Plants of A. annua were obtained from Biotech Tricopharming Research SL and cultivated in their experimental fields located in San Cristobal de La Laguna (Tenerife, Spain) during summer 2022, and leaves were harvested in early October before flowering. The leaf size was between 3-4 cm. Even more important, the leaves were at the same phenological state (maturity). Since in A. annua the bottom leaves are the oldest, and the upper leaves are the youngest, only leaves at medium height of equally tall plants were taken. Leaves from different plant species were harvested in San Cristobal de La Laguna in March 2023 to obtain a model of a complex extract mixture. The leaves were dried at 38 °C and powdered.

Artemisinin (98%) was obtained from Sigma-Aldrich (CAS 63968-64-9), as well as pure vanillin (CAS 121-33-5), thymoquinone (CAS 490-91-5), carvone (2244-16-8), verbenone (CAS 1196-01-6), santonin (481-06-1), sitosterol (83-46-5), dehydroabietylamine (CAS-1446-61-3), quinine (CAS 130-95-0), and gramine (CAS 87-52-5). The reagents sodium hydride (CAS 7646-69-7), pyrrole (CAS 109-97-7) and tert-butyldimethylsilyl chloride (CAS 18162-48-6) were also purchased from Sigma-Aldrich. Commercially available solvents were analytical grade and were purchased from Merck, as well as the deuterated solvents.

For UHPLC-MS quantification, LC-MS grade acetonitrile was obtained from Merck (Darmstadt, Germany), and deionized water was obtained from a Milli-Q system A10 (Millipore, Massachusetts, USA). Formic acid (> 95%) was from Honeywell (New Jersey, USA). To prevent carryover contamination in the laboratory, volumetric glassware was cleaned using a glass reagent from Godax Laboratories (Maryland, USA), while other glassware was heated at 550°C for 4 hours.

Different methodologies were tested to obtain higher plant extract amounts. More detail of the solvents used is shown in Table 1 .

Method A: The solvent (12.5 mL) was added to powdered dried plants (0.5 g) and the mixture was sonicated at room temperature for 1 h. The organic layer was filtered, and the filtrate was treated with active carbon to remove chlorophylls, until disappearance of the green colour. The active carbon was then filtered off. Then the filtrate was evaporated under vacuum, providing the dry plant extract residue.

Method B: The solvent (12.5 mL) was added to powdered dried plants (0.5 g) and the mixture was stirred for 18 h at room temperature, using a magnetic stirrer. The organic layer was filtered, and the filtrate was treated with active carbon as in Method A to obtain the dry plant extract residue.

Method C: Acetone (50 mL) was added to the powdered dried plants (2 g) and the mixture was stirred for 18 h at room temperature, using a magnetic stirrer. The organic layer was filtered, and the filtrate was treated with active carbon as in Method A to obtain the dry plant extract residue.

Sodium hydride (60% suspension in mineral oil; 1.20 g, 30.0 mmol, 1.50 equiv.) was added to a solution of commercially available pyrrole (1.40 mL, 20.0 mmol) in dry THF (40 ml), at 0°C and under nitrogen. The mixture was stirred at the same temperature for 15 min, and then allowed to warm to 25°C for 2 h. Then it was cooled back to 0°C and tert-butyldimethylsilyl chloride (3.92 g, 26.0 mmol, 1.30 equiv.) was added. The mixture was allowed to warm to 25°C. and stirred overnight. Then it was poured into water and extracted with EtOAc (3 x 10 mL). The combined organic layers were dried over magnesium sulfate, filtered and concentrated under vacuum. The crude product was purified via column chromatography (n-hexane:AcOEt, 95:5), to give N-(tert-butyldimethylsilyl)pyrrole (2.77 g, 77%). Its ¹H NMR values are in accordance with reported data (Dhanak and Reese, 1986; Caramenti et al., 2017): ¹H NMR (400 MHz, CDCl₃) δ 6.80 (s, 2H), 6.33 (s, 2H), 0.89 (s, 9H), 0.43 (s, 6H). ¹H NMR (600 MHz, CDCl₃) δ 6.80 (s, 2H), 6.32 (s, 2H), 0.89 (s, 9H), 0.43 (s, 6H).

The quality of this compound as NMR reference was determined with an inversion/recovery experiment using a 600 MHz Bruker spectrometer (see Supporting information), to calculate the T₁ of the standard signals. The main parameters of the experiment were: scan number = 8, Recycle Delay d1 = 50 sc, with a gain RG = 114, a pulse width pw = 6.57 µsc, an acquisition time at = 1,36 sc.

The NMR data were processed with the MestreNova application (Windows versions 14.3.1, Mestrelab Research SL), using the Data Analysis>New>Integral Graphics tool, as instructed in the manual (see Mestrelab Research, 2023). Thus, for δ_H = 6.8, G = 0.205161 and T_{1 =} 4,87 sc; for δ_H = 6.33, G = 0.527823 and T_{1 =} 1.89 sc; for δ_H = 0.44, G = 1.87069 and T_{1 =} 0.53 sc. For a good quality, the total time TT should be superior to 5 x T₁. Since TT = at + d1 = 51.36 sc, all the standard signals meet the requirement (Thus, 5 xT1 is 24,35 sc for δ_H = 6.8; then 9,45 sc for δ_H = 6.33, and finally 2,65 sc for δ_H = 0.44).

Method I: A solution of pure artemisinin or plant extract residue in CDCl₃ was placed in an NMR tube. Different amounts of the reference (about 15-20 mg) in 0.5 mL of the deuterated solvent were prepared, and then aliquots of each solution (e.g., 20 µL) were added to the NMR tubes with a Hamilton syringe. The NMR tube was vigorously shaken before running the NMR experiment.

Method II: To a flask with the pure compound (first validations) or the artemisinin-containing plant extract residue (10-15 mg) was added CDCl₃ (0.2 ml), and the solution was placed in a NMR tube using a Pasteur Pipette. The flask was washed with a second addition of CDCl₃ (0.2 ml), which was also introduced in the NMR tube. A certain amount of the reference was weighed in a small vial (2-3 mg, the amount in itself is not important, but it should be weighed accurately), well dissolved in 0.1 mL CDCl₃, and added to the NMR tube. The empty flask was washed with more solvent (0.1 mL), which was also placed in the tube. The homogeneity of the solution was secured with a brief vigorous agitation (a Vortex saves work).

The NMR spectra was recorded in a Bruker 400 MHz NMR spectrometer. In the automatic service mode, the NMR experiments (32 scans, 30° pulses to reduce acquisition time), presented the following parameters: Recycle delay d_{1 =} 1 sc, with a gain RG = 101, an acquisition time at = 4,59 sc, a pulse width pw = 8 µsc, and a spectral width SW = 17,85.

The FID (Free Induction Decay) data were processed with the MestreNova application (version 14.3.1, and 10.0.1-14719 for artemisinin analysis, Mestrelab Research SL), using the NMR>Processing tools, with the manual and/or automatic phase and baseline corrections (for more details, see Mestrelab Research, 2023).

Selected reference and artemisinin signals (δ_H = 6.33, 6.80 and 0.44 for the reference and δ_H = 5.86 for artemisinin) were integrated and their areas were compared.

Artemisinin quantification was made adding 2 ml of acetone (LC-MS grade): ultrapure water (Milli-Q^®) in a 75:25 v:v ratio to 10 mg of plant extract residue. Then, vigorous Vortex agitation was applied for 1 minute to achieve a homogeneous solution. After preparation of the samples, they were diluted as follows (artemisinin stock solutions).

Artemisinin stock solutions of 1000 mg/L were prepared in acetonitrile of LC-MS grade and used for the preparation of daily working mixtures of analytes by dilution. All solutions were stored in the dark at -18°C.

With respect to the apparatus and software, analyses were carried out in a Waters Acquity UPLC^® H-Class, consisting of a sample manager with a flow-through needle and a quaternary solvent from Waters Chromatography. The UHPLC system was hyphenated with an MS Xevo TQD (Waters Chromatography) with an electrospray ionization interface in positive mode. Mass-lynx™ software from Waters Chromatography was used to control the pumps and sample manager, as well as MS parameters and the collection and processing of spectrum data. Separation was performed on an ACQUITY UPLC HSS T3 column (100 mm x 2.1 mm, 1.8 µm) using a pre-column with the same stationary phase (5 mm x 2.1 mm, 1.8 µm), both from Waters Chromatography. The column and pre-column temperature were set at 40°C.

The mobile phase consisted of acetonitrile (ACN; solvent A) and water (solvent B), both containing 0.1% (v/v) of formic acid. The composition was initially set at 50/50 (v/v) A/B and maintained for 4.5 min. Then, it was changed to 100% A in one min, which was held during one min. Finally, the initial conditions were set up in one min and maintained for another min until the system stabilized. The flow rate was 0.5 mL/min, and the injection volume was 5 µL at 10°C.

The MS analysis was performed in multiple reaction monitoring mode using the retention time and two different transitions as identification points. A maximum tolerance of ±20% for the relative intensity of confirmation to quantification ions with respect to the reference ion ratio was established (European Commission, 2002). The source conditions were as follows: capillary voltage 3.2 kV, source temperature 150°C, desolvation temperature 600°C, cone gas (N₂) flow rate 120 L/h, desolvation gas (N₂) flow 1000 L/h, collision gas (Ar) pressure 0.5 bar. The multiple reaction monitoring transitions (MRM), cone voltage, and collision energies applied for artemisinin are shown in Table S1 (Supporting Information).

A statistical analysis was performed to compare qNMR and UHPLC-MS results in R Studio (version 2022.07.1 + 554,^© 2009-2022 RStudio, PBC). Thus, a Wilcoxon signed-rank test was applied in 5 quantified samples.

An ideal NMR reference should present high purity, be inert towards the compounds in the mixture, and have only a few signals that do not overlap with those of the extract and the natural product (Selegato et al., 2019). As illustrated for artemisinin and artemisinin-containing extracts ( Figures 2A, C , see ampliations in the Supporting Information), the reference substance should have signals only above δ_H = 6 and/or below δ_H = 1.0. Moreover, since the reference should be “universal”, no signals in the aromatic area (δ_H > 7) were desired. These requirements leave a narrow shift range (δ_H = 6-7 and δ_H<1.0) for the internal standard. Relatively few natural products have signals at δ_H = 6-7, and only semisynthetic derivatives present them at δ_H<0.6, but even so, to avoid any risk of overlapping in one of these intervals, reference signals should appear in two or more ranges.

We designed several compounds and decided on N-TBS-pyrrole (2, Figure 2B ; Dhanak and Reese, 1986; Caramenti et al., 2017). Although this compound was known, this would be its first use as an NMR standard. N-TBS-pyrrole (2) has two multiple bond signals (resembling singlets) above δ_H = 6.0, and two alkyl signals in the opposite side of the spectrum. One of them corresponds to two methyl groups attached to silicon, and therefore appears at an unusual high field (δ_H = 0.44). Since the reference signals at δ_H = 6-7 and δ_H = 0.44 are so far apart, and appear in regions with few signals, complete reference/sample overlap is extremely unlikely. In addition, the baseline and phase of the spectrum (which should be optimized to achieve accurate signal areas) can be easily corrected by comparing the relative areas of one signal at δ_H = 6-7 (2H) and the signal at δ_H = 0.44 (6H), which should be in a 1:3 ratio.

Moreover, in the unlikely case that key signals of other target natural products appeared in the δ_H = 6-7 interval, and besides, overlapped the sharp standard references, the quantification would still be possible. Thus, the signal at δ_H = 0.4 (6H) could be used to determine the area of the reference singlets at δ_H > 6.0 (2H each). Then, the area of the product signal/s could be determined by subtraction.

As additional advantages, N-TBS-pyrrole (2) is readily prepared in one step from inexpensive commercial reagents ( Scheme 1 ), so its final cost is quite competitive with respect to commercial references. Finally, it is a liquid with a moderate boiling point that ensures accurate handling under standard conditions, but also removal under high vacuum for 1 hour at least, allowing the recovery of valuable natural products from the quantification sample.

As commented before, both the culture conditions of Artemisia annua and the extraction methodology influence artemisinin yields (Ferreira et al., 2005; Jha et al., 2011; Nahar et al., 2020). Therefore, different extraction protocols were compared to find the more suitable for product quantification and for its commercial isolation (Christen and Veuthey, 2001; Lapkin et al., 2006; Castilho et al., 2008; Nahar et al., 2020).

Under some reported protocols the plants were extracted with toluene; however, this aromatic solvent has a high boiling point, and even carrying out the elimination of the solvent under vacuum, a relatively high temperature is needed to remove traces, probably resulting in losses of thermolabile artemisinin (Li et al., 1981; Lin et al., 1986). Other extraction solvents have been reported, such as hexane, diethyl ether, chloroform and acetone (Christen and Veuthey, 2001; Lapkin et al., 2006; Castilho et al., 2008; Nahar et al., 2020). Halogenated solvents are generally avoided since traces of them could be carcinogenic on the long term. Several of these methodologies are compared in Table 1 , using different solvents and extraction methodologies.

N-hexane and acetone provided the cleanest NMR spectra. The best extract amounts were obtained with low-cost and non-toxic acetone, which is an excellent solvent both for polar and non-polar compounds. In addition, acetone has the lowest boiling point, which allows extraction of thermolabile compounds such as artemisinin in good yields (Castilho et al., 2008; Kontogianni et al., 2020).

Once the preferred methodology was established (last entry in Table 1 , sample C1), similar amounts (2 g) of four dried powdered A. annua plants were extracted ( Table 2 ). An average amount of 110 mg of extract was obtained, but as expected, the amount of extract varied with the sample, with C4 providing 40% more extract than C3.

In addition, the amount of artemisinin for the same amount of extract also varied. Five extracts from different plants (including sample C1) were prepared according to Method C, and their artemisinin content was determined by UHPLC-MS/MS ( Table 3 ). The data of the HPLC results will be later compared with those obtained with the NMR methodology.

Once the extraction methodology had been established, a rapid, simple and low-cost quantification of artemisinin in extracts was addressed, using Nuclear Magnetic Resonance. The accuracy of the methodology was first studied by mixing known amounts of pure artemisinin (1; ART) and the reference (2; REF). The relationship between the amounts (in mmols) of artemisinin and the reference can be extrapolated by determining the NMR areas of one proton of artemisinin and one proton of the reference. Since each signal of the reference at δ_H > 6.0 corresponds to two protons, while the signal of artemisinin corresponds to one proton, the area of the reference is halved (alternatively, the artemisinin area is doubled). These relationships are expressed in the Equation 1 below:

Equation 1 can be adapted to calculate the amount of artemisinin in milligrams since the MW of artemisinin and the reference are 282.33 and 181.29 respectively. Thus, [Eq1] can be transformed into [Eq2]:

The example shown in Figure 3 shows that even if the ratio of reference and artemisinin is quite different, the NMR method gives reliable results. Thus, the artemisinin key signal (right one) was given a 1.0 a.u. value, and the reference signals thus obtained a 0.41 a.u. value. The application of Equation 2, for a known amount of reference (0.6 mg), gave the amount of artemisinin (4.56 mg), which matched well with the real one (4.55 mg):

Other examples with different amounts of artemisinin and reference are presented in Table 4 (more detailed in Supporting Information), with very good results in most cases, even with small amounts of artemisinin (2.2, 1.6 and 0.8 mg).

In these first validations shown in Table 4 , two procedures were compared. In the first (Method I), a solution of the reference in 0.5 mL of the deuterated solvent was prepared, and then different aliquots of this solution were used for the experiments. This method (I) has the advantage that initially a relatively large amount of reference could be weighed (15-20 mg), minimizing weighing errors, and then a small aliquot can be extracted, allowing to accurately measure a small amount of reference (such as the 0.6 mg shown in the example of Figure 3 ). However, the method required careful measuring of the final solution volume, which was the sum of the liquid reference substance and the deuterated solvent. In addition, to measure the volume of the aliquot with precision required Hamilton syringes. Despite this, this method is preferable if small amounts of the reference are needed.

The second methodology (Method II) is operationally simpler. As commented, the artemisinin (or the extract) is weighed in a vial, dissolved in the deuterated solvent, and introduced in the NMR tube; the vial is washed once to ensure completed addition. The reference is weighed in another vial and the procedure is repeated. The amount of reference and artemisinin/extract can vary as desired; the only important precaution is to weigh accurately the reference with a precision balance (which under good calibration can have a precision of ± 0.2 mg). The amount of solvent does not need to be precise since the final reference/compound ratio would be similar. Therefore, the method is operationally simple and allows processing many samples in a small time.

Once that the preliminary methods were validated with pure samples, we studied the influence of the extract in the NMR determinations. The signals of the extract could overlap with the key artemisinin and (to a lesser extent) the reference signals, hindering the precise calculation of signal areas. To have as many extract signals as possible, a model vegetal extract (not containing artemisinin) was prepared using plants from different species. The stability of added standard artemisinin in the model extract (and later natural one in the A. annua extract) including the reference was also checked. The relationship between artemisinin and the reference signals was not altered, even after 3 days in the NMR tube.

Then an accurate amount of artemisinin and the reference were mixed with the model extract, and the amount of artemisinin was again calculated using method II, as detailed in the Supporting Information. Figure 4 shows the plant extract ( 4A ) and mixture of the extract, pure artemisinin and reference ( 4B ). To our satisfaction, the reference signals did not overlap with those of the extract. Using a random amount of the reference (3.70 mg) and measuring the areas (1.0 for the artemisinin key signals and 4.65 for the reference) before applying the above formulas, an artemisinin amount of 2.47 mg was calculated, which was close to the real one (2.30 mg).

Finally, the quantification of artemisinin was carried out in real A. annua extracts to compare the results with those obtained with UHPLC-MS. As shown in Figure 5 , in the ¹HNMR spectra of the A. annua extract, the artemisinin signal at δ_H = 5.86 is clearly separated from the others. Moreover, this signal is an easy-to-integrate singlet. Other artemisinin signals can overlap with those in the extract, and therefore, are not so appropriate for quantification. A good correlation between the NMR and UHPLC results was obtained herein and for the other samples EC1-5 described in the Supporting Information, as summarized in Table 5 . The differences are less than 0.5 mg, generally in the range 0.1-0.4 mg. In general, the qNMR values were slightly higher than HPLC values, similarly to the results reported for LC-ELSD by Liu et al. (2010), which could be due to several factors. Unlike HPLC or LC, qNMR does not require separation of the extract components along a column, which minimizes small loses by retention in the column support. The type of HPLC detector can also generate small differences as shown by Liu et al. (2010).

As commented before, overlapping of the NMR and HPLC signals with those of minor impurities can also affect the values. This overlapping could result in slight over or underestimation of the real signal areas. While in HPLC there is overlapping of compound signals (due to separation problems between artemisinin and related terpenes, and even non-related compounds of similar polarity), in NMR there is overlapping of proton signals. To solve this problem, in the first case, the column and/or the elution system is optimized, in the second case, a change of d-solvent usually gives good results (as shown later).

Finally, artemisinin mean value in A. annua plant extract residues ( X ¯ ± SE) by qNMR was 1.13 ± 0.50 mg of ART which was not significantly different to UHPLC-MS result (1.14 ± 0.51 mg of ART).

To prove the utility of the NMR reference, it was used to quantify a library of structurally-diverse natural products of commercial importance, which are shown in Figure 6 . Thus, vanillin (3) is an example of aromatic natural product and is found in the extract of the vanilla bean. It is widely used as flavouring in foods and drugs (Dignum et al., 2001; Walton et al., 2003). Thymoquinone (4) was isolated from Nigella sativa, and displays cardioprotective, antidiabetic and anti-inflammatory properties (Farkhondeh et al., 2017; Leong et al., 2021). The terpenes carvone (5), verbenone (6) and santonin (7) are also bioactive products. Carvone (5) is found in the essential oils of spearmint and caraway seeds, and displays antibacterial and antifungal activity (De Carvalho and Da Fonseca, 2006). The pleasant-smelling verbenone (6) is found in a variety of plants, and since it also acts as an insect pheromone, it is used for pest control (Progar, 2005; Martini et al., 2020). Santonin (7) is another component of the Artemisia genus, and has been used for its antihelmintic (roundworms) properties, and also displays antibacterial and antipyretic activities (Sakipova et al., 2017). Sitosterol (8) is a phytosterol used in food additive E499 (Patel and Thompson, 2006; Food Standards Agency, 2023), while dehydroabietylamine (9, also called leelamine) is a diterpene amine which acts as inhibitor of pyruvate dehydrogenase kinase, and also presents promising antitumoral activity (Aicher et al., 1999; Jung et al., 2022). Leelamine has also served as precursor of several alkaloids. Examples of alkaloids are quinine (10) and gramine (11). Quinine, an alkaloid from Cinchona sp., is an antimalarial drug, a food additive and a fine chemical, as commented in the Introduction. Gramine (11) is an indole alkaloid found in several plant species mostly in Gramineae species, such as Hordeum (barley) and Phalaris grasses, the giant reed Arundo donax, or the silver maple Acer saccharinum (Pachter et al., 1959; Leete and Minich, 1977; Semenov and Granik, 2004; Grün et al., 2005). It probably plays a defensive role, since it is toxic to many insects (Corcuera, 1984). It can be used to used to control undesired algae invasions, specially Coelastrella sp. (Grün et al., 2005; Canton et al., 2019).

To validate the utility of NMR quantification, a known amount of these compounds and the reference was mixed with the model vegetal extract. Using procedure II, the amount of the natural product was calculated and compared with the real amount. Excellent results were obtained, as shown in Table 6 and the NMR spectra described in the Supporting Information. In general, the reference signals did not overlap with the extract or natural product signals.

The quantification of sitosterol (8, Figure 7 ) deserves a detailed comment. When the experiment was run in CDCl₃, the olefinic signal of sitosterol overlapped with those of the extract. Fortunately, a singlet corresponding to one methyl group at δ = 0.68 separated from the bulk, and quantification was carried out using it. However, the separation of the signals improved by replacing the deuterated solvent. Thus, using d-benzene, the olefinic signal separated from those of the extract, and the resolution of the methyl group also improved, affording better accuracy. Therefore, a change in the d-solvent can be useful to overcome problems due to signal overlapping.

Finally, the quantifications of dehydroabietylamine (9) and the alkaloids quinine (10) and gramine (11) in the model plant extract were carried out. Gramine presented some solubility problems at higher doses, and we considered using other solvent such as d-methanol for this case. However, at the small amounts used in this experiment, the quantification in CDCl₃ proceeded well, giving accurate results for these nitrogen compounds. Therefore, the feasibility of using N-TBS-pyrrole as an universal reference for structurally-diverse natural products in plant extracts was demonstrated.