A total of 42 accessions were used in this study. These accessions were from different geographic regions around the world including commercial accessions and landraces. The details of these 42 accessions were presented in Supplementary Table 1 and Table 1. The sepals and petals were collected from each accession and divided into six parts (Figure 1), sepal throat (ST), sepal eye (SE), sepal limb (SL), petal throat (PT), petal eye (PE), and petal limb (PL), according to a previously described method (Cui et al., 2019). Subsequently, each part was placed in liquid nitrogen immediately after detaching, then preserved at −80°C until the flavonoids extraction.

High-performance liquid chromatography grade methanol, formic acid, trifluoroacetic acid (TFA), and acetonitrile were purchased from Fisher Scientific (Fair Lawn, NJ). Quercetin 3-glucoside (Qu3g), myricetin, kaempferol 3-glucorhamnoside, apigenin, and two cyanidin derivatives standards, cyanidin 3-glucoside (Cy3g), and cyanidin 3-rutinoside (Cy3r) were purchased from Sigma-Aldrich (St. Louis, MO). Kaempferol (Km), quercetin (Qu), and rutin (Rt) were purchased from Solarbio (Solarbio, China). Ultrapure water from PureLab Ultra Water System (ELGA LabWater, United Kingdom) was used in this experiment. All other reagents used were of analytical grade.

Samples (0.1 g fresh weight, FW) were fully ground in liquid nitrogen, extracted in 2 mL solvent mixture (methanol/water/formic acid/TFA, 70:27:2:1, v/v/v/v), and allowed to settle for 24 h without light. The extracts were then centrifuged (12,000 rpm, 20 min), and the supernatant was filtered (0.22 μm) into vials. The HPLC analyses were carried out on a Thermo Fisher HPLC system connected with a 996 photodiode array detector (UltiMate 3000, ThermoFisher, United States), which was set in the range of 190–600 nm. Data collection and processing was accomplished by the Chameleon software version 2.0. The chromatographic separation was performed on a Venusil ASB C18 column (Agela Technologies, China) with 4.6 mm × 250 mm, 5 μm. The mobile phases comprised 2% aqueous formic acid (A) and acetonitrile (B). The gradients were programed as follows: 0 min, 8% B; 3 min, 8% B; 23 min, 20% B; 33 min, 40% B; 43 min, 40% B; 45 min, 8% B. The column temperature, injection volume, and flow rate were set at 35°C, 10 μL, and 0.8 mL/min, respectively. The flavonoids and anthocyanin chromatograms were extracted at 350 nm and 520 nm. All samples were extracted in triplicate.

Liquid chromatography–mass spectrometry (LC–MS) analysis was performed on the HPLC instrument described above, interfaced with a microOTOF Q quadrupole time-of-flight mass spectrometer (Thermo Fisher, United States) connected to either electrospray ionization (ESI) or an atmospheric pressure chemical ionization source. The HPLC analysis conditions were similar to those described above. The mass signal range was m/z 50–1,100. The ionization of flavonoids was achieved with an ESI source in both positive and negative modes, and the parameters were set as follows: capillary voltage, 3,500 V; endplate offset, 500 V; drying gas (nitrogen) flow, 8.0 L/min; drying gas temperature, 180°(C; collision rf, 200 Vpp; nebulizer pressure, 0.8 bar; prepulse storage, 8.0 (s; transfer time, 80.0 μs; and collision energy, 10.0 eV.

The quantitation was conducted by external calibration of the corresponding standards from the areas of the chromatographic peaks at 350 nm for flavonoids and 520 nm for anthocyanins. All standards were dissolved in methanol. The following equations were used: cyanidin-3-glucoside (y = 274.1046 x + 0.1328, R² = 0.99987); quercetin (y = 392.9441 x + 0.5815, R² = 0.99865); quercetin 3-glucoside (y = 339.3973 x − 0.3364, R² = 0.99997); rutin (y = 214.2924 x − 0.1913, R² = 0.98327); apigenin (y = 516.2105 x + 0.0001, R² = 0.99666); myricitrin (y = 1040.74 x − 2.1097, R² = 0.99798). The content of compounds that did not have corresponding standards was calculated from the most suitable standard calibration curve.

The mean value of each sample was obtained from three replications and used for further analysis. The statistical analysis was conducted by the R (x64 3.5.1) software. The variations in the contents of ten major flavonoids among different floral parts were determined by the Mann–Whitney U test at p < 0.001. The hierarchical cluster analysis was conducted by the Ward D method, and the flavonoids difference between clusters were determined by one-way ANOVA at p < 0.05.

The flavonoids were identified according to HPLC retention times, UV λ_max spectrum, and MS data (in both NI and PI modes), as well as previous reports (Markham, 1989). In total, 15 flavonoids were detected in the 42 accessions. The HPLC-DAD and HPLC-ESI(±)-MS² analyses results, such as molecular ion, aglycone ion, and main fragments of MS², are summarized in Table 2. In this study, eight flavonols, two flavones, and two anthocyanins were identified from the flowers of Hemerocallis, and three compositions were unknown. The PE of Hemerocallis ‘Little Bumble Bee’ was of these 15 compositions (Figures 2, 3).

Thirteen of the fifteen separated peaks showed λ_max at 350 nm, indicating they were flavonols and flavones. According to four standard retention times (Figure 2) and MS data, the peaks 9, 11, 14, and 15 were speculated as rutin (Rt), quercetin 3-glucoside (Qu3g), quercetin (Qu), and kaempferol (Km), respectively. Rt, Qu, and Km were previously published in the alabastrum of H. fulva (Lin et al., 2011; Chen et al., 2012; Kao et al., 2015). Also, Qu3g is found in H. fulva fresh leaves (Yanjun et al., 2004) and ‘Stella de Oro’ flowers (Cichewicz and Nair, 2002). Leaves of different amaranth species also have abundant Rt, Qu, Qu3g, and Km (Sarker and Oba, 2018, 2019, 2020c,d).

In methanol, kaempferol exhibits λ_max spectrum at approximately 266 nm (band II) and 367 nm (band I), while the λ_max of quercetin was approximately 255 and 370 nm (Markham, 1989). Furthermore, the glycosylation of the 3-hydroxyls could cause a hypsochromic shift of band I by about 12–17 nm, whereas the 7-hydroxyls would not change the λ_max (Singh et al., 2010). Based on these principles, peak 8 was indicated as quercetin 7-glucoside (Qu7g) with λ_max spectrum (band I) at 363.3 nm, MS data 463.09 ([M–H]⁺), and fragment m/z 271 ([Y0]⁺), with a glucoside (162 Da) loss in molecular weight. Peak 12 showed λ_max at 345 nm (band I hypsochromic shift of nearly 21 nm), which was assigned to kaempferol 3-glucoside (Km3g) with the MS data 449.01 ([M–H]⁺) and fragment m/z 287.04 ([Y0]⁺). As widely distributed flavonols in many plants (Li et al., 2009; Sarangowa et al., 2014), Qu7g and Km3g are also present in other daylily accessions (Cichewicz and Nair, 2002; Lin et al., 2011; Sun et al., 2018). Peak 4 was characterized as quercetin 3-arabinoside (Qu3ar) because it had fragments at m/z 433.08 ([M–H]^–) and 271 ([Y0]^–), indicating the loss of an arabinoside (150 Da) (Chen et al., 2010). A similar finding was previously reported in the daylily accession ‘Baihua’ (Sun et al., 2018).

Peak 7 had λ_max at 311.18 nm, the MS was 623.09 ([M–H]^–) and fragment m/z 315.02 (losing 309 Da), pointing out to isorhamnetin 3-rutinoside (Is3r) (Deng et al., 2013). The Is3r was previously reported in H. fulva (Yanjun et al., 2004) and other plant resources, such as lotus (Deng et al., 2013) and sea buckthorn (Pop et al., 2013). Peaks 10 and 13 were speculated as flavones, luteolin 7-glucoside (Lt7g), and apigenin 7-glucoside (Ap7g), respectively. Lt7g was previously reported in yellow color tree peony (Li et al., 2009) and olive leaves (Lama-Muñoz et al., 2019), while Ap7g has been published in chamomile (Bączek et al., 2019). However, this study is the first to report the two flavones in Hemerocallis.

Unfortunately, peaks 1, 2, and 6 were not inferred. Peak 1 had MS 663 ([M–H]⁺), and fragment m/z 479 ([Y0]⁺), indicating the tentative compound was isorhamnetin 3-neohesperidoside. However, the isorhamnetin 3-neohesperidoside showed λ_max at around 254 nm in previous reports (Deng et al., 2013; Pop et al., 2013), which was not confirmed (λ_max = 324.77 nm) in our study. Similarly, peaks 2 and 6 were also confirmed according to λ_max spectrum and MS data.

In our study, peaks 3 and 5 showed λ_max at 520 nm with more peak area, indicating they were anthocyanins (Mitchell et al., 1998). Two anthocyanin standards, cyanidin 3-rutinoside (Cy3r) and cyanidin 3-glucoside (Cy3g) were used to identify the different compounds by co-elution, and the retention times were 15.40 min and 16.7 min, respectively (Figure 3). Peak 5 showed the same retention time by the two standards, and further MS data indicated it was Cy3r. However, the retention time of peak 3 was earlier than the Cy3g standard with MS 611.09 ([M–H]⁺) and m/z 287.1 ([Y0]⁺). Previous studies proved that the polarity of di-glucosides is greater than that of mono-glucosides. Therefore, the elution time is always earlier than mono-glucosides (Yongcheng, 2018), indicating cyanidin 3,5-glucoside (Cy3g5g) was the speculated compound. Cy3r and Cy3g5g are common anthocyanin glycosides and are widely distributed in many plants, such as wild bananas (Kitdamrongsont et al., 2008), tree peony (Zhang et al., 2014), and rose (Zhang et al., 2015).

The flavonoid composition and contents varied dramatically among the different accessions and parts (Supplementary Table 2), ranging from 0.00 to 321.99 μg g^–1 FW. Thus, Rt, Qu3ar, and U1 were dominant flavonoids in Hemerocallis floral organ, comprising 31.58, 20.66, and 15.61% of relative content, respectively. Our study found that most of the flavonoids we detected were higher in the petal than the sepal (Figure 4). The coefficient of variation (CV) ranged from 25.58 to 186.92% (Table 3). For ST, the flavonol content of Qu had the most considerable CV value (149.57%), followed by U2 (135.21%) and Qu7g (128.16%). For SE, the most significant CV was Qu3g with 163.47%, followed by Cy3r with 143.83% and Km with 133.24%, respectively. For SL, the first three compounds with the most extensive variation range were Rt, Is3r, and Cy3r, with CV values of 141.31, 127.52, and 112.80%, respectively. For PT, the largest variation compound was Km (CV = 158.50%), followed by Qu3g and U2 (CV = 151.81% and 147.89%, respectively). For PE, Qu3g had the largest CV value (178.05%), followed by Cy3r (146.04%), and Is3r (138.68%). For PL, the first three compounds with the largest variation range were Is3r, Cy3g5g, and U2, with CV values of 186.92, 136.45, and 121.40%, respectively.

The different flavonoids showed different distribution patterns in the six floral parts. In the three parts of the sepal, including ST, SE, and SL, three compounds, Rt, Lt7g, and Km3g, could be detected in the ST of all the 42 accessions. Moreover, we did not detect Rt in SE of three accessions, including ‘Nakai’, ‘Green Mystique’, and ‘Little Wine Cup.’ Meanwhile, six accessions detected no Rt in SL, including ‘Canadian Border patrol 2’, ‘Green Mystique’, ‘Bonanza’, ‘Dahuaxuancao’, ‘Frans Hals’, and H. fulva var. kwanso var. reasata. In the three parts of the petal, Rt could be detected in PT of all the 42 accessions, but PE was not detected in ‘Canadian Border Patrol’ and ‘Little Wine Cup.’ Thirteen accessions did not detect Rt in PL.

Hence, Rt was the dominant compound (Figure 4), similar to the findings of previous reports (Sun et al., 2018), and mainly distributed in PT. In addition, we discovered that Rt was more widely distributed in the sepal than the petal, but the average content was higher in the petal than the sepal. Moreover, the Rt contents were higher in the eye (SE and PE) than limb (SL and PL). Qu7g, and Km3g also showed the same distribution pattern. Ap7g nearly had the same distribution pattern, except the limb, that SL (8.78 μg g^–1 FW) higher than PL (3.91 μg g^–1 FW).

In this study, the two cyanidin derivatives (Cy3g5g and Cy3r) could not be detected in nightlilies and daylilies with yellow flowers, while only other 20 colored accessions detected cyanidin derivatives. Moreover, the anthocyanin was nearly not detected in the throat (ST and PT) of the 42 accessions (Supplementary Table 2), indicating no anthocyanin contributed to throat pigment coloration. Throat only had three color phenotypes, yellow, yellow-green, and light yellow, similar to our previous investigation (Cui et al., 2019). As shown in Table 3, the average value of Cy3r ranged from 26.99 to 72.08 μg g^–1 FW, which is higher than Cy3g5g (12.07∼34.83 μg g^–1 FW). For the Cy3r content, the average SE (26.99 μg g^–1 FW) was lower than PE (39.91 μg g^–1 FW). Similarly, the SL also showed a lower average value (37.95 μg g^–1 FW) than PL (72.08 μg g^–1 FW). For Cy3g5g content, SE and SL showed lower average values (12.14 μg g^–1 FW and 12.07 μg g^–1 FW, respectively) than PE and PL (23.71 μg g^–1 FW and 34.83 μg g^–1 FW, respectively). Thus, petals had higher anthocyanin contents than sepals in the 20 colored accessions.

The calyx of the Hemerocallis flower organ is specialized into a bright colored sepal, which becomes the main ornamental part of the corolla together with petals. However, we revealed that the flavonoid composition and content showed a considerable difference between sepals and petals. Obviously, the flavonoids compositions were different among the various floral parts. This study used the Mann–Whitney U non-parametric test via the R language to analyze the difference in ten major compounds among the six floral parts (Table 4). Generally, the diversity in petal compositions was more than the sepal. Eight compounds showed a significant difference (p < 0.05) between PT and PL, while only three compounds showed a significant difference (p < 0.05) between ST and PT.

For sepals, the ST was significantly different from the other two parts. In the Mann–Whitney U non-parametric test results between ST and SE, six compounds showed significant differences. For instance, Cy3g5g and Rt showed a significant difference at p < 0.001, Lt7g and Km3g at p < 0.01, and two compounds (U1 and Qu3ar) at p < 0.05. At the same time, ST showed a significant difference with SL in Cy3g5g, Rt, and Km3g at p < 0.001, U1 and Qu3ar at p < 0.01, and Qu3g at p < 0.05. However, only Rt showed a significant difference (p < 0.01) between SE and SL.

For petals, the Mann–Whitney U non-parametric test results between PT and PE showed five compounds had significant differences, such as one compound (Cy3g5g) at p < 0.001 level, one compound (Is3r) at p < 0.01 level, and three combinations (Qu3ar, Lt7g, and Ap7g) at p < 0.05. Eight compounds showed significant differences between PT and PL, such as Cy3g5g, Rt, Lt7g, Qu3g, Km3g, and Ap7g at p < 0.001, U1 at p < 0.01, and Qu3ar at p < 0.05. The difference between PE and PL was also significant, which was not similar to sepals; five compounds (Rt, Lt7g, Qu3g, Km3g, and Ap7g) were significantly different at p < 0.001, and one compound, Is3r, was significantly different at p < 0.05.

However, the difference between the same parts in the sepal and petal was not substantial. Only three compounds showed a significant difference between ST and PT, including U1 (p < 0.01), Rt (p < 0.01), and Qu3ar (p < 0.05). The significantly different compounds between SE and PE were Rt and Km3g, p < 0.001, and Is3r at p < 0.05. Between SL and PL, two compounds (Lt7g and Ap7g) showed significant differences at p < 0.001, and two compounds (Qu3g and Km3g) had a considerable difference at p < 0.05.

Hierarchical cluster analysis was conducted by Ward’s D method to accurately describe the characteristics among the 42 accessions. Depending on the variations of ten identified flavonoids and anthocyanin components of the accessions, finally, these accessions were grouped into four clusters (Figure 5). Flavonoid compounds, including flavonols, flavones, etc. in leaves were also varied across the amaranth accessions (Sarker and Oba, 2020a,b; Sarker et al., 2020). One-way ANOVA showed a significant difference in the ten significant compounds between four clusters (p < 0.05). The variations in ten major compounds between the four groups are shown in Figure 6.

The accessions falling in clusters I and II had lower U1 and Qu3ar relative to other clusters in the six floral parts. The cluster I was composed of 21 accessions, which could be divided into three subgroups. The landraces of nightlilies clustered in the same subgroup, which included two branches. For instance, ‘Datonghuanghua’, ‘Malinhuanghua’, ‘Yanchihuanghua’, and ‘Qiaotouhuanghua’ fell into the same branch; ‘Dongzhuanghuanghua’, ‘Dalihuanghua’, and H. minor clustered in another branch of this subgroup. The five compounds of Rt, Lt7g, Qu3g, Km3g, and Ap7g were not detected in PL, but the orange ‘Childrens Festival’ clustered into the same branch, which was contrary to our expectation. Meanwhile, the three accessions, H. citrine, H. thunbergii, and H. altissima, and two landraces, ‘Huohuanghua’, and ‘Qiezi’, clustered in another subgroup. The third subgroup included one night lily accession, one purple daylily, two yellow daylily, and two pink daylily accessions.

The cluster II was composed of three pink accessions, ‘Canadian Border Patrol 2’, ‘Always Afternoon’, and ‘Canadian Border Patrol.’ The pink accessions showed the highest average content of Is3r in PE and PL (97.8 and 107 μg g^–1 FW, respectively) than other clusters; One-way ANOVA also supported these results (p < 0.05). The Km3g content of ST and PT was also significantly higher (p < 0.05) than other clusters, with average values 56.78 and 101.25 μg g^–1 FW, respectively. Although the Km3g content in SE and PE was the highest, the differences were not significant (p < 0.05).

A total of 17 accessions, including most orange and red accessions, fell in cluster III; their PE showed higher contents in Qu3ar and Rt than cluster II. The average Qu3ar was 104.4 μg g^–1 FW, ranging from 11.76 to 191.76 μg g^–1 FW, while the average Rt was 116 μg g^–1 FW, varying from 8.84 to 309.43 μg g^–1 FW. The cluster IV was composed of a single ‘Little wine cup’, and it was the only accession that was not detected in Rt, Lt7g, Qu3g, Km3g, and Ap7g, in SE and PE. Thus, ‘Little wine cup’ could be thought of as a natural mutant to study the loss of Rt metabolism in Hemerocallis.