Integrated HPLC and STOCSY reveal two new lanaroft type biflavonoids from Selaginella plants

Abstract

Dereplicating by standalone LC-MS remains challenging for low-abundance biflavonoids (BFVs) in Selaginella extracts, as structural isomers share identical molecular weights, possess diverse interflavonoid linkages, and exhibit highly similar fragmentation patterns, thereby increasing the risk of misannotation. Here, an integrated high-performance liquid chromatography (HPLC)-statistical total correlation spectroscopy (STOCSY) workflow was introduced to enable cross-identification of BFVs across partially resolved fractions in two Selaginella extracts, namely Selaginella plana, and S. padangensis. In the workflow, initial screening of crude extracts was performed by matching HPLC retention time against an in-house BFV library. Subsequent STOCSY analysis of partially resolved fractions obtained from S. plana, enabled orthogonal validation of known BFVs. Crucially, unmatched correlated signals guided the selection of target fractions for purification and characterization, leading to the discovery of planaflavone A, an undescribed lanaroft-type BFV. A comparable HPLC RT observed in S. padangensis further directed the isolation of an additional lanaroft-type BFV, padangeflavone A. Collectively, this integrative workflow accelerated compound discovery by enabling discrimination of BFVs with diverse linkage types that are indistinguishable by LC–MS.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-48736-4.

Introduction

Selaginella plants (Selaginella spp.) are well recognized for their richness in biflavonoids (BFVs), which display diverse medicinal properties¹. BFVs consist of two monomeric flavonoid units connected via either a carbon-carbon (C-C) bond or a carbon-oxygen-carbon (C-O-C) linkage^2,3. The diverse linkage patterns and variation in monomeric units have resulted in the identification of numerous BFVs, many of which share identical molecular weights despite differing in connectivity. For example, two apigenin units linked via C3’-C8” and C3’-C6” bonds yield amentoflavone⁴ and robustaflavone⁵, respectively, which structures require 2D NMR techniques for accurate identification. Moreover, numerous other flavonoid scaffolds with diverse linkage patterns forming BFVs continue to be reported⁶. Such structural diversity suggests that additional BFVs may exist in Selaginella species, particularly those present in low abundance and awaiting discovery.

In natural product chemistry and metabolomics dereplication strategies are essential for the rapid identification of secondary metabolites and the discovery of lead compounds or novel analogs within complex mixtures, while minimizing losses from repeated isolation^7,8. Chromatography coupled with mass spectrometry (e.g., LC-MS or LC-MS/MS) has become the dominant approach for this purpose^9,10. However, these techniques often struggle to distinguish dimeric compounds such as BFVs. Structural isomers of BFVs share identical molecular weights, frequently exhibit highly similar MS/MS fragmentation patterns. Moreover, they remain underexplored LC-MS-based studies, making BFV annotation uncertain and increasing the risk of false positive identification.

NMR-based statistical approaches have also emerged as complementary tools for metabolite annotation¹¹. Among these approaches, statistical total correlation spectroscopy (STOCSY) analysis has been widely applied in ¹H-NMR-based metabolomics for tentative identification of compounds without the need for prior separation^12,13. STOCSY reveals NMR signals originating from the same molecule through strong linear correlations between variables, assessed using Pearson correlation coefficients (r). High correlations may also highlight signals from metabolites involved in shared biological pathways¹⁴. Despite these advances, current identification of biological signals largely relies on chemical shift comparisons with existing resources such as the Human Metabolome Database: HMDB¹⁵, LOTUS: Natural product online¹⁶, nmrshiftdb2¹⁷ and a dereplication database (DEREP-NP)¹⁸. For BFVs in particular, the limited availability of spectroscopic and chromatographic databases and structural diversity arising from varying interflavonoid linkages creates a persistent analytical gap.

In this study, an integrative approach combining HPLC-DAD and STOCSY was employed to guide the isolation of previously undescribed low-abundance BFVs from two Selaginella cultivar species, S. plana (Desv. ex Poir.) Hieron. and S. padangensis Hieron. (Fig. S1). These species were selected because of the limited information available on their biological activities and BFV contents. The approach began with HPLC-DAD using a reverse-phase analytical column to screen BFV compositions in crude extracts. Subsequently, ¹H-NMR analysis was performed on all chromatographic fractions obtained from normal-phase separation. The normalized ¹H-NMR dataset was then subjected to STOCSY analysis to validate the HPLC findings, with results visualized through hierarchical clustering. This workflow enabled discrimination of BFVs with diverse linkage types and facilitated efficient selection of target fractions for subsequent isolation and characterization.

Results and Discussion

In our workflow (Fig. 1), the ethyl acetate extract of S. plana and S. padangensis (Fig. S1) were first analyzed by HPLC-DAD. BFV candidates were identified by matching experimental retention times (RTs) with those recorded in the in-house database (Table S1). The results were subsequently validated through STOCSY analysis, employing driver peaks corresponding to the isolated signals of each BFV candidate. Due to the availability of sufficient fractions (twenty-five), only S. plana was selected for STOCSY evaluation (Fig. S2). The STOCSY results were visualized as a clustered heatmap, enabling the observation of chemical shift patterns and fraction clusters. The correlated signals were subsequently matched against the suspected BFV. If a match was found, this confirmed its identity. Otherwise, fractions containing unmatched resonances were selected for isolation and structural elucidation. To enable cross-dereplication in another crude extract, the polarity of an undescribed BFV isolated from S. plana was used as a reference to guide the isolation of an additional undescribed BFV from S. padangensis at a similar polarity. The overall dereplication workflow is illustrated in Fig. 1.

Fig. 1.

Dereplication workflow.

HPLC profiling of ethyl acetate extracts

Ethyl acetate extracts of S. plana and S. padangensis were analyzed by HPLC-DAD as shown in Fig. 2. Due to the matrix effects, minor gradient variations and column aging, RT matching within a tolerance of ± 0.5 min was used together with DAD-UV spectra comparison. The results revealed that both extracts contained amentoflavone (3), robustaflavone (4), siamenflavone A (or sequoiaflavone), delicaflavone, and 2″,3″-dihydroisocryptomerin 7-methyl ether (8). In contrast, robustaflavone 4’-methyl ether (5), isocryptomerin (7), and siamenflavone C were detected exclusively in S. padangensis, as shown in the supporting information (Table S1).

Fig. 2.

HPLC chromatogram (a) S. plana, and (b) S. padangensis using a reverse-phase analytical column.

1D and 2D STOCSY analysis

Although identification can be achieved through direct RT comparison and NMR spectral matching, STOCSY analysis remains essential, as it not only confirms assignments but also uncovers less obvious BFVs or co-eluting BFVs within complex mixtures. Initially, the selected driver peaks for each BFV were evaluated using twelve artificial mixtures. All ¹H-NMR spectra of twelve artificial mixtures are presented in Fig. 3. Subsequently, the selected driver peaks were applied to identify BFV compositions in the twenty-five incompletely separated fractions from the ethyl acetate crude of S. plana (Fig. S2) using 1D and 2D STOCSY analysis.

Fig. 3.

¹H-NMR spectra (600 MHz, in DMSO-d₆) of twelve artificial mixtures.

The overall BFV profiles of S. plana were initially examined using two-dimensional (2D) STOCSY analysis without driver peak selection (Fig. S3). STOCSY correlation coefficients, ranging from 1.0 to −1.0, were visualized with a red to blue gradient (high to low correlation). Based on the characteristic ¹H-NMR features of BFVs, the analysis focused on the δ_H 8.0 to 5.0 ppm with |r| ≥ 0.7. Distinct meta-coupling patterns (J ~ 2 Hz), typically associated with A-ring protons of flavonoid units (H-6/H-8), were observed in the region between δ_H 6.5–5.5 ppm (box a). In addition, co-varying signals within δ_H 8.0–5.0 ppm indicated a fully aromatic system of the flavone system (box b), which lacks aliphatic H-2/H-3 resonances due to the presence of a C2 = C3 double bond. Whereas, strongly correlated signals at δ_H 5.6–5.4 ppm (box c) corresponded to C-2″ protons of flavanone units. These resonances were typically accompanied by proton signals from the diasterotopic H-3″ appearing at δ_H 2.6–2.9 ppm, forming the characteristic ABX spin system of saturated C-ring in flavanone units¹⁹. The coexistence of flavone and flavanone type signals indicates the presence of multiple biflavonoids with diverse linkage types, resulting in structural heterogeneity among BFVs. However, the identities of individual BFV could not be resolved solely by 2D STOCSY due to the extensive overlap and structural diversity. To address this, one-dimensional 1D STOCSY was employed to identify highly correlated signals between the selected driver peak and other resonances within the same molecule.

For 1D STOCSY analysis, the correlated NMR buckets were considered significant when |r| ≥ 0.7 and FDR ≤ 0.05. Significant correlations were visualized in a clustered heatmap, highlighting co-varying signals across fractions and the relationships between each driver peak and associated resonances. Examples of STOCSY outputs, both with and without p-value and FDR calculations, are presented in Fig. S4. These analyses demonstrated that correlated NMR buckets remained consistent at thresholds above 0.6. In contrast, lower correlation thresholds (|r| ≥ 0.5–0.6) resulted in a substantial increase in cross-correlated signals, including the NMR buckets originating from different fractions and potentially unrelated molecular entities. In this range, multiple testing correction began to influence bucket selection, as illustrated by the differences observed between the STOCSY clustered heatmaps in Fig. S4 (left panel (a) vs. right panel (b), (|r| ≥ 0.5). Furthermore, when applying |r| ≥ 0.6 to an additional STOCSY dataset using different driver peaks, the correlated signals were increased remarkably, including cross-fraction associations. These effects reduced molecular specificity, and many of these correlated signals likely reflected indirect associations rather than true intra-molecular relationships. Therefore, the correlation coefficient of at least 0.7 was considered, as signals meeting this criterion predominantly originated from the same fractions and were therefore possibly derived from the same molecule.

The STOCSY correlations were subsequently validated by comparison with ¹H-NMR chemical shifts of in-house BFVs and direct spectral matching against their reference spectra. Using δ 8.005 ppm as the driver peak (Fig. S5a), two chemical shift patterns were observed in the hierarchical clustering in fraction F19 (aromatic signals at δ 8.005 (d), 7.585, 7.575, 6.765, 6.755, 6.715, 6.705 ppm) and fraction F24 (aromatic signals at δ 8.005 (d), 7.995, 7.565, 6.795, 6.725, 6.685 ppm). Direct spectral matching confirmed that correlated signals from fractions F20 to F24 correspond to amentoflavone, while robustaflavone (driver peak at δ 7.975 ppm) was primarily detected in F19 (Fig. S5b). In contrast, siamenflavone A, sequoiaflavone, and delicaflavone were not detected in any of the analyzed fractions.

Notably, 2″,3″-dihydroisocryptomerin 7-methyl ether (8) (driver peak at δ 8.055 ppm) was identified based on correlated signals observed at δ 12.895 (hydroxy group), δ 5.615, δ 5.585, and δ 5.605 (at C-2″), δ 3.885, and δ 3.795 (dimethoxy groups), and multiple aromatic resonances at δ 8.055, δ 7.385, δ 7.015, δ 7.005, and δ 6.465 ppm. Hierarchical clustering indicated that this compound was primarily present in fraction F6 (Fig. S6a). To clearly indicate the correlated signals from the STOCSY spectra, the structure of 2″,3″-dihydroisocryptomerin 7-methyl ether (8) was overlaid on the STOCSY pseudospectrum as shown in Fig. 4a. The position numbers were highlighted in red to indicate the proton signals that exhibit significant positive correlations with the selected driver peak.

Fig. 4.

One-dimensional STOCSY spectra of (a) fraction F6 using driver peak at δ 8.055 ppm, (b) fraction F1 using driver peak at δ 3.785 ppm. The positions highlighted in red indicate proton resonances that exhibit significant positive correlations with the selected driver peaks.

Interestingly, the clustered heatmap (Fig. S6a) revealed shared resonances at δ 3.795 ppm (methoxy group), δ 7.375 ppm (aromatic proton), and δ 5.585 ppm (C-2″ proton) in F1 and F6, suggesting that these two fractions contained BFVs with a methoxy moiety and a flavanone unit. However, their placement in distinct sample clusters indicated that the BFVs in F1 and F6 represented different molecules. To further resolve these differences, the signal at δ 5.585 ppm was selected as a new driver peak for STOCSY analysis, revealing additional co-varying signals in F1 (Fig. S6b). The results highlighted the presence of two distinct BFVs in F1 and F6, both containing flavanone units. In F1, the C-2″ protons appeared at δ 5.575, δ 5.585, δ 5.595, and δ 5.555 ppm, corresponding to a doublet of doublet (dd) pattern, whereas in F6, the C-2″ protons were observed at δ 5.615 and δ 5.605 ppm. Additionally, methoxy groups were consistently detected, appearing at δ 3.795 ppm in F6 and at δ 3.785 ppm in F1. These findings underscore the structural diversity of BFVs, with distinct flavanone-containing molecules distributed across different fractions. To further investigate correlated signals in fraction F1, the resonance at δ 3.785 ppm (methoxy group of the unknown BFV) was selected as a new driver peak (Fig. S6c). STOCSY analysis revealed additional correlations with aromatic protons at δ 7.435, 7.425, 7.325, 6.855, and 6.425 ppm, as well as signals characteristic of a flavanone unit at δ 5.555, 5.565, 5.575, 5.885, and 5.895 ppm. Additionally, ortho-coupling patterns were also observed (Fig. 4b), indicating the presence of a para-disubstituted aromatic ring system in the undescribed BFV detected in F1.

¹H-NMR spectra of both fractions were compared with the in-house spectrum of 2″,3″-dihydroisocryptomerin 7-methyl ether (8). The stacked spectral comparison (Fig. S7) confirmed that both fractions contained 2″,3″-dihydroisocryptomerin 7-methyl ether. However, the ¹H-NMR spectrum of F1 also exhibited signals indicative of an undescribed BFV distinct from in-house BFVs. Notably, the resonances at δ_H 5.56 ppm (dd) suggested the presence of protons at C-2”, a spectral feature observed among our reference compounds only in siamenflavone C. However, further comparison with the ¹H-NMR spectrum of siamenflavone C revealed the absence of the characteristic meta-coupled proton signals corresponding to H-6” and H-8” signals at δ_H 5.86 and 5.87 ppm in F1. The incomplete spectral correspondence excluded the presence of siamenflavone C. Consistent with this observation, prior STOCSY analysis across twenty-five fractions did not detect correlated signals corresponding to siamenflavone C. Consequently, fraction F1 was prioritized for extensive isolation and structural elucidation of this novel BFV using comprehensive 1D and 2D NMR techniques. This approach aimed to resolve the overlapping signals and characterize the unique structural features of the undescribed compound.

STOCSY analysis of additional signals

To tentatively identify BFVs that may have been overlooked by direct RT comparison, additional ¹H-NMR signals were selected as driver peaks. This extended analysis revealed the presence of robustaflavone 4’-methyl ether (5) in fraction F17 (driver peak at δ 7.835 ppm), which was not found by using a driver peak at 5.70 ppm, and robustaflavone 4’,7”-dimethyl ether (6) in fraction F12 (driver peak at δ 7.845 ppm). The corresponding STOCSY results for these compounds are shown in Fig. S8.

Targeted isolation and purification of two previously undescribed BFVs

From STOCSY analysis, the target fraction F1 was successfully purified, yielding compound 1, named planaflavone A. The HPLC RT of compound 1 was at 85.108 min. Notably, late-eluting peaks in the S. padangensis extract were also observed at 83.311 and 84.574 min. Given the limited number of in-house BFVs detected at these RTs, only 2″,3″-dihydroisocryptomerin 7-methylether (8) was observed (RT 85.136 min). Therefore, these fractions at those RTs in S. padangensis were considered promising targets for further purification and structural characterization using 1D and 2D NMR techniques. Subsequent purification yielded another unprecedented BFV, named padangeflavone A (2), with an RT of 84.89 min. This corresponded to the peak observed at 84.57 min in the crude extract. In addition to the newly characterized compounds, several known BFVs were isolated from S. plana and S. padangensis, including amentoflavone (3), robustaflavone (4), robustaflavone 4’-methyl ether (5), robustaflavone 4’,7”-dimethyl ether (6), and isocryptomerin (7). The chemical structures of planaflavone A (1), padangeflavone A (2), and other known BFVs are presented in Fig. 5. The HPLC chromatogram recorded at 254 nm, displaying all identified BFVs, is presented in Fig. S9.

Fig. 5.

Chemical structures of isolated biflavonoids.

The structural elucidation of two undescribed lanaroft-type BFVs; Planaflavone A (1) and Padangeflavone A (2)

Planaflavone A (1) was obtained as a pale-yellow amorphous powder with an optical rotation of +1.58 (c = 1.1, MeOH). The RT from HPLC-DAD analysis was 85.108 min, closely matching that of 2″,3″-dihydroisocryptomerin 7-methyl ether (RT = 85.136 min). The molecular formula was determined as C₃₂H₂₄O_10, based on the [M-H]⁻ ion at m/z 567.1306, with a calculated double bond unit (DBU) of 21. The IR spectrum displayed characteristic absorptions for hydroxyl (3269 cm^− 1), carbonyl (1651 cm^− 1), and aromatic C = C (1498 cm^− 1 and 1442 cm^− 1) groups.

The combined MS, IR and NMR data suggested that compound 1 possesses a lanaroft-type biflavonoid framework composed of a flavone unit linked to a flavanone unit. The ¹H- NMR spectrum displayed two sets of A₂B₂ aromatic systems, indicating two para-substituted B rings, together with meta-coupled protons characteristic of A rings. The presence of an oxygenated methine proton at δ_H 5.59 (H-2″) and two diastereotopic methylene protons at δ_H 2.84 and 3.31 (H-3″) confirmed saturation of the C-2/C-3 bond, supporting a flavanone substructure.

Two strongly chelated hydroxy protons (δ_H 12.94 and 12.12) and two methoxy signals (δ_H 3.89 and 3.93) were observed. HMBC correlations established the interflavonoid ether linkage between C-4’ and C-8″. In particular, the deshielded C-8″ signal (δ_C 124.3) and correlations from H-6″ to C-7″ and C-8″ confirmed substitution at C-8″ (Fig. 6). Additional HMBC correlations from methoxy protons from C-7 and C-7″ located the methoxy substituents at these positions. These features distinguished compound 1 from lanaroflavone²⁰ by methylation at C-7″, thereby establishing planaflavone A as a new lanaroft-type biflavonoid.

Fig. 6.

Key COSY and HMBC correlations.

Padangeflavone A (2) was obtained as a yellow amorphous powder with an optical rotation of −366.4ᴼ (c = 1.1, MeOH). Its molecular formula, C₃₂H₂₄O₁₀, was established based on HRESI-MS for [M + H]⁺ m/z 569.1430, indicating the same degree of unsaturation as planaflavone A (1). The FTIR spectrum exhibited absorptions for hydroxyl (3252 cm^− 1), sp³ C-H (2924 cm^− 1), carbonyl (1713 cm^− 1), aromatic C = C (1647 and 1604 cm^− 1), and C-O (1160 cm^− 1) functionalities.

The combined spectroscopic data indicated that compound 2 possesses a lanaroft-type BFV skeleton as compound 1, composed of a flavone unit linked to a flavanone moiety. The ¹H-NMR spectrum similarly exhibited two sets of A₂B₂ aromatic systems together with two meta-coupled protons, consistent with para-substituted B rings and substituted A rings. Signals attributed to oxygenated methine proton at δ_H 5.46 (H-2”) and two diastereotopic methylene protons at δ_H 2.70 and 3.18 (H-3”) further confirmed the presence of a flavanone substructure. Two strongly chelated hydroxy protons were observed at δ_H 12.00 and 12.79.

The long-range HMBC correlation from H-3′,5’ (δ_H 6.94) with C-8” (δ_C 124.3) established the C-4’-O-C-8” interflavonoid linkage (Fig. 6). However, unlike compound 1, no methoxy was observed at C-7”. Instead, compound 2 bears a methoxy substitution at C-4’’’, indicated by the correlation of the signal at δ_H 3.76 correlated with C-4’’’ (δ_C 161.0). Therefore, these spectroscopic data indicated that compound 2 is as a positional isomer of compound 1.

All complete NMR data of planaflavone A (1) and padangeflavone A (2) are summarized in Table 1, along with the key COSY and HMBC correlations, which are shown in Fig. 6.

Table 1.

¹H-NMR and ¹³C-NMR chemical shifts of planaflavone A (1) and padangeflavone A (2).

Position	Planaflavone A (1) (600 MHz, acetone-d6)		Padangeflavone A (2) (600 MHz, acetone-d6)
	δH (J in Hz)	δC (J in Hz)	δH (J in Hz)	δC (J in Hz)
2	-	164.9	-	164.4
3	6.73, s	105.1	6.61, s	104.6
4	-	183.3	-	182.7
5	-	159.0	-	161.9
6	6.34, d (2.2)	98.8	6.22, d (2.4)	98.2
7	-	166.7	-	166.1
8	6.70, d (2.2)	93.3	6.59, d (2.4)	92.8
9	-	163.1	-	162.2
10	-	106.1	-	103.3
1′		125.5	-	125.0
2′,6′	8.03, d (8.9)	129.2	7.92, d (9.0)	128.4
3′,5′	7.06, d (8.9)	116.2	6.94, d (9.0)	115.7
4′	-	162.5	-	162.2
2″	5.59, dd (13.2, 2.9)	80.6	5.46, dd (13.0, 3.0)	80.1
3″	2.84, (under H2O peak)	43.5	2.70, dd (17.0, 3.0)	42.9
	3.31, dd (17.2, 13.3)		3.18, dd (17.0, 13.0)
4″	-	198.5	-	197.9
5″	-	166.7	-	166.1
6″	6.37, s	93.0	6.24, s	92.5
7″	-	161.7	-	158.2
8″	-	124.3	-	124.3
9″	-	161.6	-	161.0
10″	-	103.8	-	103.3
1‴	-	130.4	-	129.7
2‴,6‴	7.45, d (8.9)	129.2	7.31, d (8.4)	128.6
3‴,5‴	6.94, d (8.5)	116.3	6.80, d (8.4)	115.1
4‴	-	159.0	-	161.0
OMe-	3.89, s (OMe-7″)	57.0	3.76, s (OMe-4‴)	56.5
OMe-7	3.93, s	56.5	3.80, s	55.9
OH-5	12.94, s	-	12.79, s	-
OH-5″	12.12, br s	-	12.00, s	-

Recorded in acetone-d₆ at 600 MHz, values in ppm (δ). J (in Hz) in parentheses. br s: broad singlet, d: doublet, dd: doublet of doublet, m: multiplet, s: singlet.

The absolute configurations at C-2″ of both compounds 1 and 2 were determined by ECD calculations, as shown in Fig. 7a and b, respectively. The results confirmed that planaflavone A (1) and padangeflavone A (2) share the same stereochemistry, both being elucidated as (2"R)-planaflavone A and (2"R)-padangeflavone A, respectively. Furthermore, the optimized conformer of planaflavone A (1) showed a relatively compact geometry, in which the two flavonoid moieties have a strong intramolecular OH—O hydrogen bond, as indicated by a short H—O distance of 1.91 Å (Fig. 7c). While the conformation of padangeflavone A (2) (Fig. 7d) was more open and twisted due to the presence of the methoxy group instead of the hydroxy group, resulting in a decrease in intramolecular interaction between the two flavonoid units. This finding establishes that, despite their positional differences in methoxy substitution—C-7″ in compound 1 versus C-4‴ in compound 2—the two biflavonoids possess identical stereochemical orientation at the C-2″ center. The consistency in absolute configuration further supports their classification as structural isomers within the lanaroft-type biflavonoid family, highlighting that their variation arises solely from substitution patterns rather than differences in stereochemistry.

Fig. 7.

Simulated and experimental ECD spectra. (a) planaflavone A, (1) (b) padangeflavone A (2), and their corresponding lowest-energy conformers (c) planaflavone A (1), and (d) padangeflavone A (2).

Structural characterization of other isolated BFVs

Amentoflavone (3) was obtained as a yellow amorphous powder. MS (ESI) m/z 537.0829 [M-H]⁻, corresponding to molecular formula C₃₀H₁₈O₁₀. ¹H-NMR (DMSO-d₆, 600 MHz) δ_H 13.10 (s), 12.98 (s), 10.80 (s), 10.26 (s), 8.01 (1H, m), 7.58 (2 H, d, J = 8.9 Hz), 7.14 (1H, d, J = 9.2 Hz), 6.83 (1H, s), 6.80 (1H, s), 6.71 (2 H, d, J = 8.9 Hz), 6.46 (1H, d, J = 2.0 Hz), 6.40 (1H, s), 6.19 (1H, d, J = 2.1 Hz). The information corresponds to amentoflavone (3) reported by Ryu²¹.

Robustaflavone (4) was obtained as a yellow amorphous powder. MS (ESI) m/z 536.01862 [M-H]⁻, corresponding to molecular formula C₃₀H₁₈O₁₀. ¹H-NMR (DMSO-d₆, 600 MHz) δ_H 13.24 (s), 13.00 (s), 10.80 (s), 10.37 (s), 7.98 (2 H, 8.8), 7.92 (1H, dd, 8.7, 2.4), 7.78 (1H, d, 2.3), 7.04 (1H, d, 8.6), 6.95 (2 H, 8.8), 6.84 (s), 6.79 (s), 6.64 (s), 6.48 (1H, d, 2.1), 6.19 (1H, d, 2.0). The information corresponds to robustaflavone (4) reported by Zheng²².

Robustaflavone 4’-methyl ether (5) was obtained as a yellow amorphous powder. MS (ESI) m/z 551.0994 [M-H]⁻, corresponding to molecular formula C₃₁H₂₀O₁₀. ¹H-NMR (DMSO-d₆, 600 MHz) δ_H 13.23 (1H, s), 12.96 (1H, s), 10.83 (1H, br s), 10.38 (1H, s), 8.10 (1H, dd, J = 8.8, 2.5 Hz), 7.97 (2 H, d, J = 8.8 Hz), 7.84 (1H, d, J = 2.5 Hz), 7.25 (1H, d, J = 9.0 Hz), 6.95 (2 H, d, J = 8.8 Hz), 6.89 (s), 6.84 (s), 6.65 (s), 6.51 (1H, d, J = 2.0 Hz), 6.20 (1H, d, J = 2.0 Hz), 3.81 (3 H, s). The information corresponds to robustaflavone 4’-methyl ether (5) reported by Lin²³.

Robustaflavone 4′,7”-dimethylether (6) was obtained as a yellow amorphous powder. MS (ESI) m/z 565.1141 [M-H]⁻, corresponding to molecular formula C₃₂H₂₂O₁₀. ¹H-NMR (DMSO-d₆, 600 MHz) δ_H 13.13 (1H, s), 12.95 (1H, s), 8.10 (1H, dd, J = 8.8, 2.4 Hz), 8.03 (2H, d, J = 8.8 Hz), 7.84 (1H, d, J = 2.4 Hz), 7.26 (1H, d, J = 9.0 Hz), 6.99 (1H, s), 6.97 (2H, d, J = 8.8 Hz), 6.92 (1H, s), 6.89 (1H, s), 6.50 (1H, d, J = 2.0 Hz), 6.20 (1H, d, J = 2.0 Hz), 3.83 (3H, s), 3.80 (3H, s). The information corresponds to robustaflavone 4′,7”-dimethylether (6) reported by He²⁴.

Isocryptomerin (7) was obtained as a yellow amorphous powder. MS (ESI) m/z 551.0989 [M-H]⁻, corresponding to molecular formula C₃₁H₂₀O₁₀. ¹H-NMR (DMSO-d₆, 600 MHz) δ_H 13.10 (1H, s), 12.88 (1H, s), 8.02 (2 H, J = 8.8 Hz), 8.01 (2 H, J = 9.0 Hz), 7.13 (1H, s), 7.04 (2 H, d, J = 9.0 Hz), 6.96 (2 H, d, J = 8.8 Hz), 6.94 (1H, s), 6.85 (1H, s), 6.50 (1H, d, J = 2.0 Hz), 6.20 (1H, d, J = 2.0 Hz), 3.89 (3 H, s). The information corresponds to isocryptomerin (7) reported by Geiger and Markham²⁵.

2”,3”-Dihydroisocryptomerin 7-methyl ether (8) was obtained as a yellow amorphous powder. MS (ESI) m/z 567.1306 [M-H]⁻, corresponding to molecular formula C₃₂H₂₄O₁₀. ¹H-NMR (DMSO-d₆, 600 MHz) δ_H 12.89 (1H, s), 12.04 (1H, s), 9.65 (1H, s), 8.05 (2 H, d, J = 9.0 Hz), 7.38 (2 H, d, J = 8.6 Hz), 7.01 (2 H, d, 9.0), 6.93 (s), 6.83 (2 H, d, J = 8.9 Hz), 6.79 (1H, d, J = 2.2 Hz), 6.46 (1H, s), 6.41 (1H, d, J = 2.3 Hz), 5.60 (1H, dd, J = 13.4, 2.9 Hz), 3.88 (3 H, s), 3.80 (3 H, s), 3.43 (1H, m), 2.77 (1H, dd, J = 17.2, 2.9 Hz). The information corresponds to compound 8 reported by Chen²⁶.

Conclusion

The combined use of HPLC and STOCSY analysis proved highly effective in revealing the structural information of BFVs. Strongly correlated signals (|r| ≥ 0.9) at δ 5.565 and 5.555 ppm corresponded to the proton at C-2″ of the flavanone unit, while the methoxy group at δ 3.785 ppm also showed significant correlation (|r| ≥ 0.8). These findings matched the ¹H-NMR chemical shifts of planaflavone A (1), confirming the presence of the flavanone unit and highlighting the utility of STOCSY in identifying target fractions, even when certain signals were obscured. The unusual polarity of planaflavone A further guided the selection of fractions in the S. padangensis extract, ultimately leading to the discovery of padangeflavone A (2). Together, HPLC profiling and STOCSY analysis provided a powerful dereplication strategy that accelerated BFV identification and enabled the discovery of two previously undescribed lanaroft-type BFVs—planaflavone A (1) from S. plana and padangeflavone A (2) from S. padangensis. This approach not only facilitated the isolation and full structural elucidation of novel compounds from complex mixtures but also minimized compound loss and shortened the overall laboratory workflow. Nevertheless, the approach relied on high-quality NMR datasets and well-curated spectral references, and its effectiveness may be limited when dealing with extremely low-abundance constituents or highly complicated matrices lacking suitable reference signals. While HPLC-DAD provides physical separation and higher sensitivity, although co-elution of structurally similar BFVs with comparable polarity may still occur. Therefore, these approaches should be used in complementary rather than interchangeable.

Future studies may extend this dereplication strategy to other dimeric systems of phenolic compounds, such as biaryl polyphenols or lignin-type natural products, thereby broadening its applicability in natural product discovery. Importantly, while the structural novelty of these BFVs has been established, comprehensive biological evaluations remain necessary to fully assess their bioactive potential.

Materials and methods

Chemicals and Reagents

All BFVs in the in-house database were obtained as described in a previous work of our research group²⁷. Methanol (analytical grade) was obtained from RCI Labscan Limited (Bangkok, Thailand). Acetonitrile and formic acid (LC-MS grade) were purchased from Duksan Pure Chemicals Co. Ltd. (Gyeonggi-do, Korea), and Tokyo Chemical Industry (Tokyo, Japan), respectively. Ultrapure (type I) water was produced by a water purifying system (ELGA LabWater, UK). Precoated silica gel 60 F₂₅₄ (20 × 20 cm, Sigma-Aldrich, Darmstadt, Germany) was used for TLC analysis. Deuterated solvents (Acetone-d₆ and DMSO-d₆) for ¹H, ¹³C, and 2D-NMR analyses were purchased from Cambridge Isotope Laboratories, Inc.

Plant materials and the preparation of artificial mixtures

Plant materials used, S. plana and S. padangensis, were obtained from cultivated varieties (cultivars) maintained in compliance with institutional and national regulations, and not collected from wild populations.

S. plana

Selaginella plana (Desv. ex Poir.) Hieron. (cultivar) was collected from Baan Fa-Pa Fern, Bangkok, Thailand (13.87235, 100.58130) in January 2022 (Fig. 1a-c). The species was identified by Assoc. Prof. Dr. Nopporn Thasana and a specimen was deposited at the Chemistry Research Building, Chulabhorn Research Institute (CRI), under voucher reference number CRI-Spl-BFPF-2022. The extraction protocol was performed as previously reported²⁷. Briefly, the fresh S. plana (1.12 Kg) was chopped and extracted with 100% distilled methanol (MeOH) at room temperature in triplicate. The obtained extract solutions were later combined and concentrated to yield the crude methanolic extract. Subsequently, the crude extract was redissolved in water and consecutively partitioned with hexane, dichloromethane (DCM), and ethyl acetate (EtOAc). Due to the EtOAc extract enriched in BFVs, the crude obtained from the EtOAc concentrate (0.77 g) was selected for subsequent fractionation using normal-phase flash chromatography (Silica gel, 230–400 mesh). The fractionation was performed by using dichloromethane and methanol, starting with 100:0 to 0:100 (dichloromethane: methanol) to afford twenty-five fractions. All fractions were concentrated, dried under vacuum, and kept at −20 °C for further target isolation.

S. padangensis

Selaginella padangensis Hieron. (cultivar) was also collected in January 2022 from Baan Fa-Pa Fern, Bangkok, Thailand (13.87235, 100.58130) in January 2022. The species was identified by Assoc. Prof. Dr. Nopporn Thasana, and a specimen was deposited at the Chemistry Research Building, CRI, under voucher reference number CRI-Spa(II)-BFPF-2022 (Fig. 1d-f). The whole plant (3.2 Kg) was cut and extracted with 100% distilled MeOH in triplicate. Subsequently, the methanolic extracts were combined and concentrated, followed by redissolving the extract in water and partitioning using the three solvents as described in the S. plana part. Similarly, the EtOAc extract (4.2 g) was selected and kept at −20 °C for further analysis.

Artificial mixtures

Eight BFV standards²⁸—crysocaoloflavone, isocryptomerin, siamenflavone C, delicaflavone, robustaflavone 4’-methyl ether, siamenflavone A, robustaflavone, and amentoflavone were used to prepare twelve artificial mixtures for validating the selection of STOCSY driver peaks. Initially, all standards were dissolved in methanol-acetonitrile (1:1, v/v) at a concentration of 3 mg/900 µL. Subsequently, all standards were mixed according to their Rf values on TLC to create twelve artificial fractions. The volume ratios (µL) of standards in each fraction are shown in Table S2.

Instrumentations

HPLC-Diode Array Detection (HPLC-DAD)

EtOAc crude extracts of S. plana (Desv. ex Poir.) Hieron. (cultivar) and S. padangensis Hieron. were prepared at 1 mg/mL each in methanol (MeOH). The HPLC conditions for metabolite profiling were used similarly to the previous work²⁷. Briefly, the HPLC system was carried out on an Agilent HPLC system (Agilent Technology, USA) equipped with a diode array detector using a scan range of 200–400 nm. The column was a C18 column (Kinetex 2.6 μm polar C18, 100 Å, 150 × 2.1 mm). The mobile phases were acetonitrile (ACN) (Solvent A) and 0.1% v/v formic acid in MilliQ water (Solvent B). The gradient elution was performed with a flow rate of 0.2 mL/min, starting with 5% Solvent B (0–5 min), followed by 0.73% Solvent B (6–100 min), and then increased to 100% Solvent B until 115 min (101–115 min). The solvent B was then decreased to 5% and maintained at this gradient for 15 min (116–130 min). The column temperature was set at 40 °C, and the injection volume was 100 µL. The peaks were detected at the wavelength of 254 nm. The obtained isolated BFVs were prepared at 1 µg/mL in MeOH: ACN (1:1, v/v) and run under the same chromatographic conditions as described above.

NMR sample preparation and analysis

Twelve artificial mixtures

The ¹H-NMR samples of twelve artificial mixtures were prepared at 1 mg in 1000 µL of DMSO-d₆ containing 0.5 mM sodium trimethylsilyl propanesulfonate (DSS) as an internal standard. Subsequently, the solutions were filtered through 0.22 μm PTFE, and 600 µL of each solution was transferred to NMR tubes. The ¹H-NMR spectra of twelve artificial mixtures are presented in Fig. 3.

Prior to analysis, all spectra were aligned and bucketed to reduce data dimensionality and to minimize chemical shift variations caused by differences in pH and ionic strength²⁹. Initially, twelve ¹H-NMR spectra of artificial mixtures were segmented into bin sizes of 0.01 ppm and 0.001 ppm. The STOCSY analyses of both bin sizes provided a comparable correlation pattern, as shown in Fig. S10. Therefore, the 0.01 bin size was ultimately preferred. This bin size provided an optimal balance between spectral resolution and data robustness, while reducing dimensionality and computational times. A 0.001 ppm bin size increased the number of variables by an order of magnitude, resulting in longer computational times without additional meaningful correlations. For 1D STOCSY analysis, the driver peaks were manually selected and validated using the artificial mixtures (Table S3) to confirm that observed correlation patterns reflect the co-varying NMR signals originating from the same molecule.

Additionally, 1D STOCSY analysis of twenty-five ¹H-NMR spectra with a 0.001 ppm bin size introduced noise and baseline artifacts in the correlation. Examples of three STOCSY analyses of twenty-five ¹H-NMR spectra using the 0.001 ppm bucketed data are shown in Fig. S11. Subsequently, the selected driver peaks were applied to identify BFV compositions in the twenty-five incompletely separated fractions from the ethyl acetate crude of S. plana using 1D and 2D STOCSY analysis.

Twenty-five fractions of S. plana crude

The ¹H-NMR samples of the resulting twenty-five fractions were prepared at a concentration of 1 mg in 650 µL of DMSO-d₆ containing (v/v) 0.03% tetramethyl silane (TMS). The solutions were filtered through 0.22 μm PTFE, and 600 µL of each solution was transferred to NMR tubes.

NMR analysis

All ¹H-NMR spectra were acquired on an ASCEND™ 600 MHz NMR spectrometer (600.15 MHz and 298 K), equipped with a liquid nitrogen cooled CryoProbe™ ATM and operated using Topspin™ 4.1.4. The Bruker noesygppr1d pulse sequence was used, which applies presaturation for water suppression. All spectra were recorded using a relaxation delay (D1) of 18 s with a receiver gain (RG) of 101, a transmitter offset frequency (O1) of 1996.71 Hz, and a mixing period of 0.01 s with a number of scans (DS) of 16. The recorded spectra were manually phased and baseline-corrected. All samples were analyzed in triplicate.

Data processing

After phase and baseline correction of all ¹H-NMR spectra, including twelve artificial mixtures and twenty-five fractions, the spectra alignment was performed based on internal standard signals, set to 0 ppm using Bruker TopSpin™ 4.1.4. All ¹H-NMR spectra were bucketed using two bin sizes (0.01 and 0.001 ppm) with AssureNMR™ 2.2 (Bruker BioSpin GmbH). The region of bucketing data was performed over 13.0 to −0.01 ppm. The chemical shift regions δ_H 12.0-11.5.0.5, 9.5–8.6, and 3.7-0.7 ppm were removed to reduce noise, baseline effects, and residual water signals¹⁴.

Statistical total correlation spectroscopy (STOCSY)

The bucketed NMR dataset consisted of columns representing NMR bins (in ppm range) and rows corresponding to individual fractions. The resulting buckets were normalized by dividing each value by half of the peak area of TMS (0.01 to −0.01 ppm) using Microsoft^® Excel for Mac Version 16.100.3 (25083017). To further reduce the impact of concentration differences, two normalization methods—probabilistic quotient normalization (PQN) and constant sum normalization (CSN)³⁰—were evaluated on the ¹H-NMR dataset using R version 4.5.1 (2025-06-13). Based on STOCSY results, CSN was identified as the most suitable method and was therefore applied in all subsequent analyses.

Both one-dimensional (1D) and two-dimensional (2D) STOCSY analyses were performed. In 2D STOCSY, Pearson correlation coefficient (r) was calculated for all pairwise combinations across datasets without driver peak selection, using the metabom8³¹ and ggplot2³² R packages. Correlation values below 0.7 (|r| < 0.7) were excluded, and the resulting data matrix was visualized as a contour plot. In 1D STOCSY analysis, the required packages for the calculation were metabom8 and ggplot2 R packages. The correlation coefficient was calculated between the selected driver peak and all other variables. The driver peaks were selected based on well-resolved signals of each BFV candidate and unique NMR signals of interest. The correlation coefficients were converted to t-statistics to compute two-tailed p-values, and the Benjamini-Hochberg method was applied to control the false discovery rate (FDR)^33,34. The buckets with |r| ≥ 0.7 and FDR ≤ 0.05 were considered significant. To visualize the significant NMR buckets, intensities of all significant correlated buckets were then scaled (z-score), and both rows (samples) and columns (ppm buckets) were hierarchically clustered using Euclidean distance and complete linkage³⁵ using the ggplot2 R package.

Structural elucidation

All isolated compounds were characterized using 1D NMR (¹H-NMR (600.15 MHz) and ¹³C-NMR (150.91 MHz)) and 2D NMR (¹H-¹H COSY, HSQC, HMBC). 1D and 2D NMR techniques were recorded on a Bruker ASCEND™ 600 MHz NMR spectrometer. Tetramethyl silane (TMS) was used as an internal standard in acetone-d₆ and DMSO-d₆. The chemical shifts (δ) of ¹H-NMR spectra were recorded in parts per million (ppm) on the scale relative to the internal standard (0 ppm) in acetone-d₆ and DMSO-d₆, the solvent signals at 2.05 and 2.50 ppm, respectively. The coupling constant (J) in ¹H-NMR spectra was reported in Hz. The chemical shifts of ¹³C-NMR spectra were also reported in ppm relative to the internal solvent signals (206.68 and 29.92 ppm in acetone-d₆ and 39.5 ppm in DMSO-d₆). The coupling constant (J) was reported in Hz. The acquired 2D NMR was Fourier transformed and manually phase corrected. Similarly, acetone-d₆ and DMSO-d₆ were used as the solvents for all the samples, and the chemical shift values were referenced to the TMS signal. Fourier-transform infrared spectroscopy (FTIR) spectra were recorded on a Thermoscientific Nicolet iS 5, and high-resolution mass spectrometry (HRMS) was also performed by using Thermo Scientific, Orbitrap Q Exactive Focus via electro-spray ionization (ESI) in a positive and/or a negative mode for all compounds. Absolute configurations were determined by ECD calculations. Before the calculations, the optimized conformers were searched using CREST³⁶, followed by DFT calculations using the Gaussian 16 Rev. C.01 program. Subsequently, the obtained low-energy confirmers were further optimized ωB97XD/cc-PVDZ level of theory of the gas phase. Each conformer with a population over 1% based on the Boltzmann distribution of Gibbs free energies was subjected to the following ECD calculations. The ECD spectra were simulated by using TD-DFT at M06-2x/def2-SVP level of theory for 30 excited states with IEFPCM of methanol solvent model^37–39. The simulated ECD curves were generated by using SpecDis with Boltzmann average all conformers and overlapping Gaussian function with an exponential half-width (σ = 0.35)⁴⁰. Theoretical ECD spectra of the related enantiomers were generated from the direct inversion of simulated ECD spectra. All structures were virtualized by using CYLview20⁴¹. All relative Gibbs free energies and the percentage population of all possible conformers corresponding to planaflavone A and padangeflavone A are shown in Tables S4 and S5, respectively. All atom coordinates are shown in Tables S6-7. The two most stable conformers of planaflavone A (1) (Fig. 7c) and padangeflavone A (2) (Fig. 7d) were displayed using the PyMOL Molecular Graphics System (version 3.1.6.1). The atom distance as shown in Fig. 7c was measured using Avogadro (version 1.2.0).

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

W. K. performed STOCSY, identified the compounds within the complex mixtures and drafted the manuscript. U. L. conducted NMR (noesygppr1d) experiments for STOCSY analysis, performed HPLC analysis, isolated planaflavone A, including other known biflavonoids from *S. plana* and collected HRMS, 1D and 2D NMR for each compound. Y. A. P. carried out the isolation of padangeflavone A and other known compounds from *S. padangensis* . W. T. performed plant collection and extract, and biflavonoids database. N. S. performed ECD calculations. S. R. reviewed and edited the manuscript. N.T. supervised the experimental workflow and manuscript preparation, including edited and approved the final version of the manuscript.

Funding

This research project is supported by Chulabhorn Graduate Institute postdoctoral fellowship (Grant No. CGI(2025)/03), Chulabhorn Graduate Institute (Fundamental Fund by National Science Research and Innovation Fund (NSRF): fiscal year 2025) (FRB680079/0518 Project code 209040), Thailand Science Research and Innovation (TSRI), and Chulabhorn Research Institute (Grant No. 53503/4821870).

Data availability

Data is provided within the supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.

Consent to publish

All authors give consent for publication.