and aged proteins, osmotic and chemical safety valves, separation of anabolic and catabolic reactions and targeting of enzymatic complexes to distinct positions in the cell have hardly been addressed thus far. Some attempts, however, indicate steps in that direction, including intracellular scaffolding of new pathways (Dueber et al., 2009), capture of antitoxicity activities from metagenomes (Sommer et al., 2010) and even design of intracellular catalytic compartments based on viral capsids (Comellas-Aragones et al., 2007) and carboxysomes (Frank et al., 2013). It is also possible to entertain multistrain catalysts in which a given metabolic pathway is implemented through delivery of the different steps by different members of a microbial consortium (Kjaergaard et al., 2000; Brenner et al., 2008; de Lorenzo, 2008). This could avoid many of the metabolic conflicts of having the process engineered in a single-pot superbug. Still, metabolic engineering and Synthetic Biology could go even further beyond and provide durable solutions to metabolic puzzles of the type discussed above that nature has not been able to entirely disentangle. This will stay as one of the most exciting challenges in microbiological and biotechnological research for the next few years. In the meantime, it is increasingly clear that bioengineering of new metabolic properties is not only about stitching DNA pieces one after the other but largely about understanding the biochemical frame and the metabolic logic of the biological chassis (Porcar et al., 2014).

Authors are indebted to Claudia Rato-Silva, Conghui You and Pablo Nikel for inspiring discussions and to anonymous reviewers for their suggestions to improve the manuscript. The work in VdL's laboratory is supported by the BIO program of the Spanish Ministry of Economy and Competitiveness (MINECO), the ST-FLOW and ARISYS Contracts of the EU, the ERANET-IB Program and the PROMPT Project of the Autonomous Community of Madrid. The authors declare no conflict of interest.

Conflict of interest statement. None declared.

REFERENCES