All analytical chemistry methods used to analyze R. marginata bibenzyl cannabinoid compounds are reported in Methods S1.

Fresh samples were freeze‐dried and a subsample (20 mg) ground (Omni Bead Rupter 24) before extraction in DCM (0.5 ml) containing benzyl alcohol (0.98 mg ml⁻¹) by shaking overnight. The benzyl alcohol was a check for completeness of silylation and an internal standard for approximate quantification of bibenzyls, since separate GC analyses of each purified bibenzyl with benzyl alcohol showed similar FID responses. Extracts were filtered through cotton wool, and a subsample (20 μl) was prepared for Gas chromatography‐mass spectrometry (GC‐MS) analysis by silylation with N,O‐bis(trimethylsilyl)‐trifluoroacetamide: trimethylchlorosilane (99 : 1, 20 μl) in the presence of pyridine (20 μl). The December 2022 collections were ground while fresh and then freeze‐dried. GC‐MS analysis was performed as described in Methods S1. Cannabinoids in R. marginata samples and extracts degraded when exposed to air and/or light during drying and extraction. Plant material was freeze‐dried as soon as possible after collection and stored at −18°C until grinding immediately before sample preparation. Once in extract solution, all compounds except BB4G 10 were stable. Replicate extraction, silylation, and GC‐MS analyses from a homogenous (ground) R. marginata sample gave relative standard deviations of the cannabinoid concentrations of 2–15%.

Radula marginata plants and leaves were gently rinsed with distilled water in a minimum of three consecutive baths to remove soil residues. As R. marginata leaves are only one cell‐layer thick, slides containing leaf pieces were directly mounted in distilled water with a cover slip. Plants were then observed with an Olympus Vanox AHBT3 (Olympus Optical Co. Ltd, Tokyo, Japan) microscope using bright field or the UV filter set (excitation 330–385 nm, dichroic mirror 400 nm, emission ≥ 420 nm) for autofluorescence. Images were captured by an Olympus color camera DP74 (Olympus Optical Co. Ltd, Tokyo, Japan).

The data were log‐transformed for normalization and subsequently subjected to analyses of variance (one‐way and two‐way ANOVA) and principal component analysis (PCA). The significance of differences between means was evaluated using a pairwise multiple comparison procedure (Tukey's test). Student's t‐test was also used to compare means. The Origin software (v.2022b, OriginLab Corp., Northampton, MA, USA) was used for these analyses. PCA plot was generated using Metaboanalyst 6.0 (Pang et al., 2024).

Eight bibenzyls (Fig. 1) were purified from R. marginata collections from sites S1 and S2 (Fig. 3). Previously reported compounds 1, 2, 9, 10, 13, and 14 were identified by comparing their ¹H and ¹³C NMR data with literature values (Methods S2).

A R. marginata extract from S1 showed a major GC‐MS peak with a molecular ion of a PET 1 isomer but shorter retention time and very different fragment ion intensities (Table 1). The purified compound 5' s ¹H NMR data (Table S2; Fig. S2) differed from those of PET 1 in having two allylic methyl signals and one exocyclic methylene, vs one allylic methyl and two alkyl methyl signals for PET; and in having two very broad signals for H3′ and H5′, vs sharp meta coupled doublets in PET. The methyl signals could be explained by a monoterpene moiety without the ether ring of PET: either the ring‐open perrottetinene diol (PTD 5, Fig. 1) or a regioisomer with a limonene moiety attached ortho. Neither of these compounds has been reported previously as a natural product, but both were synthesized by Crombie et al. (1988). The R. marginata compound's NMR data (Table S2; Fig. S2) matched the data for PTD 5 (Crombie et al., 1988; Stott et al., 2021) with H3‐H4 trans. This stereochemistry means that PTD 5 from R. marginata is closely analogous to the major Cannabis compound CBD 7, which also has H3‐H4 trans and broadened NMR signals due to restricted rotation about the C3 to C1′ bond (Choi et al., 2004). The absolute stereochemistry of PET 1 from R. marginata at C4 has been shown by syntheses (Song et al., 2008; Chicca et al., 2018) to be opposite to that at C4 of THC 3 from C. sativa (Fig. 1) (Mechoulam & Gaoni, 1967) (Table S3). The absolute stereochemistry of the compound isolated from R. marginata, with optical rotation [α]_D + 39 (Table S3), is proposed as (+)‐trans‐PTD 5 shown in Fig. 1, by comparison with synthesized (+)‐ and (−)‐trans‐PTD with established stereochemistries (Stott et al., 2021). This is opposite to that established for C4 of (−)‐trans‐CBD 7 from C. sativa (Mechoulam & Hanuš, 2002), but the same as at C4 of (+)‐cis‐PET 1 from R. marginata (Fig. 1).

Atomic Pair Tanimoto coefficients were evaluated between cis‐PET and trans‐THC as well as between trans‐PTD and trans‐CBD, confirming their structural similarities with scores of 0.52 and 0.51, respectively (ChemMine Tools, Backman et al., 2011).

Having discovered PTD 5 in R. marginata, we sought the corresponding acid 6, analogous to CBDA 8. Some silylated extracts showed a GC peak with a prominent MS ion at m/z 593 (M‐CH₃) and a weak molecular ion at m/z 608, appropriate for PTD plus CO₂ and three trimethylsilyl (TMS) groups (Table 1). The compound responsible for this GC‐MS peak was isolated and comparisons of its NMR data with those of PETA 2 (Table 2; Fig. S3) showed major changes indicative of it being a ring‐open isomer. The NMR signals for the monoterpene and benzoic acid moieties were very similar to those of CBDA 8 (Table 2), so the R. marginata compound is proposed to be PTDA 6, with H3‐H4 trans as in CBDA 8 (8, like CBD 7, has H3‐H4 couplings of 9–13 Hz (Marchetti et al., 2019)). The absolute stereochemistry (at C4) is assumed to be the same as PTD 5 (Fig. 1). PTDA 6 has not previously been reported as a natural product.

To study intraspecific variation of R. marginata cannabinoids, we developed a miniaturized GC–MS analytical method, which included silylation to protect PETA against the previously reported thermal decarboxylation in the GC injector (Toyota et al., 2002), and which occurs for THCA (Nahar et al., 2020). This did stabilize PETA and assisted in the discovery of PTDA and BB4GA in R. marginata, giving good symmetric peaks characterized by their GC retention indices and their MS (Table 1).

Seventy‐five collections of R. marginata were made at three sites (Fig. 3) on six separate dates (Table S1). GC‐MS analysis results, with the absolute concentrations of six bibenzyl cannabinoids, plus the two other 2‐prenylated bibenzyls BB2P and BB2G, are in Table S4. BB2P was the major bibenzyl in all R. marginata extracts (> 50% of total bibenzyls), followed by BB2G representing 20–30% of the total bibenzyls.

The concentrations of PET and PETA varied greatly across collections (Table S4): PET ranged from not detected (ND, < 0.01 mg g⁻¹ of freeze‐dried weight (FDW)) to 11 mg g⁻¹ FDW; and PETA from ND to 7 mg g⁻¹ FDW. The newly discovered cannabinoids ranged from ND to 10 mg g⁻¹ FDW for PTD; from ND to 17 mg g⁻¹ FDW for PTDA; from ND to 11 mg g⁻¹ FDW for BB4G; and from ND to 7 mg g⁻¹ FDW for BB4GA. In terms of total bibenzyl cannabinoids (as the sum of PET/PETA, PTD/PTDA, and BB4G/BB4GA), there was an eightfold difference between the lowest (3.09 mg g⁻¹ FDW) and the highest‐ranking sample (24.4 mg g⁻¹ FDW).

Intraspecific variation is shown in Fig. 4 combining the concentrations of biosynthetically linked pairs of bibenzyl cannabinoids. A plot of PET/PETA and PTD/PTDA concentrations (Fig. 4a), with proportion of BB4G/BB4GA represented as dot size, indicated that samples highest in PET/PETA had low or no PTD/PTDA or BB4G/BB4GA, while the samples highest in PTD/PTDA had low or no PET/PETA or BB4G/BB4GA, and the samples highest in BB4G/BB4GA had low or no PET/PETA or PTD/PTDA. Most of the R. marginata samples in this study could be divided into two bibenzyl cannabinoid chemotypes, either dominated by PET 1 and PETA 2 (PET chemotype) or by PTD 5 and PTDA 6 (PTD chemotype), with some samples having a mixture of these two bibenzyl cannabinoid classes.

Principal Component Analysis (PCA) of the concentrations of the eight bibenzyl compounds across the 75 collections mostly distinguished samples from S1 from samples from S2 and S3 along PC1 (Fig. 4b). The concentration of PET was the main discriminating factor for PC1 (Table S5). Significant differences across collection sites were indeed identified in terms of total PET/PETA (F _2,75 = 19.04, P < 0.001), total PTD/PTDA (F _2,75 = 4.46, P < 0.05), and total BB4G/BB4GA (F _2,75 = 4.7, P < 0.05) (Fig. 4c–e, respectively). S1 samples were mostly of the PTD chemotype, while S2 samples were predominantly PET‐dominant chemotypes. S3 samples were mostly PET‐PTD mixed, with significantly lower amounts of BB4G/BB4GA as compared to S1 and S2 samples. There was no difference in total bibenzyl cannabinoid concentrations (sum of all three cannabinoid types) between the three sites (F _2,72 = 0.26, P = 0.77) (Fig. S4), suggesting similar native metabolic capacity for all R. marginata chemotypes.

To evaluate the seasonal variability of bibenzyl cannabinoid concentrations, samples were collected from S1 and S2 on six separate dates between February 2021 and December 2022 (Fig. 4f–k). For both sites, season of sampling had a significant impact on total cannabinoid concentrations (F _5,23 = 56.6, P < 0.001 and F _5,33 = 49.4, P < 0.001 for S1 and S2, respectively) (Fig. S5). In the PTD‐dominant R. marginata chemotypes from S1, significantly greater total PTD/PTDA concentration was identified in December 2021 (early summer) (F _5,23 = 3.75, P < 0.05), whereas there was no significant variation in terms of total PET/PETA and total BB4G/BB4GA (Fig. 4f–h). For the PET‐dominant chemotypes from S2 (Fig. 4i–k), the total PET/PETA concentrations were significantly affected by the season (F _5,33 = 4.15, P < 0.01), with greater concentrations reported in December 2021 as compared to the other collection dates. Changes in accumulation of PTD/PTDA were also significantly affected by the season but to a lesser extent (F _5,33 = 3.62, P < 0.05). Overall, both sites and seasons as well as their combinations had a significant influence on PET/PETA and PTD/PTDA concentrations (Two‐way ANOVA). For PET/PETA, sampling site accounted for the largest proportion of variance (27.6%), followed by the interaction between site and season (12.3%) and season alone (9.5%), with 38.9% of the variance remaining unexplained. By contrast, for PTD/PTDA, collection date was the dominant factor explaining variance (31.4%), with smaller contributions from the interaction (12.5%) and site (10.8%).

Ambient light spectra recorded at S1 and S2 on the same day in December 2022 indicated a significantly higher ratio of far‐red light at S1 compared to S2 (Fig. S6). Climate data between January 2020 and December 2022 (Fig. S7; Table S6) showed lower minimal temperatures during winter 2021, but there was no obvious difference over the 3 yr that could explain the higher concentrations of total bibenzyl cannabinoids in December 2021 samples.

We evaluated the growth behavior of R. marginata under artificial light, as well as the plasticity of its bibenzyl cannabinoid metabolism after transplantation to a controlled environment. Colonies of PTD‐dominant and PET‐dominant R. marginata chemotypes, collected in February 2021 from S1 and S2 respectively, were transplanted onto plastic towers covered with Sphagnum peat, and grown in cultivation units (Fig. 5a).

Over four months in cultivation cabinets with biweekly water spraying, the length of S1 R. marginata shoots increased on average by 3.5 ± 0.3 mm and 4.7 ± 0.7 mm under artificial light without and with FR light supplementation (FR− and FR+), respectively. Branch length from S2 R. marginata increased by 1.7 ± 0.5 mm and 5.8 ± 1.3 mm under FR− and FR+ light, respectively (Fig. 5b). Overall, FR supplementation significantly increased shoot length (F _1,66 = 8.6; P < 0.01), while nutrient supplementation (H₂O vs H₂O+) had no influence on growth (F _1,66 = 0.6; P = 0.45). Shoot growth did not significantly vary between samples originating from different sites (F _1,66 = 0.67; P = 0.41) and behaved similarly under FR light supplementation (site × light treatment, F _1,66 = 1.17; P = 0.28). Light microscopy revealed phenotypical changes upon FR light supplementation (Fig. 5c–f) for both sites, with increased interleaf space and leaf sizes (on average 0.34 ± 0.06 mm² per leaf and 1.43 ± 0.05 mm² per leaf under FR− and FR+, respectively, for S1 (P < 0.001) and 0.54 ± 0.06 mm² and 1.27 ± 0.08 mm² under FR− and FR+, respectively, for S2 (P < 0.001)). Oil body size or shape did not visibly differ between sites, or with light and nutrient treatments (Fig. 5g–j).

After 4 months of R. marginata growth in culture, there were no significant changes in any of the bibenzyl cannabinoid contents (Fig. 6; Table S7). However, chemical analyses of plants grown for 12 months did reveal significant differences in concentrations. Total cannabinoids for FR− treated samples (with or without nutrient supplementation) increased from 5.0 ± 0.3 mg g⁻¹ FDW in the original sample up to 25.2 ± 4.4 for S1 derived samples, and from 6.0 ± 1.1 up to 21.3 ± 0.6 for S2‐derived samples (Fig. 6a,b). The increase was mainly due to an increase of PTD/PTDA and its precursors BB4G/BB4GA amounts in the PTD‐dominant chemotype (S1 samples) (Fig. 6c–e), whereas elevated concentrations in BB4G/BB4GA and PET/PETA were mainly responsible for the increase identified in PET‐dominant S2 samples (Fig. 6f–h).

In vitro cultures of R. marginata plants were initiated from sterilized spore capsules collected from S2 in February 2021. The different stages of growth (spore germination, development of protonema, rhizoids, and finally of a foliose‐like region) were monitored monthly until sampling for chemical analysis in June 2022 (Fig. S8). These in vitro R. marginata plants presented a different morphology to plants in their natural habitat, growing in a ‘bell‐shape’ manner (Fig. 7a,b) with leaves showing irregular sinuate margins (Fig. 7b,c). They also formed ball‐like clusters on the solid media (Fig. S8d). Oil body characteristics of the in vitro plants were similar to those observed in wild plants, that is single large dark‐brown oil bodies within the medial lobe cells and multiple clearer spherical oil bodies in the margin cells (Fig. 7c,d). However, in vitro samples also presented small, round, transparent oil bodies in some internal cells, which resemble those seen at the leaf margins (Fig. S9). In vitro plants accumulated on average 1.92 times more total bibenzyl cannabinoids than wild counterparts collected in February 2021 (T ₇ = 4.05, P < 0.01) (Fig. S10; Table S8), due to significant increases in PET/PETA and BB4G/BB4GA (T ₇ = 3.08, P < 0.05 and T ₇ = 3.37, P < 0.05, respectively, Fig. 7e). The signature of the PET/PETA‐dominant chemotype of origin was still apparent, with no PTD/PTDA compounds detected and on average 7.1 ± 0.7 mg g⁻¹ FDW of PET/PETA (Fig. 7e). The largest bibenzyl concentration increase was for BB2P, with a 2.95‐fold difference between axenic and wild cultivation (T ₇ = 9.11, P < 0.001, Fig. 7f).

The discovery of PTD + PTDA and BB4G + BB4GA in R. marginata completes the parallels of bibenzyl cannabinoids with the main Cannabis cannabinoids (Figs 1, 8). These results expand phytocannabinoid knowledge, identifying two ‘new’ (i.e. first report in Nature) bibenzyl cannabinoids: trans‐PTD 5, analogous to medicinal cannabidiol trans‐CBD (Fig. 1), and its precursor acid PTDA 6. trans‐PTD has previously been synthesized chemically (Stott et al., 2021) and through yeast metabolic engineering (Naesby, 2021). BB4GA 9, known from flowering plants (Bohlmann & Hoffmann, 1979; Dat et al., 2008), was purified for the first time from any liverworts.

This discovery came from analyses of many (75) R. marginata collections from different colonies at three separate sites, as opposed to the original reports of PET and PETA from single samples of R. perrottetii (Japan) (Toyota et al., 1994) and R. marginata (A/NZ separate site, Fig. 3) (Toyota et al., 2002). It is possible that the same array of bibenzyl cannabinoids may be present in R. perrottetii and in other Radula reported to contain PET: Japanese R. campanigera and R. chinensis (Asakawa et al., 2020), and Costa Rican R. laxiramea (Cullmann & Becker, 1999). Our results emphasize that reports of important secondary metabolites must be followed up with analyses of their variation within the producing species. For example, the antimicrobial polyketide triketones in mānuka (Leptospermum scoparium (Myrtaceae)) show distinct regional chemotypes across this shrub's A/NZ geographic range (Douglas et al., 2004).

This work showed great variations in relative proportions of the bibenzyl cannabinoids in R. marginata, mostly with chemotypes dominated by PET + PETA and/or PTD + PTDA (Fig. 4). Most liverwort clusters from Site 1 were PTD‐dominant; from Site 2, PET‐dominant; while Site 3 samples had mixed PET‐PTD. These chemotypes parallel the most common Cannabis chemotypes: THC, CBD, and mixed THC‐CBD (Toth et al., 2020). A few samples across sites were dominated by the precursors BB4G/BB4GA (BB4G‐chemotype) (Fig. 4), analogous to the less common cannabigerol (CBG) chemotype of Cannabis (Fournier et al., 1987). The different chemotypes of individual Cannabis plants are due to genetic differences, with single nucleotide polymorphisms in the coding regions of cannabinoid synthases playing an important role in determining plant chemotype (Singh et al., 2021). Because liverworts are small and undifferentiated, it was not possible to determine where an individual R. marginata plant began or ended, and thus, the collected samples (c. 30 cm² of plant tissue per contiguous colony, Fig. 2) likely contained a mix of plants, rather than having a single genetic origin. Individual R. marginata plants (n = 3) grown in vitro from single spores showed PET + PETA and BB4G + BB4GA but no PTD + PTDA (Fig. 7). This could show that individual liverworts are high PET low PTD, or the opposite, analogous to the individual Cannabis plant chemotypes. Additionally, R. marginata is dioecious, and sex‐specific differences in metabolite profiles, similar to those documented by Zhou et al. (2024) in the model liverwort Marchantia polymorpha, may occur and contribute to content variability.

Overall differences between sites and seasons emphasized the potential influence of environmental conditions on bibenzyl cannabinoid metabolism. While the three sites under investigation are within a 50 km radius (Fig. 3), there were some differences in the spectral composition and intensity of the light under the canopy, with a higher proportion of far‐red light at S1 than S2 (Fig. S6). This might partly explain the significant intraspecific variability between sites (averaged cluster compositions, Fig. 4). There were also some significant intraspecific variations in concentrations of bibenzyl cannabinoid between sample dates, with the different major compounds at S1 and S2 higher in December (early summer, Fig. 4). The only previous study of intraspecific variation of liverwort chemistry that we could find was on another Radula species. Blatt‐Janmaat et al. (2023) studied R. complanata collected in Sweden, Germany, and Canada: Most variation (39%) in the metabolite profiles was attributed to the type of host tree, and 25% attributed to differences in environmental condition. However, these metabolic shifts were mainly in primary metabolites. No specific bibenzyls or cannabinoid‐like compounds were referenced in that study, even though Japanese R. complanata was previously shown to accumulate bibenzyl‐2‐geranyl (BB2G) and bibenzyl‐2‐prenyl (BB2P) (Fig. 1) (Asakawa et al., 1991). Another case for metabolite variation could be differing microbiomes, as strikingly observed for one A/NZ marine sponge species (Storey et al., 2020). Kayser and coworkers found some parallels between endophytes in R. marginata and Cannabis (Kusari et al., 2014).

We have shown that the leafy liverwort R. marginata was extremely slow growing (c. 1 mm per month) in all culture conditions investigated (Fig. 5), in contrast to the fast‐growing thallose liverwort Marchantia polymorpha, which has been seen as a promising chemical production platform (Sauret‐Gueto et al., 2020; Bowman et al., 2022). Therefore, commercial production of bibenzyl cannabinoids from cultured plants would be impractical, and natural colonies would take very long times to recover from wild harvest damage. It is important to note that R. marginata's conservation status is ‘At Risk‐Declining’ (De Lange et al., 2020).

Leaf size and branch growth were boosted by far‐red light supplementation (Fig. 5), but total bibenzyl concentrations did not differ over 4 months (Fig. 6). Extended cultivation (12 months) was needed to reveal the potential influence of cultivation conditions on bibenzyl production (Fig. 6). Liverwort grown without far‐red light supplementation showed increases in total bibenzyl cannabinoid concentrations (expressed as mg per g plant Freeze‐Dried Weight, Fig. 6). Far‐red light, largely dominating on the forest floor, has notably been shown to influence specialized metabolism in plants (Zhang et al., 2021). Adding far‐red has also been shown to increase leaf size and terpenoid production in another bryophyte, the moss Sphagnum flexuosum (Vicherova et al., 2020).

Importantly, the PET+PETA vs PTD + PTDA chemotype‐specific signatures observed in the wild persisted in glasshouse and axenic in vitro conditions (Figs 6, 7), suggesting a strong genetic component in bibenzyl cannabinoid metabolism.

The discovery of PTD + PTDA and BB4G + BB4GA in R. marginata completes the parallels of bibenzyl cannabinoids with the main Cannabis cannabinoids (Figs 1, 8). Hussain et al. (2018) have studied genes and transcription factors in R. marginata and proposed parallel pathways to the phenylethyl/β‐aralkyl (i.e. bibenzyl) cannabinoids and the alkyl cannabinoids (Hussain et al., 2019). Based on the extensive knowledge of Cannabis cannabinoid biosynthesis (Berman et al., 2023), and knowledge of bibenzyl biosynthesis in the model liverwort Marchantia polymorpha (Takahashi & Asakawa, 2017; Zhu et al., 2023), we now propose a detailed model for bibenzyl cannabinoid biosynthesis in Radula (Fig. 8). In this model, six key enzymes produce BB4GA, integrating phenylpropanoid, polyketide, and methylerythritol 4‐phosphate (MEP)/isoprenoid pathways, with separate synthases to give PETA and PTDA (Fig. 8).

PETA and PTDA formation requires stereospecific oxidative cyclization of the geranyl group of BBGA. This reaction is likely catalyzed by a berberine bridge enzyme‐like enzyme (BBE), that is flavin adenine dinucleotide (FAD)‐dependent monoxygenase (Daniel et al., 2017). In Cannabis, CBGA is cyclized to THCA and CBDA by homologous BBE‐like enzymes, which share c. 80% amino acid sequence identities. We hypothesize analogous synthases in R. marginata acting on the common acyclic precursor BB4GA to give cis‐PETA and trans‐PTDA (Fig. 8). In Cannabis, THCA and CBDA share the same H3‐H4 trans stereochemistry in the monoterpene moiety, and both synthases have been shown to produce both THCA and CBDA, albeit in very different ratios (Zirpel et al., 2018). Site‐directed mutagenesis of divergent residues surrounding the active site of these two enzymes suggested that they share similar active site geometries and most likely bind their common substrate, CBGA, in a similar way (Zirpel et al., 2018). However, the opposite stereochemistries of cis‐PETA vs trans‐PTDA in Radula (Fig. 1), imply that the common BB4GA substrate binds in very distinct conformations within the active sites of their respective synthases. Hence, it can be anticipated that the R. marginata PTDA and PETA synthase are much more distantly related than their Cannabis counterparts and share significant differences both in the amino acid composition and overall geometry of their respective active sites.

The structural parallels between the R. marginata and Cannabis compounds carry through to their bioactivities. As predicted by Toyota et al. (1994), (−)‐cis‐PET 1 had agonistic activity for the human cannabinoid receptor 1 (CB1) and further showed in vivo psychoactivity (Chicca et al., 2018). Stott et al. (2021) tested the efficacy of synthesized (−)‐trans‐PTD against seizure in a mouse model and showed significant protective activity, which was not observed for (−)‐trans‐CBD 7 in their system. Although R. marginata (+)‐trans‐PTD has opposite absolute stereochemistry, its biological potency should be explored. Further studies are also warranted to validate the ability of (+)‐trans‐PTD to modulate the endocannabinoid system, including interaction with receptors CB1 and CB2. Connor and coworkers have investigated inhibition of human recombinant T‐type calcium channels by phytocannabinoids in vitro, and showed that ‘in all cases, phytocannabinoid acids were more potent than their corresponding neutral forms’ (Mirlohi et al., 2022). Therefore, investigation of the bibenzyl cannabinoid acids PETA, PTDA, and BB4GA from R. marginata is warranted.

Another divergence between the secondary metabolisms of R. marginata and Cannabis is in the concentrations of decarboxylated products in planta. CBG, THC, and CBD do not occur in significant concentrations in intact Cannabis plants, with postextraction decarboxylation needed to give psychoactive THC (Wang et al., 2016). However, decarboxylated bibenzyl cannabinoids BB4G, PET, and PTD are present in generally higher concentrations in Radula plants than the (presumed) acid precursors (Table S4). The side chain differences between Radula and Cannabis compounds seem unlikely to affect spontaneous decarboxylation reactivity. In addition, Radula grows in cooler and darker conditions than Cannabis, and therefore, spontaneous decarboxylation in planta seems unlikely. Enzymatic decarboxylation of the stilbene acid lunularic acid has previously been reported in the liverwort Conocephalum conicum (Pryce & Linton, 1974), and a similar decarboxylative step could occur in R. marginata. The presence of BB2P and BB2G, without detectable acid precursors, also suggests the existence of orchestrated decarboxylation mechanisms. The presence in significant amount of these two molecules also suggest the presence of an alternative fate for the bibenzyl acid involving decarboxylation followed by specific prenylation on C2 (Fig. 8). Interesting, prenyl transferases in Cannabis (or H. umbraculigerum) would not accept a ‘neutral’ or decarboxylated precursor (Berman et al., 2023), indicating another case of divergent evolution for prenyl transferases.

The variability in the ratio between the cyclized (PET/PETA, PTD/PTDA) and uncyclized (BB4G/BB4GA) bibenzyl cannabinoids across seasonal R. marginata wild collections and across light experiments, while keeping a relatively constant total cannabinoid content, suggests that specific environmental or physiological parameters such as leaf age might play an important role for the likely irreversible BB4GA cyclizations.

In Cannabis, the genetic model of inheritance of THC/CBD ratio is based on CBDAS genotyping, where pure‐chemotype plants (THC‐ or CBD‐chemotype) are due to homozygosity for functional or nonfunctional alleles of CBDAS, respectively, and mixed chemotypes are due to heterozygosity for that allele (Wenger et al., 2020; Ren et al., 2021). CBDA synthase has been shown to be a superior competitor for CBGA when both synthases are present (Ren et al., 2021). While it has been suggested that the diversity of CBDA and THCA synthases arose from duplication and neofunctionalization (Daniel et al., 2017), there is a lack of reported ancient genome‐wide duplication in liverworts (Dong et al., 2022), also attested by the lower number of genes in liverwort genomes compared with most other clades of plants (Linde et al., 2023). Genome sequencing of the different R. marginata chemotypes would be an essential prerequisite to identify the number and nature of potential PETA and PTDA synthases.

Another shared feature between Cannabis and Radula is the presence of specialized storage structures for the bioactive compounds: In Cannabis, cannabinoids predominantly accumulate in external flower glandular trichomes (Andre et al., 2016), whereas in R. marginata, internal oil bodies (Fig. 2c,d) are the likely reservoirs of these specialized metabolites (Romani et al., 2020). Cannabis trichomes also contain mono‐ and sesquiterpenes, which vary greatly across different cultivars (Booth et al., 2020). By contrast, the other volatiles accompanying the bibenzyl cannabinoids in R. marginata were much less variable. BB2P 14 was the major bibenzyl in all 75 R. marginata extracts (> 50% of total bibenzyls), along with BB2G 13 representing 20–30% of the total bibenzyls (Table S4). A single sesquiterpene, tentatively assigned as trans‐selina‐4,11‐diene, was the main terpene in the GC‐MS traces of all of these extracts (Fig. S11).

The discovery of bibenzyl cannabinoids in R. marginata and their structural similarity to cannabinoids in the different Cannabis chemotypes suggests another fascinating case of convergent evolution in the plant kingdom. It adds an extra dimension to the previously described convergent evolution in Helichrysum umbraculigerum, where both CBGA and BB4GA accumulated (Berman et al., 2023). This phenomenon, where unrelated species develop similar traits independently, often in response to analogous environmental pressures or ecological niches, underscores the versatility and adaptiveness of plant specialized metabolism. The evolution of cannabinoid‐like compounds in R. marginata might reflect similar ecological roles or biochemical pathways that have evolved convergently with those in Cannabis. It is unknown whether cannabinoid‐like compounds arose recently within some Radula species or was present in more ancestral species and developed the biochemical processes earlier than Cannabis. The Radulaceae liverwort family likely diversified in the late Cretaceous period c. 100 million years ago (Cooper et al., 2012), when Rosids (including Cannabaceae) were also diversifying (Magallón et al., 2015). Ecological roles for cannabinoid synthesis could include defence mechanisms against pathogens or light adaptation (Andre et al., 2016), or they might play a part in intra‐ or interspecies signaling (Vicherova et al., 2020). Further research into the biosynthetic genes and pathways involved in the production of these compounds in R. marginata could provide deeper insights into the pressures and mechanisms driving this convergent evolution (Davies & Andre, 2023).

While this study offers a comprehensive chemical characterization of cannabinoid‐like metabolites in R. marginata across different chemotypes and environmental conditions, the next frontier lies in uncovering the biosynthetic pathways driving their production. Future genome and transcriptome analyses could reveal the enzymes responsible for the production of these unique bibenzyl cannabinoids, such as PTDA, and unlock our understanding of how R. marginata regulates bibenzyl cannabinoid biosynthesis. Such studies would not only validate our hypotheses regarding metabolic pathways and plasticity but also provide critical tools for engineering in heterologous hosts, expanding the chemical diversity and enhancing the potential pharmaceutical applications of these compounds. Exploring the microbial communities associated with R. marginata could also reveal their role in shaping bibenzyl cannabinoid biosynthesis.

All species native to A/NZ, including endemic R. marginata, are taonga/treasures to Māori who are their kaitiaki/guardian. Future harvesting, research, and any commercialization of intellectual property derived from R. marginata need to fully consider Māori rights and responsibilities for the plant. As previously outlined by T. Whare in Hussain et al. (2019), these are guaranteed by the Te Tiriti o Waitangi/Treaty of Waitangi as well as tikanga/Māori law. International frameworks such as the Genetic Resources and Associated Traditional Knowledge Treaty (Anon, 2024) or the Nagoya Protocol (Anon, 2011), further recognize Indigenous peoples' assertions of full authority over natural resources.

The work reported here was carried out with approval and assistance of the local Ngāti Hauā and Ngāti Hinerangi iwi/peoples for the collections from their rohe/tribal estates. By including a Biocultural Notice, the authors seek to demonstrate a commitment to ethical research practices, cultural sensitivity, and respect for the intellectual property rights of Indigenous and local communities (Anderson & Hudson, 2020).

All study data are included in the main text and in Supporting Information, including methods: Methods S1 and Fig. S1.