The Biosynthesis of Artemisinin (Qinghaosu) and the Phytochemistry of Artemisia annua L. (Qinghao)

The Chinese medicinal plant Artemisia annua L. (Qinghao) is the only known source of the sesquiterpene artemisinin (Qinghaosu), which is used in the treatment of malaria. Artemisinin is a highly oxygenated sesquiterpene, containing a unique 1,2,4-trioxane ring structure, which is responsible for the antimalarial activity of this natural product. The phytochemistry of A. annua is dominated by both sesquiterpenoids and flavonoids, as is the case for many other plants in the Asteraceae family. However, A. annua is distinguished from the other members of the family both by the very large number of natural products which have been characterised to date (almost six hundred in total, including around fifty amorphane and cadinane sesquiterpenes), and by the highly oxygenated nature of many of the terpenoidal secondary metabolites. In addition, this species also contains an unusually large number of terpene allylic hydroperoxides and endoperoxides. This observation forms the basis of a proposal that the biogenesis of many of the highly oxygenated terpene metabolites from A. annua – including artemisinin itself – may proceed by spontaneous oxidation reactions of terpene precursors, which involve these highly reactive allyllic hydroperoxides as intermediates. Although several studies of the biosynthesis of artemisinin have been reported in the literature from the 1980s and early 1990s, the collective results from these studies were rather confusing because they implied that an unfeasibly large number of different sesquiterpenes could all function as direct precursors to artemisinin (and some of the experiments also appeared to contradict one another). As a result, the complete biosynthetic pathway to artemisinin could not be stated conclusively at the time. Fortunately, studies which have been published in the last decade are now providing a clearer picture of the biosynthetic pathways in A. annua. By synthesising some of the sesquiterpene natural products which have been proposed as biogenetic precursors to artemisinin in such a way that they incorporate a stable isotopic label, and then feeding these precursors to intact A. annua plants, it has now been possible to demonstrate that dihydroartemisinic acid is a late-stage precursor to artemisinin and that the closely related secondary metabolite, artemisinic acid, is not (this approach differs from all the previous studies, which used radio-isotopically labelled precursors that were fed to a plant homogenate or a cell-free preparation). Quite remarkably, feeding experiments with labeled dihydroartemisinic acid and artemisinic acid have resulted in incorporation of label into roughly half of all the amorphane and cadinane sesquiterpenes which were already known from phytochemical studies of A. annua. These findings strongly support the hypothesis that many of the highly oxygenated sesquiterpenoids from this species arise by oxidation reactions involving allylic hydroperoxides, which seem to be such a defining feature of the chemistry of A. annua. In the particular case of artemisinin, these in vivo results are also supported by in vitro studies, demonstrating explicitly that the biosynthesis of artemisinin proceeds via the tertiary allylic hydroperoxide, which is derived from oxidation of dihydroartemisinic acid. There is some evidence that the autoxidation of dihydroartemisinic acid to this tertiary allylic hydroperoxide is a non-enzymatic process within the plant, requiring only the presence of light; and, furthermore, that the series of spontaneous rearrangement reactions which then convert this allylic hydroperoxide to the 1,2,4-trioxane ring of artemisinin are also non-enzymatic in nature.


Malaria
Malaria is an infectious disease which has affected human beings since the dawn of recorded history. By the middle of the last century, however, many felt that malaria was on the retreat and that one day it might even be vanquished. Two factors were primarily responsible for this perceived reduction in the severity of the malarial threat. Firstly, the Anopheles mosquito, which transmits the disease to humans, could at last be controlled by widespread application of the insecticide DDT. Secondly, the Plasmodium parasite, which causes malaria (four members of the genus infect humans: P. falciparum, P. vivax, P. malariae and P. ovale), could also be effectively be controlled by the use of synthetic analogues of quinine (itself also a natural product, obtained from the bark of the cinchona tree), such as chloroquine, which had been developed before World War II.
By the 1960's, however, malaria was back with a vengeance. The mosquitoes were developing resistance to DDT, which was soon to be banned in any case because of environmental concerns. The Plasmodium falciparum parasite, which is responsible for cerebral malaria, an often fatal complication, was also developing resistance to chloroquine. Thailand and South America were the first regions to be affected, but resistance to chloroquine soon spread to many other parts of the World. Nowadays, it is particularly serious in South East Asia. It was against this background of increasing resistance, and of the on-going wars in neighboring Cambodia and Vietnam, that the Chinese government began a major initiative to discover new antimalarials from plants used in Traditional Chinese Medicine (TCM).

Artemisia annua (Qinghao)
The herb "Qinghao" first appeared in a book entitled "Wu Shi Er Bing Fang" (Prescriptions for Fifty-Two Ailments) more than two thousand years ago. The earliest reported use for Qinghao was for the treatment of haemorroids; but "Zhou Hou Bei Ji Fang" (Handbook of Prescriptions for Emergency Treatment), written in 340 AD, describes the use of Qinghao as a treatment for fevers [1]. The first text in which Qinghao might specifically be identified as a remedy for malaria is "Ben Cao Gang Mu" (1596) in which the herb is described as "a treatment for hot and cold due to intermittent fever illness". These old pharmacopeias describe preparations in which the leaves (collected in summer or spring) are pounded with a pestle and mortar in order to express the "juice". This procedure was, perhaps, intended to improve the recovery of essential oils from trichomes on the leaf surface, in which the active principal, artemisinin, is now thought be contained.
It is not entirely clear whether "Qinghao" which is referred to in these ancient texts is solely of semi-synthetic drugs with pharmacological properties superior to those of the parent [8]. The most important such derivatives are artemether, arteether and artesunate (Figure 1), which exhibit greater potency than artemisininin itself, as well as improved solubility, and favourable metabolic and hydrolytic stabilities. Formulations based on these drugs are now at the heart of the WHO's global fight against malaria The entire world production of such semi-synthetic artemisinin derivatives is currently reliant on harvesting and extraction of artemisinin from A. annua plants, which is practiced on a multi-tonne scale in countries such as China and Vietnam. Malaria is now the most serious infectious disease in the World, with at least 300 million cases reported every year. It is estimated to be responsible for up to 2 million deaths annually -mainly amongst children -with more than half of the deaths occurring amongst the poorest 20% of the World's population. The importance of artemisinin has been founded on a continuing lack of resistance almost three decades after its introduction -although, very recently, reports of the emergence of resistance have indeed begun to appear [7,8] This slow onset of resistance may be a consequence of the unique mechanism of action for this drug, which is associated with the unusual endoperoxide group. It is thought that artemisinin becomes activated when its endoperoxide group comes into contact with Fe(II) in free haem groups, which have been liberated by the parasite's digestion of the haemoglobin contained in the red blood cell.
The appearance of resistance to artemisinin could be a devastating blow for many parts of South East Asia and Africa, where artemisinin-based drugs are currently the only effective treatment for malaria (resistance to the older generation of quinine-derived antimalarials having already become endemic to these regions). In an attempt to forestall the emergence of resistance, the WHO have been recommending that artemisnin should be taken in combination with another antimalarial drug -socalled Artemisinin Combination Therapy (ACT). This strategy is designed to slow the development of resistance, because during treatment with two drugs, the chance of a mutant emerging which is resistant to both is the product of the probabilities of resistance arising to either drug separately. It seems that, in the continuing absence of an effective malaria vaccine, the development of new antimalarial drugs -most likely derived from, or inspired by, artemisinin -will continue to be our primary weapon in the fight against malaria.
camphor and germacrene D [21] (others have confirmed the absence of artemisia ketone from Vietnamese oil) [52]. Interestingly, this difference was also reflected in the artemisinin content of Chinese and Vietnamese plants at 0.17% and 1.0% dry weight, respectively. Others have reported a similar variation in essential oil content between varieties of A. annua [53,54] and have concurred that differences in the essential oil composition should be ascribed to the existence of chemotypes (or chemical races) in this species [43] (see also Section 4.1).
GC-MS studies are particularly suited to the analysis of the more volatile components of the plant metabolome, such as the monoterpenes (Section 2.5) and some of the unfunctionalized sesquiterpene hydrocarbons which are reported in Section 2.6. The identification of a metabolite by GC-MS generally requires that its retention time and mass spectrum be matched with that of a known standard, which is recorded in a database. Using this thechnique, it is therefore possible to make a very rapid analysis of a large number of compounds employing a relatively small amount of plant material. Many of the simple aliphatic and aromatic metabolites which are reported in Sections 2.1 and 2.2 are actually comparatively minor components of the essential oil of A. annua, that have been identified solely on this basis. Although it is also possible to analyse more highly oxygenated sesquiterpenes, such as artemisinin and its biosynthetic precursors by GC-MS [55] (Section 2.6), these less volatile components are generally more easily isolated by liquid chromatography (LC) [15]. LC is definitely the technique of choice for the more non-volatile compounds of the extract of A. annua, which include flavonoids (Section 2.4), and many triterpenoids and sterols (Section 2.7). In the preparative mode, the LC techniques [56] of column chromatography and high performance liquid chromatography (HPLC) can provide sufficient material to allow for the subsequent structural elucidation of completely novel metabolites, when used in conjunction with techniques such as nuclear magnetic resonance (NMR) spectroscopy and X-ray crystallography. Most of the structures of the more highly oxygenated cadinane and amorphane sesquiterpenoids from A. annua which are reported in Section 2.6.3, were determined by this more powerful, but also more laborious approach. The majority of these components are unique to this species and several have turned out to be relevant to the biosynthesis of artemisinin.
The highly oxygenated nature of many of the terpenoid metabolites from A. annua has been stressed throughout this review, because of its perceived relevance to the biosynthesis of artemisinin. In particular, the reader's attention has been drawn to the unusually wide diversity of terpenoid allylic hydroperoxides and endoperoxides which have been recorded from this species. It is suggested that several of these peroxides result from the reaction of molecular oxygen with the tri-substituted double bond of an appropriate (and frequently abundant) mono-sesqui-or diterpene precursor. Thus, allylic hydroperoxides are found for monoterpenes (264 and 265; Section 2.5.1), sesquiterpenes (414 and 481; Section 2.6.3) and diterpenes (553; Section 2.7.1), all of which might be derived from abundant hydrocarbon precursors; in addition to one monoterpene endoperoxide (329; Section 2.5.3) and three sesquiterpene endoperoxides (465, 495 and 497; Section 2.6.3). These reactive hydroperoxides might then be responsible for the formation of a large number of the highly-oxygenated terpenes reported from A. annua.
Finally, it has recently been proposed that the the yin-yang nature of Chinese herbal medicine might equate to antioxidation-oxidation in modern parlance [57]. If this is true, then the wide-ranging antioxidant properties associated with the various terpenoidal components from A. annua (i.e. their propensity to undergo spontaneous autoxidation) would be entirely consistent with its classification as a cooling herb in the TCM system.

Aromatic Alcohols, Ketones and Acids
The simple aromatic compounds described in Tables 12, 13 and 14 form a relatively small group of natural products from A. annua. Metabolites in this section are probably biosynthesised by both the polyketide and shikimate pathways (some simple aromatics which are derived from the terpenoid pathway are discussed separately in Section 2.5.3).

Phenylpropanoids
The phenylpropanoids, all of which contain a C 3 substituent fused to a benzene ring (C 6 ), are produced by the shikimate pathway, which is unique to plants. Most of the simple phenylpropanoids reported in Table 15 were described from GC-MS studies. Compounds 166-182 (Table 16), which are esters formed by various combinations of ferulic and cinnamic acid with the four hydroxyl groups of quinic acid, were isolated from a single HPLC-MS study [71]. Some of the coumarins reported in Table 17 have also been obtained from undifferentiated tissue cultures (callus and suspension) of A.
annua (see Section 4.2). The structures of both of the 2,2-dihydroxychromene natural products 190 and 191, which appear in Table 17, are questionable on thermodynamic grounds -one might expect both to lose a molecule of water thereby forming a more highly conjugated coumarin.

Flavonoids
A. annua L. is a rich source of flavonoids, as are many other members of the Asteraceae family. It has been suggested that some of the more abundant methoxylated flavonoids from A. annua may potentiate the antimalarial activity of artemisinin in crude extracts of this plant. These flavones include: casticin (227), artemetin (228) [91], chrysosplenol D (225) and chrysoplenetin (226) [92] (interstingly, the latter two flavonoids are also reported to potentiate the activity of berberine and norfloxacin against a resistant strain of Staphylococcus aureus). Perhaps for this reason, phytochemical studies have sometimes sought to determine the distribution of bioactive flavonoids (such as chrysoplenetin (226), casticin (227), eupatin (232) and artemetin (228)) in conjunction with that of artemisinin [93].

Monoterpenoids
Monoterpenoids generally contain ten carbons (C 10 ) and are the principal components of the essential oil of A. annua obtained by steam distillation (or other techniques that are selective for more volatile natural products). It is widely believed that monoterpenes are located in the glandular trichomes -small structures, which are loosely attached to the surfaces of the leaves and flowers [108].
The regular acyclic monoterpenes which are described in Section 2.5.1 consist of an eight-carbon chain, which is often functionalized at the 1-position, with methyl substituents at the 3-and 7-positions (Table 23). They are formed by ionization of the pyrophosphate group in one of the C 10 precursors: geranyl pyrophosphate (GPP), neryl pyrophosphate (NPP) or linalyl pyrophosphate (LPP) (Figure 2) [these monoterpene precursors are, in turn, formed from the "head-to-tail" condensation of a "starter" molecule of dimetylallypyrophosphate (DMAPP) with a "chain extender" of isopentenyl pyrophosphate (IPP) (both C 5 )]. A. annua is also a rich source of irregular acyclic monoterpenoids (Section 2.5.2), which are derived from the "head-to-middle" condensation of two C 5 precursors. The "regular" acyclic monoterpenoid precursors can then undergo further intramolecular reactions to yield monocyclic monoterpenes (Section 2.5.3). The largest group of such monocyclic monoterpenes in A. annua is the p-menthane series (Table 25), which incorporate a single sixmembered ring. Further cyclization produces bicyclic monoterpenes (Section 2.5.4), which may contain an additional five-membered ring (e.g., camphanes in Table 26); or a four-membered ring (pinanes , Table 27); or a three-membered ring (thujanes, Table 28). Studies with 14 C-labeled LPP have shown that in A. annua this precursor is converted to cyclised monoterpenes such as 1,8-cineole (326) and α-pinene (347) with greater efficiency than the alternative precusors, NPP and GPP [109].

Regular Acyclic Monoterpenes
A cDNA for (3R)-linalool synthase, which converts GPP to (3R)-linalool (250) by ionization of the pyrophosphate group, has been described recently from A. annua [110]. Most of the regular acyclic monoterpenes reported in Table 23 can be derived by further functional group modifications (dehydration, reduction or oxidation) of linalool (250) or its isomers, geraniol (244) and nerol (248).

Irregular Acyclic Monoterpenes
Phytochemists were interested in A. annua before the discovery of artemisinin because it is a rich source of unusual irregular acyclic monoterpenoids, such as artemisia ketone (276) [115][116][117][118][119][120], which is the major constituent of the essential oil in some varieties of A. annua [29,37,42,73], and can account for up to 50% of the total [22,25,33,43,45,53]. It is often found in conjunction with smaller amounts of artemisia alcohol (273) [22,33,38,121]. Artemisia ketone (276) is formed by a non-standard "head-tomiddle" condensation of DMAPP [104,[122][123][124][125][126], which is thought to proceed as shown in Scheme 2 via chrysanthemyl pyrophosphate, an intermediate containing a three-membered ring. The mechanism for the formation of this unusual intermediate [127], is believed to mimic the formation of presqualene, another naturally-occuring cyclopropane, which is involved in the biosynthesis of the triterpene precursor, squalene (see Section 2.7.2) [128]. Two other classes of irregular cyclic monoterpenoids, the lavandulanes 279)-281 and the santolinanes 282 and 283, are also known from A. annua. The formation of all three skeletons has been explained in terms of different cleavage reactions occurring at each of the three carbon-carbon bonds in the cyclopropyl ring of the common precursor, chysanthemyl pyrophosphate (Scheme 2) [126].

Scheme 2. Formation of the irregular artemisyl, lavandulyl and santolinyl skeletons in A.
annua by "head-to-middle" condensation of a DMAPP (C 5 ) precursor and subsequent carbon-carbon cleavage reactions of the resulting intermediate, chryanthemyl pyrophosphate.

Sesquiterpenoids
Sesquiterpenoids (C 15 ) constitute the most abundant and most diverse group of natural products from A. annua. All are produced from farnesyl pyrophosphate (FPP; 378), which is the product of a "head-to-tail" condensation of three C 5 units (IPP and DMAPP; see Figure 2). Sesquiterpene hydrocarbons are generally more volatile than their highly oxygenated counterparts and are more suited to study by GC-MS (cf. monoterpenes in Section 2.5, which are the other major components of the essential oil from A. annua). More highly oxygenated sesquiterpenes, such as artemisinin (495), are best analysed by liquid chromatography.

Farnesane Sesquiterpenes
The farnesanes are the structurally simplest group of sesquiterpenes. This acyclic group of natural products is produced by ionization of the pyrophosphate group in the 15-carbon precursor, FPP (378), followed either by quenching with water or loss of a proton, which results in the variety of structures shown in Table 29. Alternatively, FPP can undergo further cyclization to mono-cyclic sesquiterpenes (Section 2.6.2), bicyclic sesquiterpenes (Section 2.6.3) or tricyclic sesquiterpenes (Section 2.6.4), according to enzymatically-catalysed mechanisms which bear close analogies with the biosynthesis of the cyclized monoterpenoids (Sections 2.5.3 and 2.5.4).

Monocyclic Sesquiterpenes
The six-membered ring in the bisabolane sesquiterpenes from A. annua (Table 30) is formed by cyclization of the C 15 precursor, FPP (378), in much the same way that the p-menthane monoterpenes (Section 2.5.3) arise from the corresponding C 10 precursor. Alternative cyclizations of FPP can lead to the ten-membered germacrane sesquiterpenes, which are described in Table 31, and the elevenmembered humulane sesquiterpenes in Table 33. The bicyclic caryophyllanes, which are formed by a further cyclization of a humulane precursor, have also been included in Table 33. 2.6.2.1. Bisabolanes α-and β-Bisabolene are the most abundant bisabolane sesquiterpenes from A. annua, constituting up to 5% of the essential oil [28,46]. The bisabolane sesquiterpene, α-bisabolol (385), has been found as a minor product from the enzyme amorpha-4,11-diene synthase from A. annua (Section 2.6.3) [136]. Several other bisabolanes, which have not yet been reported as natural products from A. annua, were also obtained from the cyclization of FPP which is catalysed by this enzyme [134] [these included zingiberene; β-sesquiphellandrene; and zingiberenol (see Figure 3, Section 3.1)].
The ten-membered germacrane ring frequently occurs with unsaturation at the 1(10) and 4positions, as is apparent from Table 31. The six-membered ring in the elemane sesquiterpenes (Table  32) is thought to be formed by Cope rearrangement of such 1(10),4-germacradienes. Elemanes may thus be artifacts which are introduced during the extraction and isolation process, especially, perhaps, as a result of the high temperatures often associated with GC-MS analysis (e.g., 397 may be formed thermally from 391; and 396 may be formed from 390 by just such a pericyclic reaction).  [26,37,40,43,46] and can be found at levels of up to 5-10% of the total essential oil [19,22,42]. A cDNA clone for -caryophyllene synthase, the sesquiterpene cyclase which converts FPP (378) to β-caryophyllene (405) has now been isolated from A. annua [139]. The -caryophyllene synthase gene was found to be expressed widely in most plant tissues during early development, and could be induced in mature tissue in response to a fungal elicitor. This was interpreted as evidence that β-caryophyllene (405) might play a role in plant defense [139]. The enzyme amorpha-4,11-diene synthase from A. annua has been observed to produce small amounts of γ-humulene (isomeric with αhumulene (401) in Table 33 -see also Figure 3, Section 3.1), as well as the expected bicyclic product [134].

Cadinanes, Muurolanes and Amorphanes
The amorphane/cadinane group of bicyclic sesquiterpenes is by far the largest class of sesquiterpenes found in A. annua. Unfortunately, the nomenclature and stereochemistry reported for cadinanes and amorphanes in the literature has sometimes become quite confusing. In this review, the relative stereochemistry at the 1-, 6-and 7-positions is used to define four skeletal types, according to guidelines which are set out in the Dictionary of Natural Products [16]. Thus, in the cadinane skeleton, the decalin ring is trans-fused (1α,6,7β), while in the muurolane skeleton (1,6,7β), it is cis-fused. The cadinane and muurolane sesquiterpenes found in A. annua have been grouped together in Table  35. Amorphane sesquiterpenes also incorporate a cis-decalin ring junction, but differ from the muurolane sesquiterpenes in their relative stereochemistry at the 7-position. The very large group of amorphane sesquiterpenes from A. annua is listed in Table 36. (No representative of the bulgarane sesquiterpenes (1α,6,7α), the fourth possible skeleton allowed by this classification scheme, is known from A. annua).  Both the synthesis [144] and NMR properties [145,146] of amorphane and cadinane sesquiterpenes from A. annua have been reviewed. Artemisinic acid (473) [105,147,148], arteannuin B (462) [105,147,149] and artemisinin (495) [105,147,150] are the most abundant representatives of this class of natural products and were amongst the first sesquiterpenes to be reported from A. annua. Several of the amorphane sesquiterpenes in Table 36 have since been implicated as biosynthetic precursors to artemisinin (495), which has been classified as a seco-cadinane in Table 37 (the prefix "seco-" indicates that carbon-carbon bond cleavage has occurred -in this case between C-4 and C-5). The amorphane sesquiterpene, artemisinic acid (473) ( Table 36) was first isolated in 1981 by Prof Tu's group [151]. Its structure was confirmed both by X-ray crystallography [148,152] and by NMR spectroscopy [153]; and subsequently by synthesis [154,155]. Depending on the chemotype of A. annua being studied, artemisinic acid (473) can be present at ten times the concentrations of artemisinin (495). For this reason, much research has been undertaken into the chemical conversion of artemisinic acid (473) to artemisinin (495), which can be achieved with an efficiency of greater than 40% [156,157] (see also Section 4.4 for an application of this conversion to the production of artemisinin). By varying the conditions for the oxidation step, artemisinic acid (473) can be converted to various other sesquiterpenes from A. annua, including: arteannuin B (462) [158,159], deoxyarteannuin B (477) [160] and epi-deoxyarteannuin B (478) [161,162] (note that deoxyarteannuin B (477) [129,161], epi-deoxyarteannuin (478) [68,158,160,163,164] and 6, 7-dehydroartemisinic acid (476) [165] have all also been obtained independently by chemical synthesis).
Dihydroartemisinic acid (480), which is the 11,13-dihyro analogue of artemisinic acid (473) in Table 36, was first isolated as a natural product several years after artemisinic acid [166,167] and it has also been chemically synthesised [168,169]. It is particularly significant that dihydroartemisinic acid hydroperoxide (481), the tertiary allylic hydroperoxide from dihydroartemisinic acid, has also been isolated as a natural product from A. annua [170]. This has led to the suggestion that dihydroartemisinic acid (480) might be converted to its tertiary hydroperoxide (481) in the living plant by a non-enzymatic process as shown in Scheme 5. This hypothesis has apparently been confirmed by recent in vivo and in vitro experiments [155,185] (see Section 3.3) which also suggested that (481) can undergo further non-enzymatic conversion to (495).
Although both artemisinic acid (473) and dihydroartemisinic acid (480) are the most signifcant amorphane sesquiterpenes from A. annua in regard of the biosynthesis of artemisinin (495), several other amorphanes from this species have also been implicated in this process (see Section 3.3). These amorphane sesquiterpenes appear amongst the alphabetical listings in Table 36. Arteannuin A (461) was one of the first sesquiterpenes to be reported from A. annua and it has since been synthesized on two occasions [171,172]. The structure of arteannuin B (462) was determined in 1972 by X-ray crystallography [173] in combination with 1D- [174] and 2D- [175] NMR spectroscopy, and it has also been confirmed by chemical reactions [176]. Several syntheses of arteanuin B are reported [158,171,[177][178][179] and arteannuin C [180] is now thought to be identical with arteannuin B [175]. Syntheses have also been reported of arteannuin E (463) [181] and arteannuin F (464) [182] (which is also referred to as artemisilactone) [171].
annua. All of these compounds have been reported as metabolites of dihydroartemisinic acid (480) in vivo (Scheme 5) [185], and there is evidence from in vitro studies to support the biogenetic proposal that arteannuin H (465) might be produced by spontaneous autoxidation reactions involving a secondary allylic hydroperoxide, which is derived from dihydroartemisinic acid (480) as shown in Scheme 5 [183]. The stereochemistry of the 5-hydroxyl group was wrongly assigned when arteannuins K (468), L (469) and M (470) were first reported as natural products [166]. The correct stereochemistry at the 5-OH group has now been established as α (as drawn) by 2D-NMR studies involving derivitization of synthetic arteannuins K, L and M as their Mosher esters [184]; and by chemical synthesis of both natural (-)-arteannuin M [184] and its ()-enantiomer [186][187][188]. The structure of the natural product arteannuin O (471), which is epimeric with arteannuin M (470) at the 4-position, was confirmed by X-ray crystallography; arteannuin O (471) has also been obtained by a reconstructive synthesis from artemisinin (495) via dihydro-epi-dexoyarteannuin B (485) [184]. All of the five-membered lactones, dihydro-deoxyarteannuin B (484) [129], dihydroarteannuin B (479) [166,189] and dihydro-epi-deoxyarteannuin B (485) [166,189] were also fully characterized by 2D-NMR when first reported as natural products. Dihydro-epi-deoxyarteannuin B (485) has since been obtained by synthesis on several occasions [68,169,[190][191][192] and is probably derived from the allylic hydroperoxide (481) in vivo as shown in Scheme 5. 4,5-Epoxy-6-hydroxy amorphan-12-oic acid (489) can be regarded as the lactone-ring opened analogue of dihydroarteannuin B (479) [193]; and αepoxyartemisinic acid (487) has also been obtained by synthesis [194].   It is interesting to note that nine of the structures reported in Tables 36 and 38 occur in both their 11,13-dihydro and 11,13-dehydro forms. These nine pairs are listed in Table 37. Feeding labelled dihydroartemisinic acid (480) to A. annua resulted in sixteen labelled amorphane and cadinane sesquiterpenes [185], which included all nine of the 11,13-dihydro forms in the "pairs" in Table 37 [193,218]. This has lead to the proposal that the main metabolic route to all three of these metabolites and artemisinin (495) in A. annua involves the spontaneous autoxidation of dihydroartemisinic acid (480) and the subsequent chemical reactions of the derived tertiary allylic hydroperoxide (481), as is shown in Scheme 6.

Scheme 6. The most dominant products from metabolism of dihydroartemisinic acid (480)
in vivo in A. annua plants.
When labeled artemisinic acid (473), the 11,13-dehydro analogue of dihydroartemisinic acid (480), was fed to intact A. annua plants, slightly fewer labeled metabolites were isolated [155]. However, all seven metabolites from this experiment are also known as natural products from A. annua and six of the seven feature in the pairs of metabolites discussed in Table 37. The most abundant metabolite from feeding artemisinic acid (473) was arteannuin B (462), followed by epi-deoxyarteannuin B (478) and the seco-cadinane (494) as shown in Scheme 7 (the remaining four metabolites are: annulide (457), isoannulide (490), deoxyarteannuin B (477) and artemisinic acid methyl ester (474), which were all isolated in trace amounts.
It is intriguing to note that there are exact structural homologies between six of the seven highly oxygenated 11,13-dehydro sesquiterpenes which have been isolated as metabolites of artemisinic acid (473) and a subset of the sixteen 11,13-dihydro metabolites which were obtained in the preceding study with dihydroartemisinic acid (480) (see Table 37). Clearly, the in vivo transformations of artemisinic acid (473) closely parallel those of its 11,13-dihydro analogue, dihydroartemisinic acid (480), as is shown by Schemes 5, 6 and 7. (However, note that in the case of artemisinic acid (473), no allylic hydroperoxide analogous to (481) in Scheme 6 was isolated in vivo as a natural product, and its existence as an intermediate in Scheme 7 must therefore be inferred. Such an allylic hydroperoxide can, however, be produced in the laboratory by chemical reactions with 1 O 2 and it is known to undergo in vitro several of the transformations which are depicted in vivo in Scheme 7).

Scheme 7. The most dominant products from metbolism of artemisinic acid (473) in vivo in A. annua plants.
It seems likely therefore that similar mechanisms are operative in the metabolism of both dihydroartemisinic acid (480) and artemisinic acid (473). These biological transformations have been proposed to involve spontaneous autoxidation of the  4,5 double bond in (473)/(480) and subsequent rearrangements of the resultant allylic hydroperoxides [155,185]. In addition, feeding experiments with both labeled precursors appear to show that artemisinic acid (473) and dihydroartemisinic acid (480) are NOT mutually interconvertible; rather, each is the committed precursor to the two large families of highly oxygenated 11,13-dehydro and 11,13-dihydro sesquiterpene metabolites which are known from this species (see, for example, the nine pairs of compounds in Table 37). This observation fits well with the reported occurrence of two chemical races of A. annua: a low-yielding-artemisinin chemotype, which is rich in artemisinic acid; and a high-yielding-artemisinin chemotype, which also contains significant quantities of dihydroartemisinic acid, as discussed in Section 4.1.
The formyl ester in nor-amorphane 502 is thought to have been derived from the carbon at the 12position of an amorphane precursor by oxidative rearrangements. This group has been lost altogether, presumably as the result of ester hydrolysis, in the bis-nor-amorphane sesquiterpene, norannuic acid 501 [66,199]. Similarly, the formyl group in 1-oxo-2-[3-butanone]-3-methyl-6-[2-propanol formyl ester]-cyclohexane (505) [218] is probably also derived from oxidative rearrangements, this time deriving from the 5-position of an appropriate bicyclic precursor (the related seco-cadinane natural product 504 has also been reported from the in vitro autoxidation of dihydroartemisinic acid in organic solution) [193]. It seems likely that the ethyl substituent in artemisinin G (496) might be derived from the 4-and 15-positions of a conventional amorphane precursor.
The novel carbon skeleton of compound 500 is thought to be derived from the amorphane skeleton by migration of C-8 from C-7 to C-6 (i.e., compound 500 is an 8(76) abeo amorphane), which results in a contraction of the B ring from six atoms to five. Similarly, the unusual carbon skeleton of 503 might also have arisen from an amorphane precursor, in which the A ring has been contracted from six to five atoms, as a result of carbon-carbon bond migration of C-3 from C-4 to C-5 [9].

Guaianes
Guaianes (Table 39) are bicyclic sesquiterpenes which contain fused 5-and 7-membered rings. Guaianes are often found together with eudesmanes (which contain two fused 6-membered rings -see Table 34) and may be derived by an alternative cyclization of the same germacrane precursor which gives rise to eudesmanes. Aromadendrane sesquiterpenes (Table 40) [19,38,51,66], are also representatives of guaianes which have undergone a further cyclization, with the cyclopropyl ring now being formed between positions 1-and 6-. α-Guaiane (506) is the most abundant guaiane sesquiterpene in A. annua [22], whilst spathulenol (515) is the most common aromadendrane [41] (both can reach levels up to 5% of the essential oil).

Diterpenes
The structure of the diterpene phytene-1-ol-2-hydroperoxide (553) [9,260] was wrongly assigned as phytene-1,2-diol (552), when it was first described as a natural product from A. annua [261]. Structural revision was made on the basis of studies of the photo-oxygenation of commercially-available phytol, which produced both phytene-1,2-diol (552) (in racemic form), as well as its 2-hydroperoxy analogue (553) [260]. Both phytol (550) itself and an authentic sample of phytene-1,2-diol (552) were subsequently obtained as natural products from the seeds of A. annua [9]. Natural phytol is expected to have the 7R,11R absolute configuration [16], and natural phytene-1,2-diol (552) was therefore assigned as being a mixture of epimers at the 2-position on the basis of its NMR spectra and on the assumption that the configuration at the 7-and 11-positions of (552) remained fixed, as in phytol. The observance of epimers is most easily explained if both the hydroperoxide (553) and alcohol (552) are products of the spontaneous autoxidation of phytol (550), occurring within the tissues of A. annua plants as is shown in Scheme 8.

Triterpenes and Sterols
The most abundant sterols from A. annua are stigmasterol (570) and sitosterol (568) [96], which are ubiquitous components of plant cell membranes. Squalene synthase (SQS) is the enzyme which catalyses the first committed step in the pathway leading from FPP (378) to triterpenes and phytosterols, such as these ( Table 43). The SQS gene and cDNA have been successfully cloned and sequenced from A. annua on several occasions [263][264][265]. Much of the interest in SQS stems from its position at a key point in terpenoid biosynthesis, in which FPP (378) branches either to triterpenes or sesquiterpenes. Thus, it is possible that suppression of SQS expression [267] could be used to enhance OR OH the biosynthesis of artemisinin (495), which is a product of the alternative sesquiterpene pathway from FPP (378) [268].
A β-amyrin synthase, responsible for cyclization of squalene to the tritepene skeleton, has also been obtained from A. annua. It was possible to produce the triterpene β-amyrin (559) in significant amounts when this enzyme was engineered into Saccharomyces cerevisiae (two other enzymes in the pathway: 3-hydroxy-3-methylglutaryl-CoA reductase and lanosterol synthase were also manipulated in these experiments) [269].

Nitrogen-Containing Natural Products
Only a small number of peptides and other nitrogen-containing natural products are reported from

The Biosynthesis of Artemisinin (Qinghaosu)
Terpene biosynthesis [271][272][273][274][275][276] and its regulation [277,278] in A. annua have been well reviewed, including the central role of amorpha-4,11-diene (451) in the biosynthesis of artemisinin (495) [279]  OH and the likelihood that glandular trichomes are the location wherein artemisinin biosynthesis actually occurs [244,280] (indeed, it has recently been concluded that artemisinin biosynthesis occurs in the two outer apical cells of the glandular secretory trichomes [104] -the glandular trichomes of A. annua are comprised of 10 cells in total). The biosynthesis of artemisinin will be considered in three phases, as depicted in Scheme 9. It should be emphasised that this view of the biosynthetic route to artemisinin is still not universally accepted. Although there is increasing experimental evidence in support of its general correctness, there are a significant number of experimental results (particularly in the "older" literature) which might appear to contradict Scheme 9. Thus, for many years, it was assumed that artemisinic acid (R=CH 2 for 473 in place of R=CH 3 for 480 in Scheme 9) [148,215,251,[281][282][283][284], rather than dihydroartemisinic acid (480; R=CH 3 ), was the late-stage precursor to artemisinin at the juncture between phases 2 and 3.

Scheme 9. Three phases in the biosynthesis of artemisinin (495).
A large number of other cadinane and amorphane sesquiterpenes, including: arteannuin B (462) [248,285]; epi-deoxyarteannuin B (478) [286]; dihydroarteannuin B (479) [249]; dihydro-epideoxyarteannuin B (485) [286]; the seco-cadinane (494) [217]; and artemisitene (497) [211,248,284] have also been suggested as late-stage intermediates at or around this point in the biosynthesis (in addition, α-epoxyartemisinic acid (487) has been stated not to be a biosynthetic intermediate to artemisinin [194]). The experimental evidence for each of these precursors is discussed individually in some detail in Section 3.3 (phase 3 of the biosynthesis of artemisinin). However, as the length of the foregoing list shows, there have been so many different proposals for phase 3 of the biosynthesis, that not all of them can be correct. In fact, none of 462, 478, 479, 485, 494 or 497 feature in the most likely biogenetic route to artemisinin (495) which is discussed first in Section 3.3, although many of these compounds have been implicated as side-products, in reactions which diverge away from this main biosynthetic route to (495) (see Schemes 5, 6 and 7 in Section 2.6.3 for example).

Phase 1 (Isopentenyl Pyrophosphte to Amorpha-4,11-diene)
This first phase in the biosynthesis of artemisinin is the least controversial, and most of the enzymes involved in the conversion of isopentenyl pyrophosphate (IPP) and its isomer dimethylallyl pyrophosphate (DMAPP) to amorpha-4,11-diene (451) have now been isolated and characterized from A. annua. It has been proposed that the IPP used in the biosynthesis of artemisinin comes from both the mevalonate and the mevalonate-independent pathways [287], and it has recently been demonstrated that the cental isoprenoid unit in the FPP (378) precursor to artemisinin is predominantly biosynthesized from the non-mevalonate pathway [102]. 14 C-Labelling studies have confirmed the C 5 terpenoid precursor IPP as a starting point for the biosynthetic pathway to artemisinin [288] and 13 C labeling studies have also been used to characterize the photosynthetic mechanism of A. annua as C3 [289].
The first step in phase 1 involves the conversion of IPP and DMAPP (both C 5 ) to the C 15 intermediate, farnesyl pyrophosphate (FPP; 378) by the enzyme farnesyl diphosphate synthase (FPPS). A cDNA encoding this enzyme, which catalyses the "head-to-tail" chain extension of a DMAPP starter by two molecules of IPP, has now been cloned from A. annua [290]. FPP sits at a branch point in terpenoid metabolism. Further elaboration to triterpenes and plant sterols (Table 43) requires the "head-to-head" coupling of two molecules of FPP, which is catalysed by the enzyme squalene synthase (SQS), as discussed in Section 2.7.2. Conversion of FPP (378) to sesquiterpenes, on the other hand, requires the operation of ionase and cyclase enzymes, such as: (E)--farnesene synthase (Section 2.6.1); germacrene A synthase (Section 2.6.2); -caryophyllene synthase (Secton 2.6.2); and epi-cedrol synthase (Section 2.6.4). The comitted intermediate in the biosynthesis of artemisinin (495) is the bicyclic sesquiterpene-amorpha-4,11-diene (451) [291], which is formed from FPP (378) by the action of the sesquiterpene cyclase, amorpha-4,11-diene synthase (ADS) [138,292]. cDNAs encoding ADS from A. annua have been isolated, sequenced, and expressed [293,294]. The mechanism of this particular enzymatic cyclization of FPP has now been studied in detail using a recombinant amorpha-4,11-diene synthase from A. annua. It is proposed that FPP is first isomerized to nerolidyl diphosphate; ionization of nerolidyl pyrophosphate is followed by C-1,C-6-ring closure to generate a bisabolyl cation; next, this cation undergoes a 1,3-hydride shift [295] permitting a second ring closure between the 1-and 10-positions to generate the amorphane skeleton; and, finally, deprotonation at either C-12 or C-13 affords amorpha-4,11-diene (451) [136], as is shown in Scheme 10. This mechanism is supported by GC-MS analysis of several minor sesquiterpene products which are also produced by ADS. These include the known metabolites (E)-β-farnesene (382), α-bisabolol
Historically, it has often been assumed that artemisinic acid (473) is then the starting point for the third and final phase of the biosynthesis of artemisinin. However, the most recent evidence suggests that dihydroartemisinic acid (480), produced from the saturated "lower" branch in Scheme 11, is, in fact, the true precursor to artemisinin. Kim and Kim have reported that artemisinic acid is not converted to dihydroartemisinic acid in A. annua [297] and more recent biosynthetic studies, employing both labeled artemisinic acid [155] and dihydroartemisinic acid [185], have confirmed that there is no interconversion in either direction between dihydroartemisinic acid (473) and artemisinic acid (480), as is represented by the "crossed" double-headed arrow in Scheme 11. Therefore, if dihydroartemisinic acid (480) is the true precursor to artemisinin (495) in phase 3 of the biosynthesis, then reduction of the exocyclic double bond in amorpha-4,11-diene (451) must be occurring before artemisinic acid (473) in Scheme 11. Interestingly, there is now some evidence for such a route to dihydroartemisinic acid (480), which involves two oxidations at C-12 of amorpha-4,11diene (451), producing artemisinic aldehyde (460) via artemisinic alcohol (458); followed by reduction of the  11,13 double bond in (460) to dihydroartemisinic aldehyde (483); and, finally, oxidation of (483) at C-12 to yield dihydroartemisinic acid (480) [135]. A recombinant DBR2 enzyme has been purified to approximately 90% from E. coli and found to be capable of reducing the  11 (13) double bond in artemisinic aldehyde (460) [but not in artemisinic alcohol (458), artemisinic acid (473), arteannuin B (462) or artemisitene (497)] [298]. This enzyme appears to be a member of the enoate reductase family of enzymes, with similarities to 12-oxophytodienoate reductases [298], and its discovery is potentially very significant with regard to defining the biosynthetic route from (451) to (480) in Scheme 11.

Dihydroartemisinic Acid as a Late-Stage Precursor to Artemisinin
In this review, it is assumed that the final steps in the biosynthetic pathway to artemisinin (495) proceed from dihydroartemisinic acid (480), rather than artemisinic acid (473) (or, indeed, any of the other late stage intermediates: 462, 478, 479, 485, 494 or 497 which have been proposed in the past). Discussions of phase 3 of the biosynthesis therefore commence with dihydroartemisinic acid (480)although the other possibilities are also evaluated in some detail in parts b) -e) of this section. Artemisinin is a seco-cadinane (Table 38), and carbon-carbon cleavage at C-4/C-5 in (495) therefore accompanies formation of the 1,2,4-trioxane ring in this final phase of the biosynthesis. No enzymes have yet been described for any of these putative reactions in phase 3. Indeed, experiments with classical plant peroxidases, a class of enzyme with the potential for involvement in these kinds of reactions, have failed to increase the yield of artemisinin [299,300].
It is also possible that the final transformations of dihydroartemisinic acid to artemisinin might proceed via non-enzymatic processes. In this regard, it is interesting to note that artemisinin biosynthesis has recently been correlated with increased levels of singlet oxygen -although this has been explained in terms of the upregulation of genes involved in artemisinin biosynthesis, rather than the operation of non-enzymatic processes [256]. A non-enzymatic mechanism involving molecular oxygen is particularly attractive in view of the variety of spontaneous autoxidation reactions which have already been suggested for the biogenesis of many other highly oxygenated terpenes from A. The proposal that the final transformations to artemisinin (495) in phase 3 may be non-enzymatic also receives strong support both from in vivo studies [168,185] and from experiments which have been performed in vitro [167,193,218], under conditions that are relevant to the living plant. Together, in vivo and in vitro experiments have suggested a mechanism for the conversion of dihydroartemisinic acid (480) to artemisinin (495) via a spontaneous autoxidation process, involving four steps, as is shown in Scheme 12.

Scheme 12.
A four-step mechanism for the spontaneous autoxidation of dihydroartemisinic acid (480) to artemisinin (495) in A. annua.
The four steps shown in Scheme 12 are: i) photo-sensitized reaction of the  4,5 -double bond in dihydroartemisinic acid (480) with singlet molecular oxygen (via an "ene-type" mechanism); ii) Hock cleavage of the resulting tertiary allylic hydroperoxide 481; iii) oxygenation of the enol product from Hock cleavage; and iv) cyclization of the resulting vicinal hydroperoxyl-aldehyde to the 1,2,4-trioxane system of artemisinin (495). The tertiary allylic hydroperoxide 481, which is produced in step i), has already been shown to be a biosynthetic intermediate, linking dihydroartemisinic acid (480) and artemisinin (495), by in vivo experiments with labeled dihydroartemisinic acid (480) [185,189]. Compound 481 has also been described independently as a natural product from A. annua [167,170]; and quantitative studies, which monitored the decline of dihydroartemisinic acid (480) and the increase of artemisinin (495) during leaf development and senescence have confirmed that an intermediate, such as 481, is probably involved in this conversion [301] [it is interesting to note that the alternative secondary allylic hydroperoxide from the spontaneous autoxidation of dihydroartemisinic acid (480) in step i] has also been proposed as the precursor to another amorphane sesquiterpene endoperoxide from A. annua, arteannuin H (465) [166], as is shown in Scheme 5 in Section 2.6.3).
There is -as yet -no direct evidence from in vivo studies to suggest how further transformation of 481 to the 1,2,4-trioxane ring in artemisinin might occur in A. annua. The evidence for steps ii)-iv) which are proposed in Scheme 12 comes solely from in vitro studies, conducted with the tertiary allylic hydroperoxide 481 [171,236,237,239,254], which can be obtained in good yield from photosensitized oxygenation of 480 [193,218]. These in vitro experiments have shown that 481 can indeed undergo sponataneous conversion to a transient enol as proposed in step ii) of Scheme 11 [204,238,243,302,303], and it has even been possible to fully characterize this unstable intermediate by 2D-NMR at low temperature [193]. This known reaction of allylic hydroperoxides is referred to as a Hock cleavage (or a Criegee rearrangement) and it is in this step that carbon-carbon bond cleavage occurs between C-4 and C-5, thereby producing the seco-cadinane skeleton of artemisinin (495).
The double bond in the enol which is produced by step ii) then reacts rapidly with a second molecule of oxygen in step iii). This reaction most probably proceeds via an "ene-type" mechanism similar to that for the conversion of 480 to 481, although apparently without the requirement for light (i.e. 3 O 2 rather than 1 O 2 is perhaps the active species) [193,236,304,305]. All carbons in the resultant αaldehydo hydroperoxide [238] are now at the correct oxidation level and this intermediate immediately "zips up" to the 1,2,4-trioxane ring of artemisinin (495) in step iv).
It has been speculated that the proximity of the 12-carboxylic acid group to the  4,5 -double bond in 480 and the hydroperoxide functionality in 481 might be the reason why both transformations i) and ii) proceed with such apparent ease in vitro [218]. Finally, it is worth reiterating that both in vivo and in vitro studies have shown that the tertiary allylic hydroperoxide 481 can be transformed into a wide variety of other compounds, in addition to artemisinin (495). Many of these compounds have been obtained previously as natural products from A. annua (see Schemes 5 and 6 in Section 2.6.3). Inspection of Scheme 12 suggests that some of these metabolites, such as dihydroarteannuin B (479) and dihydro-epi-deoxyarteannuin B (485) may be formed by alternative reactions to the Hock cleavage of 481 in step ii), while others such as the seco-cadinane 493 may arise by alternative reactions of the enol from Hock cleavage in step iii). annua using plantlet hydroponic and stem tip feeding methods [306]. Several experiments in the older literature (1990's and earlier) have suggested that artemisinic acid (473) can then be converted in to both arteannuin B (462) and artemisinin (495) [215,281,285]. All these experiments employed radiolabelled forms of artemisinic acid that were presented to cell-free systems, which had been derived by extensive manipulations of A. annua plants. However, the most recent biosynthetic study [155], which employed stable isotope-labeled artemisinic acid [ [185] (see discussion in a) above). In addition, it should be noted that a separate study with a cell-free system has also confirmed that dihydroartemisinic acid (480) is a precursor to artemisinin in A. annua, while artemisinic acid (473) is not [307]. One explanation for these apparently contradictory results with artemisinic acid (473) may lie in the differing experimental approaches which have been adopted. The threshold for detection by NMR of a metabolite which is labeled with a stable isotope is likely to be several orders of magnitude higher than the corresponding threshold for a radioisotpic label. Thus, it is possible that the "older" experiments (1990s and earlier), which detected radiolabel, might -in reality -be identifying only trace quantities of metabolites, which were not derived directly from the labeled precursor. It is possible that partial degradation of a radio-labeled precursor can lead to some radioactivity appearing in a "pool" of small molecules, such as acetyl-CoA (see Scheme 13). These small molecules would then serve as precursors for several biosynthetic pathways, including the terpenoid pathway, and could thereby be incorporated indirectly into many metabolites from such pathways. By contrast, the stable isotopelabeled metabolites which have been detected by NMR [155] represented at least 1% (very often between 5-30%) of the label that had been supplied and therefore provide a picture only of the most significant and direct transformations which have been undergone by the precursor. In addition, it should also be pointed out that the possibilities for introducing artifacts when performing feeding studies with whole plants, which require no external manipulation, is likely to be significantly reduced as compared to cell-free extracts, which must endure many perturbations to the biological system (homogenization, addition of buffers etc..).

Artemisinic Acid (473) as a Late-Stage Precursor to Artemisinin
Although the most recent experiments with artemisinic acid (473) failed to observe any detectable incorporation into artemisinin, a very significant conversion was observed into arteannuin B (462), as well as six other highly oxygenated sesquiterpene natural products, all of which retained the 11,13double bond, as shown in Scheme 7 in Section 2.6.3 [155] [furthermore, there was no evidence for conversion to any 11,13-dihydro metabolite, including artemisinin (495)]. Such transformations can most easily be accounted for by oxidation of the  4,5 -double bond in 473 to a hydroperoxide, analogous to that postulated in Scheme 12 for dihydroartemisinic acid (481). There is indeed ample precedent for the formation of such hydroperoxides from 473 by photooxygenation reactions in vitro [243], which lead both to arteannuin B (462) [251] and 11,13-dehydro analogues of artemisinin [254,308].
In conclusion, although the absence for any detectable transformation of artemisinic acid into artemisinin in the most recent study [155] is at variance with the earlier literature (pre-1990's) [281,284,286] for the biosynthesis of artemisinin, it is consistent with much of the more recent literature (post-1990's), in which there is now a gathering consensus that dihydroartemisinic acid (480), rather than artemisinic acid (473), is the true late-stage precursor to artemisinin (495).

Arteannuin B (462) and Dihydroarteannuin B (479) as Late-Stage Precursors to Artemisinin
As noted above, in the most recent study of the metabolism of artemisinic acid, arteannuin B (462) was obtained as the major metabolite of artemisinic acid (473), without any evidence for the accompanying formation of artemisinin (495) [155]. However, because both arteannuin B (462) and artemisinin (495) were often reported together from "older" biosynthetic studies with artemisinic acid (473), some authors have previously proposed that arteannuin B (462) might be an intermediate in the proposed conversion of artemisinic acid (473) [215,251] to artemisinin (495) [203,248,285]. It is indeed possible to convert arteannuin B into artemisinin by chemical transformations [239] and there have been various suggestions as to mechanisms by which this transformation might also occur in vivo [209,217]. A microbial system which is capable of converting arteannuin B (462) to artemisinin (495) has recently been described [309] and an enzyme with the appropriate activity has also been purified [310,311]. On the other hand, a study which found that epoxyartemisinic acid (487) could be converted to arteannuin B (462) also stated that epoxyartemisinic acid (487) could not be transformed into artemisinin (495) and therefore -by implication -arteannuin B cannot be a precursor to artemisinin [284]. It has also been claimed that the 11,13-dihydro analogue of arteannuin B (462), dihydroarteannuin B (479), can be converted into artemisinin by cell free extracts of A.

The seco-Cadinane (494) and Artemisitene (497) as Late-Stage Precursosr to Artemisinin
The seco-cadinane aldehyde (494) has been hypothesized to be a precursor to artemisinin (495) via its enol tautomer, which undergoes reaction with molecular oxygen * to produce artemisitene (497), the 11,13-dehydro analogue of artemisinin (this reaction was proposed to occur in a similar manner to the transformations which have now been established experimentally for the enol in steps iii) and iv) in Scheme 12 [217,313]). Artemisitiene (497) can then reportedly be converted into artemisinin (495) in vivo [240] [the enol of 494 was proposed to be derived from Grob fragmentation of the vicinal diol 486, which is in turn derived from arteannuin B (462)].

Strategies for the Production of Artemisinin from A. annua and Derived Systems
Although artemisinin can be produced by chemical synthesis [171,[228][229][230][231][232][233][234][235][236][237][238][239], the structural complexity of this target and the large number of steps involved in all published synthetic routes render this approach far too expensive for many in the Third World who are most affected by malaria. Various alternative possibilities for the production of artemisinin from A. annua and derived systems have therefore been extensively investigated, and this topic has been amply reviewed [313][314][315][316][317][318][319][320]. Four strategies are considered in this review, all of which might benefit from a knowledge of the full biosynthetic pathway to artemisinin. These are: 1. Plant breeding programmes; 2. Plant tissue culture; 3. Endophytic fungi; and 4. Genetic engineering. The most successful strategy to date has been plant breeding, resulting in the cultivar "Artemis", which was registered in Switzerland in 1999, and contains up to 1.4% artemisinin (by comparison, the extremely variable yields obtained from wild varieties of A. annua typically range between 0.01-0.5% artemisinin [321]). There is a potential for even higher yields in the future from on-going plant breeding programmes, such as that of the CNAP Artemisia project at the University of York (UK) [195].
Recent attempts to produce artemisinin through fermentation, by genetically engineering several enzymes from A. annua in to a microbial host, would seem to have the greatest promise for the cheap and reliable production of artemisinin. This approach, pioneered by the synthetic biologist Jay Keasling at UC Berkeley (USA), has so far yielded microbially-produced artemisinic acid (473), which is then transformed to artemisinin (495) in a separate chemical process [see Section 3.3b for a discussion of the confusion surrounding the true biosynthetic status of artemisinic acid (473)]. Amyris Biotechnologies and Sanofi-aventis are currently developing a commercial-scale manufacturing process, which is based on this semi-synthetic approach to artemisinin.
The discussions in Sections 4.1-4.4 emphasize how a full understanding of both the phytochemistry of A. annua and the biosynthesis of artemisinin can be helpful for improving the production of this important antimalarial drug by each of these four strategies.

Plant Breeding Programmes
Plant selection and breeding programmes have often sought to combine the properties of artemisinin-rich clones of A. annua (such as the Chinese and Vietnamese varieties), with more vigorous but lower-yielding clones (such as some of the European varieties) in order to achieve a robust hybrid with a high yield of artemisinin (typically 1% or more -corresponding to an agricultural yield of approximately 200 Kg/ha for dry leaves) [322][323][324]. There is increasing evidence that artemisinin production in A. annua is localized in the biseriate glandular trichomes: specialized structures, found predominantly on the surface of leaves and flowers [244,325]. Several authors have reported that the artemisinin content is highest just before flowering [326][327][328] (interestingly, this trend is apparently not followed by other metabolites: the content of artemisia ketone (276) decreases before flowering and then increases afterwards; caryophyllene (405) shows the opposite trend; while levels of monoterpenes, such as cineole (326), bornyl acetate (335) and camphor (341) all remain relatively constant throughout growth [112]) and it has also been noted that the content of artemisinin can be up to ten-fold higher in flowers, as compared with leaves [19,329,330]. These observations may simply reflect an increased density of glandular trichomes in the flowers [301], and hence a correspondingly higher yield of artemisinin. By contrast, artemisinic acid (473) seems to be obtained in maximum yield at the late vegetative stage (however, the concentration of artemisinic acid in A. annua (473) is strongly dependant on chemotype -it can be up to ten-fold higher than that of artemisinin in European and Chinese varieties, although this is reversed in Vietnamese strains).
It has also been observed that increased illumination or sunlight results in a higher yield of artemisinin [331][332][333]. This would be consistent with the hypothesis that the final steps in the biosynthesis of artemisinin proceed in the trichomes by spontaneous autoxidation reactions of dihydroartemisinic acid (480), requiring 1 O 2 , which is produced in the presence of light and a photosensitizer (e.g., chlorophyll -see step i) in Scheme 12). Several researchers have endeavoured to quantify not just the amount of artemisinin (495) [334], but also its presumed biosynthetic precursors, dihydroartemisinic acid (480) [335] and artemisinic acid (473) [335][336][337], when studying the developmental biology of A. annua. One particularly interesting study has concluded that correlated variations in all three natural products prove the existence of chemotypes of A. annua. Thus, plants with a high artemisinin (495) level were found also to have a high dihydroartemisinic acid (480) level, but a relatively low concentration of artemisinic acid (473); while chemotypes with low levels of artemisinin (495) and dihydroartemisinic acid (480) contained a correspondingly high concentration of artemisinic acid (473) [338]. These findings were explained [20] on the assumption that the enzymatic reduction of artemisinic acid (473) to dihydroartemisinic acid (480) might represent a "bottle neck" in the biosynthetic pathway to artemisinin (495). However, given that it seems now to be proven that artemisinic acid (473) is not a precursor to dihydroartemisinic acid (480) (see crossed arrow in Scheme 11) [155,185], the existence of these two chemotypes might instead be explained on the assumption that artemisinic acid (473) is a "dead end" metabolite, leading "away" from artemisinin (495) in phase 2 of the biosynthetic route, as is implied in Scheme 11.
Finally, there have been several recent attempts to create transgenic A. annua plants with elevated levels of artemisinin [339][340][341] by transferring genes for various enzymes in the biosynthetic pathway to artemisinin (e.g. HMGR, FPPS, ADS and CYP71AV1 -see Scheme 13).
Much of the current research activity into the production of artemisinin by plant tissue culture revolves around hairy root cultures of A. annua [349] which have been transformed by Agrobacterium tumefaciens [350,351] (elicitors from endophytic fungi -see the next section -have also been employed in A. annua plant tissue cultures [202,352,353]). Correlations have been observed between peroxidase activity and artemisinin levels in hairy roots [354], although it has also been suggested that a high peroxidase activity in cell cultures may be partly responsible for their very low artemisinin contents observed [355]. (579) R 1 =OMe; R 2 =H (580) R 1 =OH; R 2 =OMe (581) R 1 = (HOCH 2 )CH(CH 2 OH); R 2 =OMe (582) R 1 = H 2 C=C(CHO); R 2 =OMe (583)

Genetic Engineering
Metabolic engineering has attracted increasing attention over recent years as an alternative means for the production of plant-derived drugs. Both artemisnin from A. annua and taxol from Taxus brevifolium are attractive targets for this emerging science of synthetic biology since both drugs are available in only limited quantities from the natural source and are also difficult to synthesize chemically [364]. Research into the production of artemisinin from such genetically modified microorganisms has been well reviewed [365][366][367][368][369].
The goal of synthetic biology is to reprogram a microorganism for the efficient production of a natural product by establishing new metabolic pathways, which lead to the formation of the desired metabolite, whilst simultaneously removing existing metabolic pathways which detract from the formation of such a product. To date, investigators have concentrated on expressing the enzymes from phases 1 and 2 of the biosynthetic route to artemisinin in transgenic organisms (see Scheme 9 in Section 3). These have included: HMG-CoA reductase (HMGR) [370]; farnesyl diphosphate synthase (FPPS) [371][372][373][374]; amorpha-4,11-diene synthase (ADS) [375][376][377]; and the P 450 enzyme which oxidises amorpha-4,11-diene to artemisinic acid (CYP71AV1) [378] (Scheme 13). Both Saccharomyces cerevisiae [379,380] and Escherichia coli [381][382][383] have been used as hosts when attempting to establish a viable recombinant microbial pathway for the biosynthesis of artemisinin.
The approach outlined in Scheme 13 has succeded in producing high levels of amorpha-4,11-diene (451) from E. coli [384] and artemisinic acid (473) from S. cerevisiae in quantities which exceed those from A. annua itself (100 mg -1 ·g·L -1 ) [385,386]. The semi-synthesis of artemisinin (495) from microbially-produced artemisinic acid (473) then requires two further chemical steps: reduction of the exocyclic double bond; and photo-sensitized oxidation of the endocyclic double bond to produce the 1,2,4-trioxane ring in (495) [156,237]. Fortunately, the genetically-engineered yeast transports artemisinic acid (473) outside of the cell, where it is retained on the cell wall, and can easily be released simply by altering the pH. It is then necessary to purify microbially-produced artemisinic acid (473) prior to these chemical transformations. Scheme 13. Genetic engineering of enzymes for the production of artemisinic acid (473) from A. annua in to a microbial host.
In order to achieve the heterologous production of artemisinin (495) by a large-scale fermentation process, it will be necessary to express every step in the biosynthetic pathway to artemisinin in a microbe. Keasling has observed that "the production of drugs via heterologous pathways in microbial hosts is frequently hindered by insufficient knowledge of the native metabolic pathway and its cognate enzymes; often the pathway is unresolved, and the enzymes lack detailed characterization" [387]. These comments are certainly pertinent to the biosynthesis of artemisinin in A. annua, as is evident from the foregoing discussions in Sections 3.2 and 3.3. The author would argue that the prerequisite for the reconstruction of the complete biosynthetic pathway to artemisinin (495) in a transgenic yeast or bacterium is the elucidation of the full details of phase 2 of the biosynthesis in A. annua, in which amorpha-4,11-diene (451) is converted to dihydroartemisinic acid (480); as well as a better understanding of phase 3, in which 480 is then converted to 495 (see Scheme 9). If this can be achieved, then it should -at the very least -be possible to produce dihydroartemisinic acid (480), rather than artemisinic acid (473), by the fermentation of simple sugars (in fact, the production of dihydroartemisinic acid (480) in yeast using the Dbr2 gene has now been described [298] -see also discussion at the end of Section 3.2 -and the use of substrate-promiscuous enzymes as an alternative means for the microbial production of (480) is also under investigation [387]).
If the final stages in the conversion of dihydroartemisinic acid (480) to the 1,2,4-trioxane ring in artemisinin (495) (phase 3 in Section 3.3) do turn out to be enzymatically catalysed, then it may indeed be possible to genetically engineer all the enzymes for the complete biosynthetic pathway to artemisinin (495) into a microbial system. The current evidence, however, points to the final steps in the biosynthesis of artemisinin (495) being non-enzymatic -most probably proceeding by spontaneous autoxidation reactions, which occur in the hydrophobic environment of a glandular trichome. If this is the case, then the simplest approach might be to engineer a microorganism to produce dihydroartemisinic acid (480) which is exported out of the cell, where it can more easily be isolated and purified [in much the same way that genetically-engineered yeast accumulates artemisinic acid (473) on the cell surface]. Inevitably, a chemical oxidation step would still then be required in order to convert microbially-produced dihydroartemisinic acid (480) to artemisinin (495), but this would certainly represent a saving in time and cost by comparison with the two-step chemical transformation (both reduction and oxidation), which is required for microbially-produced artemisinic acid (473).
An alternative more radical approach would be to attempt to mimic within the microbial fermentation medium the spontaneous autoxidation reactions of dihydroartemisinic acid (480) which are believed to occur in the glandular trichomes of A. annua plants. This would require the inclusion of some kind of hydrophobic phase in intimate association with the fermentation medium. Previously, dodecane has been incorporated into the culture medium, in order to provide a separate hydrophobic layer in which volatile amorpha-4,11-diene (451) produced by microbial fermentation can be trapped [382,385]. It would be interesting to investigate whether alternative hydrophobic phases can be found which provide a lipophilic environment similar to that of the glandular trichome, in which spontaneous autoxidation is favoured. (Such a phase would also need to be able to efficiently "trap" dihydroartemisinic acid and to be non-toxic to the fermenting organism). It should be noted that our understanding of how the spontaneous autoxidation of natural products such as 480 occurs in the absence of a photosensitizer [388] (or, indeed, any other additional chemical reagents [218]), leading to highly oxidized products such as 495, is still incomplete. This is a relatively unexplored area which requires fundamental research, but the ultimate prize would be the production of artemisinin, both cheaply and reliably, by a single fermentation process, which incorporates an "in-built" spontaneous autoxidation step.