Engineering of Metabolic Pathways Using Synthetic Enzyme Complexes^,

Abstract

Prospects are reviewed for the use of synthetic enzyme complexes as a metabolic engineering tool.

Plants provide a source of enzymes for metabolic engineering to produce valuable or useful products in micro-organisms; furthermore, plants can be engineered (Andre et al., 2016; Vickery et al., 2016; Moses et al., 2017). Production of high-value compounds (e.g. pharmaceuticals) and nutraceuticals (e.g. omega-3 fatty acids, carotenoids, tocochromanols, ascorbate, and anthocyanins) involves either the introduction of innovative pathways into a convenient host species or optimization of endogenous pathways. Other manipulations include engineering protective, secondary-compound production for pest and pathogen resistance and osmolytes for stress resistance. Manipulation of central metabolic pathways such as photosynthesis (e.g. Calvin-Benson cycle, alternative carbon sinks, introduction of CO₂ concentrating mechanisms, photorespiratory bypasses, and xanthophyll cycle) or starch and lipid synthesis has much potential to contribute to yield improvement. The use of plants as metabolic engineering vehicles to produce valuable compounds, as opposed to transferring plant pathways to microbes, will depend on feasibility and economic factors. Specialized cells and tissues (glandular trichomes, resin ducts and lactifers, or oilseeds) adapted to synthesize and store toxic and hydrophobic compounds involved in defense may make production of certain classes of compounds (e.g. isoprenoids and alkaloids) more advantageous in plants (Huchelmann et al., 2017). On the other hand, plants present bottlenecks in terms of the number of genes that can be conveniently manipulated and a long time frame for optimizing pathway engineering (Sweetlove et al., 2017). As an alternative to stable transformation, the method of transient expression (for example in Nicotiana) provides a rapid route to optimizing engineering and could act as a production platform (Reed and Osbourn, 2018). Also, it has become apparent that cambial (stem) cells are easily cultured and produce high yields of secondary compounds, such as taxol from yew (Taxus cuspidata; Lee et al., 2010). This finding could lead to a resurgence in the use of plant cell cultures. Recent developments in metabolic engineering and the application of a synthetic biology approach have been summarized (Stewart et al., 2018). Key tools and requirements for metabolic engineering in plants are a set of promoters for various functions. For example, they drive purposes such as expression in specific cell types; the ability to introduce multiple enzymes that are expressed at the appropriate level; and ensuring that the supply of reductant and cofactors is not limiting. Furthermore, they drive the targeting of the pathway to specific subcellular locations/organelles; an example of the importance of location is illustrated by the production of dhurrin in transgenic tobacco (Nicotiana benthamiana). Dhurrin is a cyanogenic glycoside produced by sorghum, and the enzymes are normally anchored to the endoplasmic reticulum (ER). Targeting the enzymes to the thylakoid membrane in a complex enables ferredoxin to be used as an alternative reductant and improves performance of the pathway (Gnanasekaran et al., 2016; Henriques de Jesus et al., 2017). This example also serves as an introduction to the potential of synthetic enzyme complexes to assist metabolic engineering.

ENZYME COMPLEXES: OCCURRENCE AND SIGNIFICANCE

Metabolons and Substrate Channeling

The possibility that enzymes are not randomly distributed but are associated into potentially dynamic complexes consisting of enzymes in a metabolic pathway (metabolons) has a long history. The term “metabolon” was introduced by Srere (1985) to denote a “supramolecular complex of sequential metabolic enzymes and cellular structural elements” (Srere, 1987). He proposed that metabolons would enable channeling of pathway intermediates between enzymes. (The next paragraph defines “channeling.”) The original definition included ribosomes and the DNA replication complex. However, more recent usage excludes these highly organized structures, and there is a tendency for enzyme complexes to be termed metabolons in the absence of evidence for channeling or other functional attributes. There are immense technical challenges in detecting potentially loose and dynamic enzyme interactions (for example, by pulldowns, yeast two-hybrid, and in vivo using fluorescent proteins) and assessing their in vivo functionality. In plants, there are examples of enzyme associations detected by these various methods, and these examples have been reviewed (Laursen et al., 2015; Sweetlove and Fernie, 2018). Examples include flavonols/isoflavonols (Achnine et al., 2004; Crosby et al., 2011; Lee et al., 2012b; Dastmalchi et al., 2016; Diharce et al., 2016), polyamines (Panicot et al., 2002), sporopollenin (Lallemand et al., 2013; Qin et al., 2016), alkanes (Bernard et al., 2012), indole acetic acid (Müller and Weiler, 2000; Kriechbaumer et al., 2016), carotenoids (Nisar et al., 2015), and dhurrin (Møller and Conn, 1980; Laursen et al., 2016). In central metabolism, the best studied examples are glycolysis and the tricarboxylic acid (TCA) cycle (Giegé et al., 2003; Graham et al., 2007; Zhang et al., 2017). The glycolytic enzymes are associated with the mitochondrial membrane and show dynamic behavior; complex formation increases with high respiratory demand (Graham et al., 2007). Similarly, in mammalian cells, the purinosome, an assembly of enzymes involved in purine biosynthesis, assembles when there is high demand for product (Pedley and Benkovic, 2017; Baresova et al., 2018). In only a few cases has the functional significance of these enzyme complexes been established. In this review, “metabolon” will be used in cases where channeling is demonstrated. The phrase “enzyme complex” will be used where two or more enzymes in a metabolic pathway are physically associated.

The functional significance of metabolons has been debated, but the principles are becoming clearer, in part because of additional insights derived from synthetic enzyme complexes. To be effective, an enzyme complex must enable channeling (Castellana et al., 2014; Sweetlove and Fernie, 2018). Channeling is the movement of an intermediate between active sites of successive enzymes with much decreased escape into the bulk cytoplasmic solution (Fig. 1A). Channeling could involve direct tunneling of intermediates between active sites and/or electrostatic guidance (Elcock et al., 1997). Channeling occurs in highly organized complexes such as Trp synthase (Dunn et al., 2008), malate dehydrogenase/citrate synthase (Bulutoglu et al., 2016), and bacterial Pro oxidation. In the latter example, Pro is converted to Glu through Pro dehydrogenase (PRODH), which produces 1-pyrroline-5-carboxylate (P5C). P5C spontaneously hydrates to form l-Glu-semialdehyde (GSA), which is then oxidized by P5C dehydrogenase (P5CDH) to form Glu. In many bacteria, PRODH and P5CDH comprise a bifunctional enzyme, and kinetic studies indicate direct channeling of P5C/GSA between the active sites. However, in other cases, such as Thermus thermophilus, the enzymes PRODH and P5CDH are on distinct proteins. Kinetic studies, substrate trapping, and surface plasmon resonance analysis of protein-protein interaction showed an orientation-dependent association between the enzymes and substrate channeling (Sanyal et al., 2015). Therefore, weak but specific interactions between these enzymes have evolved to enable channeling. This concept is an example of the Rosetta Stone hypothesis (Marcotte et al., 1999). The hypothesis suggests that if two separate proteins have homologs in another genome that are located on a single polypeptide, then the separate proteins are likely to interact with each other. A large proportion of the identified fusion proteins are enzymes (Enright et al., 1999; Marcotte et al., 1999). From an engineering point of view, synthetic fusion enzymes may or may not be effective. This is most likely because the enzymes have not coevolved complementary structures that enable effective channeling. Or, fusion could interfere with correct folding.

Figure 1.

Enzyme Assemblies and Their Influence on Substrate Channeling. A, Two closely associated (“co-evolved”) enzymes enabling direct channeling of the intermediate between active sites. The active sites could be located on separate proteins or on a single bifunctional protein. B, Tagged enzymes attached to a synthetic scaffold protein, nucleic acid scaffold, or lipid scaffold (Table 1). There is little channeling because the diffusion rate of the intermediate is much faster than enzyme activity. C, The same assembly as (B) but showing how multimeric scaffolded enzymes can form larger aggregates of high enzyme concentration. D, A large assembly of enzymes providing high local enzyme concentration enables probabilistic channeling. Here, the high enzyme concentration increases the chance that the intermediate binds to an enzyme active site before diffusing away. E, An encapsulated enzyme assembly is identical to (D), but a self-assembling protein coat provides an additional diffusion barrier with pores at the vertices to allow (selective) exchange of substrates and products. The enzymes could be tethered to the coat proteins. Examples are bacterial microcompartments (BMCs) specialized for use of carbon sources in pathways involving reactive intermediates (metabolosomes) and for CO₂ fixation with encapsulated carbonic anhydrase and Rubisco (carboxysomes). Eukaryotes lack BMCs but have pyrenoids (an aggregation of Rubisco surrounded by a loose starch sheath found in algae and hornworts) and peroxisomes. Peroxisomes house enzymes that produce toxic products and could be considered analogous to metabolosomes. They are bounded by a membrane that is relatively permeable to small molecules, and they exhibit channeling.

Considering less-organized metabolons, the essential and comprehensive analysis by Sweetlove and Fernie (2018) identifies the key point that the close association of sequential enzymes (in the absence of specific interactions) cannot be effective at channeling because substrate diffusion rate is much faster than enzyme catalysis, so the intermediate can escape (Fig. 1B). Only a few enzymes operate at diffusion limited rates (k_cat/K_M ∼ 10⁹ M⁻¹ s⁻¹), including triose phosphate isomerase, carbonic anhydrase, superoxide dismutase, catalase, and acetylcholine esterase. Therefore, simply pairing noncoevolved enzymes will not in itself be effective; and, even if it could be effective, it would increase initial rate but not steady-state rate (Sweetlove and Fernie, 2018). As noted previously, direct channeling requires coevolved enzymes. This is unlikely to be the case when heterologous enzymes are used for engineering. Channeling requires that intermediates are not in equilibrium with the bulk solvent, and this situation could be achieved by a large cluster of enzymes, not necessarily arranged in a specific manner, so that “probabilistic” channeling occurs (Castellana et al., 2014; Sweetlove and Fernie, 2018). Because of localized high enzyme concentration, the probability that a substrate binds to an active site before it leaves the cluster is increased, and an increase in flux is also predicted (Fig. 1D). It is suggested that high enzyme concentration can influence the thermodynamic feasibility of a pathway and its direction (Angeles-Martinez and Theodoropoulos, 2015). Evidence for effective channeling in enzyme complexes in vivo is scarce, although the wide range of central and secondary metabolism pathways with interacting enzymes suggests that it is likely. Demonstration of channeling is challenging, and the various approaches have been reviewed (Zhang et al., 2017; Sweetlove and Fernie, 2018). Isotopic dilution is a useful technique: If channeling is occurring, an added unlabeled pathway intermediate will not equilibrate with the labeled intermediate derived from a labeled precursor. Channeling has been demonstrated in vitro for the isolated ER-bound dhurrin biosynthesis metabolon (Møller and Conn, 1980) and validated in vivo when a biosynthetic complex is introduced into chloroplasts (Henriques de Jesus et al., 2017). In plants, isotope dilution experiments have shown channeling in the glycolytic pathway bound to the surface of mitochondria (Giegé et al., 2003; Graham et al., 2007). A comprehensive study of the plant TCA cycle showed 158 binary protein-protein interactions that were confirmed by the channeling of citrate and pyruvate using isotope dilution experiments (Zhang et al., 2017). This key study provides strong evidence for physical association between enzymes and the occurrence of channeling. It is also of significance because TCA cycle enzymes were the first enzymes involved in the initial characterization of metabolons (Srere, 1987; Vélot et al., 1997; Bulutoglu et al., 2016). The other two consequences of channeling include decreasing the loss of potentially reactive and toxic intermediates into the bulk solution and influencing flux at branchpoints (Zhang et al., 2017; Sweetlove and Fernie, 2018). It is notable that a large proportion of metabolons are membrane-associated. As well as the prior examples, glycolytic enzymes associate with the cytoskeleton in yeast and Arabidopsis (Araiza-Olivera et al., 2013; Garagounis et al., 2017). It is possible that channeling is aided by the physical and chemical properties in the cytoplasm in proximity to surfaces such as membranes or cytoskeletal elements (Theillet et al., 2014). It is proposed that bacterial cytoplasm is divided into a supercrowded “cytogel” extending 20 to 70 nm from the plasma membrane and more dilute cytosol (Spitzer and Poolman, 2013). While not likely to influence the diffusion rate of small molecules significantly, the formation of protein complexes may be favored near surfaces, suggesting that anchoring synthetic complexes to a membrane could be an advantageous strategy. Protein-protein interactions are driven by several mechanisms not covered here (Williamson, 2012). A fresh suggestion is that enzymes show chemotactic movement along their substrate gradient, an effort that could drive their colocalization (Wu et al., 2015; Illien et al., 2017; Agudo-Canalejo et al., 2018; Zhao et al., 2018). These experiments use fluorophore-tagged enzymes to follow movement in microfluidic devices. However, the interpretation of the fluorescence correlation spectroscopy, on which the conclusions are based, has been criticized (Günther et al., 2018).

SYNTHETIC ENZYME COMPLEXES

Construction and Functioning

The existence of enzyme complexes and the possibility that they are important in influencing metabolic pathways have provided the drive to explore the use of synthetic enzyme complexes in metabolic engineering. There are essentially two approaches: anchoring enzymes on scaffold molecules of various kinds or encapsulating enzymes in protein-coated microcompartments based on bacterial microcompartments and viral capsids. Many reviews have discussed and advocated synthetic enzyme complexes and possibly outnumber actual examples of its application. The reader is referred to these reviews for more information: (Conrado et al., 2008; Boyle and Silver, 2012; Lee et al., 2012a; Singleton et al., 2014; Chessher et al., 2015; Pröschel et al., 2015; Siu et al., 2015; Polka et al., 2016; Plegaria and Kerfeld, 2018; Qiu et al., 2018). A selection of examples of synthetic enzyme complexes is reviewed here in relation to the methods used and outcome.

Protein and Protein-Lipid Scaffolds

The initial report of the construction of a synthetic enzyme complex in metabolic engineering was the assembly of three enzymes required for synthesizing mevalonic acid (acetoacetyl-CoA thiolase, hydroxy-methylglutaryl-CoA synthase, and hydroxymethylglutaryl-CoA reductase) on a synthetic protein scaffold (Dueber et al., 2009; Table 1). The enzymes were linked to scaffolds using high-affinity mammalian protein-protein interaction domains (SH3, GBD, and PDZ) assembled in various combinations in a synthetic scaffold protein with cognate binding domains. Each enzyme was fused to SH3, GBD, and PDZ ligands. Expression in E. coli resulted in assembly of the scaffolded proteins and an increase in mevalonate accumulation when scaffolded. Following this success, further scaffolding experiments have been reported (Table 1). These display an increasing diversity and ingenuity of methods used to scaffold enzymes. Other high-affinity protein-protein interaction domains have been harnessed (e.g. dockerin-cohesin, Leu zippers, synthetic coiled-coil proteins). Generally, two to three enzymes have been assembled on the scaffolds. In cases where enzymes are multimeric, attachment of one enzyme to several scaffolds could allow cross-linking to form larger structures (Fig. 1C). Larger conglomerations have been achieved by scaffolding to very large proteins (Price et al., 2016) or to proteins liable to form inclusion bodies (Han et al., 2017). A very promising approach is the production of a network of cytoskeleton-like synthetic protein filaments to which enzymes are scaffolded (Lee et al., 2018a, 2018b). Some naturally occurring metabolons are membrane bound as noted previously. In this context, enzyme complexes anchored in lipid droplets have been produced by using scaffold proteins that associate with lipid-binding proteins, such as oleosin (which is the coat protein for lipid droplets in oilseeds) and certain virus coat proteins (Myhrvold et al., 2016; Lin et al., 2017).

Table 1. Examples of Metabolic Pathway Engineering Using Scaffolded Enzyme Complexes.

Product	Host Organism	Approach	Outcome	Assay Conditions	Reference
Ethyl acetate	S. cerevisiae	Dockerin tags + cohesion/oleosin scaffold.	Enzymes colocalize to lipid droplet membranes (FRET). 1.8-fold increase in product in cell lysate assay. Channeling not tested.	Initial activity in vitro?	Lin et al., 2017
Ethanol (pyruvate decarboxylase and alcohol dehydrogenase)	E. coli	Filamentous scaffold proteins formed from bacterial microcompartment coat protein (PduA) fused to synthetic self-assembling coiled-coil proteins. Enzymes tagged with coiled-coil proteins. Also attached to inner membrane.	Network of cytoplasmic filaments visualized by TEM. Protein colocalization confirmed by tagged fluorescent proteins and microscopy.	Ethanol yield increased 2-fold by 20 h, but initial rate of increase same with or without scaffold.	Lee et al., 2018b
Dhurrin	N. benthamiana	Fusion of three enzymes to TatB and TatC (thylakoid membrane proteins). Transient expression, chloroplast targeted.	Dhurrin increases 5-fold. Channeling suggested by decreased side products. Confirmed thylakoid location, but not enzyme proximity.	Products measured 5 d post-Agrobacterium infiltration	Henriques de Jesus et al., 2017
Butan-1-ol	E. coli	Enzymes attached to Clostridium exoglucanase cellulose-binding domain-induced inclusion bodies via Leu zipper tags.	2-fold increase in butanol formation. Enzymes shown to be present in inclusion bodies.	Stable transgenic lines.	Han et al., 2017
Indole-3-acetic acid	E. coli	Enzymes (or split GFP) fused to DNA-binding TALE proteins assembled on a plasmid with various distances between DNA-binding sites.	GFP fluorescence indicates assembly on DNA scaffold. IAA production increased up to 8-fold in scaffold and binding site spacing-dependent manner.	Overnight IPTG induction.	Zhu et al., 2016
Methanol to Fru-6-phosphate via formaldehyde	E. coli	A multisubunit malate dehydrogenase fused to SH3 plus a two-enzyme fusion protein with a SH3 ligand.	Assembly into a complex confirmed by TEM and dynamic light scattering. A 97-fold increase in F6P in vitro and a 2.4-fold increase in methanol consumption in vivo.	Faster initial MeOH consumption rate in vivo up to 5 h post addition.	Price et al., 2016
Indigo	E. coli	Bacteriophage Ø P9 and P12 proteins assembled into protein-lipid vesicles. Enzymes or fluorescent proteins fused to N terminus of P9.	Colocalization of fluorescent proteins and fractionation of cell extracts show assembly of lipid-protein droplets (∼ 20 nm diameter). Indigo production increased 2.5-fold in the complex (P12-dependent).	Enzyme expression induced “overnight.”	Myhrvold et al., 2016
2,3-butanediol from phosphoenolpyruvate (PK) and α-acetolactate synthase	S. cerevisiae	Enzymes tagged with cohesin and dockerin to assemble via cohesin-dockerin interaction.	Complex formation confirmed by immunoprecipitation. A 1.3-fold increase in butanediol. Evidence for channeling: pyruvate produced by PK is less available for ethanol formation.	Faster product yield (g/L culture) up to 24 h after initiating a culture by dilution.	Kim et al., 2016
Branchpoint between carbamoyl phosphate (carbamoyl phosphate synthetase) use for Arg (Orn carbamoyltransferase) and carbamoyl-Asp synthesis (Asp carbamoyltransferase)	E. coli	Carbamoyl phosphate synthetase and Asp carbamoyltransferase fusion protein expressed at high level.	Increase in phase-bright cytoplasmic structures typical of protein-dense clusters. Evidence for diversion of carbamoyl phosphate away from the competing Arg synthesis pathway dependent on clustering.		Castellana et al., 2014
Alkanes (acyl-ACP-reductase and fatty aldehyde decarbonylase)	E. coli	Fusion protein or enzymes tagged with zinc-finger DNA binding proteins assembled on a plasmid DNA scaffold.	Enzyme scaffold assembly not assessed. Fusion protein increases alkanes 4.8-fold and DNA scaffold up to 8.8-fold (dependent on enzyme stoichiometry).	24 h post-IPTG induction.	Rahmana et al., 2014
Resveratrol	S. cerevisiae	GBD, SH3, and PDZ combined in protein scaffolds and enzymes tagged with their ligands.	Resveratrol increased by up to 5-fold.	Measured 36 h (5-fold) and 96 h (2-fold) after induction.	Wang and Yu, 2012
Resveratrol, 1, 2-propanediol, and mevalonate	E. coli	Enzymes tagged with zinc-finger DNA binding proteins assembled on a plasmid DNA scaffold. Random scaffold control.	Assembly shown in vitro (split YFP) and in vivo. Up to 5-fold increase in product depending on enzyme proximity/pathway.	24 h post induction (resveratrol and propane diol). 50 h post induction (mevalonate).	Conrado et al., 2012
Hydrogen production (ferredoxin and hydrogenase)	E. coli	RNA scaffolds with aptamers plus proteins tagged with aptamer-binding proteins.	Assembly indicated by split GFP. Mutant aptamer site controls. Up to 48-fold increase in product (dependent on scaffold geometry).	16 h after induction.	Delebecque et al., 2011
Glucaric acid	E. coli	GBD, SH3, and PDZ combined in protein scaffolds and enzymes tagged with their ligands.	Up to 5-fold increase (g/L). Enzyme stoichiometry effects observed.	48 h post induction.	Moon et al., 2010
Mevalonate	E. coli	GBD, SH3, and PDZ combined in protein scaffolds and enzymes tagged with their ligands.	Up to 77-fold increase (g/L). Enzyme stoichiometry effects observed..	Up to 3 d postinduction.	Dueber et al., 2009

Nucleic Acid Scaffolds

DNA and RNA have been explored as enzyme scaffolds (Delebecque et al., 2011; Conrado et al., 2012; Rahmana et al., 2014; Zhu et al., 2016). Enzymes are tagged with DNA-binding proteins (e.g. zinc finger proteins and transcription activator-like effectors [TALEs]) and DNA scaffolds are synthesized with specific binding site arrangements and spacing. These have been expressed in E. coli with the scaffolds on plasmids. The plasmid copy number determines the amount of scaffold. RNA scaffolds have also been tested on the basis that RNA can fold into potentially useful geometries to provide binding sites (aptamers) for tagged proteins (Delebecque et al., 2011). These approaches are appropriate for bacteria but would be more problematic in plants due to the need for accessible DNA and potential RNA instability.

Why Is Scaffolding Successful?

The examples of scaffolding in Table 1 show an increase in product because of enzyme scaffolding, although the benefit is sometimes modest. E. coli and Saccharomyces cerevisiae are the predominant hosts for testing scaffolds, with the only plant example being the targeting of dhurrin biosynthesis enzymes to the thylakoid membrane by transient expression in N. benthamiana (Henriques de Jesus et al., 2017). Why does scaffolding work? As discussed earlier, dispersed scaffolded enzyme units would be unlikely to exhibit channeling and, if they did, faster initial reaction rates—but not increased steady-state rates—would be expected (Sweetlove and Fernie, 2018). Considering the examples shown in Table 1, it is generally not possible to determine if the system is at steady state because many of the measurements are made hours or days after inducing scaffolding. Therefore, it is tempting to propose that most of the manipulations inadvertently induce the formation of sufficiently large complexes that increase local enzyme concentration, enabling probabilistic channeling and increased rate at steady state (Fig. 1C). This proposal is essentially how pyrenoids work (as will be shown later in this study). The example of Castellana et al. (2014); Table 1) is also important because it shows that high expression of a bifunctional enzyme to form a complex big enough to visualize enables channeling and diverts intermediates at a branchpoint. Controlling flux at a branchpoint is also seen in the example of butanediol formation (Kim et al., 2016). Channeling was demonstrated when dhurrin biosynthesis enzymes were anchored to the thylakoid membrane. The pathway intermediates are reactive; and when the enzymes are not anchored, LC-MS analysis detects many compounds derived from them, and anchoring greatly reduces their accumulation (Henriques de Jesus et al., 2017). This point is important for engineering pathways that involve reactive intermediates, where the benefit could be protection against toxicity, a value could be equal to the benefit of greater yield. It is evident that normal metabolism causes “metabolite damage” (the production of unintended compounds), and it is suggested that metabolite repair enzymes could be part of the metabolic engineering tool kit (Sun et al., 2017). Channeling between critical enzymes would also contribute to damage-limitation. In the scaffold examples, it is likely that channeling is enabled by aggregation of the individual scaffolds into larger clusters. In some of the cases, this prospect has been demonstrated (Table 1). IAA and alkane biosynthesis enzymes have been detected in complexes in plants (Müller and Weiler, 2000; Bernard et al., 2012; Kriechbaumer et al., 2016) but without specific evidence for channeling, so it is noteworthy that scaffolding increases production of these compounds in micro-organisms (Table 1).

Most of the examples shown in Table 1 demonstrate that assembly has occurred by using techniques such as coimmunoprecipitation (co-IP), fluorescent proteins (bimolecular fluorescence complementation [BiFC] and Förster resonance energy transfer [FRET]) and transmission electron microscopy (for larger assemblies and encapsulated enzymes). Possibly, super-resolution microscopy and transmission electron cryomicroscopy (cryoEM) will be useful in providing more detailed information on the size and structure of complexes. Another factor, not explicitly tackled in any of the studies, is the possibility that the scaffolding has a favorable influence on the total amount of enzyme (perhaps by decreased rate of proteolysis) or influences specific activity and kinetic properties. Again, the characterization is rarely sufficiently detailed to assess these possibilities. Finally, it is reasonable to suppose that unsuccessful attempts at scaffolding have not been published, making it impossible to assess the probability of success.

SYNTHETIC MICROCOMPARTMENTS

Nanoreactors

Many bacteria produce microcompartments (BMCs), protein-coated nanostructures that encapsulate enzymes (Fig. 1E). Their structure and functions have been well-reviewed (Kerfeld et al., 2018). They can be divided into two categories by function. Metabolosomes contain catabolic enzymes for the use of carbon sources through pathways that produce reactive intermediates. Carboxysomes function as part of the CO₂-concentrating mechanism (CCM) of photosynthetic bacteria, and they contain ribulose bisphosphate carboxylase-oxygenase (Rubisco) and carbonic anhydrase. They occur in two distinct forms: α-carboxysomes in proteobacteria, and some cyanobacteria and β-carboxysomes in cyanobacteria. BMCs function by concentrating enzymes (Rubisco and carbonic anhydrase) in a restricted space that enables channeling (as described previously for other enzyme complexes). The shell presumably evolved because there is an additional benefit to a diffusional barrier. Our understanding of how BMCs assemble, encapsulate the correct enzymes, and allow substrate and product exchange via pores has advanced to the point where synthetic BMCs that self-assemble have been expressed in bacterial cells. Assembly of enzymes for encapsulation is assisted by incorporation of encapsulation peptides (EPs; Gonzalez-Esquer et al., 2016; Plegaria and Kerfeld, 2018). In plant research, there has been a focus on the possibility of introducing carboxysomes into chloroplasts to mimic the cyanobacterial CCM. As in cyanobacteria, this process would also need transporters to concentrate bicarbonate into the chloroplast stroma. Bicarbonate would enter the carboxysomes, where CO₂ production is catalyzed by encapsulated carbonic anhydrase. The high local concentration of Rubisco drives rapid CO₂ fixation and outcompetes the oxygenase reaction (Rae et al., 2013). The beginnings of this goal have been achieved by successful assembly of β-carboxysome shells in chloroplasts by transient expression of five shell proteins in Nicotiana (Lin et al., 2014). YFP, tagged with a small targeting peptide from the carboxysome-organizing protein, CcmN, was incorporated into the shells. In a recent breakthrough, a minimal functional carboxysome was expressed in tobacco chloroplasts (Long et al., 2018). This result was achieved by introducing two α-carboxysome coat proteins and the large and small subunits from the cyanobacterium Cyanobium. Chloroplasts were transformed to enable knockout of the endogenous Rubisco large subunit. The resulting plants were able to grow, carrying out CO₂ assimilation with the encapsulated Cyanobium Rubisco. The results show that this minimal carboxysome allows encapsulated Rubisco to function; therefore, the pores in the protein coat enable exchange of substrates and products.

BMCs are related to viral capsid proteins. Capsid proteins can self-assemble in heterologous hosts and have the potential to be used to encapsulate enzymes. An interesting recent example is encapsulation of an indigo biosynthesis pathway from Trp in a virus capsid protein. The enzymes were anchored to the capsid proteins using SpyTag/SpyCatcher protein fusions (Giessen and Silver, 2016). This system is based on the CnaB2 domain from the fibronectin-binding protein FbaB from Streptococcus pyogenes, and the process works by a spontaneous reaction between a Lys residue on SpyCatcher and an Asp on SpyTag to form an isopeptide bond (Reddington and Howarth, 2015). Tagging the enzymes with SpyCatcher and bacteriophage MS2 capsid protein with SpyTag resulted in assembly of particles in E. coli that increased indigo production by 60% compared to controls (Giessen and Silver, 2016). Pores of BMCs and capsids can be engineered to control substrate uptake specificity (Glasgow et al., 2015). Isolated capsids showed that the enzymes were markedly more stable in vitro because of the covalent linkages.

HARNESSING AND MODIFYING OTHER NATURALLY OCCURRING STRUCTURES

Synthetic Organelles

Pyrenoids and peroxisomes could be considered large enzyme complexes that enable probabilistic channeling. Peroxisomes contain oxidases that produce hydrogen peroxide along with catalase, which decomposes the peroxide to water. In leaves, photorespiration generates a large flux of glycolate, which is oxidized in peroxisomes to produce glyoxylate (a reactive aldehyde) and hydrogen peroxide. By cooperation between peroxisomes and mitochondria, photorespiration produces glycerate for recycling into the Calvin-Benson cycle (Hagemann and Bauwe, 2016). Isolated spinach leaf peroxisomes produce glycerate at the same rate with and without an intact membrane; and, in both cases, the intermediates glyoxylate and hydroxypyruvate are not detected in the suspension medium (Heupel and Heldt, 1994). The results indicate that the leaf peroxisome is a protein complex that maintains its integrity without the membrane boundary and that exhibits channeling. More recently, interaction between glycolate oxidase and catalase was shown by BiFC and co-IP (Zhang et al., 2016). The relative simplicity of peroxisomes (lack of a genome and a single membrane permeable to small molecules) makes them a tempting basis for production of a synthetic organelle housing engineered metabolic pathways. As noted previously, the evidence for channeling in leaf peroxisomes (even in the absence of a membrane) provides a useful starting point. Various pathways have been engineered into peroxisomes, for example, to produce polyhydroxyalkanoates in Arabidopsis (Mittendorf et al., 1999; Kessel-Vigelius et al., 2013). Yeast peroxisomes have been engineered to efficiently produce alkanes and fatty alcohols from acyl-CoAs, with evidence that the high enzyme concentrations enabled channeling (Zhou et al., 2016). These pathways use the acyl-CoA metabolizing capacity of peroxisomes. Modification of the existing peroxisomal protein import system can increase its efficiency for importing enzymes (DeLoache et al., 2016). However, the recent creation of a different protein import system that runs in parallel with the endogenous system provides a step toward synthetic peroxisomes (Cross et al., 2017). Deeper understanding of proliferation mechanisms and the protein-protein interactions that hold the peroxisomal matrix together will also assist in reaching this goal.

The pyrenoid could provide another starting point for producing a protein aggregate with high enzyme concentration that enables channeling. This structure consists of an aggregate of Rubisco in the chloroplasts of algae and some liverworts, and it is required for their CO₂ concentrating mechanism (CCM). It traps CO₂ produced by carbonic anhydrase, allowing improved Rubisco activity (Meyer et al., 2017; Küken et al., 2018). The protein components of this structure have been identified. The protein EPYC1, present in high concentration, interacts with Rubisco to form a scaffold (Mackinder et al., 2016, 2017). Additionally, the resulting structure is liquid, rather than crystalline, and it undergoes a phase transition and fission during cell division (Freeman Rosenzweig et al., 2017). These structures suggest the possibility of making synthetic organelle-like structures without walls or membranes for metabolic engineering. The introduction of pyrenoids into plant chloroplasts is a potential route for improving photosynthesis (Mackinder, 2018).

CONCLUSION

It is evident that the amount of final product in engineered metabolic pathways can be increased by various ingenious scaffolding approaches. The most likely explanation is that the resulting enzyme aggregates are (often inadvertently) large enough to enable probabilistic channeling due to increased local enzyme concentration. Pyrenoids work in the same manner, while encapsulation (in BMC and capsid coat proteins) increases enzyme concentration and provides an additional (potentially selective) diffusion barrier. Leaf peroxisomes are robust protein complexes that can hold together without their membrane and exhibit channeling. Peroxisomes and pyrenoids could form the basis for engineering multienzyme metabolic pathways that benefit from channeling. Ultimately, the widespread use of channeling in plant metabolic engineering will be determined by a balance between the extra time required to introduce and optimize enzyme assemblies compared to the potentially modest benefit in product yield. Channeling could provide a critical advantage if it enables production of compounds with highly reactive and toxic pathway intermediates or improves diversion of central metabolism intermediates into the engineered pathway.