Terpenoids constitute a large and diverse class of natural products that serve many functions in nature. Most of the tens of thousands of the discovered terpenoids are synthesized by plants, where they function as primary metabolites involved in growth and development, or as secondary metabolites that optimize the interaction between the plant and its environment. Several plant terpenoids are economically important molecules that serve many applications as pharmaceuticals, pesticides, etc. Major challenges for the commercialization of plant-derived terpenoids include their low production levels in planta and the continuous demand of industry for novel molecules with new or superior biological activities. Here, we highlight several synthetic biology methods to enhance and diversify the production of plant terpenoids, with a foresight towards triterpenoid engineering, the least engineered class of bioactive terpenoids. Increased or cheaper production of valuable triterpenoids may be obtained by ‘classic’ metabolic engineering of plants or by heterologous production of the compounds in other plants or microbes. Novel triterpenoid structures can be generated through combinatorial biosynthesis or directed enzyme evolution approaches. In its ultimate form, synthetic biology may lead to the production of large amounts of plant triterpenoids in in vitro systems or custom-designed artificial biological systems.
Plants synthesize and accumulate a wide range of small molecules or natural products that are involved in fundamental physiological and ecological processes. Some of these natural products have therapeutic potential which has been exploited by humans for thousands of years in the form of traditional herbal medicine. In recent years, with our growing understanding of their biosynthesis, regulation and functioning, plant-derived natural products have emerged as high-value therapeutics, flavors and fragrances, colorants and health-promoting agents. Based on their structure and biosynthetic origin, plant natural products can be classified into different groups, such as the terpenoids, alkaloids and phenolic compounds (Croteau et al., 2000). This review focuses on the terpenoids, of which tens of thousands of compounds have been characterized from plants. The terpenoids or isoprenoids comprise structurally diverse compounds that are associated with primary as well as secondary metabolism. Gibberellin, abscisic acid and brassinosteroid phytohormones, phytosterols and carotenoid pigments are primary metabolic terpenoids involved in basic functions, such as the regulation of plant growth and development, photosynthesis, membrane permeability and fluidity (Bohlmann & Keeling, 2008; Vranová et al., 2012). However, the majority of the plant terpenoids are secondary metabolites that play a crucial role in the interaction of the plant with its environment, for instance by serving as pollinator attractants, herbivore repellents, anti-feedants, toxins or antibiotics (Gershenzon & Dudareva, 2007).
The structural variety and inherent biological activities of many plant terpenoids have rendered them widely applicable. With an annual production of 107 tons, natural rubber is the most abundant terpenoid produced. Because of its unique properties, it serves as a biological material in the non-food industry for the production of heavy-duty tires, vibration dampers or latex products, such as surgical gloves (van Beilen & Poirier, 2007). Other examples of plant terpenoids with significant economic value include: menthol, a monoterpenoid extracted from peppermint and used in the flavor and fragrance industry; abietic acid, a diterpenoid isolated from conifer rosin that is used in lacquers, varnishes and soap; and the anti-malarial and anti-cancer drugs artemisinin and taxol, respectively (Bohlmann & Keeling, 2008).
A major hurdle in the commercialization of plant terpenoids is that they often accumulate in very low concentrations in planta, thereby hindering their purification in large amounts from the natural source. When the extraction of a natural product from its natural source is not sufficient, several alternative approaches can be explored, including: (1) plant breeding and genetic engineering to generate cultivars or transgenics accumulating higher levels of the desired compounds; (2) the development of scalable plant cell or root cultures; and (3) the engineering of microbial hosts to produce the compound. Commercially viable alternative production systems have already been established for some terpenoids, which is reflected in the emergence of companies, such as Phyton Biotech (http://www.phytonbiotech.com/), a global provider of chemotherapeutics, including paclitaxel extracted from Taxus cell cultures, and Amyris (http://www.amyris.com/), which uses a synthetic biology platform for the production of artemisinin in yeast.
Furthermore, the (pharmaceutical) industry is in constant search for novel molecules, primarily as a result of the discovery of new drug targets, the emergence of new diseases and, in the case of infectious diseases, the growing resistance of microbes to the currently marketed antibiotics (Pollier et al., 2011). In addition, the business model of pharmaceutical companies is under threat, as leading blockbuster drugs will soon lose patent protection and become available for market competition, which often leads to lower market prices, thereby rendering the production of the drug non-profitable to the original developer. As traditional pharmacological screening of medicinal plants is time consuming and expensive, and the output of combinatorial chemistry libraries is low in terms of new drugs, alternative approaches to generate new molecules or scaffolds are required (Koehn & Carter, 2005; Welsch et al., 2010). Combinatorial biosynthesis accelerates the process of natural evolution and multiplies the natural diversity by generating novel enzyme–substrate combinations. Thereby, it can be rationally applied to custom design new compounds (Kirschning et al., 2007; Pollier et al., 2011).
In this review, we provide a futuristic view into the engineering of triterpenoids, the least engineered class of terpenoids with pharmaceutical potential, by drawing inspiration from the current status of engineering of other terpenoid classes in plants and microbial hosts. We highlight the latest approaches for enhancing the production and increasing the structural diversity of natural compounds, and frame the potential of the booming trends in synthetic biology in the perspective of triterpenoid production.
II. ‘Natural’ terpenoid biology
A basic understanding of the biosynthesis and regulation of a compound is strategic to any bioengineering initiative. Therefore, we set the base for triterpenoid biology by providing an insight into their synthesis and regulation in plants. A correct perception of their native production habitat and machinery permits the translation of this knowledge to the bioengineering of native plants or heterologous hosts.
1. Classification and biosynthesis of plant terpenoids
Despite their enormous structural diversity, terpenoids share a common biosynthetic origin and follow similar synthesis routes. All terpenoids are derived from the repetitive fusion of isoprene (C5H8) units, and the number of isoprene units determines their classification. In higher plants, the biosynthesis begins with the generation of isopentenyl pyrophosphate (IPP), the principal precursor, through the mevalonate (MVA)/3-hydroxy-3-methylglutaryl-CoA reductase (HMGR) pathway or the 2-C-methyl-d-erythritol 4-phosphate (MEP)/1-deoxy-d-xylulose 5-phosphate (DOXP)/non-MVA pathway. The IPP is isomerized to its allylic isomer dimethylallyl pyrophosphate (DMAPP). The consecutive condensation of IPP and DMAPP units leads to the formation of prenylated pyrophosphates, the immediate precursors of the different terpenoid classes (Fig. 1). These condensation reactions are catalyzed by specific prenyltransferases which are named according to the product they generate. Specific terpenoid synthases then modify these precursors to terpenoid skeletons (Chen et al., 2011), which are subsequently decorated by various enzymatic modifications to generate the structural and functional diversity of terpenoids. Plants also exhibit a clear compartmentalization for the generation of IPP and the synthesis of terpenoids (Croteau et al., 2000; Vranová et al., 2012; Fig. 1).
Here, we focus on plant triterpenoids, comprising primary metabolites, such as the phytosterols and the brassinosteroid hormones, and secondary metabolites, such as the saponins. The IPP for triterpenoid biosynthesis is generated through the cytosol, peroxisome and endoplasmic reticulum-localized MVA pathway. The ‘head-to-tail’ condensation of two IPP units with a DMAPP unit yields the C15 farnesyl pyrophosphate (FPP), two of which subsequently fuse ‘head-to-head’ to generate the linear C30 triterpenoid precursor, squalene. This compound is further epoxidized to 2,3-oxidosqualene (Augustin et al., 2011), which, in turn, is typically cyclized by specific oxidosqualene cyclases (OSCs) to tetra- or pentacyclic structures to form the dammarenes, tirucallanes and phytosterols, or the oleananes, ursanes, lupanes and taraxasteranes, respectively (Phillips et al., 2006). In some plant species, 2,3-oxidosqualene can also be cyclized to mono- and tricyclic triterpenoid backbones (Xue et al., 2012; Fig. 2).
The cyclization of 2,3-oxidosqualene forms the branch point between primary and secondary triterpenoid metabolism. Cycloartenol, formed by the cycloartenol synthase (CAS)-mediated cyclization of 2,3-oxidosqualene, is the committed precursor for phytosterol biosynthesis. Higher plants synthesize a mixture of various sterols from cycloartenol, which can accumulate in a free form or as esters or glycosides (Nes, 2011). In turn, the phytosterols cholesterol, campesterol and sitosterol are the precursors of the C27, C28 and C29 brassinosteroid hormones, respectively (Fujioka & Yokota, 2003). In addition, cholesterol can also undergo a series of oxygenations and glycosylations to form C27 secondary metabolites, the steroidal saponins (Dewick, 2001). The other cyclization products of 2,3-oxidosqualene form committed precursors for secondary metabolite biosynthesis (Fig. 2). These cyclized precursors are further oxidized by one or many cytochrome P450s (CytP450s) to form sapogenins. In some plants, such as the birch (Betula pubescens) and olive tree (Olea europaea), the sapogenins form the final accumulating secondary metabolite, whereas, in others, the sapogenins are glycosylated by UDP-dependent glycosyltransferases (UGTs) to generate amphipathic glycosides, the saponins (Augustin et al., 2011).
2. Regulation of terpenoid biosynthesis in plants
The biosynthesis of terpenoids is tightly controlled in plants, as they serve many functions in plant growth, development and response to biotic and abiotic environmental factors (Tholl, 2006; Nagegowda, 2010; Vranová et al., 2012). Terpenoid synthesis occurs within specific tissues or at specific plant developmental stages (Nagegowda, 2010). For instance, many plant species have glandular trichomes, specialized structures for the synthesis of secreted terpenoid natural products (Lange & Turner, 2013). The triterpenoid saponin glycyrrhizin accumulates only in the underground organs, stolons and roots of licorice (Glycyrrhiza) plants (Seki et al., 2008). Avenacins, the bioactive saponins in oat (Avena sativa), accumulate only in the root epidermis, where they provide resistance to phytopathogenic fungi (Haralampidis et al., 2001). Such specific terpenoid synthesis is mainly regulated at the transcriptional level. The avenacin biosynthesis genes are tightly co-regulated and expressed exclusively in the root epidermis in which the avenacins accumulate (Haralampidis et al., 2001; Qi et al., 2006; Field & Osbourn, 2008).
In addition to this spatiotemporal regulation, induced terpenoid biosynthesis is often observed in response to herbivore feeding, pathogen attack or various abiotic stresses (Nagegowda, 2010; Vranová et al., 2012). For instance, 7 d after Spodoptera littoralis larvae fed on Medicago sativa leaves, the total saponin content of the damaged foliage increased by 84%, causing a deterrent effect on the larvae. Accordingly, larval performance was reduced when forced to feed on the damaged leaves (Agrell et al., 2003, 2004). The increased accumulation or release of terpenoids in response to various (a)biotic stresses is often mediated by an increased transcriptional activity of the specific terpenoid biosynthetic genes (Tholl, 2006; Nagegowda, 2010). This transcriptional response is controlled by a complex signaling cascade in which jasmonate hormones (JAs) play a crucial role. Hence, the treatment of plants or plant cell cultures with JAs often causes transcriptional and metabolic changes comparable with pathogen or herbivore attack. The exposure of Medicago truncatula cell suspension cultures to methyl jasmonate (MeJA) leads to increased saponin accumulation, as a consequence of transcriptional activation of the saponin biosynthetic genes (Suzuki et al., 2005).
The concerted transcriptional activation of entire secondary metabolic pathways by JAs is conserved across the plant kingdom. However, downstream of the conserved JA perception and initial signaling cascade, species-specific transcriptional machineries exist that regulate the transcriptional activity of the specific biosynthetic genes (Pauwels et al., 2009; Pauwels & Goossens, 2011; De Geyter et al., 2012). A few transcription factors regulated by the JA signaling cascade that activate the transcription of (sesqui)terpenoid biosynthetic genes have already been characterized (De Geyter et al., 2012), but none for triterpenoids so far. It should be noted, however, that JAs are not the only regulators of terpenoid metabolism in plants and that complex cross-talk between various stress- and development-related signaling cascades occurs (De Geyter et al., 2012).
In addition to the transcriptional, developmental and spatiotemporal modulation of terpenoid biosynthetic genes, post-translational regulatory mechanisms also exist in terpenoid biosynthesis. The activity of HMGR, the enzyme that catalyzes the key regulatory step of the MVA pathway, is controlled at the protein level through the action of protein phosphatase 2A (Leivar et al., 2011) or by the E3 ubiquitin ligase SUD1 (Doblas et al., 2013).
3. Bioengineering of terpenoids in planta
Because of their strict regulation, most terpenoids are produced in very small amounts in their natural sources. The low yield makes extraction expensive, which is eventually reflected in their market value. Consequently, there is a wide gap between demand and supply of terpenoids, which hampers their widespread application. The classical approach to ensure a constant or improved yield is the selection and propagation of high-producing cultivars or the production and/or elicitation of (transgenic) plant (cell) cultures (Zhao et al., 2005; Georgiev et al., 2009, 2012; Lambert et al., 2011; Lim & Bowles, 2012; Wilson & Roberts, 2012). Our growing understanding of terpenoid biosynthesis, together with the development of functional genomics and systems biology toolkits, has enabled the metabolic engineering of whole plants and plant cultures to enhance productivity and alter terpenoid distribution in planta (Roberts, 2007; Dudareva et al., 2013).
As terpenoid biosynthesis is strictly regulated and often controlled by specific transcription factors, one way to increase productivity is to modulate the expression of such or other regulatory factors (Broun, 2004; De Geyter et al., 2012). However, despite the identification of transcription factors that steer the biosynthesis of terpenoids, the overexpression of a single transcription factor often does not lead to a higher production of the compounds. For instance, the overexpression of ORCA3, an APETALA2/ETHYLENE RESPONSE FACTOR (AP2/ERF) transcription factor that controls the expression of several terpenoid indole alkaloid (TIA) biosynthetic genes, is not sufficient to induce TIA production in Catharanthus roseus cell cultures, indicating that only a part of TIA biosynthesis is under the control of this transcription factor (van der Fits & Memelink, 2000). Hence, further elucidation of the complex signaling cascades that lead to an increased accumulation of terpenoids is mandatory for large-scale metabolic engineering of terpenoid production using transcription factors. To date, in planta triterpenoid engineering has been hampered by the lack of knowledge about the regulatory mechanisms controlling gene expression (Sawai & Saito, 2011). Hence, a challenge for future triterpenoid research will be to identify the transcription or other regulatory factors that steer their biosynthesis.
A second way to increase productivity is by the specific overexpression of rate-limiting enzymes in the pathway. The overexpression of genes encoding enzymes such as HMGR, deoxyxylulose 5-phosphate synthase (DXS) and prenyltransferases, has been used to elevate terpenoid levels in plant tissue cultures (Degenhardt et al., 2003). Enhanced terpenoid production has also been observed on alteration of the subcellular localization of enzymes, presumably resulting from the uncoupling of biosynthesis and regulation (Bouwmeester, 2006; Wu et al., 2006; Farhi et al., 2011; Kumar et al., 2012). A single study has reported an attempt to engineer triterpenoid synthesis in tobacco (Nicotiana tabacum) by the heterologous expression of an avian FPP synthase (FPPS) and a yeast squalene synthase (SQS) gene targeted to the cytoplasm or plastid. No differences in squalene accumulation caused by specific targeting of the enzymes were observed. However, when the enzymes were directed to the trichomes through a trichome-specific promoter, higher squalene accumulation was accompanied by negative effects on plant growth and physiology. Remarkably, these additional effects were not observed when the same genes were expressed from a constitutive viral promoter (Wu et al., 2012). Nonetheless, this study underscores the potential to engineer triterpenoids in planta by relocation of the biosynthetic pathway and enhancement of the precursor flux, and encourages future research on this terpenoid class.
In addition to enhancing terpenoid production yields, in planta engineering has also been used as a tool to modulate the terpenoid composition of plants for other purposes, such as β-carotene to engineer crop nutritional value (Farré et al., 2011) and volatile terpenoid compounds to improve plant defense, pollinator attraction, scent or aroma (Dudareva et al., 2013), amongst others.
III. ‘Synthetic’ terpenoid biology
1. Definition of synthetic biology concepts
Metabolic engineering was defined in 1991 as ‘the directed improvement of production, formation, or cellular properties through the modification of specific biochemical reactions or the introduction of new ones with the use of recombinant DNA technology’ (Bailey, 1991). Since then, metabolic engineering has enabled spectacular advances in the production of a myriad of small compounds, including terpenoids, particularly in microbes. Now, demands have increased and biological functions that do not exist in nature are also desired. This can be achieved through synthetic biology, which can be defined as ‘the design and construction of new biological components, such as enzymes, genetic circuits, and cells, or the redesign of existing biological systems’ (Keasling, 2008). More elaborately, synthetic biology refers to the redesign of complex natural living systems in a rational and systematic way to simplified, predictable and controllable modules that can be modeled and manipulated to generate industrially scalable systems with a defined purpose. For many years, the term ‘synthetic biology’ was used to describe concepts that would be classified today as metabolic engineering. However, the definitions are not sharp-edged, and hence metabolic engineering might still be considered as the simplest form of synthetic biology (Channon et al., 2008).
Based on the approach used for synthetic biology, two main branches, commonly referred to as ‘top–down’ and ‘bottom–up’ synthetic biology, can be recognized. The top–down approach involves the introduction of exogenous genes into a host and the engineering of its native metabolic networks to reprogram cellular behavior by employing engineering and mathematical modeling toolkits. The bottom–up approach utilizes the biochemical toolkit for the de novo construction of synthetic genomes and unnatural components that behave in an analogous manner to their natural counterparts, and thereby allows the genesis of artificial living systems. The top–down approach of metabolic engineering for the production of useful products pertains to one of the most established concepts in the field of synthetic biology. Metabolic engineering combines transgene expression with the analysis of metabolic networks to optimize genetic and regulatory processes within cells for the production of a desired product. Metabolic engineering in a heterologous host may also involve the mathematical modeling of the host's native metabolic networks to calculate the yield of the desired product, the measurement of metabolic fluxes to pinpoint parts of the network that constrain production, genetic engineering of the host network to relieve these constraints and modeling of the modified network to calculate the product yield until an industrially applicable level is obtained (Koffas et al., 1999).
Contrary to cell-based synthetic biology, in which the cell's growth and survival objectives might interfere with the engineering objective, that is, the production of a desired compound, cell-free ‘in vitro synthetic biology’ provides a bottom–up platform, in which all available resources are concentrated on a user-defined objective, which could eventually result in improved production systems (Harris & Jewett, 2012). A cell-free environment is highly flexible and devoid of genetic regulation or transport barriers, facilitating substrate addition and product purification.
Alongside the engineering of organisms for enhanced production, synthetic biology also aims to create novel compounds with useful properties. One way to achieve this is by ‘combinatorial biosynthesis’, which allows the generation of new-to-nature compounds through the assembly of genes from different organisms, but catalyzing reactions in related pathways in a native or heterologous host, thereby establishing new enzyme–substrate combinations in vivo (Julsing et al., 2006). An alternative way to create novel compounds is by ‘directed evolution’ or ‘enzyme engineering’. The concept of directed enzyme evolution mimics the process of natural evolution and employs a set of methodologies to enhance or modify the function of a progenitor enzyme to accept an unnatural substrate or to catalyze a new biosynthetic reaction, thereby resulting in the formation of novel products (Dalby, 2011). Obviously, this concept can also be used in metabolic engineering for enhanced production by improving enzyme performance with its natural substrates.
2. Metabolic engineering and microbial biosynthesis of plant terpenoids
Compared with plant production systems, microorganisms are attractive alternatives as heterologous hosts because of their rapid doubling time, robustness under process conditions, ease of scalability, simplicity of product purification because of the absence of competing contaminants and cost-effectiveness resulting from the conversion of inexpensive feedstock to valuable compounds (Zhang et al., 2011). The choice of a suitable host (or ‘chassis’) is critical and should be based on multiple factors, including the chemical nature and complexity of the product to be synthesized, the genetic amenability of the host, the intrinsic availability of precursors for product biosynthesis, the codon usage bias of the host, the need for post-translational modifications and the feasibility to metabolically engineer the host to boost productivity (Keasling, 2010). Microbial synthesis of any plant natural product can be achieved by ‘precursor-mediated product synthesis’, in which an existing host pathway is altered to incorporate a heterologous pathway, or by ‘de novo synthesis’, in which new-to-host biosynthetic routes are imported, thereby avoiding feedback regulation (Chang & Keasling, 2006). After the establishment of heterologous synthesis, it is usually imperative to metabolically engineer the host to optimize the production yield and rate (Chemler & Koffas, 2008).
The colloquial hosts Escherichia coli and Saccharomyces cerevisiae have been employed for both precursor-mediated and de novo synthesis of mono-, di-, sesqui-, tri- and tetraterpenoids (Misawa, 2011), with artemisinic acid, the precursor of the antimalarial drug artemisinin, as the showcase for plant-derived terpenoids (Keasling, 2012). The prokaryotic E. coli has an inherent MEP pathway and the eukaryotic S. cerevisiae has the MVA pathway to produce IPP and its isomer DMAPP. Theoretically, terpenoid biosynthesis can be incorporated into these hosts by expressing the corresponding genes, but low yields may be obtained because of the limited intracellular IPP pool. The IPP and subsequent precursor levels have been supplemented by metabolic engineering of: (1) the MVA pathway in E. coli (Campos et al., 2001); (2) the MEP pathway and prenyltransferases in E. coli (Kajiwara et al., 1997); (3) the MVA pathway by a feedback regulation-deficient HMGR in S. cerevisiae (Ro et al., 2006); (4) the MVA pathway by decreasing downstream enzymes to accumulate precursors in S. cerevisiae (Paradise et al., 2008); (5) the global transcription factor regulating sterol biosynthesis in S. cerevisiae (Davies et al., 2005); and (6) protein scaffolds for the MVA pathway in S. cerevisiae (Dueber et al., 2009; Fig. 3).
Alongside targeted engineering, global approaches have been applied to improve the terpenoid pathway flux in microbial hosts. A ‘chromosomal promoter engineering’ strategy was used to express some of the endogenous MEP genes from a strong bacteriophage T5 promoter in an E. coli strain harboring β-carotene biosynthetic genes, resulting in the enhanced production of β-carotene relative to the parental strain (Yuan et al., 2006). Similarly, a ‘global transcription machinery engineering’ on the rpoD gene encoding σ70, the primary sigma factor, resulted in increased lycopene production in E. coli (Alper & Stephanopoulos, 2007).
Once precursor synthesis has been optimized, another major hurdle to overcome is to achieve functional expression of the pathway genes downstream of the precursor, particularly CytP450s. Plant CytP450s are endoplasmic reticulum-localized enzymes with a prerequisite for a CytP450 reductase (CPR) partner for efficient functioning (Podust & Sherman, 2012). In this regard, S. cerevisiae, with its native CytP450s and CPR, has an advantage over E. coli for the expression of complex terpenoid pathways (Hamann & Møller, 2007). Nevertheless, plant CytP450s supplemented with a plant CPR have been successfully expressed in both E. coli and S. cerevisiae (Arsenault et al., 2008).
Saccharomyces cerevisiae has already been employed for the expression of triterpenoid saponin biosynthetic genes. Through its native ergosterol biosynthesis, S. cerevisiae produces oxidosqualene, the precursor of saponins. In engineered strains optimized to accumulate oxidosqualene, different OSCs and CytP450s have been expressed, mainly for their functional characterization (Augustin et al., 2011; Fig. 2). Engineering efforts have been limited to the production of β-amyrin only. Through a conventional pathway engineering approach, a final titer of 6 mg l−1 was demonstrated in an S. cerevisiae strain expressing a β-amyrin synthase (bAS) from Artemisia annua (Kirby et al., 2008). Subsequent to a genotype-to-phenotype linking study, a 500% improvement in β-amyrin production was achieved by overexpression of the native genes, ERG8, ERG9 and HFA1, in an S. cerevisiae strain expressing a Pisum sativum bAS, resulting in a final titer of 3.93 mg l−1 (Madsen et al., 2011). The β-amyrin levels produced by the parent strains in the above reports reflect the cyclization efficiency of the enzymes employed. Therefore, by employing a more efficient bAS (or any other saponin biosynthetic gene), followed by targeted and/or global engineering, it should be possible to further enhance β-amyrin (or triterpenoid) levels.
The β-amyrin-producing S. cerevisiae strains have been utilized as a tool for the in vivo expression and characterization of novel CytP450s. The co-expression of a CytP450 with a plant-derived CPR resulted in the generation of yeast strains producing different sapogenins. The expression of M. truncatula CYP716A12, together with the Lotus japonicus bAS and the L. japonicus CPR, resulted in the production of oleanolic acid in yeast (Fukushima et al., 2011; Fig. 2). β-Amyrin has also been modified to natural and rare triterpenoids by the combination of multiple CytP450s in yeast. The expression of M. truncatula CYP72A68v2 and CYP93E2 in the oleanolic acid-producing strain resulted in the production of gypsogenic acid and 4-epi-hederagenin, respectively (Fukushima et al., 2013). In addition to β-amyrin-producing strains, α-amyrin-, lupeol- and dammarenediol-producing yeasts have been employed for the functional characterization of CytP450s. The M. truncatula CYP716A12 also catalyzes the C-28 oxidations of α-amyrin to ursolic acid and lupeol to betulinic acid in yeast (Fukushima et al., 2011). Similarly, the C-6 and C-12 hydroxylations of dammarenediol by CYP716A53v2 and CYP716A47, respectively, have been demonstrated in yeast (Han et al., 2011, 2012).
To complement metabolic engineering, synthetic biology offers a plethora of tools through the generation of minimal hosts, standard biological parts, regulatory elements, vectors, assembly methods and in silico computer-aided design tools (Keasling, 2012). The first and main requirement for the production of any natural product is the availability of a robust host. Synthetic biology facilitates the generation of ‘minimal hosts’ that contain only the genes essential for their growth to synthesize macromolecules from simple and inexpensive feedstock. Minimal hosts of E. coli have been generated with c. 15% genome reduction by the deletion of non-essential genes (Pósfai et al., 2006). For S. cerevisiae, the synthetic yeast genome project Sc2.0 aims to design fully synthetic minimized hosts without transposable elements and telomeric sequences, with relocated tRNAs and with site-specific recombination sites incorporated into the genome. Two partially synthetic S. cerevisiae chromosomes with genome reductions of 15–20% have been generated and successfully reincorporated (Dymond et al., 2011). The Streptomyces avermitilis linear chromosome was reduced to 81.46% of the wild-type chromosome by stepwise deletion of a region of > 1.4 Mb, including genes coding for the synthesis of all endogenous secondary metabolites. The minimized strain was able to produce artemisin precursors on expression of a synthetic codon-optimized A. annua amorphadiene synthase gene (Komatsu et al., 2010). In addition, the feasibility of generating completely artificial synthetic hosts with a desired set of genes has been demonstrated by the cloning of a chemically synthesized and assembled Mycoplasma genitalium genome in S. cerevisiae (Gibson et al., 2008).
Most often metabolic engineering focuses on the maximization of the production of a final compound with less attention to the behavior of intermediates. Contrary to this, the bottom–up synthetic biology approach allows the deconvolution of metabolic pathways to independent parts that are optimized for host-specific expression, and are subsequently incorporated rationally to build production modules. The repositories of functional parts (promoters, ribosomal binding sites, protein domains, terminators, etc.), generated within synthetic biology initiatives, facilitate the assembly of metabolic pathways (Boyle & Silver, 2012). Two depositories with codon-optimized parts for pathway engineering in E. coli (The Registry of Standard Biological Parts, partsregistry.org/Main_Page) and terpenoid engineering in S. cerevisiae (Serber et al., 2012) have been described. Synthetic biology also promotes the variable expression of related biosynthetic genes to avoid metabolic bottlenecks. Robust synthetic promoter libraries with defined promoter strengths enable modular gene expression in bacteria and yeast (Hammer et al., 2006; Nevoigt et al., 2006). Tunable intergenic regions that generate mRNA secondary structures and RNase recognition sites have been employed for the differential stabilization of segments of mRNA encoding multiple enzymes in the form of operons (Pfleger et al., 2006). Synthetic protein scaffolds that are particularly efficient in overcoming rate-limiting steps have been generated to increase flux through metabolic pathways by tethering enzymes together (Dueber et al., 2009).
Natural product biosynthesis typically involves multigene pathways, thus implementing the necessity for the simultaneous expression of multiple genes in a microbial chassis. Both in vitro and in vivo methods facilitate multigene assembly in E. coli and S. cerevisiae (Ellis et al., 2011; Wang et al., 2012), some of which have been employed to assemble carotenoid biosynthetic pathways (Shao et al., 2009; Lemuth et al., 2011). In parallel, viral mechanisms, such as internal ribosome entry sites and 2A oligopeptide sequences, have been adapted for polycistronic gene expression (de Felipe, 2002). However, the latter tools have not yet been implemented for the expression of terpenoid pathway enzymes.
3. Combinatorial biosynthesis of plant terpenoids
Combinatorial biosynthesis-based reconstitution of pathways is a useful tool to generate known and novel natural products, which can be further modified by semi-synthesis. In its simplest form, combinatorial biosynthesis is the process of generating different, but structurally related, molecules through the assembly of genes from different organisms in a single host (Kirschning et al., 2007; Fig. 4a). Plants possess an immense potential for combinatorial biosynthesis (Pollier et al., 2011). However, apart from a pioneering study, in which the expression of a bacterial halogenase in C. roseus resulted in the generation of novel chlorinated TIAs (Runguphan et al., 2010), there have been no reports on a directed combinatorial biosynthesis approach for any other terpenoid or metabolite in plants to date. Nonetheless, the existing chemical diversity, together with our growing understanding of their biosynthesis, renders (tri)terpenoids appealing compounds for the combinatorial generation of novel analogs. For instance, the screening of a synthetic triterpenoid combinatorial library derived from betulinic and ursolic acid led to the identification of compounds with an enhanced anti-malarial activity relative to the parent compounds (Pathak et al., 2002).
Combinatorial biosynthesis of triterpenoid saponins holds great potential, as they exhibit a plethora of biological activities. Bardoxolone methyl, a semi-synthetic derivative of oleanolic acid, has been clinically evaluated for the treatment of chronic kidney disease. The synthesis of bardoxolone methyl occurs through chemical modifications of the three active portions of oleanolic acid that render the derivative biologically more potent than the starter molecule (Sporn et al., 2011). The enzymatic addition of extra functionalities to the triterpenoid backbone through combinatorial biosynthesis could increase the number of sites that can be accessed for further synthetic modifications (Pollier & Goossens, 2012).
A major drawback of the generation of novel molecules in planta lies in the complexity of plant metabolite extracts and the complications of purifying a compound of interest from a large pool of different molecules, including compounds with similar structures and physicochemical properties. Therefore, combinatorial biosynthesis of plant secondary metabolites has also been performed in microorganisms, which lack the production of compounds similar to the target compound (Fig. 4b). Novel carotenoid structures with an enhanced antioxidative activity have been generated in E. coli by the combinatorial expression of bacterial and plant genes (Sandmann, 2002). Recently, rare triterpenoids have been combinatorially produced in S. cerevisiae (Fukushima et al., 2013). A current obstacle to the wider utilization of combinatorial biosynthesis for plant-derived compounds is the limited availability of plant genes encoding the enzymes that catalyze the biosynthetic reactions. In the future, these bottlenecks may be solved by gene discovery in (medicinal) plants or, alternatively, by directed evolution of enzymes towards novel functions (Kwon et al., 2012).
4. Enzyme engineering or directed evolution of terpenoid biosynthetic enzymes
Small-molecule drugs, considered to be relevant as lead molecules, often have a high degree of chemical complexity with multiple functional groups and defined stereochemistry (Nannemann et al., 2011). In their natural source, these small molecules are most often synthesized by enzymes that have a high regio- and stereoselectivity, high catalysis rate and relaxed substrate specificity. Nonetheless, natural enzymes often cannot meet the requirements of industrial chemists in terms of substrate tolerance, efficiency, process tolerance and economic viability. Hence, enzymes have been engineered by directed evolution to improve one or more of their properties under defined conditions (Dalby, 2011; Fig. 4c). Directed enzyme evolution has progressed tremendously lately, and it is now feasible to engineer enzymes to accept unnatural substrates and to catalyze regio- and stereospecific reactions with an efficiency comparable with that of the natural enzymes (Goldsmith & Tawfik, 2012). The promiscuous nature of proteins gives them an inherent ability to generate novel or altered functions with a small number of amino acid substitutions (Aharoni et al., 2005), and computational methods, such as catalytic active site prediction (CLASP) and directed evolution using CLASP: an automated flow (Chakraborty et al., 2011; Chakraborty, 2012), utilize virtual screening for spatial, electrostatic and scaffold matching to identify target progenitor proteins. Enzymes catalyzing branch-point reactions in multi-branched pathways, in which a substrate is converted to multiple products, have a high evolvability. In addition, evolvable enzymes exhibit multiple mutational residues and are ‘locally specific’ as they recognize a common motif on structurally diverse substrates (Umeno et al., 2005).
Oxidosqualene, the immediate precursor of triterpenoid biosynthesis, is a versatile molecule that is cyclized into multiple products by different OSCs. Several of these OSCs are multifunctional in nature and generate multiple products in a single reaction (Phillips et al., 2006), highlighting the promiscuity, and thus evolvability, of the enzymes. Through directed evolution, the major cyclization product of a multifunctional OSC could be redefined to a specific or novel product. For other terpene synthases, this has already been successfully attempted. Following a site-saturation mutagenesis, the specificity of a carotenoid synthase was altered to generate unnatural C45 and C50 backbones in E. coli (Umeno & Arnold, 2004). The product specificity of a γ-humulene synthase from Abies grandis that cyclizes FPP to 52 different sesquiterpenoids was evolved by site-saturation mutagenesis to generate independent synthases, each producing one or a few products derived from a predominant reaction pathway (Yoshikuni et al., 2006).
This evolution approach could also be extended to downstream triterpenoid biosynthetic enzymes, in particular the CytP450s. Triterpenoid saponin backbones are made up of 30 carbons, c. 20 of which are accessible for CytP450-mediated modifications, as deduced from known saponins (Dinda et al., 2010). In addition, diverse functional groups are observed at the modifiable carbons, pointing to the existence of specific CytP450s that catalyze these specific reactions. For instance, the C-11 position of many triterpenoid backbones can be oxidized with an α- or β-hydroxy group, and a CytP450 that specifically catalyzes the α-hydroxylation has already been characterized (Seki et al., 2008). To date, only a few CytP450 families involved in triterpenoid modifications have been identified (Fig. 2). Through directed CytP450 evolution, it should be possible to: broaden their substrate acceptance to divergent backbones, target specific carbon positions, and specify the functional group to be added to the triterpenoid skeleton. Such approaches have been implemented on carotenoid desaturases that have been evolved by random mutagenesis to accept unnatural C35 carotenoid backbones in E. coli (Umeno & Arnold, 2003).
Protein engineering based on molecular evolution also serves as a tool to enhance enzyme efficiency or to abolish feedback regulation on enzymes. Through adaptive evolution, the unfavorable in vivo properties of truncated yeast HMGR were minimized for optimal functioning in E. coli, thereby also enhancing the final product yield by c. 1000-fold (Yoshikuni et al., 2008). Key to directed evolution studies is a profound understanding of sequence-to-structure-to-function relationships of a protein. Integrated databases of triterpenoid cyclases (TTCED; Racolta et al., 2012) and CytP450s (CYPED; Sirim et al., 2009) facilitate the identification of functionally relevant and selectivity-determining amino acid residues within members of a protein family by extensive sequence analysis. Therefore, the boosting of protein engineering efforts could enhance synthetic biology efforts in triterpenoid engineering in the future.
5. In vitro synthetic biology: an evolving tool
In vitro synthetic biology systems can comprise ‘synthetic enzymatic pathways’ (SEPs), in which purified enzymes are combined in an aqueous environment to convert a substrate to a product through a series of reactions. Alternatively, ‘crude extract cell-free’ (CECF) systems, in which cells are grown, harvested and lysed to obtain a crude extract, can be utilized for the conversion of a substrate to a product (Hodgman & Jewett, 2012; Fig. 5a). The choice between SEP and CECF is influenced by time, cost and the need for cellular reinforcement to support the desired network. A CECF approach, for instance, is more suited for a reaction requiring a constant supply of energy, such as protein synthesis (Carlson et al., 2012); however, unlike SEP, CECF reactions can exhibit undesirable activities because of the crude nature of the cellular extract.
The multireaction nature of biochemical networks, low protein concentrations, enhanced substrate diffusion, low enzyme proximity and low reaction rates as a result of unbalanced enzyme activity still hamper the efficiency of cell-free synthetic biology. In vitro compartmentalization (IVC) is one way of achieving proximity of reaction components. In IVC, genes are coupled to a substrate and encapsulated in water-in-oil emulsions, together with transcription and translation machineries, to facilitate enzyme synthesis and consequent product formation (Fig. 5b). Novel enzymes have been uncovered by linking product formation to genes in a confined microenvironment through IVC (Rothe et al., 2006). IVC is also being employed as a screening approach for the directed evolution of enzymes (Arnold & Volkov, 1999; Forster & Church, 2007). In addition to enclosing reaction components in a defined environment through IVC, metabolic channeling has been employed as an alternative to reduce substrate diffusion lengths (Idan & Hess, 2013). Protein scaffolding (Fig. 5c), surface tethering of enzymes (Fig. 5d), covalently linked enzyme aggregates (Fig. 5e) and foam dispersion of enzymes with liposomes using surfactants (Fig. 5f) have been employed to facilitate the spatial organization of pathway components (Hodgman & Jewett, 2012).
Current applications of in vitro synthetic biology are limited to proteins, nucleic acids and small-molecule ligands. Nonetheless, these tools can undoubtedly be extended to natural product or (tri)terpenoid engineering in the future. For instance, IVC could be employed as a tool for the directed evolution of CytP450s. A potential hurdle is the membranous nature of CytP450s, which prevents their solubilization in the aqueous reaction environment, but which may be overcome by the utilization of nanodisc membranes (Denisov & Sligar, 2011). A great advantage of using in vitro synthetic biology in triterpenoid engineering is the simplicity and ease of catalysis of precise regio- and stereospecific reactions with a high efficiency in a relatively pure form, which may overcome the drawbacks of chemical synthesis, metabolic engineering and product purification.
IV. Perspectives: ‘exploration of triterpenoids: the road ahead’
Triterpenoid saponins comprise a wide range of bioactive compounds, some of which (mainly pentacyclic triterpenoids) can be readily isolated from plant sources in considerable amounts for pharmacological studies or to serve as scaffolds for the semi-synthesis of new lead bioactive agents. Semi-synthetic derivatives of the natural pentacyclic triterpenoids oleanolic, ursolic and betulinic acid (Fig. 6) are a thousand-fold more active than the parent compound, and have been utilized in in vitro and in vivo studies for a broad range of clinical applications (Liby et al., 2007a; Liby & Sporn, 2012; Salvador et al., 2012). Such compounds certainly hold great potential, but many challenges remain. In this concluding section, we address some of the most prominent.
1. Triterpenoids that have entered clinical trials
Two types of pentacyclic triterpenoid derivatives have recently been clinically evaluated. First in class was bardoxolone (CDDO, Fig. 6), an intravenously administered semi-synthetic derivative of oleanolic acid, which was evaluated as an anti-cancer agent in patients with metastatic disease (Tsao et al., 2010). Following this study, further efforts focused on the more potent, orally administered derivative bardoxolone methyl (CDDO-Me, Fig. 6) in patients with advanced solid tumors and lymphomas. Interestingly, 90% of the patients showed significant improvements in kidney function, without developing any serious adverse drug effects (Hong et al., 2012), which prompted a phase II trial in patients with moderate to severe chronic kidney disease and type 2 diabetes. Unfortunately, the improvements in kidney function were accompanied by adverse drug effects (Pergola et al., 2011). Nonetheless, a worldwide phase III trial was initiated to access the long-term clinical benefit of CDDO-Me in slowing the progression of end-stage renal disease and lessening cardiovascular death in patients with advanced chronic kidney disease and type 2 diabetes. This trial was halted in October 2012 as a result of severe adverse effects and mortality in patients taking the drug (http://www.clinicaltrials.gov/show/NCT01351675). Synthetic oleanane triterpenoids, such as CDDO and CDDO-Me, are multifunctional drugs with potent anti-inflammatory, anti-oxidative, anti-proliferative, pro-apoptotic and differentiating effects (Liby et al., 2007b). They probably interact with multiple targets or entire regulatory networks, rather than with single molecular targets; hence, they might be most effective in the early stages of disease when a homeostatic agent is desired, contrary to an application as treatment for late-stage disease when irreversible tissue damage and cell death have occurred (Sporn et al., 2007; Liby & Sporn, 2012).
The second synthetic triterpenoid to be clinically evaluated was bevirimat (Fig. 6), a betulinic acid derivative and an orally administered, novel inhibitor of human immunodeficiency virus (HIV) maturation. Bevirimat inhibits HIV type 1 (HIV-1) replication by binding to the Gag polyprotein, thereby blocking its processing and resulting in the production of non-infectious virions (Zhou et al., 2005). Phase I and II clinical studies with bevirimat showed dose-proportional pharmacokinetics and no serious adverse events in HIV-1-infected adults (Smith et al., 2007). However, the clinical development of bevirimat was halted in June 2010 (http://www.clinicaltrials.gov/show/NCT01026727). Bevirimat has been questioned with respect to its effectiveness when used in a combined therapeutic regimen with other drugs and with regard to the ability of HIV to evolve resistance (Malet et al., 2007; Nijhuis et al., 2007; Martínez-Cajas et al., 2008; Verheyen et al., 2010).
2. Is there a future for bioactive triterpenoids in therapeutics?
Many triterpenoids still hold great potential as future therapeutics in myriad applications. The synthetic oleanane triterpenoids bardoxolone imidazolide (CDDO-Im, Fig. 6) and bardoxolone ethylamide (CDDO-EA, Fig. 6) are being studied for their ability to induce chondrogenic differentiation, which, together with their potent anti-inflammatory effect, could serve to prevent or treat osteoarthritis (Suh et al., 2012). CDDO-Me has the potential to be developed as a chemopreventive drug, as demonstrated by the delayed tumorigenesis in mouse cancer models (Tran et al., 2013). Celastrol (Fig. 6), another oleanane triterpenoid, could be of therapeutic value for the treatment of chronic diseases, such as asthma, arthritis, neurodegenerative diseases and cancer (Kannaiyan et al., 2011).
The natural and semi-synthetic derivatives of ursane triterpenoids, such as ursolic, β-boswellic, asiatic, corosolic and pomolic acid (Fig. 6), have been investigated in cancer research for their anti-proliferative and apoptotic effects (Salvador et al., 2012). A phase I study with intravenously administered ursolic acid nanoliposomes showed a linear pharmacokinetic profile and good tolerance in healthy volunteers and patients with advanced solid tumors (Zhu et al., 2013).
The betulin scaffold is also still being explored for the development of new anti-HIV agents. Betulin derivatives have been recently conjugated to other anti-HIV agents to generate multi-target single agents which could simplify treatment regimens and reduce risks caused by drug–drug interactions. Hybrid conjugates of betulin and dihydrobetulin (Fig. 6) with the nucleoside reverse transcriptase inhibitor 3′-azido-3′-deoxythymidine (AZT) have been found to be more potent than bevirimat (Xiong et al., 2010). Furthermore, an ointment containing the natural triterpenoid betulinic acid is being evaluated in a phase II study for the treatment of dysplastic melanocytic nevus, a likely precursor to melanoma (http://www.clinicaltrials.gov/show/NCT00346502).
Currently, the most promising immunological adjuvant undergoing clinical investigation is QS-21 (Fig. 6), a fraction of soluble triterpenoid glycosides from the soap bark tree (Quillaja saponaria; Sun et al., 2009). It can augment antibody and T-cell response to a variety of antigens involved in infectious diseases, degenerative disorders and cancers. Adjuvant systems containing QS-21 in combination with other immunostimulants have been formulated to promote protective immune responses following vaccination (Garçon & Van Mechelen, 2011). Clinical studies utilizing a QS-21 adjuvant system for a candidate malaria vaccine have advanced to phase III trials, where modest protection against clinical and severe malaria was observed in African infants (RTS et al., 2012). Another QS-21 adjuvant system has been employed in a phase I/II study for a candidate HIV-1 vaccine which induced T-cell response in seronegative volunteers, thus supporting further clinical investigation (Van Braeckel et al., 2011).
Tetracyclic triterpenoids have been hitherto less explored, but also exhibit great therapeutic potential. Withanolides, such as withaferin A (Fig. 6), display anti-inflammatory, immunoregulatory, anti-tumor, anti-angiogenic and chemopreventive activities (Mirjalili et al., 2009; Mayola et al., 2011; Zhang et al., 2012). Cucurbitacins have been studied for their ability to induce apoptosis in cancer cell lines (Chen et al., 2012) and, like most triterpenoids, target multiple signaling networks, highlighting their usefulness as cytostatic agents (Ríos et al., 2012). The ginsenosides, the tetracyclic triterpenoid glycosides from Ginseng (Panax spp.), have been demonstrated to possess anti-cancer activities through the modulation of diverse molecular mechanisms in various pre-clinical and clinical studies (Nag et al., 2012).
The multiple mechanisms by which triterpenoids can instigate cell death impede the development of resistance against them and maintain their status as attractive candidates for drug development. Nonetheless, true proof-of-concept for their utility as effective drugs, and ultimately market blockbusters, can only be brought about via a series of well-designed pre-clinical studies that use triterpenoid compounds in well-characterized models to unambiguously establish structure-to-activity relationships. Such information can then be exploited further to semi-synthesize even more efficacious derivatives with superior ADMET (absorption, distribution, metabolism, excretion, toxicity) properties. In addition, an in-depth understanding of the molecular mechanisms that underlie their biological activities will be necessary to harness their full potential.
3. The need for more ‘plant’ knowledge
In addition to the cost and effort involved in the drug discovery and development process itself, pharmaceutical companies often face another major challenge, which is to be able to scale up the production of the active principle and make the process cost-efficient, and, last but not least in the case of natural products, sustainable!
Although some triterpenoids, such as oleanolic acid, can be extracted from by-products of the olive (oil) industry, and thus are available in ample amounts (Pollier & Goossens, 2012), many others, such as the ginsenosides, are scarce, and extraction from plants alone is insufficient. In addition, the triterpenoid profiles of plants are variable and often influenced by environmental factors, which may affect the quantity and quality of the bioactive principle that can be extracted from the same biomass. Furthermore, triterpenoid-producing plants may have a slow growth rate or be difficult to grow, which makes cultivation non-profitable to farmers. Even when a natural product drug can be produced in large amounts in planta, there can be supply and demand imbalances, which may feed back to fluctuations in cultivation acreages and yields. A metabolic engineering or synthetic biology platform may provide an alternative and sustainable prospect to agricultural supply by creating a complementary non-seasonal, high-quality source for valuable bioactive triterpenoids. Clearly, the development of alternative performing heterologous production platforms will be accompanied by multiple challenges, which need to be balanced against the concerns about the (stability of the) market value of the drug to be produced.
The artemisinin case has shown that synthetic biology can reach industrial-scale deployment for drug production (Keasling, 2012; http://www.nature.com/news/malaria-drug-made-in-yeast-causes-market-ferment-1.12417). In the case of triterpenoids, bioengineering may follow the beaten track established for semi-synthetic artemisinin. However, it may also involve distinct host optimization for large-scale triterpenoid production. Obviously, yeasts will remain potent vehicles, but microalgae or plants amenable to culture in bioreactors and engineering technologies certainly represent attractive alternative hosts for a triterpenoid-oriented synthetic biology program.
A major restraint to the successful bioengineering of plant-derived triterpenoids is the scarcity of indispensable knowledge about their biosynthesis, which hampers both plant and microbial engineering. Most triterpenoid saponins are known to accumulate in a tissue-, organ- or signal-specific manner in plants, but there is virtually no insight into the mechanisms responsible for this pattern. Multiple OSCs catalyzing the cyclization of 2,3-oxidosqualene to different triterpenoid precursor backbones have been isolated already, but only a handful of genes corresponding to the ‘decorating’ enzymes have been identified, whereas hundreds must exist when considering the structural diversity of triterpenoids in the plant kingdom. Similarly, although the biosynthetic enzymes are mostly microsomal in nature, triterpenoids typically localize to the epidermal wax layer or the vacuoles, suggesting the existence of yet undiscovered transporter systems. Hence, there is a great need to unravel the molecular mechanisms involved in triterpenoid saponin production in planta to assist their exogenous engineering.
Fortunately, the booming number of functional genomics technologies with ever-increasing resolution and coverage of the genome, transcriptome, proteome, interactome and metabolome will offer the necessary power to list all the possible elements involved in the synthesis of plant terpenoids in the near future. In particular, the linking of signal- and tissue-dependent metabolome and transcriptome analysis will remain a powerful principle to pinpoint biosynthetic genes, transporters and transcription factors. If successful triterpenoid-related gene discovery can profit from the numerous tools and platforms that are meanwhile being developed in the field of synthetic biology to reduce the cost and time required to engineer biological systems, triterpenoid bioengineering awaits a bright future.
We thank Annick Bleys for help in preparing the manuscript. This work was supported by the European Union Seventh Framework Programme FP7/2007–2013 under grant agreement number 222716 – SMARTCELL. T.M. is indebted to the VIB International PhD Fellowship Program for a predoctoral fellowship. J.P. is a postdoctoral fellow of the Research Foundation-Flanders.