The DESIGNER Approach Helps Decipher the Hypoglycemic Bioactive Principles of Artemisia dracunculus (Russian Tarragon)

Abstract

Complementing classical drug discovery, phytochemicals act on multiple pharmacological targets, especially in botanical extracts, where they form complex bioactive mixtures. The reductionist approach used in bioactivity-guided fractionation to identify single bioactive phytochemicals is inadequate for capturing the full therapeutic potential of the (bio)chemical interactions present in such complex mixtures. This study used a DESIGNER (Deplete and Enrich Select Ingredients to Generate Normalized Extract Resources) approach to selectively remove the known bioactives, 4′-O-methyldavidigenin (1; 4,2′-dihydroxy-4′-methoxydihydrochalcone, syn. DMC-1) and its isomer 4-O-methyldavidigenin (2; syn. DMC-2), from the mixture of phytochemicals in an ethanol extract from Artemisia dracunculus to determine to what degree the more abundant 2 accounts for the established antidiabetic effect of the A. dracunculus extract. Using an otherwise chemically intact “knock-out extract” depleted in 2 and its regioisomer, 1, in vitro and in vivo outcomes confirmed that 2 (and likely 1) acts as major bioactive(s) that enhance(s) insulin signaling in skeletal muscle, but also revealed that 2 does not account for the breadth of detectable biological activity of the extract. This is the first report of generating, at a sufficiently large preparative scale, a “knock-out extract” used as a pharmacological tool for in vitro and in vivo studies to dissect the biological impact of a designated bioactive in a complex phytochemical mixture.

Graphical Abstract

Type 2 diabetes is a progressive disease characterized by hyperglycemia due to insulin resistance in peripheral tissues coupled with impaired insulin production and increased hepatic glucose output. As insulin resistance is a cornerstone in developing type 2 diabetes, clinical approaches to managing this condition focus on improving peripheral tissue responsiveness to the action of insulin. Although skeletal muscle is the primary site of insulin-mediated glucose disposal, therapeutic strategies targeting skeletal muscle to improve insulin sensitivity have had limited success. Small organic molecules have been tested as potentiators of insulin receptor activation or inhibitors of protein tyrosine phosphatase 1B as approaches to increase insulin sensitivity in skeletal muscle. However, concerns about selectivity, the route of administration, and the risk of hypoglycemia have raised questions about the therapeutic effectiveness of these compounds.^1,2 Nonetheless, given the central role of insulin resistance as an early factor in developing type 2 diabetes, improving insulin sensitivity in skeletal muscle remains an important clinical strategy.

Many antidiabetic agents have phytochemicals as lead structures. This includes metformin, a mainstay of type 2 diabetes treatment, which was inspired by a guanidine from Galega officinalis.³ Plants from the genus Artemisia, in particular, A. annua (qinghaosu; the source of artemisinin) and A. dracunculus (Russian tarragon), also have a long history of antidiabetic medicinal use.⁴ The importance of Artemisia species in providing phytochemicals for medicinal purposes is highlighted by the 2015 Nobel Prize for Medicine or Physiology being awarded to Youyou Tu for the discovery and development of artemisinin as an antimalarial drug.⁵ An ethanol extract from A. dracunculus L. (Asteraceae), termed PMI-5011 (for nomenclature, see Table S1, Supporting Information), reduces blood glucose levels and improves insulin levels in murine models of obesity-induced insulin resistance.^6–8 PMI-5011 also enhances insulin signaling in murine and primary human skeletal muscle cells.^7,9–11 Bioactivity-guided fractionation identified five compounds in PMI-5011 that affect glucose metabolism in vitro: 4,5-di-O-caffeoylquinic acid, davidigenin, 6-demethoxycapillarisin, 4′-O-methyldavidigenin (1; 4,2′-dihydroxy-4′-methoxydihydrochalcone, syn. DMC-1) and its isomer 4-O-methyldavidigenin (2; 2′,4′-dihydroxy-4-methoxydihydrochalcone, syn. DMC-2).^12–16 Among them, purified 2 has demonstrated pronounced hypoglycemic effects by reducing blood glucose levels within 6 h of administration in obese, insulin-resistant mice.⁸ Hence, 2 is the most promising designated bioactive marker of PMI-5011.

However, this designation remains accompanied by important questions that arise from the concept of bioactivity-guided fractionation, which by definition deconstructs the chemistry of an investigated extract. Being reductionist in nature, bioactivity-guided fractionation typically fails to capture the therapeutic potential related to compound interactions in complex mixtures, such as PMI-5011. Backed by ethnobotanical knowledge, evidence is growing that herbal extracts exert their health effects holistically. This means that botanicals work as intact, complex chemical mixtures, in which multiple compounds most likely act on multiple pharmacological targets: a concept termed “polypharmacology”.^17,18 Therefore, advancing the development of PMI-5011 and its bioactives as rational therapeutic agents in treating insulin resistance requires an approach that decomposes A. dracunculus polypharmacology through selective modification rather than single-target decomposition of the crude extract.

To this end, the present study used the DESIGNER^19,20 approach to Deplete and Enrich Select Ingredients to Generate Normalized Extract Resources as a means of determining more precisely which compounds and their combinations within the extract have the desired biological activity. The new polypharmacological tool consisted of a knock-out extract (KOE) of PMI-5011, in which the previously identified bioactives, 2 and 1, were removed selectively. The biological properties of this KOE were then compared with those of unaltered PMI-5011, a mixture of primarily 2 and its regioisomer 1 termed methyldavidigenin knocked-out fraction (KOF), and synthetic 2. The ultimate objective of the present study was to determine the pharmacological contribution of the two methyldavidigenin derivatives to the overall effects of PMI-5011 on glucose homeostasis, both in vitro and in vivo. Located at the chemistry–biology interface, this study employed orthogonal phytochemical methodology and took a highly integrated in vitro/in vivo pharmacological approach to rationalize the therapeutic potential of a complex botanical extract in triggering diabetes resilience mechanisms.

RESULTS AND DISCUSSION

Designated Bioactive Principles in Russian Tarragon

A standardized ethanol extract of A. dracunculus, PMI-5011, reduced blood glucose levels and improved insulin sensitivity in mouse models of genetic and diet-induced obesity.^7,21 Enhanced insulin signaling in murine and primary human skeletal muscle cells in vitro by PMI-5011 suggested that this extract affects glucose metabolism in skeletal muscle.^9,11,14 These findings led to the hypothesis that bioactive small molecules from PMI-5011 can enter skeletal muscle cells. A kinetic study on myotube cells with PMI-5011 revealed that the most promising bioactive chalcone, 2 (for structure see Figure 2), binds to the cell surface within 1 h, is internalized between 1 and 6 h, but is undetectable after 16 h (Figure 1). Along with the quantification of 2 by LC-MS during the cell culture experiment, it should be noted that other designated bioactive compounds from PMI-5011, namely, sakuranetin, 6-demethoxycapillarisin, and davidigenin, were also detected in the samples from treated cells. These compounds were not quantified due to the nonavailability of reference standards for calibration. However, their presence approximately followed the changes observed in the concentrations of 2, with relatively large amounts in the cell culture medium and much smaller amounts bound to cell surfaces and present in the cell lysate of treated cells (data not shown). Like 2, these compounds were not detected in the pellets of treated cells.

Figure 2.

UHPLC-UV (A) and ¹H NMR spectroscopic (B) analyses of the PMI-5011 crude extract (in black) vs the KOE (in red) and the KOF (in green). PMI-5011 and KOE were analyzed at the same concentration of 10 mg/mL for UHPLC-UV and 34 mg/mL for ¹H NMR spectroscopy. The outcomes demonstrate that 2, marked with *, and its regioisomer 1, were removed effectively from the original PMI-5011 crude extract. The KOE preserved the metabolomic profile of the original PMI-5011, except for the removed 2 and 1, as well as the two volatile compounds identified as 3 and 4 (marked with Δ), for which concentrations were reduced due to inevitable solvent evaporation.

Figure 1.

4-O-Methyldavidigenin (2) is present in skeletal muscle in vitro. C2C12 myotubes were incubated with vehicle control (DMSO) or PMI-5011 (10 μg/mL) for 1, 6, or 16 h as indicated. At each time point, (A) the medium and washes were collected at neutral pH; (B) this was followed by collection of surface-bound compounds at acidic pH and cell lysates that were separated into soluble (Lysate) and insoluble fractions (Pellet) for analysis of content by UPLC/MS. (A, B) The mass of 2 recovered in each fraction is shown, and (C) the mass balance indicates total 2 was recovered as the parent compound (*, no 6-demethoxycapillarisin; ND, not detected).

Preparation of Knockout Extracts as Pharmacological Tools

Two countercurrent separation steps (CCS; see also Figures S2 and S3, Supporting Information) transformed the original PMI-5011 into a knockout extract, in which 1 and 2 were removed selectively (as per quantitative UHPLC-UV; Table 1). As 2 forms a critical pair of separation with its regioisomer, 1, both compounds were co-depleted from the original extract. The two CCS steps also led to the production of a methyldavidigenin knocked-out fraction containing the two isomeric compounds 1 and 2. In order to evaluate the effect of the CCS process on the phytochemical integrity of PMI-5011 and its KOE derivative, a reconstituted extract was prepared by performing the full CCS followed by recombining all eluents including the KOE and the KOF. The reconstituted extract (RE) served as both a phytochemical and biological control. NMR and UHPLC metabolomic profiles documented the preservation of the chemical integrity of the KOE compared to PMI-5011 and evaluated the effectiveness of 1 and 2 depletion (Figures 2 and S5 and S6, Supporting Information).

Table 1.

Concentrations of 1 and 2 in the Tested Extracts and Fraction^a

	mean ± stdv (% [w/w])
sample	1 (4′-O-methyldavidigenin)	2 (4-O-methyldavidigenin)
PMI-5011 CE	0.81 ± 0.00	1.76 ± 0.00
PMI-5011 RE	0.81 ± 0.00	1.77 ± 0.00
KOE	∼0.04 ± 0.01 below LOQ	below LOD
KOF	21.86 ± 0.09	68.73 ± 0.49

^aAll extracts were analyzed at 10 mg/mL, and the KOF was analyzed at 0.108 mg/mL. The reconstituted PMI-5011 extract (RE) contained the same amount of 1 and 2 as the original PMI-5011 crude extract (CE). LOD: limit of detection for 1 and 2 (2.06 ± 0.26 μg/mL). LOQ: limit of quantification (6.23 ± 0.80 μg/mL).

Comparison of all UHPLC-UV and ¹H NMR fingerprints (Figure 2) showed that the KOE retained all phytochemical features of the original PMI-5011, with the exception of two volatile isomeric phytoconstituents, namely, elemicin (3) and iso-elemicin (4). Both compounds, identified herein by NMR and MS analyses (Figures 2C, S5, S6, and S11, Supporting Information), are reported to be abundant methoxy-allylbenzenes in Russian tarragon and, thus, PMI-5011.⁴ Their concentrations were decreased in both KOE and PMI-5011 RE (Table S8.3, Supporting Information), as a result of the inevitable solvent evaporation for the production of dried extract necessary for conducting bioassays. In a similar manner, the multiple steps of solvent evaporation and sample reconstitution during any bioactivity-guided fractionation process would lead to a significant decrease of both 3 and 4 concentrations in the produced fractions, thereby hampering their detection and isolation as potential bioactive compounds. In fact, data pertaining to 3/4 bioactivities are scarce, despite their relative abundance in aromatic plants.^22,23

Complemented by the phytochemical methods, the in vivo bioactivity profiles obtained originally with unaltered PMI-5011 were compared to those obtained with KOE, the KOF, synthetic 2, and the RE. Tested at the same concentration, a KOE that will demonstrate a loss of activity compared to the original PMI-5011, together with a recovery of the activity with the KOF, will, therefore, confirm that 2 and possibly its regioisomer, 1, are responsible for the measured effect.

Insulin Signaling

Insulin signaling in palmitate-treated insulin-resistant C2C12 myotubes was assayed after treatment with the PMI-5011, the KOE, the KOF, and the synthetic 2 (Figure 3), each at 10 μg/mL. PMI-5011-mediated upregulation of insulin-dependent AKT phosphorylation at threonine 308 and serine 273 in the presence of palmitate was lost with the KOE, but restored with the KOF and purified 2. Depletion of 1 and 2 was also associated with reduced steady-state levels of IRβ, IRS-1, mTOR, and FOXO3a, along with further decreased AS160 levels compared to PMI-5011. In addition, tyrosine phosphorylation of IRS-1 and serine phosphorylation of mTOR were diminished with the KOE.

Figure 3.

Bioactive marker 4-O-methyldavidigenin (2) is required for PMI-5011-mediated enhanced AKT phosphorylation. C2C12 or L6 myotubes were incubated with palmitate overnight to induce insulin resistance in the absence (−) or presence (+) of PMI-5011, the extract without 2 (KOE), the removed fraction (KOF), or synthetic 2 at 10 μg/mL each. The cells were harvested after insulin was present for 10 min. (A, B) Western blot analysis of total and phosphorylated forms of proteins in the insulin signaling pathway. (C) Glycogen content assayed as ng released glucose/μg protein and (D) Western blot analysis of GSK3 activation.

In contrast, KOE treatment still supported PMI-5011-mediated inhibition of IRS-1 serine phosphorylation as well as increased AS160 phosphorylation, just as observed with unaltered PMI-5011 (Figure 3A/B). This indicated that compounds other than 1 and 2 are indeed responsible for these activities. While increased glycogen content resulting from PMI-5011 treatment was absent with the KOE, KOF or synthetic 2 restored this activity (Figure 3C). This correlated with the increased GSK3 phosphorylation observed with the KOE compared to synthetic 2 (Figure 3C).

Glucose Homeostasis

The next major question was whether the effect of PMI-5011 on glucose homeostasis in vivo could correlate with the levels and activities of 2. To address this, C57BL/6J male mice were fed a very high fat diet (60% kcal fat) over 16 weeks to induce obesity, prior to testing the ability of PMI-5011 (500 mg/kg), KOE (500 mg/kg), KOF (100 mg/kg), and purified 2 at 100 or 300 mg/kg body weight to alter blood glucose levels (Figure 4A). Metformin (300 mg/kg) was used as a positive control. The mice were assigned randomly to each treatment group, and there was no difference in body weight among them (Figure 4B). As shown in Figure 5A/B, a 6 h treatment with metformin, PMI-5011, PMI-5011-RE, and synthetic 2 significantly reduced blood glucose levels in the obese male mice. Blood glucose levels were not changed by the KOE and were also unchanged by the KOF, possibly due to the variability and low number of animals (N) related to the dose and limited sample.

Figure 4.

Experimental design to test the metabolic effect of 4-O-methyldavidigenin (2) in vivo. (A) C57BL/6J male mice were fed a 60% kcal fat diet for 16 weeks prior to a 6 h treatment with labrasol (vehicle), metformin, PMI-5011, reconstituted PMI-5011 (RE = KOE + KOF), KOE, KOF, and synthetic 2. (B) Body weights at 20 weeks of age for mice randomized to each treatment group. Statistical significance is reported as mean ± standard deviation, compared to (vehicle) control.

Figure 5.

Putative bioactive marker 4-O-methyldavidigenin (2) lowers blood glucose levels in a murine model of obesity-induced insulin resistance. (A) Absolute (mg/dL) and (B) percent change in blood glucose levels from baseline for each treatment group 6 h after treatment. (C) Insulin levels (ng/mL) and (D) C-peptide levels (ng/mL) for each treatment group. Statistical significance is reported as mean ± standard deviation, compared to (vehicle) control; ns, not significant; *p < 0.05; **p < 0.01; ***p < 0.001.

Unexpectedly, treatment with KOE and the higher dose of the synthetic 2 increased insulin levels significantly (Figure 5C). This was accompanied by increased C-peptide levels resulting from treatment with synthetic 2, but not with the KOE (Figure 5D). Interestingly, despite the inevitable decrease in 3 and 4 concentrations, the in vivo activities of the RE were statistically equivalent to those of the original PMI-5011. This indicated that these two major volatile constituents do not contribute to the effects of PMI-5011 on blood glucose and insulin levels.

To determine if the change in blood glucose levels corresponded to an effect on insulin responsiveness in skeletal muscle, Western blot analysis was used to interrogate insulin signaling in mixed gastrocnemius. As found in the C2C12 myotubes (Figure 3A), steady-state levels of IRβ decreased with KOE treatment, but were maintained by KOF (Figure 6A). IRβ tyrosine¹¹⁵⁰ phosphorylation was enhanced by the KOF, but not by synthetic 2, independent of concentration, suggesting the involvement of 1 in this activity. AKT levels were not altered substantially by the treatments, but there was an upregulation of AKT serine²⁷³ phosphorylation with PMI-5011 that was absent with KOE treatment, restored with the KOF, and more pronounced with pure 2. Compared to the vehicle control, AKT threonine³⁰⁸ phosphorylation was not enhanced by PMI-5011, KOE, or KOF (Figure 6A). Unlike their in vitro effect (Figure 3A), KOF and pure 2 in vivo at 100 mg/kg increased levels of the PI3K-p85 subunit in skeletal muscle within 6 h (Figure 6A). To determine if the changes in insulin signaling were specific to skeletal muscle, PI-3K p85, AKT, threonine³⁰⁸, and serine²⁷³ phosphorylation in the liver and gonadal adipose tissue was assayed. In the liver, a clear pattern of changes in the PI-3K p85 or AKT protein levels did not emerge. Regulation of AKT serine²⁷³ phosphorylation by KOE and KOF was less robust in the liver, but unlike skeletal muscle, AKT threonine³⁰⁸ phosphorylation increased with PMI-5011 treatment (Figure 6B). AKT serine²⁷³ phosphorylation was not regulated by PMI-5011, KOE, or KOF in adipose tissue. AKT threonine³⁰⁸ phosphorylation was substantially reduced by PMI-5011, KOE, and KOF, but not by treatment with the synthetic 2. An upregulation in total AKT also occurred in adipose tissue when treating with the KOF or 2 at 300 mg/kg body weight.

Figure 6.

Tissue-specific effect of PMI-5011, KOE, and 2 on insulin signaling. (A) Compound 2 enhances AKT phosphorylation and maintains IRβ total protein levels in mixed gastrocnemius skeletal muscle. (B) AKT phosphorylation at serine²⁷³, but not threonine,³⁰⁸ is regulated by 2 in liver, but not (C) in the epididymal adipose tissue of the obese, insulin-resistant male mice.

Elucidation of Complex Mixtures with Complex Modes of Action

A number of preclinical studies have reported that a complex mixture of compounds found in the PMI-5011 ethanol extract of A. dracunculus improves glucose metabolism and insulin responsiveness in obesity-induced insulin resistance.^6,7,9–11 However, the lack of a more detailed understanding of how the complex mixture of phytochemicals in PMI-5011 interacts to improve insulin sensitivity has hampered the development of PMI-5011 as a treatment of insulin resistance in type 2 diabetes. This study used the DESIGNER approach to elucidate the impact of individual compounds present in PMI-5011 on skeletal muscle insulin signaling pathways critical to maintaining insulin sensitivity. Notably, this is the first study to demonstrate that a DESIGNER-generated KOE can be produced in sufficient quantities to dissect in vivo the physiological impact of an individual compound present in a complex mixture of phytochemicals.

While the classical bioactivity-guided fractionation approach is reductionistic and aimed at pairing bioactivity with typically one or a few isolated compounds, the DESIGNER approach enables new correlations by connecting the bioactivity of target compounds with their presence and absence and concentration in complex mixtures, such as in PMI-5011, but also KOF, as exemplified in the following. Both KOF and synthetic 2 had to be dosed ~40 and ~60 times higher, respectively, than the corresponding concentration of 2 in the crude extract, in order to obtain a biological effect equivalent to the effect measured with the crude PMI-5011. In fact, when testing the crude PMI-5011 at 10 μg/mL, the effective concentration of 2 is only 0.17 μg/mL (Table 1), which is 60-fold less than purified 2 tested at 10 μg/mL. This indicates that the effects of 2 on glucose homeostasis are nonspecific. However, when comparing the activities between the crude PMI-5011 and KOE, the specific loss of activities could still be pinpointed to the depletion of the methyldavidigenins. Collectively, these results suggest that methyldavidigenins are much more active within the PMI-5011 matrix than as pure compounds. Another plausible hypothesis is that other PMI-5011 constituents promote the bioavailability or PK behavior of methyldavidigenins so they can exert their biological effect at lower concentrations. Similarly, testing KOF at 100 mg/kg led to an effective concentration of 68.7 mg/kg of 2 and 21.8 mg/kg of 1, when considering the methyldavidigenin concentrations in this fraction (Table 1). Hence, the isomeric methyldavidigenin mixture displayed identical bioactivities compared to purified 2 at 100 mg/kg, thereby suggesting a potential combinatorial effect of these regioisomers on the measured bioactivities. Further studies are needed to uncover the effect of other PMI-5011 constituents on 2 and 1 bioavailability and to determine whether their effects on glucose homeostasis are synergistic, additive, or related by other mechanisms.

Evaluation of the polypharmacological properties of KOE confirmed that the selectively removed 2 and 1 contribute to certain, but certainly not all, biological activities of the whole extract. The two-step CCS-based production scheme removed 2 and 1 and preserved equivalence to the original PMI-5011 by maintaining the proportionality between the multiple phytoconstituents (metabolomic profile), except for the labile/volatile methoxy-allylbenzenes, 3 and 4. However, results obtained in this study suggested that these constituents do not contribute to the effects of PMI-5011 on blood glucose and insulin levels.

The outcomes demonstrate that 2 (and likely 1) is the major bioactive principle in PMI-5011 that enhances AKT activation in vitro (Figure 3) as well as in vivo in skeletal muscle with insulin resistance (Figure 6). In vivo AKT activation occurred in skeletal muscle when acutely exposed to PMI-5011 or KOF, but was not observed with KOE and did not occur in adipose tissue or the liver. Interestingly, the insulin levels were increased by both KOE and pure 2 at 300 mg/kg within 6 h. The higher dose of 2 at 300 mg/kg corresponding to increased C-peptide levels were indicative of stimulated insulin secretion. Hence, 2 at high concentration had an additional impact on pancreatic function in obesity-induced insulin resistance that was absent at lower levels of this compound and/or when 2 acts in combination with other phytochemicals present in the unaltered PMI-5011 extract. Conversely, C-peptide levels were unchanged by KOE, raising the possibility that the increased insulin levels with KOE reflect a modulating impact of methyldavidigenins on hepatic and extrahepatic insulin clearance.^24–26

All of these findings correlate with changes in the effect of PMI-5011 on AKT phosphorylation in skeletal muscle. This primarily occurs in skeletal muscle within 6 h, but may not be entirely specific to the skeletal muscle, as increased AKT phosphorylation at threonine³⁰⁸ occurred in the liver as well. Although adipose tissue is important in determining glucose homeostasis and insulin sensitivity in obesity, the glucose-lowering effect of PMI-5011 and KOF did not involve enhanced insulin signaling in the visceral adipose tissue.

The DESIGNER extract approach also revealed that methyldavidigenins possess previously unrecognized mechanisms of action within PMI-5011. The steady-state levels of several proteins are differentially regulated by 2 and KOE, suggesting the balance between protein synthesis and degradation is influenced by 2. Earlier studies showed that PMI-5011 inhibits proteasome activity.²⁷ Reduced levels of IRS-1, AS160, or mTOR upon KOE treatment indicate that methyldavidigenins are responsible for the effect of PMI-5011 on protein steady-state levels. However, treatment by both PMI-5011 and KOE affected the phosphorylation of IRS-1 almost equally in the C2C12 skeletal muscle cells. This suggests that other phytochemicals in PMI-5011 are responsible for this anti-inflammatory effect, which is in line with the polypharmacological paradigm of botanicals. Considering the relatively high amount of 3 and 4 in PMI-5011 (see Table S8, Supporting Information), future directions could explore its possible biological contribution to any polypharmacological effects of PMI-5011.

Concluding Remarks

In summary, this study exemplifies the potential of the DESIGNER approach, with the production of a KOE as a polypharmacological tool for the decryption of biological activities of compounds within a complex mixture such as PMI-5011. The KOE helped decipher the contribution of 2 and its regioisomer 1 on glucose homeostasis. The outcomes confirm that 2 is a major bioactive principle that enhances AKT activation in vitro and in vivo in skeletal muscle in the presence of insulin resistance. In vivo AKT activation occurs in skeletal muscle when acutely exposed to PMI-5011 or the KOF, but is not observed with the KOE and does not occur in adipose tissue or the liver. The DESIGNER approach also enhanced our understanding of the role of the designated bioactives in the overall in vivo effects of PMI-5011. The demonstrated multiple biological effects of 2 and its regioisomer, 1, on glucose homeostasis support the development of standardized PMI-5011 for the complementary treatment of type 2 diabetes.

EXPERIMENTAL SECTION

Plant Material

Artemisia dracunculus was grown in a Rutgers University greenhouse facility in New Brunswick, NJ (40°28′41.9″ N 74°26′15.7″ W) using commercially available seed obtained from Sheffield’s Seed Company, Locke, NY, USA. Voucher specimens are retained at Rutgers University Chrysler Herbarium under the guidance of a taxonomy specialist.²⁸

Extraction and Isolation

The preparation of the PMI-5011 botanical extract from A. dracunculus and detailed information about its quality control and biochemical characterization have been reported previously.^{6,8,12,13,29,30} Earlier studies led to the isolation of five compounds with in vitro activity.¹² Compound 2 was also synthesized as previously described.⁸

Preparation of the Knockout Extract and the Knocked-Out Fraction

The depletion of 2 from PMI-5011 was achieved by two successive steps of countercurrent separation (CCS) using orthogonal biphasic solvent systems (SSs) composed of hexanes, ethyl acetate, ethanol, and water (5:4:4:4, v/v) as SS1 and hexanes, ethyl acetate, methanol, and water (6:4:6:4, v/v) as SS2 (see Figures S2 and S3, Supporting Information). The partition coefficients (K; a value expressing the ratio of the concentration of the compound between the upper and lower phase of the two-phase solvent system) of 2 were 4.53 in SS1 and 1.32 in SS2. The first CCS step used a hydrostatic Spot-Prep SCPC-250-B (Armen Instrument SAS, Gilson, Inc.), equipped with a 250 mL column, which was operated in reversed-phase mode using SS1 (25 mL/min at 2800 rpm). Under these conditions, the stationary phase fraction (Sf) at equilibrium was 0.52. For injection, 1.306 g of PMI-5011 was dissolved in 7.5 mL of each upper and lower phase. A fraction enriched with 2 eluted from K = 3.30 (550 mL) to 5.80 (875 mL). The rest of PMI-5011 was recovered fully by recombining the liquid stationary and all mobile phase that eluted before and after 2.

To generate the KOE, the fraction enriched in 2 (42 mg) underwent a second CCS step using a hydrodynamic high-speed countercurrent chromatography apparatus (TBE-300B, Tauto Biotech Co., Shanghai, People’s Republic of China), equipped with a 300 mL column, operated in reversed-phase mode with SS2 (1.5 mL/min at 800 rpm; Sf = 0.85). Even in this orthogonal system, 2, eluting at 300 (K = 1.0, 3 h 20 min) to 425 mL (K = 1.5) postinjection, showed coelution with its regioisomer, 1. Collectively, the two-step process yielded 1.122 g of KOE and 25.3 mg of the KOF that contained all of 2 and 1 (Figure 2). While the mass recovery of the initial PMI-5011 was 87.8% (see Table S4, Supporting Information), the observed loss was not compound specific and only due to the sample filtration and handling during injection (Figures S3–S5, Supporting Information, provide the analytical details). As a control, a second batch of PMI-5011 (1.274 g) underwent the same two CCS steps, but all fractions including those containing compounds 1 and 2 were recombined.

Comparative UHPLC-UV Analyses and Quantitation of 2

UHPLC analyses utilized a Shimadzu UHPLC equipped with a Kinetex XB-C18 column (2.1 × 50 mm, 1.7 μm, part# 00B-4498-AN, Phenomenex) and a diode array detector (DAD, Shimadzu SPD-M20-A). The elution gradient (0.8 mL/min) was composed of (A) water and (B) acetonitrile, both with 0.1% formic acid, as follows: 5% B from 0 to 2 min, going to 25% B at 12 min and during an additional 3 min, then reaching 70% B by 21 min and maintaining for an additional 3 min. Extracts were dissolved at 10 mg/mL, and KOF at 0.10 mg/mL, in 70% HPLC-grade acetonitrile. Under these conditions, the retention times (t_R) of 1 and 2 were 17.20 and 17.50 min, respectively. The elemicins 3 and 4 coeluted at 14.50 min. For quantification (see Figures S6, S7, and S12, Supporting Information), 2 and 3/4 calibration curves were analyzed at 277 and 225 nm, respectively.

Comparative ¹H NMR Fingerprints

For NMR analysis, exactly weighed (±0.01 mg) samples of the original crude extract and KOE (8–10 mg) as well as the KOF (ca. 1.5 mg) were dissolved in 300.0 μL of CD₃OD (99.8% D; Cambridge Isotope Laboratories Inc., Andover, MA, USA). Using calibrated glass pipets, 200.0 μL of each solution was transferred into 3 mm NMR tubes (S-3-HT-7, Norell Inc., Landisville, NJ, USA). The 1D ¹H NMR spectra were acquired at 25 °C under quantitative conditions (qHNMR) using a 90° excitation pulse experiment on a JEOL ECZ 400 MHz instrument equipped with a 5 mm multinuclear Royal probe. The probe frequency was automatically tuned and impedance matched before each acquisition. Off-line data analyses were performed using Mnova software (v.11.0.3, MestreLab Research S.L., A Coruña, Spain). The Supporting Information (Figure S9) provides the purity determination of synthetic 2.

Animal Experiments

Male C57BL/6J mice were obtained from Jackson Laboratories (Bar Harbor, ME, USA). All animal experiments were approved by the Rutgers University Animal Care and Use Committee (Protocol #04–023). The animals were housed singly with a 12 h light–dark cycle at 24 °C. At 4 weeks of age, the mice were placed on a very high fat diet (60 kcal% fat, D12492, Research Diets) for 16 weeks. Body weight was measured weekly. The mice were acclimated to gavage over a 2-week period prior to administering the experimental materials, vehicle control (Labrasol), or metformin via gavage. Blood glucose levels were assayed at baseline and each hour after administering the extracts. The mice were euthanized and tissue was collected for analysis 6 h after gavage.

Cell Culture

Murine C2C12 (ATCC; #CRL-1771) or rat L6 myoblasts (ATCC, #CRL-1458) were cultured in Dulbecco’s modified Eagle’s medium (DMEM), high glucose (25 mM) with 10% fetal bovine serum, 2 mM glutamine, and antibiotics (100 units/mL penicillin G and 100 μg/mL streptomycin), in a humidified chamber at 37 °C and 5% CO₂. To obtain fully differentiated myotubes, the medium was exchanged for DMEM, high glucose with 2% horse serum, glutamine, and antibiotics, when the myoblasts reached confluence. The medium was replaced every 48 h, and the cells were maintained in this medium until fully differentiated, when the medium was exchanged for DMEM, low glucose (5 mM) with 2% horse serum. The myotubes were fully formed by the fourth day postinduction. The experiments analyzing the impact of each experimental material on insulin signaling in the presence of palmitate-induced insulin resistance were performed as described previously.³¹ To determine uptake of bioactives into the myotubes, the cells were treated with PMI-5011 (10 μg/mL) for 0, 1, or 16 h prior to medium collection. At each time point, the cells were thoroughly rinsed with D-Hanks, pH 7.4, at 4 °C, followed by incubation with D-Hanks buffer, pH 4.0, at 4 °C for 30 min with gentle rocking to remove surface-bound compounds. Cells were then collected in denaturing buffer containing 50 mM Tris-Cl, pH 7.4, with 150 mM NaCl, 1 mM EDTA, 1% Igepal, 0.5% Na-deoxycholate, 0.1% sodium dodecyl sulfate (SDS), and protease inhibitors (1 μM phenylmethylsulfonyl fluoride (PMSF), 10 μg/mL aprotinin, 1 μg/mL pepstatin, 5 μg/mL leupeptin). The harvested cells were sonicated on ice, and the supernatant and pellet were collected by centrifugation at 13400g for 10 min at 4 °C.

UPLC/MS Analysis of Cell Preparations

Each sample of cell culture medium (or portion thereof) or cell culture preparation was partitioned in triplicate with an equal volume of ethyl acetate, dried by rotary and subsequent SpeedVac evaporation, and resuspended in 250 μL of 90% ethanol. The pellet samples were extracted directly with 90% ethanol. UPLC/MS analysis utilized a Dionex UltiMate 3000 RSLC ultra-high-pressure liquid chromatography system, a photodiode array detector, DAD-3000RS, and a Q Exactive Plus Orbitrap high-resolution high-mass-accuracy mass spectrometer (MS). Mass detection with an electrospray interface was full MS scan with low energy collision-induced dissociation from m/z 100 to 1000 in either positive or negative ionization mode. Substances were separated on a Phenomenex Kinetex C₈ reversed-phase column (100 × 2 mm, 2.6 μm/100 Å particles). The mobile phase consisted of two components: solvent A (0.5% ACS grade acetic acid in LCMS grade water, pH 3–3.5) and solvent B (100% LC-MS grade acetonitrile). The mobilephase gradient consisted of 95% A and 5% B to 5% A and 95% B over 30 min at 0.20 mL/min. Quantification used a calibration curve of 2 to convert the sample starting volumes into total amounts of 2, except for the pellets, where nothing was detected. Identification of the other bioactive compounds utilized an in-house library of spectral data and database searches (reaxys.com, Elsevier RELX Intellectual Properties SA; SciFinder, American Chemical Society).

Protein Expression Analysis

Skeletal muscle, liver, and gonadal adipose tissue lysates were prepared from powdered tissue by homogenization in 25 mM HEPES, pH 7.4, 1% Igepal CA630, 137 mM NaCl, 1 mM PMSF, 10 μg/mL aprotinin, 1 μg/mL pepstatin, and 5 μg/mL leupeptin and centrifugation at 14000g for 10 min at 4 °C. Protein concentrations were determined using BCA assays (Thermo Fisher Scientific, Rockford, IL, USA). The tissue supernatants (50 μg) were resolved by SDS-PAGE and subjected to immunoblotting using chemiluminescence detection (Thermo Fisher Scientific) and quantified as described.³² Nitrocellulose membranes were incubated with antibodies (Table S13, Supporting Information) for 1–2 h at RT or overnight at 4 °C.

Blood and Tissue Chemistry

Fasting glucose levels were measured in whole blood using a Breeze2 glucometer (Bayer, Leverkusen, Germany). Murine fasting insulin and C-peptide levels were assayed via ELISA (Crystal Chem, Downers Grove, IL, USA). Glycogen content was assayed as directed by the manufacturer (Sigma).

Statistical Analysis

Distributions of body weight, blood glucose, insulin, and C-peptide levels were assessed using the D’Agostino–Pearson omnibus normality test. Statistical significance was determined using an unpaired two-tailed t test of the mean and standard deviation. GraphPad Prism 5 software was used for statistical analysis.

Safety Statement

No unexpected or unusually high safety hazards were encountered.