Reproductive development modulates gene expression and metabolite levels with possible feedback inhibition of artemisinin in Artemisia annua L.

The relationship between the transition to budding and flowering in A. annua and the production of the antimalarial sesquiterpene, artemisinin (AN), the dynamics of artemisinic metabolite changes, AN related transcriptional changes, and plant and trichome developmental changes were measured. Maximum production of AN occurs during full flower stage within floral tissues, but that changes in the leafy bracts and non-bolt leaves as the plant shifts from budding to full flower. Expression levels of early pathway genes known to be involved in isopentenyl diphosphate and farnesyl diphosphate biosynthesis leading to AN were not immediately positively correlated with either AN or its precursors. However, we found that the later AN pathway genes, amorpha-4, 11-diene synthase (ADS) and the P450, CYP71AV1 (CYP), were more highly correlated with AN’s immediate precursor DHAA within all leaf tissues tested. In addition, leaf trichome formation throughout the developmental phases of the plant also appears to be more complex than originally thought. Trichome changes correlated closely with the levels of AN but not its precursors. Differences were observed in trichome densities that are dependant both on developmental stage (vegetative, budding, and flowering) and on position (upper and lower leaf tissue). AN levels declined significantly as plants matured as did ADS and CYP transcripts. Spraying leaves with AN or artemisinic acid (AA) inhibited CYP transcription; AA also inhibited ADS transcription. These data allow us to present a novel model for differential control of AN biosynthesis as it results to developmental stage and trichome maturation and collapse. trichome-specific cDNAs reveal CYP71AV1, a cytochrome P450 with a


Introduction
The medicinal plant, A. annua L., used as a traditional Chinese medicine for more than 2000 years (Acton and Klayman, 1985;Hsu, 2006), produces artemisinin (AN; Figure 1), the sesquiterpene lactone endoperoxide that is the central component of AN Combination Therapy (ACT). ACT is currently the most effective malaria drug and is the World Health Organization's currently recommended treatment (Bhattarai et al., 2007). AN also shows promise as a potential therapeutic for other parasitic and viral diseases as well as for the treatment of certain cancers and the reduction of angiogenesis (Effereth et al., 2002;Romero et al., 2005;Utzinger et al., 2004;Singh et al., 2004). AN has traditionally been isolated from the shoot tissue of field-grown plants, but yield of the drug is characteristically low and does not currently meet worldwide demand. Production through synthetic chemistry although possible, is not economically feasible.
In contrast, certain metabolic engineering approaches appear promising (Arsenault et al., 2008, Ro et al., 2006. Through the recent work of several groups, the biosynthesis of AN is almost completely elucidated ( Figure 1). The isopentenyl phosphate (IPP) in AN originates from both the cytosol and plastid arms of the terpenoid biosynthetic pathway (Towler and Weathers, 2007;Schramek et al., 2010). Like all sesquiterpenes, AN is composed of 15 carbons that likely derive from the condensation of three 5-carbon isoprene molecules to farnesyl diphosphate (FPP), by farnesyl diphophate synthase (FPS). Recently, however, Schramek et al. (2010) fed 13 CO 2 to A. annua plants and obtained label artemisinic intermediates, which suggested that FPP may instead be condensing a geranyl diphosphate of mixed IPP origin with a cytosolic IPP. FPP is converted to amorpha-4, 11-diene through the activity of amorpha diene synthase (ADS;Bouwmeester et al., 6 1999;Picaud, 2005). Amorpha-4,11-diene is subsequently oxidized in three steps to artemisinic acid (AA) through the action of a single enzyme, CYP71AV1 (CYP;Teoh et al., 2006;Ro et al., 2005). Recently a double bond reductase (Dbr2; Zhang et al., 2008) and an aldehyde dehydrogenase (Aldh1; Teoh et al., 2009) were also isolated ( Figure 1). They appear to function in the conversion of artemisinic aldehyde to its dihydro form and then to the dihydro acid (DHAA), respectively; Aldh1 appears to also convert artemisinic aldehyde to AA, an activity also ascribed to CYP (Zhang et al., 2008). AA may then be converted to either arteannuin B (AB) or potentially to dihydroartemisinic acid (DHAA) and then to AN (Sy and Brown, 2002;Brown and Sy, 2004; Figure 1). The mechanisms of these final steps are not currently known but it has been suggested that they may be the result of a non-enzymatic, photo-oxidation reaction (Sy and Brown, 2002). Many of these genes are highly expressed in glandular trichomes from whose cDNA libraries they were cloned (Teoh et al., 2009;Zhang et al., 2008Zhang et al., , 2009).
Artemisinin is produced in 10 cell glandular trichomes located on leaves, floral buds, and flowers (Olsson et al., 2009;Tellez et al., 1999;Ferreira et l., 1995), and sequestered in the epicuticular sac at the apex of the trichome (Olsson et al., 2009). The 2 apical secretory cells lack chlorophyll, and thus may not have chloroplasts, but may have other plastids. The 4 secretory cells beneath the apical cells, however, have chlorophyll and seem to have fully functional chloroplasts. In non-flowering A. annua plants (i.e., during vegetative growth), trichome numbers increased per unit area on the adaxial leaf surface until leaf expansion ceases at which point trichome numbers begin to decline, apparently a result of their collapse (Lommen et al., 2006); to our knowledge no data have been reported for changes in trichome counts on the abaxial leaf surface, for floral bolt leaves (the leafy bracts), or in relation to the epidermal cells. Lommen et al. (2006) further observed that AN levels, which rise with increasing trichome 7 numbers, continue to rise even after trichome populations begin collapsing. They attributed the increase in AN to maturation effects within the trichome. Recently Olsson et al. (2009) used laser dissected trichomes to show that transcripts of genes for AN biosynthetic enzymes post FPP, ADS, CYP, and DBR2, were present in the 2 apical cells, but not in the 4 sub-apical cells of the glandular trichomes. On the other hand, transcripts of DXR, a plastid-localized (Fung et al., 2010) enzyme, were not present in the apical cells, but were found in the remainder of the trichome. FPS transcripts were found throughout the trichome.
Many studies have shown that AN content can vary widely among different cultivars or ecotypes of A. annua (Waallart et al., 2000). It has also been reported that AN content is responsive to the time of harvest, light intensity, and developmental stage (Ferrerira and Janick, 1995). Specifically regarding developmental stage, AN levels are reported to reach their peak either just before flowering, or at full flower (Acton et al., 1985;Woerdenbag, 1993). Flowering, however, may not be necessary for increasing AN content as plants transformed with the flower promoting factor 1 (Fpf1) were induced to flower much earlier, but did not produce significantly increased levels of AN (Wang et al., 2004). Consequently, other factors linked to the reproductive developmental phase change are likely more involved in increasing AN levels.
To provide more insight into the regulation of artemisinin biosynthesis in A. annua, this study measured changes in steady state mRNA levels, trichome populations, and artemisinic metabolite levels in response to the shift in A. annua from the vegetative to the reproductive growth phase (Figure 2). Transcripts of 6 key genes in AN biosynthesis were measured: 3hydroxy-3-methyl-glutaryl-CoA reductase (HMGR) from the cytosolic mevalonic aciddependant IPP pathway, 1-deoxyxylulose 5-phosphate synthase (DXS) and 1-deoxyxylulouse 5phosphate reductoisomerase (DXR) from the plastidic mevalonate-independent IPP pathway, 9 A similar distribution pattern for AB was also observed ( Figure 3). Again, every tissue type from reproductive plants showed higher levels than leaves from vegetative plants. Within the reproductive plants, the pattern of AB production was almost identical to that observed for AN ( Figure 3). The highest levels were once again observed in floral tissues (FLW; 1.94 mg gFW -1 ) followed closely by the leafy bracts on the bolt of budding plants (LBB; 1.38 mg gFW -1 ); floral buds and the leafy bracts on the bolt of plants in full flower (LBF) had concentrations of AB at 0.28 mg gFW -1 and 0.45 mg gFW -1 , respectively. This was only marginally more than the amount found in vegetative leaves (

Transcript Analysis
To determine if transcript levels correlated with measured metabolite levels, tissue samples from vegetative, budding and flowering plants were harvested and separated into their constitutive parts for analysis by real time RT-PCR. Sample transcript levels were standardized to 18s rRNA and compared relative to vegetative plants. Six major genes were measured. They represent the mevalonic (HMGR) and non-mevalonic (DXS and DXR) pathways for IPP production, sesquiterpene specific IPP condensation (FPS), and two artemisinic metabolite specific transcripts (ADS and CYP).
Interestingly, the changes in transcript levels for the studied genes in these tissues varied by more than 3 orders of magnitude ( Figure 4). The genes that showed the least variation in transcript levels from vegetative leaves were the two non-MVA pathway genes, DXS and DXR These both showed peak expression only about 6 fold higher in buds and approximately 4 fold higher in leafy bracts relative to vegetative tissue. Similarly in floral buds, HMGR and FPS showed peak transcript levels at 27 and 10 fold higher than in vegetative tissues, respectively.
The AN specific genes, ADS and CYP, showed dramatic declines in transcription relative to vegetative tissue with transcript levels in some tissues more than 100 fold below that in vegetative tissue in all tissues except buds which showed significant increases.
In looking for patterns of expression, it can be seen that in general, the lower leaves showed minimal changes in gene expression, except for the AN pathway specific genes, which showed dramatic declines in expression in these tissues ( Figure 4). The bract leaves had higher levels of transcript during budding and less at flowering for all genes tested except FPS, which showed no change. A similar pattern was seen for the floral buds; all transcript levels decreased at full flower.

Trichome Levels
AN is produced and stored in the glandular trichomes on A.annua plants, so trichome populations were measured on different leafy tissues taken from plants in each developmental phase. In leaves from vegetative plants the number of trichomes in relation to epidermal cells was highest when the leaves were young and then declined significantly once leaves were fully expanded ( Figure S1). This suggests that trichomes are actually eliminated or collapsed rather In contrast the youngest leaves from both vegetative and reproductive stage plants showed no significant difference in their trichome to epidermal cell ratio (compare ULV to LBB to LBF). However, the number of trichomes mm -2 did increase significantly in the adaxial population of the youngest leaves of budding and flowering plants relative to vegetative plants.
Trichome populations (number mm -2 ) showed close correlation with AN and AB content of each type of leaf analyzed ( Figure 5).

Exogenous AN and AA Affect AN-related Transcript Levels
Lommen et al. (2006) previously suggested that upon maturation, trichomes collapse. We considered that this could in turn release the sequestered AN, and possibly exert a feedback control on AN production. This hypothesis was tested by spraying mature vegetative plants with either AN or AA in 70% ethanol (100 µg ml -1 ). Compared to 70% ethanol controls, AN sprayed plants showed no change in transcripts for HMGR, FPS and ADS ( Figure 6). CYP, however, showed > 10 fold decline. Plants sprayed with AA also showed no significant change in transcripts of HMGR and FPS, however, both ADS and CYP decreased 10 and 5 fold, respectively ( Figure 6).

Discussion
Similar to some earlier accounts (Ferrerira et al., 1995), the highest levels of AN in the A.  2008) showed, using the same 001 strain, that peak levels of AN occurred more or less throughout the entire flowering stage from bolt initiation to full flower; AA dropped substantially as the bolt began to develop, but rose somewhat at full flower.
As the wild type 001 strain transitioned from vegetative growth to full flower, the AN/DHAA ratio was similar to our results. In contrast, the AB/AA ratio remained stable. Together these data underscore the hypothesis that different sub-populations exist with variations in the temporal and tissue distribution of these metabolites and that, quite simply, the debate over when the highest level of AN occurs may be due to cultivar selection, and to when and which tissues are selected for extraction.
AN and AB are believed to be the terminal products of the pathway and generally increase throughout development; this is seen in the lower leaves and in the buds and flowers.
However, the leafy bracts show a distinctly different pattern. In these tissues, AN, AB, AA and DHAA concentrations are higher during budding than flowering. This leads to the obvious question of the fate of these compounds during that transition: are they converted to other secondary metabolites, transported to other tissues, or released to the environment? The decline in AN and AB, as well as the decline in ADS and CYP transcripts may be related to collapsing trichome populations and feedback inhibition by AN as it is released from the epicuticular space.
In A. annua, breeding for increased trichome number results in increased AN levels (Graham et al., 2010), which is consistent with our data. In Arabidopsis the density of trichomes on epidermal tissues undergoes major changes with plant maturation and then again with the transition to reproductive growth (Chien and Sussex, 1996;Telfer et al., 1997). A similar pattern of change in trichome distribution may help explain the patterns of AN concentration observed here. Although trichome counts have been reported for A. annua, data were obtained only from the adaxial surface of leaves, and apparently only from plants in the vegetative phase of growth (Liu et al., 2009;Lommen et al., 2006). Likewise, instead of correlating trichome numbers to their neighboring epidermal cell population, counts have been generally reported as trichomes per unit leaf area and range from 20-80 trichomes mm -2 , the number usually declining with leaf maturity (Lommen et al., 2006) measured trichome populations with regard to leaf position, developmental stage, and both adaxial and abaxial surface changes. Vegetative plants in our study showed similar trichome population dynamics for adaxial surfaces with 11.8 ± 1.25 and 1.2 ± 0.2 (mean±SE) trichomes mm -2 for young vs. mature leaves, respectively. In addition, the numbers on the adaxial and abaxial surfaces were quite similar for a given leaf type.
Trichome maturation from LLB LLF tissues could explain the increases in AN content as trichomes increase in size and density and subsequently allow for the conversion of DHAA or AA to AN or AB, respectively. As glandular trichomes mature, their fragility increases and eventually they burst; younger trichomes with less expanded cuticles are more stable (Lommen et al., 2006). This would help explain the decreases seen in the LBB LBF transition, both in transcript abundance and in AN levels. The trichomes on the leafy bracts on budding plants are quite young, so they are able to accumulate and sequester AN effectively and thus putative feedback to CYP would be somewhat mitigated despite the high AN levels that were measured.
As these same leaves transition to the full flower stage, trichomes begin to disappear, possibly breaking and releasing their contents.
Exogenous AN has the ability to inhibit primary root elongation in seedlings (Arsenault et al., 2010) so it is capable of influencing intracellular processes even when presented extracellularly. The rupture or senescence of trichomes and release of AN or other possibly other downstream metabolites from the epicuticular trichome sack to the surroundings probably accounts for the >100 fold decrease in CYP and ADS transcript levels measured at this stage.   (Table 1). Furthermore, the downstream putative end products, AN and AB, showed a high correlation with trichome density while their precursor molecules, DHAA and AA, were instead better correlated with ADS and CYP transcript abundance (Table 1). This is likely explained by a feedback mechanism acting on CYP and/or ADS expression limiting production of the relatively more polar AA and DHAA.
The more hydrophobic AN and AB, on the other hand, may be largely sequestered to glandular trichomes where their phytotoxic effects are limited and thus, their ability to affect transcription also may be limited. Olsson et al. (2009) suggested this is the case at least for AN; when present in the growth media, levels far below those measured in these tissues (≥ 50 µg ml -1 ) are sufficient to inhibit ≥ 75% of root elongation in A. annua seedlings (Arsenault et al., 2010). Sy andBrown (2002) showed that DHAA may undergo spontaneous auto-oxidation in highly lipophillic environments, which lends support to the notion that the final conversion of DHAA to AN may occur within the subcuticular space of the glandular trichome where hydrophobic AN appears to be sequestered. In the above scenario, bursting of the outermost cuticular layer due to mechanical stress, or age would allow for AN to be released. Overall only ADS and CYP genes showed significant correlations with any of the metabolites or trichomes (Table 1). There were no significant correlations between HMGR, FPS, DXS, and DXR with any of the measured artemisinic metabolites or trichome populations. ADS, and CYP expression correlated well with the precursors, AA and DHAA, throughout development, but not with their respective end products, AB and AN (Table 1). This is consistent with the apparent feedback inhibition that AA exerts on both genes. Trichome number, on the other hand, showed a high correlation with both AN and AB levels, but not with their respective precursors, DHAA and AA (Table 1). These are end products of two branches of the AN pathway and thus, this correlation is consistent with the storage role of the trichome for at least AN and possibly also AB.

Conclusions
This is the first report correlating transcription of six key genes involved in AN biosynthesis, artemisinic metabolite levels, and trichome populations with A. annua growth phase. Significantly, changes in trichome numbers correlated well with AN/AB content of a given leaf type, i.e., lower leaves steadily increased the amount of AN/AB and simultaneously showed a steady increase in trichome number. Precursor concentrations did not show this correlation. Although evidence is presented that supports a negative feedback loop of AN and AA on ADS and CYP genes, other transcript levels do not correlate well with AN production, suggesting that control is not maintained through transcription alone. monitoring mode for relevant molecular ions and identity was verified using the spectra and retention times of authentic external standards.

Glandular Trichome and Epidermal Cell Measurements
Glandular trichomes were counted using the method of Lommen et al., (2006). Briefly, terminal leaflets of fully expanded leaves were fixed to glass slides with double-sided adhesive tape. Trichomes were counted using a binocular microscope at 100 magnification within a 10x10 grid corresponding to 1 mm 2 . Counts were taken twice per leaf at random locations across the surface and averaged for each biological repeat; at least 4 leaves were measured per condition.
Subsequent to trichome counts, epidermal cells were counted using the technique of Cappeledes et al. (1990). The same leaf sections were then coated with a thin layer of clear nail enamel and allowed to dry completely. The epidermal layer was then peeled off the leaves using clear adhesive tape that was then adhered to a new slide and visualized at 400 magnification with a 10x10 grid corresponding to 0.25 mm 2 . Epidermal cells within each grid were counted using the same method as trichomes, with the average of 2 counts per leaf acting as a single replicate using a total of 4 leaves.

Statistical Analyses
Experiments were independently replicated at least three times and significant differences in metabolite concentration and trichome density were evaluated using the Student's t-Test. The q PCR data were analyzed using the Mann-Whitney U test (Yuan et al., 2006).