A total of 60 Corsican species, including 14 liverworts and 46 mosses, were collected. Additionally, three Japanese liverworts, as part of a collaboration with the Tokushima Bunri University (Japan), were added to the collection. Supplementary Table S1 summarizes the data related to the extract collection, including taxonomical information and yield ( Supplementary Figure S1 ; Supplementary Table S1 ).

Bryophytes, in particular mosses, are known to produce many fatty acids. These primary metabolites are common to many vascular plants, and in this study, they were not profiled (Lu et al., 2019). They were depleted to obtain an enrichment of secondary metabolites, which are known to be more species-specific. To reduce their amounts and improve the detection of metabolites of medium polarity, the dry sample material was extracted successively with hexane (HEX), methylene chloride (DCM), and methanol (MeOH). Only the DCM and MeOH extracts were further analyzed. The means of extraction yields of liverworts and mosses are similar, respectively: 0.21% and 0.25% for HEX extracts, 0.58% and 0.53% for DCM extracts, and 1.93% and 2.01% for MeOH extracts. However, boxplots of extraction yields ( Supplementary Figure S2 ) show high variability between species independently of the phylum. Several species appear as outliers: three liverworts (Frullania tamarisci, Porella arboris-vitae, and Porella obtusata) with extraction yields of 1.7%, 1.5%, and 3.4%, respectively, to DCM extracts and 2.9%, 4.7%, and 4.4%, respectively, to MeOH extracts and two mosses (Bartramia pomiformis and Orthotrichum rupestre) with 6.3% and 6.0%, respectively, to MeOH extract yields, three times more than the average.

The composition of the HEX extracts was evaluated by ¹H NMR profiling. This revealed that they contained mainly fatty acids. The DCM and MeOH extracts were submitted to Solid Phase extraction (SPE) to reduce the amount of residual fatty acids and highly polar metabolites (this process is referred to as “sample clean-up” and continued in Supplementary Figure S3 ). After sample clean-up, on average, DCM moss extracts lost 83% of their mass, while liverwort extracts lost 60%, which highlights the presence of lipids even after hexane defatting of the dried plant material. MeOH moss extracts lost 61% of their mass after cleaning, while liverwort extracts lost only 45%. These last losses may be attributed to the presence of polar residues (mainly sugars based on ¹H NMR profiling).

The comprehensive approach to secondary metabolite enrichment indicates that natural products of medium polarity (DCM + MeOH) comprise 12.9 mg per gram of dry liverworts and 7.8 mg per gram of dry mosses. This shows that mosses produce about half the amount of secondary metabolites compared with liverworts. In addition to the ends in the mean values, extraction yields were found to be very species-specific; for example, P. arboris-vitae yielded 35.9 mg per g of the dry plant (MeOH extract). For most details, refer to the boxplot figure in Supplementary Figure S2 .

All enriched DCM and MeOH extracts were systematically profiled by reversed-phase UHPLC–HRMS/MS in positive (PI) and negative ionization (NI) modes using a generic 9-min linear gradient. This yielded 252 chromatographic profiles: 126 in PI and 126 in NI.

Despite the sample clean-up procedure, evaluation of the metabolite profile of an initial set of species still revealed residual features associated with fatty acids after 6 min of elution and sugars between 0 and 1 min of elution. To focus on secondary metabolites, only features detected within a chromatographic window of 1 to 6 min were retained for further data exploration.

Both PI and NI data were gathered in two separate feature-based molecular networks (MNs): PI, 8,843 nodes including 5,252 grouped in 370 clusters; and NI, 4,572 nodes including 2,660 nodes grouped in 190 clusters ( Figure 1 ). For ease of discussion, clusters have been numbered from largest to smallest (PI clusters are preceded by P and NI by N).

All HRMS/MS spectra of the detected features were annotated using the SIRIUS software (Dührkop et al., 2015, 2021; Ludwig et al., 2020; Hoffmann et al., 2022) and, in particular, the CANOPUS module, which provides confident chemical class information using the deep neural network-based structural classification tool NPClassifier that provides pathways, superclasses, and class information (Dührkop et al., 2021).

To establish a list of individual candidate structures for all features detected with an MS² spectrum, a workflow that initially performed a spectral matching against an in-silico MS/MS database of NP spectra was used. For this, a modified cosine score was calculated between experimental and simulated spectra (Allard et al., 2016). Candidate structures were then reweighted based on the taxonomy of the source organism and in the light of previously reported occurrences gathered by the LOTUS initiative (Rutz et al., 2022). This automated workflow is known as TIMA (Rutz et al., 2019). All these annotations are available in the MASSIVE repository of the data (see Materials and methods).

For all features of interest displayed in the figures, only the best structural candidates proposed by TIMA matching the NPClassifier superclass from SIRIUS were considered valid. In cases where TIMA was not giving coherent annotation proposals, the GNPS annotations automatically obtained during the molecular network construction were considered.

This information was mapped on the MN using color-coded nodes according to the SIRIUS pathway annotation, giving a quick overview of the chemical diversity of the extracts ( Figure 1 ).

It is generally accepted that the annotations of pathways and superclass via SIRIUS have a high level of confidence (expressed as probability hereafter) (Dührkop et al., 2021). In this study, only pathway and superclass annotations with a probability greater than 0.8 in PI and 0.7 in NI were exploited. Using this protocol, 52% of all detected PI features could be attributed to a given NP pathway with good confidence (4,467 features in the dataset). Among them, 51% were annotated with structures derived from TIMA.

Over the whole dataset, significantly more PI than NI features were detected. While the PI mode provided detection of most compound types, for the fatty acids, shikimate, and phenylpropanoids (flavonoids, stilbenoids, etc.), better ionization and detection were obtained in NI. Thus, because of this complementary, both modes were kept in the data meaning to obtain a comprehensive overview of the metabolite repartition.

To summarize most of the annotation results obtained and to facilitate comparison across all samples, the number of features annotated in each superclass was gathered by species.

The 3,120 PI and 1,049 NI features were associated with 28 and 14 superclasses, respectively (all the 14 NI superclasses were also annotated in PI). For each sample, the feature intensities of each superclass were summed, assessing the occurrence of the superclass in each organism. To obtain a clearer understanding of the results, the proportion of chemical superclass annotations per species was displayed as a heatmap in the PI mode that was supplemented specifically for fatty acids, shikimates, and phenylpropanoids in the NI mode. For ease of reading, only superclasses whose sum area is greater than 5E5 were kept in the heatmap ( Figure 2 ). In addition, a bar plot representing the number of annotated vs. unannotated features per species is displayed, indicating the richness of metabolites in each species and their annotation ratio.

The interpretation of these heatmap results and comparison across mosses and liverworts together with a detailed discussion of specific clusters of the network are provided in section 2.2.

Most of the results obtained are based on spectral comparisons using computational approaches. Before entering a more detailed discussion of the MN and the specific annotation of metabolites, targeted isolation of compounds was performed to obtain a maximum number of standards for unambiguous identification of features in the molecular networking (nodes).

The nodes corresponding to isolated standards were used as anchor points to check the consistency of annotations within MN clusters and for annotation propagation purposes (Nothias et al., 2020).

To this end, five species of interest were selected based on their MN, heatmap, and the availability of plant material, and therefore extract, available in the collection. Specifically, two liverworts, F. tamarisci and Plagiochila porelloides, known to produce terpenoids such as eudesmanolides (Pannequin et al., 2017) and plagiochilines, respectively (Pannequin et al., 2023), which are strongly represented in clusters P1 and P2 ( Figures 1 , 2 ), were chosen for the targeted isolation of the sesquiterpenoids. Pellia epiphylla is another liverwort producing specific stilbenoids and was selected for the isolation of perrottetin-type bis-bibenzyls (Cullmann et al., 1997).

Finally, as mosses have been little studied in the literature, two mosses, Antitrichia curtipendula and Dicranum scoparium, were selected to complete the selection. The metabolite profiling results indicated that they produce flavonoids and phenanthrenes.

To isolate the main constituents of these five bryophyte species, a generic isolation approach based on a one-step fractionation of the extracts using high-resolution semi-prep HR-HPLC was developed.

Before separation and to maximize separation efficiency, the extracts of the five selected species were subjected to a protocol allowing significant secondary metabolite enrichment. This “sample enrichment” protocol combined a two-step liquid–liquid extraction followed by SPE enrichment of the medium polarity partition (see Supplementary Figure S3 ). This protocol enables sugars and fatty acids to be removed efficiently on a gram scale from the crude extract.

For example, the “sample enrichment” protocol applied to 1 g of crude MeOH extract of D. scoparium (moss) yielded 40 mg of enriched extract (i.e., a loss of 96% of the initial mass) ( Supplementary Figure S3 ). The enriched secondary metabolites represented less than 0.4 mg per g of dry plant material (0.04%) (for all extracts, see Supplementary Table S1 ).

The chromatographic conditions were first optimized on a reversed-phase C18 column on an HPLC–photodiode array (PDA)–evaporative light scattering detector (ELSD) instrument at the analytical scale. This optimized gradient was geometrically transferred to the semi-preparative HPLC scale (Guillarme et al., 2008). The enriched extracts were injected using a dry loading method to maintain high chromatographic resolution (Queiroz et al., 2019).

For ease of understanding, the code of each molecule consists of the acronym of the origins (Ac, A. curtipendula; Ds, D. scoparium; Ft, F. tamarisci; Pe, P. epiphylla; Pp, P. porelloides; and St, standard origin) followed by a number.

In the case of D. scoparium, four compounds were isolated and fully characterized by HRMS, NMR, and UV. Among them, three flavanones previously reported in D. scoparium were identified: apigenin 7-O-[2,4-di-O-(α-l-rhamnopyranosyl)]-β-d-glucopyranoside (Becker et al., 1986), Ds_1; 7-[(O-6-Deoxy-α-l-mannopyranosyl-(1→2)-O-[6-deoxy-α-l-mannopyranosyl-(1→4)]-β-d-glucopyranosyl)oxy]-5-hydroxy-2-(3-hydroxy-4-methoxyphenyl)-4H-1-benzopyran-4-one (Osterdahl, 1978), Ds_2; and kaempferol-3-β-d-(6-O-trans-p-coumaroyl)glucopyranoside (Tsukamoto et al., 2004), Ds_3. A new phenanthrene was also isolated (Ds_4), and the ¹H NMR spectrum showed the presence of three aromatic groups. A 1,3,4-trisubstituted benzene was identified from the aromatic protons at δ_H 6.93 (1H, dd, J = 9.2, 2.7 Hz, H-6), 7.12 (1H, d, J = 2.7 Hz, H-8), and 9.17 (1H, d, J = 9.2 Hz, H-5). A tetrasubstituted one was characterized by the two meta-coupled protons at δ_H 6.87 (1H, d, J = 2.6 Hz, H-3) and 7.29 (1H, d, J = 2.6 Hz, H-1). A singlet at δ_H 7.06 (1H, s, H-9) belongs to the third cycle. The Rotating-frame Overhauser Effect (ROE) correlations from the methoxy group at δ_H 3.89 to H-1 and H-3 positioned it in C-2; the one at δ_H 4.04 was placed in C-4 thanks to its correlation to H-3, while the third one at δ_H 4.02 was in C-10 due to its correlation with H-1 and H-9. The ROE spectroscopy (ROESY) from the hydroxyl at δ_H 9.54 to H-6 and H-8 located it at C-7. The heteronuclear multiple-bond correlations (HMBC) confirmed the identification of Ds_4 as 7-hydroxy-2,4,10-trimethoxyphenanthrene. The structure of Ds_4 with carbon numbering is shown in Figure 4 , which illustrates all new structures.

The same method was applied to the other species. From A. curtipendula, the procedure enabled the isolation and identification of a previously undescribed phenanthrene. The ¹H NMR spectrum showed the presence of a 1,3,4-trisubstituted benzene characterized by the three aromatic protons at δ_H 7.20 (1H, dd, J = 9.0, 2.8 Hz, H-3), 7.33 (1H, d, J = 2.8 Hz, H-1), and 8.33 (1H, d, J = 9.0 Hz, H-4); two ortho-coupled protons (J = 8.8 Hz) at δ_H 7.49 and 7.59, which belong to the second ring; and two singlets at δ_H 7.18 and 7.89 positioned therefore in para form the third ring. The HMBCs from the two singlets, H-5 and H-8, to the carbons at δ_C 145.9 (C-7) and 147.0 (C-6) indicated the presence of hydroxy groups in these positions. The correlations from H-4 and the methoxy group at δ_H 3.88 to the carbon at δ_C 156.9 (C-2) placed the methoxyl in C-2. The ROESY correlations from H-1 to H-10, H-9 to H-8, and H-5 to H4 and from the methoxyl to H-1 and H-3 confirmed that Ac_1 was a 6,7-dihydroxy-2-methoxyphenanthrene.

Six compounds have been isolated from the liverwort F. tamarisci, five of which correspond to eudesmanolide sesquiterpenes and one to the known triterpene ursolic acid Ft_6. Three eudesmanolide derivatives were known in the Frullania genus: oxy-frullanolide Ft_2 was previously isolated in Frullania dilatata (Asakawa et al., 1981), and frullanolide Ft_3 and γ-cyclocostunolide Ft_4 were commonly reported as F. tamarisci components (Pannequin et al., 2017).

A new cis-fused isomer of oxo-cyclocostunolide Ft_1 was characterized. The cis configuration of the lactone was in agreement with the ³ J _H6-H7 coupling constant of 6 Hz, whereas it was 10 Hz in the trans-fused lactone-like oxo-cyclocostunolide (Nadgouda et al., 1978). The ¹³C chemical shift values of CH-6 and CH-7 were also a good indicative of the cis or trans configuration: they were reported in trans-eudesmanolides like arbusculin A (Fan et al., 2016) at δ_C 82.2 and 51.4, respectively, and in cis series like frullanolide (Ft_3) at δ_C 76.1 and 41.4, respectively. In Ft_1, the CH-6 and CH-7 carbons were observed at δ_C 75.6 and 40.6, respectively, and confirmed the identification as oxo-frullanolide.

Interestingly, a new eudesmanolide dimer Ft_5 with close similarities to muscicolide A previously described in Frullania musciola (Kraut et al., 1994) was also identified. In HRMS, the spectrum of Ft_5 did not show [M+H]⁺, but the ions at m/z 465.3061 and m/z 500.3391 corresponded to [M–H₂O+H]⁺ and [M+NH₄]⁺, respectively. An ion at m/z 233.1542 appears in the HRMS spectrum and MS/MS spectra. This can be corresponded to monomeric fragments of Ft_5. Due to the high overlap of NMR signals of each monomer, the relative configuration of Ft_5 was difficult to assign. However, the NOE correlations from H-6 at δ_H 4.05 to H₃-14 at δ_H 1.09 and H₃-15 at δ_H 1.30 indicated the axial position of these protons and that they were on the same side. For the same reasons that explained those previously for oxo-frullanolide Ft_1, the ¹³C chemical shift values of CH-6 and CH-7 (and CH-6′ and CH-7′) at δ_C 82.2 and 49.6 (and 82.2 and 51.4), respectively, indicated that H-6 and H-7 (and H-6′ and H-7′) were in a trans configuration. The ¹³C NMR chemical shift values of CH-5 and CH₃-14 were also a good indicative of their relative configuration since they were observed in muscicolide A at δ_C 55.2 and 19.9 for CH-5 and CH₃-14, respectively (H-5 and H₃-14 being trans), and 49.3 and 21.2 for CH-5′ and CH₃-14′, respectively (H-5′ and H₃-14′ being trans), and in muscicolide B at δ_C 55.4 and 18.3 for CH-5 and CH₃-14, respectively (H-5 and H₃-14 being trans), and 45.5 and 32.0 for CH-5′ and CH₃-14′, respectively (H-5′ and H₃-14′ being cis) (Kraut et al., 1994). In Ft_5, the CH-5 and CH₃-14 (CH-5′ and CH₃-14′) were observed at δ_C 59.8 and 19.2 (53.4 and 21.6), respectively, indicating a trans configuration of H-5 and H₃-14 as well as H-5′ and H₃-14′. The ¹³C chemical shift values of CH₃-15 (δ_C 24.7) and CH₃-15′ (δ_C 24.1) indicated that they were both in an axial configuration. Indeed, in 4-(6-hydroxy-12-oxo-11(13)-eudesmen-4-yloxy)-11(13)-eudesmen-12,6-olide, a 4-O-4′ dimer between 4-epi-arbusculin A and its open form at the lactone, isolated from the liverwort F. tamarisci (Toyota et al., 1998), the axial methyl CH₃-15 was observed at δ_C 22.7 and the equatorial CH₃-15′ at δ_C 31.7. Finally, the multiplicity of H-1 (d, J = 10.4 Hz) has determined its axial position. Ft_5 was a new eudesmanolide dimer that was consequently named tamariscolide A.

This targeted isolation procedure was applied to the liverwort P. porelloides to yield various sesquiterpenes. Plagiochiline derivatives were isolated, which are characteristic compounds of the genus. Among them, plagiochiline D Pp_1 (Nagashima et al., 1994), plagiochiline R-15-yl octanoate Pp_2, and plagiochiline R-15-yl dec-4-enoate Pp_3 (Toyota et al., 1994) were previously described in this species, and plagiochiline R-15-yl hexanoate Pp_4 has not yet been described. Complete structural elucidation of Pp_4 was carried out by extensive NMR analysis; it should be noted that the NMR data of Pp_4 were similar to those of Pp_2 except for the lateral carbon chain. Complementarily, the eudesmanolide, diplophyllin Pp_5, previously described in the genus Plagiochila (Spörle et al., 1991), was also isolated.

Finally, the same chromatographic procedure applied to the liverwort, P. epiphylla, allowed the isolation of four perrottetin-type bis-bibenzyls already known in the species: 10′-hydroxyperrottetin E Pe_1, perrottetin E Pe_2, 10′-hydroxy-11-methoxy-perrottetin E Pe_3, and 11-methoxy-perrottetin E Pe_4 (Cullmann et al., 1997).

These 20 isolated fully characterized compounds, as well as 12 commercial standards and three compounds previously isolated from Japanese liverworts, were analyzed under the same conditions used for extract metabolomic profiling.

These 35 standards (25 terpenoids, including 18 sesquiterpenoids, three diterpenoids, and four triterpenoids, as well as five bis-bibenzyl stilbenoids, three flavonoids, and two new phenanthrenes) enabled a formal identification of the corresponding nodes and propagation of the annotation in 11 clusters in PI and three in NI. These structures, related to MS and NMR data, are summarized in Supplementary Table S2 and Supplementary S6 .

To obtain a good overview of the distribution of chemical superclasses in the different samples, the data were summarized in the form of a heatmap. The heatmap displays the sum intensities of all features corresponding to a given superclass ( Figure 2 , see Section 2.1.3).

The chemical superclasses were divided according to their pathway (alkaloids, fatty acids, terpenoids and shikimates, and phenylpropanoids). They are represented vertically and the species concerned horizontally according to their taxonomic proximity within two sections: liverworts in the upper part and mosses in the bottom part. A color gradient, from red to yellow in PI and dark blue to light blue in NI, represents the superclass according to their abundance.

The number of features detected by species, which is a measure of the extent of the chemodiversity, varies greatly from one species to another, independently of their phylum. These variations for the PI mode are represented in a bar plot in Figure 2 . The species with the most features detected are the liverworts P. arboris-vitae, Marsupella emarginata, and P. obtusata with more than 1,461, 1,182, and 1,167 features, respectively, and they are followed by one moss, Brachythecium rivulare, with 990 features. Species with fewer features detected are Pellia endiviifolia and Trichostomum crispulum with only 18 and 43 features, respectively. This variation in detected features indicates an important variability among species and no real trend between phyla. Overall, SIRIUS was able to annotate more than 80% of the features (in light gray) with at least one chemical superclass.

The heatmap in the PI mode ( Figure 2 ) differentiates between liverworts (upper panel) and mosses (lower panel). Liverworts are mainly characterized by the presence of abundant sesquiterpenoids. Within liverwort, however, most of the thallus liverworts displayed stilbenoids principally detected in the NI mode and, in general, fewer sesquiterpenoids.

The mosses are marked by the presence of fatty acids, which were also evidenced by the extraction yield during the cleaning process. Concerning terpenoids, and in comparison with liverworts, di- and tri-terpenoids were often detected. Sesquiterpenoids were also present but were, in general, less abundant. Mosses were also found to contain flavonoids mainly highlighted in the NI mode.

In general, compounds from the shikimate and phenylpropanoid pathway are poorly represented, with the exception of stilbenoids. Those are found in liverworts but were also surprisingly detected in three moss species of the family Polytrichaceae.

Remarkably, of all the samples, only one bryophyte, the liverwort Riccia perrenis, produced alkaloids and, at the same time, biosynthesized only a few terpenes.

More detailed information could be retrieved from the MN. In PI, the MN generated 370 clusters of more than two nodes and 4,368 single nodes (Figure 1A). Particular attention was paid to the main cluster, which generally groups together compounds of the same structural and indicates the main trend in chemical composition. The SIRIUS annotation shows that the first 16 clusters (first two lines of the MN) correspond to terpenoids (eight clusters: P1, P2, P3, P5, P9, P10, P11, and P12), fatty acids (three clusters: P4, P7, and P15), and stilbenoids (two clusters: P6 and P13). As suggested by the heatmap, the MN analysis shows that the most important clusters common to mosses and liverworts are annotated by SIRIUS as terpenoids (e.g., P1 and P9) and fatty acids (e.g., P4). In NI, the MN is made up of 4,604 nodes grouped into 190 clusters of more than two nodes and 2,245 single nodes (Figure 1B). In this mode, the annotations were less reliable, except for stilbenoids and flavonoids, which ionize better. MN reveals a few homogeneous clusters from the shikimate and phenylpropanoid pathway (e.g., N60) and some flavonoids (e.g., N9). The annotations for the other clusters were more heterogeneous and therefore difficult to use at this level.

To obtain a more detailed view of the metabolite composition of the bryophyte species in the collection, the annotations will be analyzed by compound class, starting with those with the greatest number of features.These data will be presented in relation to the taxonomic relationships that exist between species, particularly between the two phyla, and will be discussed in greater detail in light of the published data in the Discussion section.

Samples of the 60 bryophytes were collected at random locations in Corsica. Sampling was performed in 2021. Botanical determination was performed according to the botanical determination keys summarized in Bryophyte Flora (Augier, 1966) by Achille Pioli, a specialist in mosses, and voucher specimens were deposited in the herbarium of the University of Corsica, Corte (France). Three Japanese liverworts collected as part of a collaboration with the Tokushima Bunri University, Japan, were added to the collection in Tokushima. The sample numbers, the geographical origin of the different samples, and the voucher codes for each specimen analyzed are listed in Supplementary Table S1 . All samples were harvested in the wet season of 2021 (January–April and September–December 2021).

After 15 days of drying at room temperature, plant material was powdered by cryo-grinding and successively extracted with hexane, methylene chloride, and methanol for 24 h each. Solvent volume corresponded to v = mass of sample × 100. The extracts were dried using a rotary evaporator or lyophilization. The masses and the yields obtained are summarized in Supplementary Table S1 .

The solid-phase extraction procedure used C18 cartridges (50 µm, 12 mL, 1,000 mg; Finisterre, Teknokroma, Spain) in conjunction with a Teknokroma extraction manifold system (12-position manifold with a 13 × 75 mm test tube rack). The vacuum pressure on the manifold was maintained at ≤5 inches (12.7 mmHg) throughout the duration of the SPE protocol. The cartridges were conditioned using 10 mL of methanol–water (50:50) and 10 mL of pure methanol.

A sample with a mass of 50 mg was dissolved in 5 mL of mixture MeOH:H₂O 95:5. Analytes were eluted with 10 mL methanol–water (95:5). Eluates were evaporated using Genevac. The samples were reconstituted in methanol at 5 mg·mL⁻¹, and they were transferred into a microplate for LC–MS/MS analysis.

The selected extracts were dissolved in a mixture of MeOH/H₂O (7:3) at 5 mg·mL⁻¹. They were extracted by liquid–liquid extraction (LLE) with hexane to equal volume. The two phases were separated, dried, and evaporated. The extract obtained from the organic phase (MeOH/H₂O) was dissolved in butanol at 5 mg·mL⁻¹ and extracted by LLE with water to equal volume. After drying and evaporation, the extract obtained from the organic phase was cleaned by SPE.

SPE C18-cartridges (50 µm, 12 mL, 1,000 mg; Finisterre, Teknokroma, Spain) were conditioned with a 10-mL mixture of methanol–water (50:50) and 10 mL water pure. Samples were mixed with deactivated silica and disposed of on the head of the SPE cartridge after conditioning. Samples passed into the cartridges at a flow rate of approximately 5 mL/min under vacuum. Analytes were eluted successively with 10 mL of methanol–water (90:10), 10 mL of water, and 10 mL of ethyl acetate. Eluates were evaporated using Genevac.

Analysis and chromatographic data were obtained on an ultra-high-performance liquid chromatography system equipped with a photodiode array, an evaporative light-scattering detector, and a single quadrupole detector using heated electrospray ionization (UHPLC–PDA–ELSD–QDA) (Waters, Milford, MA, USA). The ESI parameters were as follows: capillary voltage 800 V, cone voltage 15 V, source temperature 120°C, and probe temperature 600°C. The acquisition was performed in positive ionization mode with an m/z range of 150–1,000 Da. The chromatographic separation was performed on an Acquity UPLC BEH C18 column (50 × 2.1 mm i.d., 1.7 μm; Waters) at 0.6 mL/min, 40°C with H₂O (A) and MeCN (B), both containing 0.1% formic acid as solvents. The gradient was carried out as follows: 5%–100% B in 7 min, 1 min at 100% B, and a re-equilibration step at 5% B for 2 min. The ELSD temperature was fixed at 45°C, with a gain of 9. The PDA data were acquired from 190 to 500 nm, with a resolution of 1.2 nm. The sampling rate was set at 20 points/s.

Analysis, data processing, and feature-based molecular network generation chromatographic data with high-resolution MS were obtained on a Waters Acquity UHPLC system equipped with a Q-Exactive Focus mass spectrometer (Thermo Scientific, Bremen, Germany), using heated electrospray ionization source (HESI-II). The chromatographic separation was carried out on an Acquity UPLC BEH C18 column (50 × 2.1 mm i.d., 1.7 μm; Waters) at 0.6 mL/min, 40°C with H₂O (A) and MeCN (B), both containing 0.1% formic acid as solvents. The gradient was carried out as follows: 5%–100% B in 7 min, 1 min at 100% B, and a re-equilibration step at 5% B in 2 min. The ionization parameters were the same as those used in Rutz et al. (2019).

The raw UHPLC–HRMS/MS files were converted into mzXML files using the MSConvert software. The mzXML files were then processed using the open software MZmine (3.4.16) (Schmid et al., 2023). The mass detection step was performed using a centroid mass detector with a noise level set at 1E⁶ for MS1 and 1E⁴ for MS² in PI and 1E⁴ for MS1 and 1E⁴ in MS² in NI. The ADAP chromatogram builder was employed with a minimum group size of scans of 4, a minimum group intensity threshold of 1E⁶, a minimum highest intensity of 1E⁶ in PI and 5E⁴ in NI, and an m/z tolerance of 10 ppm. The deconvolution was carried out with the ADAP (Wavelets) algorithm, using a signal-to-noise threshold of 50, a minimum feature height of 1E⁶, a coefficient/area threshold of 100, a peak duration range of 0.01–0.9 min, and a wavelet range between 0.01 and 0.08 min. The m/z and retention time (RT) for MS² scan pairing were, respectively, set to 0.005 Da and 0.1 min. The isotopes were grouped using the isotope peak grouper algorithm with an m/z tolerance of 3 ppm, an RT tolerance of 0.05 min, and a maximum charge of 2, using the most intense isotope as the representative one. The alignment was carried out with the join aligner with an m/z tolerance of 15 ppm, an RT tolerance of 0.1 min, and a weight tolerance for m/z and RT of 10 each. Ion identity networking parameters were set to m/z tolerance, 0.002 m/z or 5 ppm; check, one feature; min height, 1E³ with ion identity library parameters set to MS mode, positive; maximum charge, 2; maximum molecules/cluster, 2; adducts, M+H, M+Na, M+K; modifications, M–H₂O, M–NH₃.

The MZmine aligned table was exported in MGF format for the processing of the Feature-based Molecular Networking (FBMN). The spectral data were uploaded on the GNPS platform (Nothias et al., 2020). A network was generated with a minimum cosine score of 0.85 and a minimum of five matching peaks. The experimental spectra were searched against GNPS’s spectral libraries. The obtained network was visualized in the software Cytoscape (3.9.1, Institute for Systems Biology, Seattle, WA, USA) (Smoot et al., 2011).

The mass spectrometry data were deposited on the MassIVE public repository nos. MSV000093186 (PI) and MSV000093188 (NI) with different GNPS job parameters, and resulting data are available at the following addresses:

PI network: ID=895c9f23e6df4c30a42d8774af4121c1.

NI network: ID=5e783c91e5e74ed9aee07cc59fc12830.

Filtered features detected in PI and NI in the general dataset were annotated using a computational approach integrating SIRIUS (molecular formula) (Ludwig et al., 2020), CSI:fingerID (probabilistic molecular fingerprint by machine learning substructure prediction and in silico annotation) (Dührkop et al., 2015), and CANOPUS (systematic class annotation) (Dührkop et al., 2019, 2021). In the second step, the spectral file and attribute metadata obtained after the MN step were annotated by matching the MS1 and MS² spectra data with the LOTUS-ISDB [in-house database containing the in silico fragmentation spectra of all the compounds present in the Dictionary of Natural Products (DNP) and LOTUS databases] complemented with structure–organism pairs coming from the DNP (Allard et al., 2016; Rutz et al., 2019). The following parameters were used: spectral match parameters: parent mass tolerance 0.01 Da, MS/MS tolerance 0.01 Da, minimum cosine score 0.2, and minimum peaks 6. Spectral match of MS/MS spectra against the database provided a list of 50 chemical structure candidates for every feature. The candidates were re-ranked by taxonomic reweighting after ponderation of their spectral score inversely proportional to the taxonomic distance between the biological source of the candidate and that of the one-off analyzed sample(s) in which the feature is detected.

The separation conditions of the DCM or MeOH extracts were optimized on an HP 1260 Agilent high-performance liquid chromatography equipped with a photodiode array detector and an ELSD detector (HPLC–PDA–ELSD) (Agilent Technologies, Santa Clara, CA, USA). The chromatographic separation was performed on an XBridge C18 column (250 × 4.6 mm i.d., 5 μm; Waters) equipped with a C18 pre-column at 1 mL/min, with H₂O (A) and MeCN (B), both containing 0.1% formic acid as solvents. The UV absorbance was measured at 280 and 360 nm, and UV–Vis spectra were recorded between 190 and 600 nm (step 2 nm). The optimized gradient used for the F. tamarisci DCM extracts and P. porelloides MeOH extracts was as follows: 5 mn at 30% B, 30%–100% B in 45 min, and 10 min at 100% B. The optimized gradient P. epiphylla DCM extract used for the was as follows: 5 mn at 35% B, 35%–100% B in 45 min, and 10 min at 100% B. The optimized gradient A. curtipendula MeOH extract used for the was as follows: 5 mn at 15% B, 15%–75% B in 45 min, and 10 min at 100% B. The optimized gradient D. scoparium MeOH extract used for the was as follows: 5 mn at 15% B, 15%–75% B in 45 min, and 10 min at 100% B.

These chromatographic methods were geometrically transferred (Guillarme et al., 2008) to the semi-preparative scale on a Shimadzu system equipped with an LC-20, module pumps, an SPD-20 A UV/VIS, a 7725I Rheodyne^® valve, and an FRC-40 fraction collector (Shimadzu, Kyoto, Japan). The separation was performed on an XBridge C18 column (250 mm × 19 mm i.d., 5 μm; Waters) equipped with a C18 pre-column cartridge holder (10 mm × 19 mm i.d., 5 μm; Waters) at 17 mL/min, with H₂O (A) and MeCN (B) both containing 0.1% formic acid as solvents. The UV detection was set at 280 and 360 nm. The mixtures were injected into the semi-preparative HPLC column using a dry-load methodology developed in our laboratory (Queiroz et al., 2019). Masses injected and fractionation details are summarized in Supplementary Figure S5 .

A Bruker Avance Neo 600 MHz NMR spectrometer equipped with a QCI 5 mm Cryoprobe and a SampleJet automated sample changer (Bruker BioSpin, Rheinstetten, Germany) was employed for 1D-NMR (¹H and ¹³C-NMR) and 2D-NMR [correlation spectroscopy (COSY), multiplicity editing heteronuclear single-quantum correlation (edited-HSQC), HMBC, ROESY, and total correlation spectroscopy (TOCSY)] spectroscopy. Chemical shifts are reported in parts per million (δ) using the residual solvent signal at δ_H 7.26; δ_C 77.2 for CDCl₃, δ_H 3.31; δ_C 49.0 for CD₃OD and δ_H 2.50; and δ_C 39.5 for DMSO-d ₆. Chemical shifts (J) are reported in Hz.

Metabolite standards were purchased from Biopurify (Chengdu, China): alantolactone St_1, isoalantolactone St_2, parthenolide St_4, costunolide St_5, dehydrocostus lactone St_6, artemisinin St_7, cinnamolide St_8, tanshinone I St_10 and IV St_11, betulin St_12, betulinic acid St_13, and celastrol St_14. Telekin St_3, clerod-3,13(16),14-trien-17-oic acid St_9, and riccardin G St_15 were isolated by Prof. Nagashima and Prof. Asakawa from Japan liverworts.

The NMR descriptions of known compounds (Ft_2, Ft_3, Ft_4, Ft_6, Pp_1, Pp_2, Pp_3, Pp_5, Pe_1, Pe_2, Pe_3, Pe_5, Ds_1, Ds_2, and Ds_3) are summarized in the Supplementary Material. All spectra are according to the literature. The NMR spectra of unknown compounds (Ft_1, Ft_5, Pp_4, Ds_4, and Ac_1) are summarized in the Supplementary Material ( Supplementary Figures S7 - S33 ).

Six compounds were obtained from F. tamarisci (the yields for each compound are the sum of the compound isolated from the two extracts) in DCM extract from 20 mg [Ft_1 (1 mg)] and in MeOH Extracts Ft_2 (2.8 mg), Ft_3 (6.1 mg), Ft_4 (6.2 mg), Ft_5 (3.9 mg), and Ft_6 (3 mg)].

Ft_1: oxo-frullanolide ¹H NMR (CDCl₃, 600 MHz) δ 1.28 (3H, s, H₃-14), 1.42 (1H, overlapped, H-9ax), 1.60 (1H, overlapped, H-9eq), 1.68 (1H, overlapped, H-8ax), 1.70 (1H, overlapped, H-1eq), 1.80 (1H, overlapped, H-8eq), 1.90 (3H, s, H-15), 1.97 (1H, td, J = 14.2, 13.7, 5.1 Hz, H-1ax), 2.52 (1H, dt, J = 12.7, 5.1 Hz, H-2eq), 2.73 (1H, m, H-2ax), 3.17 (1H, td, J = 7.2, 6.4 Hz, H-7), 5.37 (1H, d, J = 6.4 Hz, H-6), 5.71 (1H, s, H-13″), 6.30 (1H, s, H-13′); ¹³C NMR (CDCl₃, 151 MHz) δ 11.1 (CH₃-15), 24.7 (CH₃-14), 25.4 (CH₂-8), 34.2 (CH₂-2), 34.5 (C-10), 36.5 (CH₂-9), 37.5 (CH₂-1), 40.6 (CH-7), 75.6 (CH-6), 122.6 (CH₂-13), 136.8 (C-4), 140.5 (C-11), 151.7 (C-5), 170.3 (C-12), 198.8 (C-3); HR-ESI/MS, see Supplementary Table S2 .

Ft_2: oxy-frullanolide (Sangsopha et al., 2016) HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Ft_3: frullanolide (Chou and Liao, 2013) HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Ft_4: γ-cyclocostunolide (Kraut et al., 1994) HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Ft_5:tamariscolide ¹H NMR (CDCl₃, 600 MHz) δ 1.09 (3H, s, H₃-14), 1.10 (3H, s, H₃-14′), 1.12 (1H, overlapped, H-1′b), 1.30 (3H, s, H₃-15), 1.31 (3H, s, H₃-15′), 1.37 (1H, m, H1′-a), 1.41 (2H, m, H₂-9′), 1.47 (4H, m, H-2′b, H-2b, H-3′a, H-8b), 1.53 (1H, m, H-3′b), 1.59 (1H, m, H-2′a), 1.63 (2H, m, H-3b, H-8′b), 1.70 (1H, m, H-9b), 1.77 (1H, m, H-2a), 1.80 (1H, m, H-3a), 1.94 (1H, d, J = 10.5 Hz, H-5′), 1.98 (1H, m, H-8′a), 2.00 (1H, m, H-5), 2.05 (1H, d, J = 13.4 Hz, H-8a), 2.12 (1H, d, J = 10.4 Hz, H-1), 2.27 (1H, d, J = 12.2 Hz, H-9a), 2.64 (2H, brs, H-7, H-7′), 4.05 (2H, m, H-6, H-6′), 5.38 (1H, s, H-13′b), 5.43 (1H, s, H-13b), 6.08 (1H, s, H-13′a), 6.10 (1H, s, H-13a); ¹³C NMR (CDCl₃, 151 MHz) δ 18.3 (CH₂-2′), 19.2 (CH₃-14), 21.6 (CH₃-14′), 22.5 (CH₂-8′), 22.7 (CH₂-8), 23.1 (CH₂-2), 24.1 (CH₃-15′), 24.7 (CH₃-15), 35.0 (CH₂-3′), 38.3 (C-10′), 40.8 (CH₂-3), 41.0 (CH₂-1′), 42.0 (CH₂-9), 42.5 (C-4′), 45.2 (CH₂-9′), 45.4 (C-10), 49.6 (CH-7), 51.4 (CH-7′), 53.4 (CH-5′), 55.8 (CH-1), 59.8 (CH-5), 71.8 (C-4), 82.2 (CH-6′), 82.4 (CH-6), 117.3 (CH₂-13′), 118.1 (CH₂-13), 138.7 (C-11), 139.7 (C-11′), 169.8 (C-12), 170.3 (C-12′); HR-ESI/MS, see Supplementary Table S2 .

Ft_6:ursolic acid (Acebey-Castellon et al., 2011): HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Five compounds were obtained from P. porelloides: (the yields for each compound are the sum of the compound isolated from the two extracts) in MeOH extract from 100 mg: Pp_1 (9.2 mg), Pp_2 (3.4 mg), Pp_3 (3.6 mg), Pp_4 (3.6 mg), and Pp_5 (1.3 mg).

Pp_1: plagiochiline D (Asakawa et al., 1979): HR-ESI/MS see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Pp_2: plagiochiline R-15-yl octanoate (Toyota et al., 1994; Ramírez et al., 2017): HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Pp_3: plagiochiline R-15-yl dec-4-enoate (Toyota et al., 1994; Ramírez et al., 2017): HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Pp_4: plagiochiline R-15-yl hexanoate ¹H NMR (DMSO-d ₆, 600 MHz) δ 0.85 (3H, t, J = 7.1 Hz, H₃-6′), 0.99 (1H, t, J = 10.0 Hz, H-6), 1.07 (1H, m, H-9α), 1.20 (1H, m, H-8α), 1.25 (4H, m, H₂-5′, H₂-4′), 1.30 (1H, m, H-7), 1.51 (2H, p, J = 7.4 Hz, H₂-3′), 1.62 (1H, dd, J = 10.0, 3.2 Hz, H-1), 1.99 (1H, m, H-9β), 2.00 (3H, s, H₃-14b), 2.00 (1H, overlapped, H-8β), 2.03 (3H, s, H₃-12b), 2.11 (3H, s, H₃-2b), 2.27 (1H, dd, J = 9.7, 3.6 Hz, H-5), 2.28 (2H, td, J = 7.4, 2.2 Hz, H₂-2′), 2.40 (2H, AB, H-11), 3.74 (1H, d, J = 11.4 Hz, H-15″), 4.05 (1H, d, J = 11.4 Hz, H-15′), 4.21 (2H, AB, H₂-14), 4.42 (1H, d, J = 12.4 Hz, H-12″), 4.56 (1H, dd, J = 12.4, 1.4 Hz, H-12′), 6.46 (1H, s, H-3), 6.67 (1H, d, J = 10.2 Hz, H-2); ¹³C NMR (DMSO-d ₆, 151 MHz) δ 13.8 (CH₃-6′), 20.6 (CH₃-14b), 20.7 (CH₂-8), 20.8 (CH₃-2b, CH₃-12b), 21.7 (CH₂-5′), 24.1 (CH₂-3′), 20.9 (CH₃-2b), 24.3 (CH-7), 25.7 (C-13), 27.5 (CH-6), 29.7 (CH-5), 30.5 (CH₂-4′), 33.4 (CH₂-2′), 33.5 (CH₂-9), 49.1 (CH-1), 51.2 (CH₂-11), 59.2 (C-10), 60.9 (CH₂-14), 62.3 (CH₂-12), 68.6 (CH₂-15), 91.1 (CH-2), 116.0 (C-4), 139.6 (CH-3), 169.2 (C-2a), 170.5 (C-12a), 170.6 (C-14a), 172.8 (C-1′); HR-ESI/MS, see Supplementary Table S2 .

Pp_5: diplophyllin (Ohta et al., 1977): HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Four compounds were obtained from P. epiphylla (the yields for each compound are the sum of the compound isolated from the two extracts) in DCM extract from 60 mg: Pe_1 (15.2 mg), Pe_2 (4.8 mg), Pe_3 (3.3 mg), and Pp_4 (2.3 mg).

Pe_1: 10-hydroxyperrottetin E (Cullmann et al., 1997), HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Pe_2: perrottetin E (Cullmann et al., 1997), HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Pe_3: 10-hydroxy-11-methoxy-perrottetin E, HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Pe_4: 11-methoxy-perrottetin E, HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Four compounds were obtained from D. scoparium (the yields for each compound are the sum of the compound isolated from the two extracts) in DCM extract from 40 mg: Ds_1 (0.5 mg), Ds_2 (0.7 mg), Ds_3 (0.5 mg), and Ds_4 (0.5 mg).

Ds_1: apigenin 7-O-[2,4-di-O-(α-l-rhamnopyranosyl)]-β-d-glucopyranoside (Becker et al., 1986), HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Ds_2: 7-[(O-6-deoxy-α-l-mannopyranosyl-(1→2)-O-[6-deoxy-α-l-mannopyranosyl-(1→4)]-β-d-glucopyranosyl)oxy]-5-hydroxy-2-(3-hydroxy-4-methoxyphenyl)-4H-1-benzopyran-4-one (Osterdahl, 1978), HR-ESI/MS, see Supplementary Table S2 ; ¹H and ¹³C NMR data, see Supplementary Material S1 .

Ds_3: tiliroside = kaempferol-3-β-d-(6-O-trans-p-coumaroyl)glucopyranoside (Tsukamoto et al., 2004), HR-ESI/MS: see Supplementary Table S2 , ¹H and ¹³C NMR data, see Supplementary Table S2 .

Ds_4: 7-hydroxy-2,4,10-trimethoxyphenanthrene ¹H NMR (DMSO-d ₆, 600 MHz) δ 3.89 (3H, s, 2OCH₃), 4.02 (3H, s, 10OCH₃), 4.04 (3H, s, 4OCH₃), 6.87 (1H, d, J = 2.6 Hz, H-3), 6.93 (1H, dd, J = 9.2, 2.7 Hz, H-6), 7.06 (1H, s, H-9), 7.12 (1H, d, J = 2.7 Hz, H-8), 7.29 (1H, d, J = 2.6 Hz, H-1), 9.17 (1H, d, J = 9.2 Hz, H-5), 9.54 (1H, s, 7OH); ¹³C NMR (DMSO-d ₆, 151 MHz) δ 55.1 (2OCH₃), 55.5 (10OCH₃), 55.8 (4OCH₃), 95.0 (CH-1), 99.8 (CH-3), 103.4 (CH-9), 110.5 (CH-8), 114.3 (CH-6), 116.0 (C-12), 118.9 (C-13), 127.5 (C-11), 128.6 (CH-5), 133.7 (C-14), 152.2 (C-10), 154.9 (C-7), 157.1 (C-2), 158.7 (C-4); HR-ESI/MS, see Supplementary Table S2 .

One compound was obtained from A. curtipendula (the yields for each compound are the sum of the compound isolated from the two extracts) in MeOH extract from 100 mg: Ac_01 (0.2 mg).

Ac_01: 6,7-dihydroxy-2-methoxyphenanthrene ¹H NMR (DMSO-d ₆, 600 MHz) δ 3.88 (3H, s, 2OCH₃), 7.18 (1H, s, H-8), 7.20 (1H, dd, J = 9.0, 2.8 Hz, H-3), 7.33 (1H, d, J = 2.8 Hz, H-1), 7.49 (1H, d, J = 8.8 Hz, H-10), 7.55 (1H, d, J = 8.8 Hz, H-9), 7.89 (1H, s, H-5), 8.33 (1H, d, J = 9.0 Hz, H-4); ¹H NMR (DMSO-d ₆, 151 MHz) δ 55.1 (2OCH₃), 106.6 (CH-5), 108.3 (CH-1), 112.1 (CH-8), 116.5 (CH-3), 123.3 (CH-10), 123.7 (CH-4), 124.3 (C-13), 125.2 (C-14), 126.5 (CH-9), 131.9 (C-11), 145.9 (C-7), 147.0 (C-6), 156.9 (C-2); HR-ESI/MS, see Supplementary Table S2 .