As mentioned in the preceding section, two extractions solvent systems, methanol (100%) and ethanol-water (1:1, v/v), were compared for extraction efficiency of phenolic acids and flavonoids in T. formosanum. After HPLC analysis, 1:1 (v/v) ethanol-water solvent mixture was found superior to 100% methanol, as a larger number of phenolic acid and flavonoid peaks appeared in the chromatogram (Figure 1A,B). Thus, the solvent system of ethanol-water (1:1, v/v) was adopted for subsequent experiments.

Usually, the mobile phases employed for separation of phenolic acids and/or flavonoids include water in combination with methanol or acetonitrile in gradient mode. Additionally, a modifier such as formic acid, acetic acid, trifluoroacetic acid or phosphoric acid was added to avoid peak tailing [17,20]. Therefore, in our experiment, various mobile phase combinations containing methanol or acetonitrile with 0.1% formic acid as modifier were compared. After various trial studies, the most appropriate solvent system was composed of 0.1% formic acid in water (A) and acetonitrile (B) with the following gradient elution: 92% A and 8% B initially, maintained for 10 min, raised to 14% B in 24 min, 23% B in 35 min, 24% B in 44 min, maintained for 12 min, increased to 32% B in 60 min, 37% B in 66 min and returned to 8% B in 68 min. A total of 29 phenolic acids and flavonoids in T. formosanum were separated within 68 min with flow rate at 1.0 mL/min and detection at 280 nm (Figure 1B). Table 1 shows retention time (t_R), retention factor (k), separation factor (α) and peak purity of various phenolic acids and flavonoids in T. formosanum. The k value for all the peaks ranged from 2.06 to 21.19, revealing a proper solvent strength was controlled. Likewise, the α values for all the peaks were greater than 1 (1.01–1.84), indicating that a good selectivity of mobile phase to sample components was achieved. With the exception of caffeoyl hexoside (peak 2) and luteolin hexoside hexoside (peak 17), the purities of all the other phenolic acids and flavonoids were higher than 90% (Table 1).

Various elution volumes involving 5–20 mL deionized water for phenolic acids and 2–5 mL methanol (100%) for flavonoids were compared for elution efficiency by using a Strata-C18-E cartridge and subjected to HPLC analysis. The most appropriate elution volume for complete elution of phenolic acids and flavonoids was 20 mL of deionized water and 5 mL of methanol, respectively. Figure 2 shows the HPLC chromatograms of phenolic acids in phenolic fraction (A) and flavonoids in flavonoid fraction (B). It was found after identification that most phenolic acids were eluted in the phenolic acid fraction with the exception of caffeoyl-d-glucose (peak 3), quinic acid derivative (peak 7) and 3,5-di-O-caffeoylquinic acid (peak 25).

Peaks 5, 8, 9, 25 and 26 were positively identified as cis-caftaric acid, chlorogenic acid, caffeic acid, 3,5-di-caffeoylquinic acid and chicoric acid respectively, while peak 23 as luteolin-7-O-glucoside by comparing the retention time, absorption and mass spectra with that of commercial standards. Figure 3 shows the chemical structures of phenolic acids and flavonoids positively identified by comparison with commercial standards. In addition, a post-column derivatization technique was used for further identification of flavonoids. The absorption characteristics of a flavonoid compound may vary depending on the type of complex it forms with the shift reagents used in the post-column derivatization method. Accordingly, the presence of flavonoid compounds was identified in our study based on the shift in absorption maximum caused by addition of sodium acetate or aluminum chloride without or with acid [20,22]. Sodium acetate, being a weak base, can complex with 7-OH in flavones or flavonols causing a bathochromic shift (red shift) of about 5–20 nm in band II. Nevertheless, an ambiguity exists in the identification of 3′- or 4′-OH of flavones or flavonols due to formation of a shoulder and irregularity in the bathochromic shift of band I [23]. On the contrary, aluminum chloride can react with 3- or 5-OH of flavones and flavonols to form acid-stable complex. Yet, an acid-liable complex is formed on reaction with two hydroxyl groups in the ortho position. Moreover, both 3,5-dihydroxyflavones and 5-deoxy-3-hydroxyflavones can result in a red shift after reaction with aluminum chloride in the presence of acid [22]. The presence of two ortho hydroxyl groups can be identified on the basis of a bathochromic shift of 30–40 nm in band I after addition of aluminum chloride plus acid, whereas the three ortho hydroxyl groups in the B ring can be detected by a red shift of only 20 nm. Similarly, a red shift of about 60, 35–55 and 50–60 nm in band I can indicate the flavonoids to be 3-hydroxy flavones, 5-hydroxy flavones and 3,5-dihydroxy flavones, respectively [22].

Figure 4 shows the overlaid UV spectra of each flavonoid peak before and after post-column addition of sodium acetate reagent. A bathochromic shift of 18, 20, 16, 14, 6 and 6 nm in band I was shown for peaks 11, 12, 14, 17, 23 and 24, respectively, implying the presence of hydroxy group at 3′C or 4′C, while a red shift of 14 nm (peak 24) in band II indicated the hydroxyl group at 7C position. For the other peaks, a sugar moiety may be attached to 7C. The shift in UV spectral characteristics of flavonoid peaks after post-column addition of aluminum chloride without neutralization is shown in Figure 5. The peaks 1, 12, 14, 16, 17, 20, 21, 22, 23 and 24 registered a bathochromic shift of 36, 36, 34, 38, 38, 36, 32, 34, 40 and 34 nm in band I, revealing the occurrence of hydroxyl group at C5 position. Similarly, a red shift of 44, 44, 44, 44, 52, 40, 42, 42, 50 and 44 nm observed after addition of aluminum chloride plus acid for peaks 11, 12, 14, 16, 17, 20, 21, 22, 23 and 24, respectively, revealed the existence of two hydroxyl groups at 3′C and 4′C positions (Figure 6) [24]. The afore-discussed results are all summarized in Table 2 showing the possible identities of 10 flavonoids.

For identification of phenolics and flavonoids in T. formosanum without commercial standards, both single quadrupole mass spectrometer with ESI mode and triple quadrupole tandem mass spectrometer (MS/MS) with multiple-reaction monitoring mode (MRM) were employed. It has been well established that the MRM mode can provide high specificity and sensitivity, and thereby the structures of unknown phenolic acids or flavonoids may be assessed based on the m/z of both precursor ion and fragment ion obtained through mass transition [25]. Moreover, the absorption spectral data were compared with that reported in the literature. Tables 3 and 4 show the mass spectral data and UV absorption maximum, respectively, for both phenolic acids and flavonoids separated from T. formosanum. Peak 1 showed an absorption maximum at 278 nm and an [M − H]⁻ ion at m/z 315 as well as a fragment ion at m/z 153 due to the loss of a hexose moiety, all revealing the compound to be protocatechuic acid hexoside [26,27]. Mass spectra of peak 2 displayed a parent ion at m/z 341 and two fragment ions with one at m/z 179 for caffeic acid through the loss of a hexose moiety, and the other at m/z 135 for decarboxylated caffeic acid after elimination of both hexose and CO₂, conclusively indicating the compound to be caffeoyl hexoside. Peak 3 was identified as caffeoyl-d-glucose based on comparison of [M − H]⁻ of parent ion (m/z 339) with that reported by Shakya et al. [28]. Likewise, based on comparison of absorption spectra (232, 280, 310 nm) and [M − H]⁻ value (m/z 137) with that reported by Atoui et al. [29] and Arranz et al. [30], peak 4 was characterized as p-hydroxybenzoic acid. Peak 6 was assigned to be caffeoyl hexoside as a similar MS pattern as that of peak 2 was obtained. By comparison of absorption data with that reported by Schütz et al. [21], peak 7 was tentatively identified as a derivative of quinic acid and the absence of MS data may be probably caused by interference with impurities. Peak 10 was tentatively assigned to a derivative of hydroxycinnamic acid based on the [M − H]⁻ value at m/z 421 and comparison of absorption spectrum (236, 314 nm) with that reported by Sakakibara et al. [31]. Peak 11 was characterized to be quercetin-pentoside-hexoside as a parent ion at m/z 595 and fragment ions at m/z 433 and 301 were obtained due to a sequential loss of hexose [M − H − hexose] and pentose [M − H − hexose − pentose] moieties [21,32]. The mass pattern of peak 13 showed an [M − H]⁻ ion at m/z 441 and fragment ions at m/z 279 and 235, representing elimination of caffeic acid moiety and caffeic acid plus CO₂ molecule, respectively, and therefore the compound was assigned to be a derivative of quinic acid [21]. The compound quercetin-pentoside-hexoside was assigned for peak 14 as the MS pattern was similar to that of peak 11. For peak 15, MS pattern depicted a parent ion at m/z 491, which upon fragmentation produced a daughter ion at m/z 329, indicating loss of a caffeic acid moiety. Additionally, a product ion at m/z 293 was mainly due to loss of dihydroxyphenyllactic acid (198 Da) plus a H₂O molecule, as indicated by Schütz et al. [21] and thus the peak was tentatively characterized as caffeoyl-dihydroxyphenyllactoyltartaric acid. This identification was based on a similar compound reported to be present in commercial dandelion root and herb juices and the identification was based on the fragment ions at m/z 329, 311 (caffeoyltartaric acid), 149 (tartaric acid), 135 (decarboxylated caffeic acid) obtained in MS² experiment and a product ion at m/z 293 in MS³ experiment after removal of a H₂O molecule [21]. It was further reported that the exclusion of H₂O molecule is energetically favored amid the elongated conjugated π-system in the structure [21]. However, in our study, the product ions corresponding to tartaric acid and its derivatives are missing which may be accounted for by the difference in the collision energy involved during fragmentation.

Peak 16 was characterized as quercetin-7-O-hexoside-3-O-(malonyl) hexoside based on the parent ion at m/z 711 and fragments ions at m/z 667 and 301, representing an elimination of a CO₂ molecule and a quercetin aglycone, respectively [33]. The presence of two hexose moieties linked to a luteolin aglycone in the structure for peak 17 was identified by the parent ion at m/z 609 and fragment ions at m/z 324 and 285 [21]. An [M − H]⁻ ion at m/z 635 and λ_max at 216, 246, 328 nm obtained for peak 18 was tentatively identified as a derivative of caffeic acid by comparison with the literature [21]. Peak 19 with a parent ion at m/z 473 was characterized to be a derivative of chicoric acid (dicaffeoyltartaric acid), as daughter ions were generated at m/z 311 and 293 owing to the loss of caffeic acid moiety and caffeic acid plus H₂O molecule, respectively [21]. MS profiling of peak 20 yielded a parent ion at m/z 593 and daughter ions at m/z 308 and 285 corresponding to rutinose moiety and luteolin aglycone, revealing the compound to be luteolin-7-O-rutinoside. Likewise, quercetin pentoside was assigned for peak 21 based on the [M − H]⁻ value at m/z 433 and the product ion at m/z 301 representing quercetin aglycone and at m/z 132 the pentose moiety. Fragmentation of the parent ion at m/z 463 for peak 22 produced two daughter ions with one at m/z 301 corresponding to quercetin aglycone, and the other at m/z 162 to glucose moiety, revealing the compound to be quercetin hexoside [34]. Similarly, quercetin pentoside was assigned for peak 24 based on the [M − H]⁻ value at m/z 433, and a product ion at m/z 299 for quercetin aglycone due to loss of pentose. Peak 27 with a parent ion at m/z 473 was characterized as a derivative of chicoric acid based on a similar MS pattern as that of peak 19. Upon fragmentation of [M − H]⁻ ion at m/z 357 for peak 28, a fragment ion at m/z 179 was produced which corresponded to caffeic acid, indicating the compound to be a derivative of caffeic acid [35,36]. Peak 29 was identified to be caffeoyl-hexose-deoxyhexoside based on the parent ion at m/z 487 and the fragment ion obtained at m/z 308 by expulsion of caffeic acid moiety and at m/z 179 due to loss of deoxyhexose plus hexose moieties [21,37].

The intra-day and inter-day variability of various phenolic acids and flavonoids in T. formosanum are shown in Table 5, with the RSD values ranging 1.0%–6.8% and 2.0%–7.7% for phenolic acids and 3.7–7.4% and 1.5–8.1% for flavonoids, respectively. This outcome clearly indicated that a high reproducibility can be achieved by employing the developed analytical method. The recovery of various phenolic acid and flavonoid standards shown in Table 6 revealed that a high recovery (>90%) was obtained for most standards, including cis-caftaric acid, chlorogenic acid, caffeic acid, 3,5-di-caffeoylquinic acid, quercetin and chicoric acid, while for luteolin-7-O-glucoside, the recovery was 84.9%. The recovery data was found to correlate well with the values reported by Schütz et al. [21], Niranjan et al. [38] and Inbaraj et al. [17].

The limit of detection (LOD) and limit of quantitation (LOQ) for cis-caftaric acid, chlorogenic acid, caffeic acid, 3,5-di-caffeoylquinic acid, chicoric acid, luteolin-7-O-glucoside and quercetin were 50.7 and 152.2, 47.9 and 143.6, 18.8 and 56.3, 88.4 and 265.1, 95.3 and 285.9, 28.3 and 85.0 and 99.6 and 298.7 ng/mL, respectively. Both LOD and LOQ in our study were substantially lower than those reported in literature [38,39]. The LOD and LOQ of chicoric acid in dried press juice of purple coneflower were shown to be 1100 and 3500 ng/mL, respectively [39]. In another study, Niranjan et al. [38] determined polyphenols in Artemisia pallens L. and the LOD and LOQ for chlorogenic acid, caffeic acid and quercetin were reported to be 1220 and 2260, 980 and 1460 and 1300 and 2400 ng/mL, respectively, Thus, it is apparent that the LC-MS method developed in our study is more sensitive than the other reported methods.

For quantitation, the following linear regression equations obtained from the calibration curve of each standard was used: y = 0.7597x − 0.2688 (R² = 0.9968) for chicoric acid, y = 0.7385x − 0.0102 (R² = 0.9999) for caffeic acid, y = 0.4953x − 0.1277 (R² = 0.9968) for chlorogenic acid, y = 0.3854x − 0.176 (R² = 0.9957) for 3,5-di-caffeoylquinic acid, y = 0.5237x − 0.0569 (R² = 0.9959) for cis-caftaric acid, y = 0.3868x + 0.0105 (R² = 0.9987) for luteolin-7-O-glucoside and y = 0.3356x − 0.018 (R² = 0.9979) for quercetin. The contents of various flavonoids and phenolic acids are shown in Table 4, with the former ranging 9.9–325.8 μg/g and the latter 14.1–10870.4 μg/g (Table 4). Of the various phenolic acids and flavonoids quantified, chicoric acid and quercetin-pentoside-hexoside dominated contributing 10,870.4 and 325.8 μg/g to the total content, respectively.

Taraxacum formosanum Kitam was procured from a local drug store in Taipei, Taiwan. After freeze-drying, T. formosanum was powdered and packed in several plastic bags. Then, they were sealed under vacuum and stored at −20 °C until further use. Phenolic acid standards, including cis-caftaric acid and chicoric acid, were purchased from Chromadex Co. (Santa Ana, CA, USA) and Extrasynthese Co. (Genay, France), respectively. 3,5-Di-caffeoylquinic acid was from Alexis Co. (San Diego, CA, USA), while chlorogenic acid, caffeic acid and syringic acid were from Sigma (St. Louis, MO, USA). Flavonoid standards luteolin-7-O-glucoside was from Extrasynthese Co., and both quercetin and (±) naringenin were from Sigma (St. Louis, MO, USA).

The HPLC-grade solvents methanol and acetonitrile were from Merck Co. (Darmstadt, Germany), while both ethanol and ethyl acetate were from Lab-Scan Co. (Gliwice, Poland). Formic acid was from Riedel-de Haën Co. (Seelze, Germany). Ethanol (95%) was from Taiwan Tobacco and Wine Bureau (Tainan, Taiwan). Deionized water was obtained by a Milli-Q water purification system from Millipore Co. (Bedford, MA, USA). The Strata C18-E cartridge (500 mg/3 mL, 55 μm, 70 Å) was from Phenomenex Co. (Torrance, CA, USA). Sodium hydroxide was from Riedel-de Haën Co. Aluminum chloride (AlCl₃·6H₂O) and sodium acetate were from Nacalai Tesque Co. (Kyoto, Japan). Glass filter paper GA-55 (diameter 110 mm, particle size 0.6 μm) was from Advantec Co. (Saijyo, Ehime, Japan). Polypropenyl cotton was from Applied Separation Co. (Allentown, PA, USA). Two HPLC columns, Gemini C18 (250 × 4.6 mm I.D., particle size 5 μm) was from Phenomenex Co. and Vydac 201TP54 C18 (250 × 4.6 mm I.D., particle size 5 μm) was from Vydac Co. (Hesperia, CA, USA).

The HPLC-MS system (Agilent Technologies 1100 series, Palo Alto, CA, USA) is composed of a G1379A on-line degasser, a G1316A column temperature controller, a G1311A quaternary pump, a G1312A binary pump, a G1315B photodiode-array detector, a G1314A UV/Vis detector, and a 6130 single quadrupole MS detector with multi-mode ion source (ESI and APCI). Also, an API 3200 triple quadrupole LC-MS/MS from Applied Biosystem Co. (Carlsbad, CA, USA) was used. The Sorvall RC5C high-speed centrifuge was from Du Pont Co. (Wilmington, DL, USA), the 2210R-DTH model sonicator from Branson Co. (Danbury, CT, USA), the FD 24 freeze-dryer 24 from Gin-Min Co. (Taipei, Taiwan) and the rotary evaporator (N-1) from Eyela Co. (Tokyo, Japan).

A method based on Kao et al. [20] was modified and used for extraction of flavonoids and phenolic acids from T. formosanum. A 0.25 g of T. formosanum powder sample was mixed with 15 mL of methanol (100%) or ethanol-water (50:50, v/v) to compare the extraction efficiency. The mixture was then shaken at 60 °C for 3 h, centrifuged at 10,000 rpm for 30 min and the supernatant was evaporated to dryness under vacuum using a rotary evaporator. The residue was dissolved in 10 mL of deionized water. Next, the crude extract was subjected to purification in a SPE cartridge based on a method described by Inbaraj et al. [17]. Initially, 5 mL of crude extract was poured into a vial and adjusted to pH 7 with 2% sodium hydroxide. Then 1 mL was collected and poured into a C₁₈ cartridge (500 mg/3 mL, 55 μm, 70 Å), which was previously activated sequentially with 10 mL each of methanol and deionized water. The phenolic acid fraction was eluted with 15 mL of deionized water, whereas the flavonoid fraction with 5 mL of methanol (100%). The volume of eluents was optimized for complete elution by evaluating 5, 10, 15 and 20 mL of deionized water for phenolic acids and 2, 3, 4 and 5 mL of methanol (100%) for flavonoids. Each eluate was then evaporated to dryness and dissolved in 1 mL of deionized water for phenolic acid fraction and 1 mL of 50% methanol for flavonoid fraction.

Two HPLC columns, namely Vydac 201TP54 C18 and Phenomenex Gemini C18, were compared for separation efficiency of phenolic acids and flavonoids from T. formosanum. The conditions for HPLC separation was based on a previous study by Kao et al. [20] and are: 91% of 0.1% formic acid (A) and 9% methanol (B) initially, raised to 12% B in 3 min, 28% B in 10 min, 33% B in 15 min, 39% B in 23 min, 45% B in 27 min, 48% B in 30 min, 49% B in 35 min, 68% B in 40 min and returned to the initial solvent ratio in 45 min, with flow rate at 1 mL/min and detection wavelength at 280 nm and column temperature at 35 °C. Both k (retention factor) and α (separation factor) were used to assess the separation efficiency of various mobile phases. The purity of each peak was automatically determined from the Agilent G2180A Spectral Evaluation Software Data Management System.

Various phenolic acids and flavonoids in T. formosanum were identified by comparing the retention times, absorption spectra (200–600 nm) and mass spectra of unknown peaks with the reference standards. A single quadrupole mass spectrometer with ESI mode (negative mode) was used for detection with scanning range between m/z 100 and 1000, drying gas flow 6 mL/min, nebulizer pressure 60 psi, dry gas temperature 300 °C, vaporizer temperature 250 °C, capillary voltage 3500 V, charging voltage 2000 V and fragmentor voltage 200 V. In addition, a triple quadrupole LC-MS/MS with ESI mode was used for further identification with curtain gas 20 arbitrary units, collision gas 5 arbitrary units, ion spray voltage 4500 V, dry gas temperature 550 °C, ion source gas pressure 1 (60 psi), ion source gas pressure 2 (50 psi), declustering potential 25 V, entrance potential 10 V, collision energy 20 V and collision cell exit potential 5 V. In addition, a post-column derivatization technique was employed for further identification of flavonoids [24]. In brief, two HPLC pumps were connected in series after the column for pumping the derivatizing agent containing 0.3 M of aluminum chloride solution or 0.5 M of sodium acetate solution, with pH being adjusted to neutral with 0.02 M of sodium hydroxide. After HPLC separation, the eluate of each peak was mixed with the derivatizing agent and allowed to enter into the reaction coil (1 m × 0.5 mm I.D.) for reaction at 80 °C and into the photodiode-array detector for UV detection.

An internal standard (IS) syringic acid was used to quantify phenolic acids by dissolving in acetonitrile/water (1:1, v/v), whereas (±) naringenin used to quantify flavonoids by dissolving in the same solvent. Next, 5 concentrations of 7.8, 15.6, 31.3, 62.5 and 250 μg/mL of cis-caftaric acid or chicoric acid standard were prepared in acetonitrile/water (1:1, v/v) separately, while 7.8, 15.6, 31.3, 62.5 and 125 μg/mL of chlorogenic acid standard was prepared in the same solvent. Likewise, 5 concentrations of 0.81, 1.6, 12.5, 25 and 50 μg/mL for caffeic acid standard and 7.8, 12.5, 25, 62.5 and 250 μg/mL for 3,5-di-caffeoylquinic acid were prepared separately. Then, to each standard solution, syringic acid was added to make-up a final IS concentration of 25 μg/mL. For quantitation of flavonoids, 5 concentrations of 3.9, 7.8, 15.6, 31.3 and 62.5 μg/mL of luteolin-7-O-glucoside in 70% methanol and 1.3, 2.5, 5.0, 10.0 and 20.0 μg/mL of quercetin in 100% methanol were prepared separately. Then, each standard solution was mixed with (±) naringenin to obtain a final IS concentration of 20 μg/mL. Next, 20 μL of each concentration sample was injected into HPLC-DAD twice, and the standard curves were prepared by plotting concentration ratio against area ratio. Each phenolic acid and flavonoid in T. formosanum was quantified using a formula reported by Inbaraj et al. [17] as given below:

wherein, As: peak area of phenolic acid or flavonoid; Ai: peak area of internal standard; a: slope of calibration curve; b: intercept of calibration curve; Ci: concentration of internal standard; V: volume of extract; DF: dilution factor; R: recovery; Ws: weight of sample (g).

According to International Conference on Harmonization [40], both intra-day and inter-day variability were measured for assessing the reproducibility. The intra-day variability was determined by injecting a sample 3 times each in the morning, afternoon and evening on the same day for a total of 9 replicates, whereas the inter-day variability was estimated by injecting a sample 3 times in a day and repeated for 3 days. Both standard deviation (SD) and relative standard deviation (RSD %) were calculated for inter-day and intra-day variability results.

Accuracy of the method was validated by measuring the recovery of 1 ml of 100 μg/mL each of cis-caftaric acid, chlorogenic acid, caffeic acid, 3,5-di-caffeoylquinic acid, chicoric acid, quercetin and luteolin-7-O-glucoside spiked into 0.25 g of T. formosanum sample separately. After extraction, purification and HPLC analysis, the recovery of each phenolic acid or flavonoid was obtained based on the amount after HPLC (spiked amount minus original amount) divided by the amount before HPLC (spiked amount).

For determination of LOD and LOQ, 3 concentrations of cis-caftaric acid (80, 100 and 1000 ng/mL), chlorogenic acid (500, 2000 and 4000 ng/mL), caffeic acid (40, 500 and 1000 ng/mL), 3,5-dicaffeoylquinic acid (750, 1000 and 2000 ng/mL), chicoric acid (250, 750 and 1500 ng/mL), luteolin-7-O-glucoside (40, 80 and 375 ng/mL) and quercetin (650, 1000 and 2000 ng/mL) were prepared. Each concentration was injected into HPLC 3 times and the standard curves were obtained by plotting concentration against peak height. Both LOD and LOQ were determined based on the following formula [40]:

wherein, N_p-p is the maximum noise height and S is the slope of each standard curve.

All the analyses were carried out in duplicate and the data are expressed as mean ± standard deviation. The regression equations and correlation coefficient (R²) were obtained directly from the Microsoft Excel 2003 software data management system [41].