Using Sephadex LH-20 resin for fractionation (GE Healthcare, Chicago, IL, USA) and YMC-triart C18 column (150 × 2.0 mm, 5 μm; YMC KOREA Co., Seongnam, Korea) for analysis, column chromatography was conducted. Preparative LC-MS analytical grade solvents were provided by J.T. Baker (Philipsburg, NJ, USA) and Burdick & Jackson (Muskegon, MI, USA). For the extraction of plant material, we employed solvents purchased from SK Chemical Co. Ltd. in Seongnam, Korea, and for NMR spectroscopy, we used solvents purchased from Cambridge Isotope Laboratories (Andover, MA, USA). The analytical high-performance liquid chromatography (HPLC) measurements of LC-MS were carried out using an Agilent 1290 UHPLC system (Agilent Technologies, Santa Clara, California, United States), which was outfitted with a 1290 Infinity binary pump. An Agilent 6530 iFunnel quadrupole-time of flight mass spectrometer (Q-TOF-MS, Agilent Technologies) equipped with an electrospray ion (ESI) source was used for mass spectrometry. The MS experiment was carried out under continuous conditions with positive and negative mode. HR‐ESI‐MS data was processed with Mass Hunter Data acquisition and Qualitative analysis software. An Agilent 1100 series capillary LC system linked with a Waters micromass ZQ mass spectrometer (Waters Co.) was used to perform semi-preparative LC-MS on a J’sphere ODS-H80 (20 × 250 mm, 4 μm, Waters Co., Milford, MA, USA). NMR spectra were collected using a Bruker 800MHz NMR spectrometer outfitted with a Bruker CryoPlatform (CD₃OD-d ₄, 800 MHz for ¹H and 200 MHz for ¹³C) at the Korea Basic Science Institute in Ochang, Korea, as well as a Jeol 600MHz NMR spectrometer (CD₃OD-d ₄, 600 MHz for ¹H and 150 MHz for ¹³C) at Yonsei University. MestReNova ver. 12.0.1 was used to process the data for the NMR spectra. Combinational data from 2D NMR experiments such as correlation spectroscopy (COSY), heteronuclear single quantum coherence (HSQC), and heteronuclear multiple bond correlation (HMBC) rotating frame overhause effect spectroscopy (ROESY) and total correlation spectroscopy (TOCSY) were interpreted to assign chemical structures.

Six plant materials, P. communis, S. japonica, S. maritima, S. europaea, S. brachyotus, and T. pannonicum, were harvested from a tidal mudflat marsh in Songdo, Incheon, South Korea (37° 23’ 7.7″ N, 126° 40’ 44″ E) in October 2021 for fundamental metabolite screening experiments. In September 2022, the target plant material, T. pannonicum, was gathered on bulk-weighing scales at the same location. The entire T. pannonicum was placed in a dehydrator and heated to 60°C for 12 h. The dried sample was initially ground into powder, and then the material was extracted twice with methanol (12L) through a sonication-assisted extraction process (40°C, 3 h). To obtain a crude extract, the filtered solution was concentrated in vacuo. The extract (12.8 g) was processed using Sephadex LH-20 resin and eluted with methanol to yield eight fractions (FR1-FR8, every 15 min, for a total of 120 min). Fraction 2 (15–30 min) was further purified by preparative reverse phase high-performance liquid chromatography with mass spectrometry (RP-LC-MS) with elution of 50% aqueous acetonitrile (0.1% formic acid). The targeted compounds eluted at a retention time of 21.2, 26.9, 35.7, 18.4, and 23.5 min were specified as pannosides A (3.2 mg), B (2.6 mg), C (2.1 mg), D (1.8 mg), and E (1.4 mg), respectively.

pannoside A (1) White amorphous powder; IR ν _max (KBr) 3309, 2944, 2833, 1651, 1115, 1026 cm^-1; UV (MeOH); HR‐ESI‐MS m/z 1249.5588 [M + Na]⁺ (calcd. for C₆₀H₉₀O₂₆Na, m/z 1249.5623); for ¹H (CD₃OD, 800 MHz) and ¹³C NMR (CD₃OD, 200 MHz) spectroscopic data, see Tables 1 , 2 .

pannoside B (2) White amorphous powder; IR ν _max (KBr) 3309, 2944, 2834, 1651, 1115, 1024 cm^-1; UV (MeOH); HR‐ESI‐MS m/z 1335.5984 [M + Na]⁺ (calcd. for C₆₄H₉₆O₂₈Na, m/z 1335.5991); for ¹H (CD₃OD, 800 MHz) and ¹³C NMR (CD₃OD, 200 MHz) spectroscopic data, see Tables 1 , 2 .

pannoside C (3) White amorphous powder; IR ν _max (KBr) 3321, 2944, 2833, 1651, 1113, 1026 cm^-1; UV (MeOH); HR‐ESI‐MS m/z 1421.6340 [M + Na]⁺ (calcd. for C₆₈H₁₀₂O₃₀Na, m/z 1421.6359); for ¹H (CD₃OD, 800 MHz) and ¹³C NMR (CD₃OD, 200 MHz) spectroscopic data, see Tables 1 , 2 .

pannoside D (4) White amorphous powder; IR ν _max (KBr) 3326, 2944, 2833, 1651, 1117, 1026 cm^-1; UV (MeOH); HR‐ESI‐MS m/z 1119.4947 [M + Na]⁺ (calcd. for C₅₄H₈₀O₂₃Na, m/z 1119.4993); for ¹H (CD₃OD, 800 MHz) and ¹³C NMR (CD₃OD, 200 MHz) spectroscopic data, see Tables 1 , 2 .

pannoside E (5) White amorphous powder; IR ν _max (KBr) 3309, 2944, 2833, 1651, 1117, 1026 cm^-1; UV (MeOH); HR‐ESI‐MS m/z 1511.6364 [M + Na]⁺ (calcd. for C₇₀H₁₀₄O₃₄Na, m/z 1511.6312); for ¹H (CD₃OD, 800 MHz) and ¹³C NMR (CD₃OD, 200 MHz) spectroscopic data, see Tables 1 , 3 .

Acid hydrolysis of pannoside A (1) was carried out using 6N HCl at 120°C for overnight to obtain 3-HB residues. The reactant was then rapidly cooled by immersing the reaction vial in ice water, the reaction solvent was evaporated under a vacuum, and residual HCl was eliminated using a freeze dryer overnight. After drying, the hydrolysate containing 3-HB residues and authentic standards of S- and R-3-hydroxybutyrate (Angene Chemical, India) were dissolved in 2 mL of tetrahydrofuran (THF, Sigma-Aldrich, USA) and treated with 10 mg of S-phenylglycine methyl ester (S-PGME) (98.0%, Combi-Blocks, USA), 10 mg of 1-(3-Dimethylaminopropyl)-3-ethylcarbodiimide hydrochloride (EDC, Combi-Blocks, USA), and 10 mg of 4-dimethylaminopyridine (4-DMAP). The reaction mixtures were stirred at room temperature for 1 h. The reactant mixtures were analyzed by QTOF-MS using a gradient system (10–50% aqueous acetonitrile with 0.1% formic acid over 40 min).

Molecular networks based on tandem mass spectrometry were constructed using the Global Natural Product Social Molecular Networking (GNPS) platform (http://gnps.ucsd.edu ) (Wang et al., 2016; Nothias et al., 2020) to analyze the metabolites of T. pannonicum. The dried crude extract was diluted in methanol at a concentration of 250 μg/mL and analyzed with LC/quadrupole time-of-flight (Q-TOF)-MS (A: water, B: acetonitrile, a gradient system: 0–20 min (10–100% B), 20–25 min (100% B), and 25–30 min (10% B), 0.3 mL/min of flow rate, and injection volume of 20 μL). MS-Convert was used to convert the data to mzML file format. Using the extracted ion mode (EIC), the production of each metabolite was measured by comparing the peak areas. The analysis was processed by the following condition; precursor ion mass tolerance: 2.0 Da, fragment ion mass tolerance: 0.5 Da, cosine score to generate molecular networks: 0.7, minimum matched peaks: 6, and minimum cluster size: 2. After analysis, the data were visualized using Cytoscape software ver. 3.9.0 (Shannon et al., 2003).

HT29 and PC3 cells were purchased from the Korean Cell Line Bank (Seoul, Korea) and PC9 cells were obtained from Yohan Seo (Daegu Gyeongbuk Medical Innovation Foundation, Korea). HT29, PC3, and PC9 cells were cultured in Roswell Park Memorial Institute (RPMI) 1640. RPMI1640 media was supplemented with 10% fetal bovine serum (FBS), 100 U/mL penicillin, and 100 μg/mL streptomycin, and cells were maintained at 37°C in a humidified atmosphere with 5% CO₂.

Cell viability was estimated using the Cell Titer 96^® AQueous One Solution Assay Kit (Promega, Madison, WI, USA). HT29, PC3, and PC9 cells were grown in 96-well plates in growth medium supplemented with 2% FBS for 24 h, and then cells were treated with test compounds, and the growth medium and test compounds were changed every 24 h. After 72 h, a cell viability assay was conducted according to the manufacturer’s instructions. The absorbance was measured at 490 nm using an Infinite M200 microplate reader (Tecan, Männedorf, Switzerland).

Pannoside A (1) was isolated as a white, amorphous powder. Its molecular formula was calculated as m/z 1249.5588 [M + Na]⁺ (calcd. for C₆₀H₉₀O₂₆Na, m/z 1249.5623) with 16 degrees of unsaturation based on the HR-ESI mass and ¹³C NMR spectrometric data ( Figure S2 ). The infrared (IR) spectrum showed absorptions at 3309 and 1651 cm^–1 indicating carbonyl and hydroxyl groups, respectively. The ¹H NMR spectrum of 1 exhibited one olefinic signal at δ _H 5.28 (t, J = 4.0 Hz) and six tertiary methyl signals at δ _H 1.38. 1.29, 1.15, 0.94, 0.91, and 0.83 (18H, s, CH₃ × 6), and nine methylene proton signals between δ _H 2.10 and 1.10. In addition, ¹H NMR data showed two oxygenated methine proton signals at δ _H 4.33 and 4.07, with three additional methine proton resonances at δ _H 2.82, 1.60, and 1.59. Subsequent ¹H NMR data analysis revealed three anomeric proton signals at δ _H 5.45 (d, J = 8.0 Hz, H-1-rha), 5.38 (d, J = 2.0 Hz, H-1-rhaA), and 4.36 (d, J = 7.5 Hz, H-1-gluA) and overlapped protons between δ _H 5.33 and δ _H 3.26, indicating the presence of three sugar moieties. Furthermore, the two doublet resonances δ _H 1.34 and 1.19 were assigned to secondary methyl protons (6H, d, CH₃ × 2), while the two multiplet proton signals at δ _H 5.31 and 4.31 were assigned to methine protons. The four methylene proton resonances at δ _H 2.82, 2.72, 2.44, and 2.40 were connected to the methine protons (δ _H 5.31 and 4.31) with vicinal couplings, suggesting the presence of two 3-hydroxybutyrate (3-HB) residues in 1. The ¹³C NMR spectrum revealed 60 carbon resonances and the presence of a glucuronyl moiety (δ _C 105.1, 77.6, 76.4, 75.1, and 73.8), rhamnosyl moiety (δ _C 95.1, 75.5, 75.4, 74.3, 71.1, and 16.1), and acetylated rhamnosyl moiety (δ _C 99.2, 73.5, 71.4, 71.1, 70.5, and 18.3) ( Figure 1 ).

Finally, 1D and 2D NMR analyses successfully established the aglycone structure of 1 ( Figures S1–S7 ). Among the remaining 30 carbon resonances, six methyls, nine methylenes, six methines (including two oxygenated and one olefinic), and nine quaternary carbons (including two hydroxyl) were identified. The proton couplings between H₂-1/H-2/H-3, H-5/H₂-6/H₂-7, H₂-11/H₂-12, H₂-15/H₂-16, H-18/H₂-19, and H₂-21/H₂-22 were established using the correlation spectroscopy (COSY) spectrum. The heteronuclear multiple bone correlations (HMBCs) from H₃-24 (δ _H 1.38) to C-3 (δ _C 86.7) and C-23 (δ _C 182.0) and from H₃-25 (δ _H 1.38) to C-1 (δ _C 45.1), C-4 (δ _C 53.2), and C-10 (δ _C 37.6) revealed a partial structure of ring A in 1. Notable HMBCs between H₃-25, C-5, and C-9 (δ _C 49.7) in conjunction with the ¹H–¹H COSY correlations unambiguously constructed the partial structure of ring B connecting rings A and B. The long-range heteronuclear couplings from H₃-26 to C-7 (δ _C 33.7), C-8 (δ _C 41.1), and C-9 (δ _C 49.7) and from H₂-11 to C-9 (δ _C 49.7), olefinic C-12 (δ _C 123.7), and C-13 (δ _C 144.9) indicated the presence of ring C. The connectivity of rings B and C was determined by HMBCs from H₃-26 to C-7, C-8, and C-9. Ring D was deduced by HMBCs from H₃-27 (δ _H 1.15) to C-13, C-14 (δ _C 43.3), and C-15 (δ _C 29.2), with the correlation spectroscopy (COSY) correlations between H₂-15/H₂-16, and HMBCs from H₃-27 to C-8, C-13, C-14, and C-15 also allowed the connection of ring D with C. The last ring in the aglycone moiety was determined by further HMBC couplings from the tertiary methyl groups H₃-29 and H₃-30 (δ _H 0.94 and 0.91) to C-20 (δ _C 31.7) and from H₂-19 (δ _H 1.73 and 1.16) to C-17 (δ _C 48.1) and C-18 (δ _C 2.82). The two carbonyl groups, C-23 (δ _C 182.0) and C-28 (δ _C 178.2), were connected to rings A and E, respectively, assigned by HMBCs from H₃-24 and H₂-22 (δ _H 1.78 and 1.61), accounting for nine degrees of unsaturation of 1 ( Figure 2 ). As a result, the planar structure of the aglycone in 1 was proposed as a pentacyclic triterpenoid glycoside called medicagenic acid.

The rotating frame Overhause effect spectroscopy (ROESY) spectrum of 1 allowed us to partially determine the relative configuration. In the ROESY spectrum, the ROE correlations of H-3 with H-5, H-5 with H-9, and H-9 with H₃-27 indicated that H-3, H-5, H-9, and H₃-27 were in the same orientation. Moreover, based on its ROE correlations of H₃-25 with H₃-26, H₃-26 with H-18, and H₃-25 with H₃-24 revealed that H-18, H₃-24, H₃-25, and H₃-26 were in the identical orientation. Lastly, the ROE correlations of H-2 with H-1a and H-5 with H-1a were observed while the ROE cross-peak between H₃-25 and H-1b was detected on the ROESY data, determining chemical structure of the aglycone in 1 as (2S, 3R, 4S, 5R, 8R, 9R, 10R, 14S, 17S, 18S)-2,3-dihydroxyolean-12-en-23,28-dioic acid or 2β,3β-dihydroxyolean-12-en-23,28-dioic acid ( Figure 2 ).

Two anomeric protons showed large J coupling constants with their neighboring protons (7.5 for H-1_gluA δ _H 4.36, 8.0 for H-1_rha δ _H 5.45), indicating the β-configuration, and one small J coupling constant (2.0 for H-1_rhaA δ _H 5.38), indicating the α-configuration. Presence of one β-d-glucuronic acid (gluA), one β-d-rhamnose (rha), and one acetylated α-l-rhamnose (rhaA) was revealed (Angyal, 1969). ¹H and ¹³C signals belonging to the same sugar units were assigned by analysis of the total correction spectroscopy (TOCSY) and HMBC data. The sequences and attachment of saccharides were determined using HMBC. The HMBC cross-peaks from H-1_gluA (δ _H 4.36) to C-3 (δ _C 86.7) and from H-1_rha (δ _H 5.45) to C-28 (δ _C 178.2) indicated that glucuronic acid and rhamnose were connected to C-3 and C-28 of the aglycone, respectively. Similarly, the HMBC correlation H-1_rhaA (δ _H 5.38) and C-2_rha (δ _C 75.4) revealed a linkage between the two sugar units. The glycosyl chain (rha and rhaA) at C-28 of the aglycone consists of rhamnose linked to two 3-hydroxybutyrate (3-HB) units. The HMBC correlations between H-1_rha (δ _H 5.45), C-28 (δ _C 178.2), H-4_rha (δ _H 5.10), and C-1’ (δ _C 171.9) revealed a linkage between rha and 3-HB at C-2’. The presence of two 3-HB units was confirmed by the presence of eight additional carbons, two carbonyls, two methines, two methylenes, and one methyl, which showed HMBC correlations between H-3’ (δ _H 5.31), C-1” (δ _C 172.6), H-2” (δ _H 2.42, overlapped), C-1”, H-4’ (δ _H 1.34), C-3’ (δ _C 68.9), H-3” (δ _H 4.14), C-1”, H-4” (δ _H 1.19) and C-3” (δ _C 65.7) revealed a relationship of 3-HB.

The hydrolysate of 1 was obtained by performing acid hydrolysis with 6 N HCl in order to determine the absolute configurations of the 3-HB units in 1. The hydrolysate was derivatized with S-PGME and the retention times were compared with the S-PGME derivatives of the authentic standards of S- and R-3-hydroxybutyrate by LC-MS. The S-PGME derivative of hydrolysate of 1 eluted at the identical retention time as S-PGME derivative of the authentic standards of S-3-hydroxybutyrate determining the absolute configuration of the 3-HB units as S form. ( Figure S36 ) Based on the above data, the structure of 1 was determined to be 3-O-β-d-glucuronic acid-(28-O-β-d-rhamnopyranosyl-4-O-(S)-3’-oxopentanyl-(S)-3”-hydroxybutanoate-[O-α-l-(2,3-O-diacetyl)-rhamnopyranosyl-(1→2)]-medicagenic acid ( Figures 1 , 2 ).

Pannoside B (2) was obtained as a white amorphous powder, and its molecular formula was determined to m/z 1335.5984 [M + Na]⁺ by HR-ESI-MS (calcd. for C₆₄H₉₆O₂₈, 1335.5991). A comparison of the chemical shifts of ¹H and ¹³C NMR data and 2D NMR spectra showed that 2 contains the identical medicagenic acid and sugar units (β-d-glucuronic acid, β-d-rhamnose, and acetylated α-l-rhamnose) as those of 1 with additional 3-hydroxybutyrate (3-HB) ( Figures S8–S14 ). Comparing the ¹³C NMR spectrum of 2 with that of 1, four additional carbons, one carbonyl, one methine, one methylene, and one methyl, were detected in 2 ( Table 2 ). Additional HMBC correlations between H-3” (δ _H 5.30) and C-1’” (δ _C 172.6) revealed the presence of an additional 3-HB. Based on these results, the structure of pannoside B was determined to be 3-O-β-d-glucuronic acid-(28-O-β-d-rhamnopyranosyl-O-4-[(S)-3’-oxo-(((S)-3”-oxopentanyl)oxy)butanyl-(S)-3’”-hydroxybutanoate]-[O-α-l-(2,3-O-diacetyl)-rhamnopyranosyl-(1→2)]-medicagenic acid ( Figure 1 ).

Pannoside C (3) was obtained as a white amorphous powder, and its molecular formula was determined as m/z 1421.6340 [M + Na]⁺ by HR HR-ESI-MS (calcd. for C₆₈H₁₀₂O₃₀Na 1421.6359). Comparison of the chemical shifts of ¹H and ¹³C NMR data and 2D NMR spectra showed that 3 contained the same medicagenic acid and sugar units (β-d-glucuronic acid, β-d-rhamnose, and acetylated α-l-rhamnose) as those of 1 and 2, with additional 3-HB units ( Figures S15–S21 ). The presence of four 3-HB units was confirmed by the presence of eight additional carbons, two carbonyls, two methines, two methylene, and one methyl, which showed additional HMBC correlations between H-3’” (δ _H 5.25), C-1”“ (δ _C 172.6), H-2”“ (δ _H 2.41 and 2.44), and C-1”“. Based on these results, the structure of 3 was determined to be 3-O-β-d-glucuronic acid-(28-O-β-d-rhamnopyranosyl-4-O-[(S)-3’-oxo-(((S)-3”-oxo-(((S)-3’”-oxopentanyl)oxy)butanyl)oxy)butanyl-(S)-3”“-hydroxybutanoate]-[O-α-l-(2, 3-O-diacetyl)-rhamnopyranosyl-(1→2)]-medicagenic acid ( Figure 1 ).

Pannoside D (4) was obtained as a white amorphous powder, and its molecular formula was determined as m/z 1119.4947 [M + Na]⁺ by HR HR-ESI-MS (calcd. for C₅₄H₈₀O₂₃Na 1119.4993). Comparison of the chemical shifts of the 1D and 2D NMR spectra showed that 4 contains the same medicagenic acid and sugar units (β-d-glucuronic acid, β-d-rhamnose, and acetylated α-l-rhamnose) as those of 1, 2, and 3, with no presence of 3-HB ( Figures S22–S28 ). The absence of the 3-HB unit was confirmed by the terminal carbonyl, which showed HMBC correlations between H-4_rha (δ _H 5.09) and C-1’ (δ _C 172.6) and H-2’ (δ _H 2.16), C-1’. In addition, six carbonyls supported the structure of pannoside D. Accordingly, pannoside D was assigned as 3-O-β-d-glucuronic acid-28-O-β-d-rhamnopyranosyl-4-O-acetyl-[O-α-l-(2, 3-O-diacetyl)-rhamnopyranosyl-(1→2)]-medicagenic acid ( Figure 1 ).

Pannoside E (5) was obtained as a white amorphous powder, and its molecular formula was determined as m/z 1511.6364 [M + Na]⁺ by HR HR-ESI-MS (calcd. for C₇₀H₁₀₄O₃₄Na 1511.6312). A comparison of the chemical shifts of ¹H and ¹³C (DEPT-135) data and 2D NMR spectra showed that 5 contained the same medicagenic acid as those of 1–4, with a modified sugar unit and an additional monosaccharide. Assignments of the four monosaccharides were established using extensive 2D NMR experiments ( Figures S29–S35 ). From the ¹³C (DEPT-135) NMR data, the four anomeric carbon signals were identified as C-1_gluA (δ _C 104.7), C-1_rha (δ _C 94.9), C-1_rhaB (δ _C 101.4), and C-1_xyl (δ _C 106.9). In addition, the HMBC correlations of H-3_xyl (δ _H 4.86) with C-1’_xyl (δ _C 172.7) and H₃-2’_xyl (δ _H 2.14) with C-1’_xyl indicated acetylated xylose. Four anomeric protons showed large J coupling constants with their neighboring protons (7.5 for H-1_gluA δ _H 4.37, 8.0 for H-1_rha δ _H 5.40, and 7.5 for H-1_xyl δ _H 4.53), indicating the β-configuration and one small J coupling constant (2.0 for H-1_rhaB δ _H 5.34), indicating the α-configuration. Rhamnose (rhaB) and acetylated xylose (xyl) were different from those of compounds 1, 2, 3, and 4 and were revealed with COSY, HSQC, and HMBC. With the above data, the presence of one β-d-glucuronic acid (gluA), β-d-rhamnose (rha), α-l-rhamnose (rhaB), and one acetylated β-d-xylose (xyl) was revealed. The attachment sequences of the saccharides were determined using HMBC. The HMBC cross-peaks from H-1_gluA (δ _H 4.37) to C-3 (δ _C 86.9) and from H-1_rha (δ _H 5.45) to C-28 (δ _C 178.0) indicated that gluA and rha were connected to C-3 and C-28 of the aglycone, respectively. In addition, the HMBC correlations of H-2_rha (δ _H 3.76), C-1_rhaB (δ _C 101.4), H-4_rhaB (δ _H 3.52), and C-1_xyl (δ _C 106.9) revealed the linkage of three sugar units, rha-rhaB-xyl. The structure of pannoside E was determined as 3-O-β-d-glucuronic acid-(28-O-β-d-rhamnopyranosyl-4- O-[(S)-3’-oxo-(((S)-3”-oxo-(((S)-3’”-oxopentanyl)oxy)butanyl)oxy)butanyl-(S)-3”“-hydroxybutanoate]-[O-α-l-rhamnopyranosyl-(1→2)-[O-β-d-(3-O-acetyl)-xylopyranosyl-(1→4)]-medicagenic acid ( Figure 1 ).

The existence of a broad range of probable yet undiscovered pannoside congeners prompted this study, we used a standardized HRMS² analysis approach to determine their abundance by using the GNPS platform. HRMS²-based metabolomic analysis of the methanol crude extract of T. pannonicum and GNPS-based visualization of MS²-data were performed according to a standardized procedure. The GNPS analysis identified 36 clusters and 1,709 nodes from the crude methanol extract of T. pannonicum. Each node represents an individual metabolite of T. pannonicum extracts detected from QTOF-MS and the nodes with the similar tandem mass patterns created the cluster. The GNPS networks were evaluated for molecular ion peaks of compounds 1–5 and potential structural congeners with different sugar and/or 3-HB residue lengths. The new terpenoid saponins (1–4) formed a large cluster ( Figure 3 ), which revealed structural similarities between the compounds, including m/z 1244.6 [M + NH₄]⁺ (1), 1330.6 [M + NH₄]⁺ (2), 1416.6 [M + NH₄]⁺ (3), and 1114.5 [M + NH₄]⁺ (4), with the exception of 5 (m/z 1506.6 [M + NH₄]⁺). The detected m/z subnetwork associated with compounds 1–4 had 21 connected nodes with molecular ion peaks that were partially ascribed to congeners with additional sugar moieties or modified sugars (m/z 1230.5, 1260.5, 1346.5, 1374.5, 1388.5, 1390.5, 1420.5, 1432.5, 1436.5, 1458.5, 1466.5, 1474.5, 1480.5, 1518.5, and 1522.5 [M + NH₄]⁺), and derivatives with different 3-HB residue lengths (m/z 1158.5, 1244.5, 1330.5, 1416.5, and 1502.5 [M + NH₄]⁺). Pannoside E, as revealed by structural analysis, differed somewhat in structure from the other compounds in the cluster owing to the presence of an extra sugar. The GNPS results show that metabolites of T. pannonicum are classified into several classes and the biggest cluster was triterterpenoid saponins cluster. All 21 congeners in the cluster with structural resemblance to pannosides A–D (1–4) are expected to have the same aglycone moiety, but given that they are so few in quantity, new derivatives will be identified by additional separation in the future.

To investigate the cytotoxic effects of pannosides A–E, we observed the effects of compounds 1–5 on cell viability in HT29 colorectal adenocarcinoma, PC3 prostate carcinoma, and PC9 non-small cell lung carcinoma cell lines ( Table 4 ). Pannoside E (5) significantly decreased cell viability of HT29, PC3, and PC9 cells in a dose-dependent manner, and particularly potently reduced cell viability of PC3 and PC9 cells with IC₅₀ values of 0.34 and 0.89 μM, respectively ( Figure 4 ).