Inheritance and covariation of specialised metabolites among cannabis chemotypes

Assignment of chemical phenotype (chemotype) to parental and filial plants

In order to understand metabolite coupling among the major Δ⁹-THC/CBD chemotypes, a genetically female Δ⁹-THC-type (chemotype I; P1) plant and a genetically female CBD-type (chemotype III; P2) plant were crossed by silver thiosulfate (STS)-induced staminate flowering. The dried pistillate inflorescences of the resulting filial F₁ plants, as well as inflorescences from F₂ plants cultivated from seeds from a selfed F₁ plant, were examined by ultra-high-performance liquid chromatography–electrospray ionisation–high-resolution mass spectrometry (UHPLC-ESI-HRMS) using data-dependent MS2 acquisition (DDA). The use of a feminised bi-parental population allowed us to score the metabolomes of unisexual female inflorescences for both parents. This circumvented the use of either a male or hermaphroditic paternal parent, which produce morphologically and chemically distinct staminate inflorescences.²

To accurately assign chemotypes to the parental and filial plants, 19 PCs were quantitatively determined by comparison with certified reference standards, and the results are summarised in Table 1 and Supplementary Table S1. Of the 19 PCs, 10 were the acid forms Δ⁹-tetrahydrocannabinolic acid A (Δ⁹-THCAA, 1, commonly referred to as Δ⁹-THCA), CBDA (2), CBCA (3), CBGA (4), cannabinolic acid (CBNA, 5), Δ⁹-tetrahydrocannabivarinic acid (Δ⁹-THCVA, 6), cannabidivarinic acid (CBDVA, 7), cannabichromevarinic acid (CBCVA, 8), cannabigerovarinic acid (CBGVA, 9) and cannabivarinic acid (CBVA, 10); and nine were the decarboxylated analogues Δ⁹-THC (11), CBD (12), CBC (13), cannabigerol (CBG, 14), cannabinol (CBN, 15), Δ⁹-tetrahydrocannabivarin (Δ⁹-THCV, 16), cannabidivarin (CBDV, 17), cannabigerovarin (CBGV, 18) and cannabivarin (CBV, 19). This allowed us to examine both the segregation of the major wild-type pentyl- (e.g. Δ⁹-THCAA, Δ⁹-THC, CBDA and CBD) and propyl- (Δ⁹-THCVA, Δ⁹-THCV, CBDVA and CBDV), Δ⁹-THC(V)- and CBD(V)-type PCs associated with the activity of THCAS and CBDAS,³¹ as well as to quantify, for the first time, the co-segregation of pentyl and propyl side chain homologues of CBC(V)-type (e.g. CBCA, CBC and CBCVA) and CBG(V)-type (e.g. CBGA, CBG, CBGVA and CBGV) PCs among parental and filial populations.

Acidic PCs are decarboxylated non-enzymatically following exposure to light and heat.³² To simplify comparisons between the PC species, inflorescence dry weight concentrations (mmol/100 g) of the acidic and decarboxylated PCs were combined and expressed as a total (e.g. the addition of Δ⁹-THCA and Δ⁹-THC is equal to Δ⁹-THC_tot). This was applied to all pentyl and propyl PC species except for CBCVA, as the decarboxylated standard for this PC (CBCV) was not available at the time of analysis.

Parental chemotypes are consistent with chemotype I and III

Although the absolute PC content within cannabis inflorescences is a polygenic trait and subject to environmental stimuli, the relative proportions of Δ⁹-THC-/CBD-type PCs is under strict genetic control and dependent on the inheritance of discrete loci encoding for THCAS and CBDAS (Fig. 1).³ Analysis of the PC profiles of the unisexual pistillate inflorescences of clonally propagated parentals grown alongside 10 F₁ clones is consistent with this model. The chemotype I parent P1 was predominant in Δ⁹-THC_tot and its propyl C₃ homologue Δ⁹-THCV_tot (Fig. 2a), of which THCAS catalyses both homologues from either pentyl CBGA or propyl CBGVA substrates respectively (Fig. 1).³¹ For P1, these two compounds made up 92% of the 19 PCs assessed, with Δ⁹-THC_tot as the predominant PC at 57%. Conversely, the chemotype III parent P2 was predominant in only CBD_tot (86% total PCs) formed by CBDAS catalysis of CBGA (Fig. 1), with >95% of PCs having pentyl, not propyl, side chains, as evidenced by the low proportion of CBDV_tot in this chemotype (Fig. 2a).

Fig. 2.

Composition of phytocannabinoids (PCs) among the parentals (P1, P2) and filial generations: F₁ (n = 10) and F₂ (n = 108). (a) Interleaved scatter plot of PC proportions. (b) Violin plot of F₂ PC concentrations (concentrations expressed in millimoles per 100 g, dry matter basis). The subscript ‘tot’ refers to the sum of the PC acid form and its respective decarboxylated form.

F₁ hybrid progeny have PC values intermediate to the parental plants

Analogous with a co-dominant model of inheritance for THCAS and CBDAS, the pentyl Δ⁹-THC_tot and CBD_tot values were uniformly intermediate as compared with P1 and P2, and, as expected, are consistent with chemotype II (Fig. 2a). A similar trend among the F₁ plants was observed for the other propyl and pentyl PC species, except for CBDV_tot, which exceeded levels observed for both parentals (Fig. 2a). Transgressive segregation of CBDV_tot, together with the intermediate F₁ Δ⁹-THCV_tot values, is interpreted as a sign of homozygosity at the Aⁿ locus for P1 and P2. This multi-locus complex is hypothesised to govern pentyl and propyl PC proportions,³³ and elevated levels of CBDV_tot among F₁ hybrids are consistent with the inheritance of Apropyln alleles from P1 that form CBGVA, as well as CBDAS from P2 that allows catalysis of CBGVA, forming CBDVA (Fig. 1).

Next, we selfed a single female F₁ plant by silver thiosulfate (STS) application to generate a segregating F₂ population, of which 108 plants were grown from seed within an environmentally controlled glasshouse. The mature 6-week-old inflorescences of the F₂ progeny were then harvested, dried and subjected to UHPLC-ESI-HRMS analysis (see Experimental). As was observed for the parents and F₁ progeny, Δ⁹-THC_tot and CBD_tot were the predominant PCs (Fig. 2b), of which the acid forms made up the majority and ranged from 0.08 to 16.95 mmol/100 g for Δ⁹-THCAA and 0.01–10.50 mmol per 100 g for CBDA. Overall, the median concentrations of the acid PCs were in the order of CBDA > Δ⁹-THCAA > CBDVA, CBCVA > THCVA > CBCA ≫ CBGA > CBVA, CBGVA > CBNA (Table 1 and Supplementary Table S1).

Minor PC composition among F₂ segregants deviates from the parentals

Although the absolute inflorescence PC contents between parental/F₁ clones and seed-propagated F₂ individuals are not directly comparable (see Experimental), the relative PC proportions between these populations indicated subtle changes in the abundance of the minor PCs (Fig. 2a). This was evident for the propyl CBCVA and the pentyl CBC_tot, which were both elevated in the F₂ relative to the parents (P1 and P2) and F₁. Among the F₂, the average propyl CBCVA content was >2-fold that of the pentyl CBC_tot content (1.0 v. 0.41 mmol per 100 g) (Fig. 2b), despite showing the reverse trend in previous generations (e.g. the pentyl CBC_tot proportion was higher in P1, P2 and F₁) (Fig. 2a). Differences were also observed for the PC intermediate propyl- and pentyl- CBG-types, with F₂ proportions for CBGV_tot and CBG_tot higher and lower, compared with the parentals and F₁ respectively (Fig. 2a).

F₂ progeny segregate into three chemotypes

As expected, three distinct Δ⁹-THC-/CBD-type chemotypes were observed among F₂ plants. Chemotypes were demarcated by plotting Δ⁹-THC_tot and CBD_tot contents (dry matter basis), and these were in agreement with the initial chemotype classifications of the parents and F₁ (Fig. 3a). Three chemotypes co-segregating with Δ⁹-THC_tot and CBD_tot contents were also distinguishable by comparison of propyl PCs Δ⁹-THCV_tot and CBDV_tot contents (Fig. 3a). This confirmed that THCAS and CBDAS can accept CBG-type PCs with either pentyl or propyl alkyl side chains, and that changes in alkyl length occur prior to the oxidative cyclisation step performed by PC synthases. Chemotype I Δ⁹-THC(V)-predominant F₂ plants had log(Δ⁹-THC_tot/CBD_tot) ≥ 2.0, whereas chemotype III CBD(V)-predominant F₂ plants had log(Δ⁹-THC_tot/CBD_tot) ≤ −1.0 (Fig. 3b). For chemotype II plants, log(Δ⁹-THC_tot/CBD_tot) values were ~−0.2, and the F₂ progeny was not significantly different from the 1:2:1 Mendelian expectation of chemotype I:II:III (25:51:32 F₂ plants, χ² = 0.59, 2 d.f. (P = 0.746)).

Fig. 3.

F₂ chemotypic distribution patterns. (a) Scatterplot of propyl and pentyl Δ⁹-THC(V)-type and CBD(V)-type contents (concentrations expressed in millimoles per 100 g, dry matter basis). Black arrows indicate PC contents of the F₁ hybrid individual, which was selfed to produce the F₂ population. (b) Histogram of THC(V)-type, CBD(V)-type and propyl/pentyl PC ratios. Numerals within parentheses indicate chemotype. The subscript ‘tot’ refers to the sum of the PC acid form and its respective decarboxylated form.

Pentyl:propyl PCs deviate from Mendelian expectation

Surprisingly, the pentyl (Δ⁹-THC_tot + CBD_tot + CBG_tot + CBC_tot) and propyl (Δ⁹-THCV_tot + CBDV_tot + CBGV_tot + CBCVA) PC contents were skewed towards the high propyl PC parent P1, for both F₁ and F₂ generations, whereas the recovery of the low propyl–high pentyl contents of P2 was not observed among the 108 F₂ segregants (Fig. 3a). Moreover, the ratio of pentyl:propyl PC contents showed a continuous distribution not inconsistent with a Gaussian (normal) distribution (D’Agostino–Pearson K2 normality test = 3.891, P = 0.1429) (Fig. 3b). Taken together, these observations indicate a di- or oligo- genic mode of inheritance for pentyl:propyl PC composition and dominance at either one or more Aⁿ loci.

CBC-type and CBG-type PCs associate with chemotypes

To understand covariation of PC classes associated with PC synthase activity, we calculated the Spearman r correlation coefficient of 108 F₂ plants using the proportional data of PCs that act as precursors (CBGA-CBG, CBGVA-CBGV), as well as products of THCAS (Δ⁹-THCA, Δ⁹-THC, Δ⁹-THCVA and Δ⁹-THCV), CBDAS (CBDA, CDA, CBDVA and CBDV) and oxidative pericyclic cyclisation (CBCA, CBC and CBCVA) (Fig. 4a, Supplementary Table S2). Concurrent with the co-segregation of Δ⁹-THC_tot/CBD_tot and Δ⁹-THCV_tot/CBDV_tot ratios among chemotypes I, II and III, strong positive correlations were observed for propyl/pentyl PC homologues of Δ⁹-THC(A) and Δ⁹-THCV(A) (Spearman r ≥ 0.72, P < 0.001), as well as for CBD(A) and CBDV(A) (Spearman r ≥ 0.62, P < 0.001) (Fig. 4b, Supplementary Table S2).

Fig. 4.

Covariation of PC proportions among F₂ progeny (n = 108). (a) Heatmap of Spearman r correlation coefficient. Colour indicates correlation coefficient: blue indicates negative correlation and red indicates positive correlation. (b) Comparison of CBCA and CBCVA proportions among chemotypes I, II and III. (c) Comparison of CBGA and CBGVA proportions among chemotypes I, II and III.

Unexpectedly, patterns of covariation were broadly identified with individual CBC(V)-type PCs, with r indicating a positive trend for individual CBC(V)/CBD(V)-type PCs and a negative or inverse trend for individual CBC(V)/Δ⁹-THC(V)-type PCs (Fig. 4a). This was most obvious for CBCVA/CBDVA (Spearman r 0.71, P = 7.13 × 10⁻¹⁸) and CBCVA/Δ⁹-THCA (Spearman r −0.68, P = 1.48 × 10⁻⁵) (Fig. 4a), which may indicate changes in the product specificity of CBDAS as compared with THCAS. To examine this, we compared the proportions of CBCVA and CBCA among chemotypes and observed a ~2-fold increase in CBC(V)A among chemotypes II and III, as compared with THCAS-associated chemotype I plants (Kruskal–Wallis one-way, P < 0.0001, Dunn’s test, P < 0.05) (Fig. 4b).

Unlike the inheritance of THCAS and CBDAS, a fixed non-allelic C locus is proposed for CBCA synthase (CBCAS), a third oxidoreductase isolated from the young leaves of cannabis that catalyses the oxidative pericyclic cyclisation of CBG(V)A, forming CBC(V)A.^34,35 While CBC(V)A maxima and presumably CBCAS activity are reached at the juvenile stages of plant development, in vitro enzyme assays using recombinant THCAS and CBDAS indicate that both of these enzymes are capable of forming small amounts of CBCA, with product specificity towards CBCA being pH-dependant (>pH 6 and <6 for THCAS and CBDAS respectively).¹⁷ Changes in the product specificity of THCAS and CBDAS within the microenvironment of the trichome secretory cavity, where this reaction takes place,³⁶ could partially account for the lower proportions of CBC-type PCs among the mature flowers of chemotype I THCAS expressing plants.

Discrete patterns of covariation were also observed for the acid forms of CBG(V)-type PCs, with r indicating a positive trend for CBG(V)A-/Δ⁹-THC(V)A-types and a negative or inverse trend for CBG(V)A-/CBD(V)A-types (Fig. 4a). This could potentially indicate a reduction in the enzymatic efficiency of THCAS, leading to an excess of CBG(V)A among THCAS-only expressing chemotype I plants, which is consistent with the reported lower catalytic efficiency of THCAS (k_cat/K_m = 1382 M⁻¹ s⁻¹) as compared with CBDAS (1492 M⁻¹ s⁻¹).³⁵ Comparisons between Δ⁹-THC- and CBD-type chemotypes confirmed this trend, with small but measurable increases in CBG(V)A proportions in chemotype I plants as compared with chemotypes II and III (Kruskal–Wallis one-way, P < 0.005, Dunn’s test, P < 0.05) (Fig. 4c).

Untargeted metabolomic analysis

Deconvolution of the raw MS data was carried out on the F₂ dataset using the software Compound Discoverer (CD, ver. 3.3, ThermoFisher Scientific, see https://www.thermofisher.com/au/en/home/industrial/mass-spectrometry/liquid-chromatography-mass-spectrometry-lc-ms/lc-ms-software/multi-omics-data-analysis/compound-discoverer-software.html). Out of the 24,166 compounds detected, 10,366 were putatively annotated, which represented 42.89% of the total compounds (Supplementary Table S3). The complexity of the metabolite composition of cannabis is a challenge for correctly identifying the compounds present in the absence of a reference standard. Since we quantitatively analysed 19 PCs with reference standards, we were able to compare the MS² spectral information and retention time to identify the correct PCs. Despite the high-scoring matches with databases, the annotation of multiple compounds assigned to the same compound name but with different retention times clearly shows the shortcomings in metabolomics studies, especially for a non-model plant such as Cannabis. Wishart et al.³⁷ reported 6172 chemical entries in the freely available online database (Cannabis Compound Database, CCDB, see https://cannabisdatabase.ca/)³⁷ and this database was also used for compound identification.

Principal component analysis (PCA) of the 24,166 detected compounds showed the separation of the three chemotypes in the scores plot of principal component 1 (PC1) v. principal component 2 (PC2), wherein 28.7 and 25.8% respectively accounted for the variability in the dataset (Fig. 5 and Supplementary Fig. S1). The PCA scores plot demonstrated a similar pattern to the classification shown in Fig. 3b. Samples classified as chemotype I (Δ⁹-THCA-dominant) clustered separately from chemotypes II and III samples. Chemotype II is closely related to chemotype III, which is CBDA-dominant, and the PCA scores plot showed the closeness of these two groupings. From the scores plot, five samples and the respective replicates (×3) classified as chemotype III were observed to cluster with chemotype II samples (Fig. 5). The effect of the 15 sample points was initially tested by excluding them in the analysis, but their exclusion did not affect the number of compounds in the list and thus, they were included in the analysis. The proximity of the grouping of chemotype II to chemotype III suggests a similarity in the biosynthetic origin of the metabolites in the two chemotypes. The PCA scores plot clearly shows differentiation among the samples between chemotype I and chemotypes II and III, and the two groupings that showed distinct separation (chemotypes I and III) were selected for further analysis in this study.

Fig. 5.

Principal component analysis (PCA scores plot of PC1 v. PC2) of 24,166 metabolites from flower tissues analysed by UHPLC-ESI-HRMS in positive ion mode. Samples were from an F₂ cannabis population (n = 108).

Differential analysis of chemotypes III and I (III v. I) indicated that 330 metabolites were significantly decreased or downregulated (P ≤ 0.05, log₂ fold ≤ −2.0) and 578 metabolites were significantly increased or upregulated (P ≤ 0.05, log₂ fold ≥ 2.0) (Fig. 6, Supplementary Tables S4 and S5 respectively). Of the 19 PCs that were analysed quantitatively, Δ⁹-THCAA (1), Δ⁹-THC (11), THCVA (6), CBNA (5) and CBVA (10) were significantly downregulated, and CBDA (2), CBD (12), CBDVA (7) and CBDV (7) were significantly upregulated. This result was not surprising since chemotype I is Δ⁹-THCA-dominant and chemotype III is CBDA-dominant. The downregulation of THCVA (and THCV) and upregulation of CBDVA (and CBDV) supported the homology in the biosynthesis of the (pentyl) C₅ alkyl (THCA, CBDA) with the (propyl) C₃ alkyl (THCVA, CBDVA) PCs.⁸

Fig. 6.

Differential analysis of compounds in chemotype III v. chemotype I. Samples were from an F₂ cannabis population (n = 108). Blue circles and black squares represent the significantly downregulated compounds in chemotype I (P ≤ 0.05 and log₂ fold change ≤ −2), and the teal circles and black triangles represent the significantly upregulated compounds in chemotype III (P ≤ 0.05 and log₂ fold change ≥ 2). The black squares and black triangles respectively represent the significantly downregulated and significantly upregulated compounds that were putatively annotated (Tables 2 and 3).

A mass search of the significantly downregulated and upregulated metabolites was carried out using the Cannabis Compound Database (CCDB, https://cannabisdatabase.ca/),³⁷ which provided greater confidence in their putative annotation since the database contains information on metabolites found in Cannabis. After consensus evaluation of the CD and CCDB results (Supplementary Appendix S1), there were 15 significantly downregulated and 28 significantly upregulated metabolites with putative annotations as summarised in Tables 2 and 3 respectively. In addition to the five downregulated PCs (1, 5, 6, 11 and 16) and four upregulated PCs (2, 7, 12 and 17) that were quantitatively analysed, 20 other metabolites were putatively identified as PCs. These were Δ⁹-tetrahydrocannabiorcol (20) and its acid form Δ⁹-tetrahydrocannabiorcolic acid (21), Δ⁹-tetrahydrocannabinol-C₄ (22), 11-nor-9-carboxy-tetrahydrocannabinol isomer (23), cannabielsoic acid isomers (25, 27 and 30), two isomers of 10-oxo-Δ^6a-tetrahydrocannabinol (28 and 29), three CBDA isomers (33, 42 and 50), cannabiglendol-C₃ (35), two isomers of cannabidiol-C₄ (36 and 45), isotetrahydrocannabinol (38), CBCV isomer (39), CBCVA isomer (40) and two CBNA isomers (44 and 48). Compounds 20, 21, 22, 23, 25, 27, 28 and 29 were downregulated. Compounds 30, 33, 35, 36, 38, 39, 40, 42, 44, 45, 48 and 50 were upregulated.

The downregulated metabolites Δ⁹-tetrahydrocannabiorcol (20, Δ⁹-THC-C₁, C₁₇H₂₂O₂, [M + H]⁺ m/z 259.1693), Δ⁹-tetrahydrocannabiorcolic acid (21, Δ⁹-THCA-C₁, C₁₈H₂₂O₄, [M + H]⁺ m/z 303.1591) and Δ⁹-tetrahydrocannabinol-C₄ (22, Δ⁹-THC-C₄, C₂₀H₂₈O₂, [M + H]⁺ m/z 301.2163) belong to the Δ⁹-THC type of cannabinoids³⁸; hence, it was not surprising that these compounds were also downregulated. Compounds 20 and 21 are homologues of Δ⁹-THC and Δ⁹-THCA respectively, with a methyl substituent at the C-3 position instead of a pentyl (Fig. 7a). Compound 22 is also a homologue of Δ⁹-THC that has a butyl substituent at the C-3 position (Fig. 7a). Compound 23 (C₂₁H₂₈O₄, [M + H]⁺ m/z/345.2062) was annotated as 11-nor-9-carboxy-Δ⁹-tetrahydrocannabinol (THC-COOH, Fig. 7a). This compound has been reported as a breakdown product of Δ⁹-THC in the human body after cannabis consumption,³⁹ but this is the first report from the plant. Thus, compound 23 was presumed to be an isomer of 11-nor-9-carboxy-tetrahydrocannabinol. It is important to note that the major THC-type compound in Cannabis is (−)-trans-Δ⁹-THCAA (1), shortened to Δ⁹-THCA throughout this manuscript. There are four stereoisomers of Δ⁹-THCAA (and subsequently Δ⁹-THC)^40,41 as shown in Fig. 8a, and these stereoisomers could be present in the plant as minor constituents.

Fig. 7.

Chemical structures of downregulated and upregulated compounds in cannabis. (a) THC-type cannabinoids. (b) Cannabielsoic acid (CBEA)-type compounds. (c) Oxo-cannabinoid compound. (d) CBD-type cannabinoids. (e) Iso-hydroxy cannabinoid compound with propyl side chain. (f) Iso-cannabinoid compound with pentyl side chain. (g) Stilbene-type compound. (h, i) Phenanthrene-type compounds. (j) Phenethylamine alkaloid compound.

Fig. 8.

Chemical structures of the stereoisomers of (a) Δ⁹-THCAA (1),^40,41 and (b) CBDA (2),⁴⁷ showing the configuration of the stereocentres (circled in red).

Compounds 25 and 27, which eluted at respectively 11.0 and 11.7 min, were both annotated as cannabielsoic acid (CBEA-C₅, C₂₂H₃₀O₅, [M + H]⁺ m/z 375.2167) and belong to the CBE-type cannabinoids.³⁸ There are two isomers of CBEA-C₅ reported, CBEA-C₅ A and CBEA-C₅ B, that are differentiated by the position of the carboxyl substituent at C-2 in the former and at C-4 in the latter (Fig. 7b). There was doubt whether these compounds are natural products since these can be formed by either oxidation or pyrolysis of CBDA,^38,42,43 but cannabielsoic acids A and B and their respective methyl esters were isolated from hashish⁴⁴ and recently, Parveen et al.⁴⁵ reported the identification of CBEA-C₅ isomers from a hemp extract. Compounds 28 and 29, which eluted at respectively 14.0 and 14.7 min, both had a molecular formula of C₂₁H₂₈O₃ ([M + H]⁺ m/z 329.2112) and were annotated as 10-oxo-Δ^6a-tetrahydrocannabinol (OTHC), which was reported from a cannabis extract by Friedrich-Fiechtl and Spiteller.⁴⁶ OTHC is an oxidised derivative of Δ⁶-THC that has a ketone group at the C-10 position (Fig. 7c).³⁷

In addition to the upregulated CBDA (2), there were three metabolites, 33, 42 and 50, with the same molecular formula as CBDA (C₂₀H₂₈O₅, [M + H]⁺ m/z 359.2219) and with similar MS profiles that eluted at respectively 10.2, 13.0 and 14.3 min (Table 3). These three metabolites were presumed to be CBDA isomers. Similar to Δ⁹-THCA, CBDA has four stereoisomers (Fig. 8b) and the dominant form in the plant is the (−)-trans-CBDA (2).⁴⁷

Two metabolites, 36 and 45, with a molecular formula of C₂₀H₂₈O₂ ([M + H]⁺ m/z 301.2163) and retention times of respectively 11.5 and 12.8 min, were putatively annotated as cannabidiol-C4 (CBD-C₄, Fig. 7d) and would most probably be isomers of each other. Compound 35 had a molecular formula of C₁₉H₂₈O₃ ([M + H]⁺ m/z 305.2113) and was annotated as cannabiglendol (Fig. 7e). Compound 38, which eluted at 13.6 min and had a molecular formula of C₁₉H₂₈O₃ ([M + H]⁺ m/z 315.2320), was annotated as isotetrahydrocannabinol or Δ⁴-iso-tetrahydrocannabinol (Fig. 7f). Compound 39, with retention time of 12.0 min and a molecular formula of C₁₉H₂₆O₂ ([M + H]⁺ m/z 287.2005), was annotated as cannabichromevarin (CBCV). However, this did not match the 13.0 min retention time of the reference standard in our in-house database and thus was assigned to be a CBCV isomer (Table 3). Compound 40, with a molecular formula of C₂₀H₂₆O₄ ([M + H]⁺ m/z 331.1905) and retention time of 12.8 min, was initially annotated as CBCVA, but from our quantitative analysis, CBCVA (8) eluted at 13.6 min. The similarity of the MS profiles of these two compounds suggested that compound 40 is a CBCVA isomer. Compound 42 had the same molecular formula as the downregulated compounds 28 and 29 and was thus annotated to be another OTHC isomer. Compounds 44 and 48 that eluted at respectively 9.3 and 10.7 min had the same molecular formula and similar MS profile as CBNA (5) and were deemed to be CBNA isomers.

Several non-PC metabolites were putatively identified. Compound 26, with retention time of 13.1 min, had a molecular formula of C₂₀H₂₄O₃ ([M + H]⁺ m/z 313.1799) and was annotated as cannabistilbene I (Fig. 7g), which is one of the 12 known dihydrostilbenes in cannabis.⁴⁸ Three upregulated metabolites, 31, 37 and 41 with retention times of respectively 10.8, 9.8 and 11.7 min, were annotated similarly, and mass spectral information suggested these four compounds to be cannabistilbene I isomers.

A downregulated metabolite at 11.2 min had a molecular formula of C₂₀H₂₆O₅ ([M + H]⁺ m/z 347.1855) (Table 2) and was putatively annotated as a gibberellin that could either be one of the isomers gibberellin A37 or gibberellin A44. Similarly, an upregulated metabolite, compound 32, that eluted at 9.1 min had the same mass and was also putatively annotated either as gibberellin A37 or gibberellin A44 (Table 3). Another upregulated metabolite, compound 34, with retention time of 9.6 min and molecular formula of C₂₀H₂₈O₅ ([M + H]⁺ m/z 349.2011), was annotated as gibberellin A53 (Table 3). Gibberellins (GAs) are plant hormones that play various roles in plant growth and development⁴⁹ and they have been reported to affect inflorescence development in cannabis.^50,51

Compound 49, which eluted at 8.9 min, was putatively annotated as 4,7-dimethoxy-1,2,5-trihydroxyphenanthrene (Fig. 7h) with a molecular formula of C₁₆H₁₄O₅ ([M + H]⁺ m/z 287.0915), and compound 51, eluting at 9.4 min, was annotated as 4,5-dihydroxy-2,3,6-trimethoxy-9,10-dihydrophenanthrene (Fig. 7i) with a molecular formula of C₁₇H₁₈O₅ ([M + H]⁺ m/z 303.1228) (Table 3). Compound 52, which eluted at 1.4 min with [M + H]⁺ m/z 166.1226 indicating that it was a nitrogen-bearing metabolite, was annotated as the phenethylamine alkaloid hordenine (Fig. 7j) with a molecular formula of C₁₀H₁₅NO (Table 3, Supplementary Table S5). A compound that eluted at 3.30 min (43) with molecular formula of C₈H₈O ([M + H]⁺ m/z 121.0645) was annotated as phenylacetaldehyde (Table 3, Supplementary Table S5), which is commonly found in plants.⁵²

Two closely eluting metabolites at 6.0 min were deconvoluted with [M + H]⁺ m/z 139.0389 and [M + H]⁺ m/z 291.0864 for compounds 46 and 47 respectively (Table 3, Supplementary Table S4). Compound 46 was putatively annotated as protocatechuic aldehyde (or 3,4-dihydroxybenzaldehyde) with amolecular formula of C₈H₈O, and compound 47 was annotated as epicatechin, which is commonly found in plants, including cannabis.^53,54

Among the annotated metabolites, the main differentiation between the two chemotypes I and III are the major PCs Δ⁹-THCA/THC and CBDA/CBD and their suite of THC-type cannabinoids and CBD-type cannabinoids respectively. However, it is interesting to note that among the upregulated metabolites (chemotypes III v. I), eight compounds were non-PCs and included two phenanthrenes, two phenolics, two aldehydes, one phenyl ketone and one alkaloid (Table 3). This suggested that there could be non-PC biomarkers that can be used to differentiate cannabis chemotypes.

Genetic materials and cultivation of medicinal cannabis

Reagents and standards

All cannabinoid certified reference materials were purchased from Novachem Pty Ltd (Melbourne, Vic., Australia) as distributor for Cerilliant Corporation (Round Rock, TX, USA). Reagent and analytical standard preparation were as reported previously.²

UHPLC-ESI-HRMS analysis

Floral samples from the F₂ plants (n = 108) were air-dried under forced ventilation at room temperature for 14 days (~78–80% moisture loss), then dried in a Thermo Scientific Heratherm gravity convection oven at 40°C for 72 h. Sample preparation and UHPLC-ESI-HRMS analysis were as reported previously² with minor modifications. Briefly, analysis was carried out on a ThermoFisher Vanquish Flex UHPLC system with solvent degasser, quaternary pump, temperature-controlled sampler/auto injector and column compartment, and photodiode array detector (PDA) coupled to an Orbitrap IDX Tribrid high-resolution mass spectrometer (ThermoFisher Scientific Inc., Waltham, MA, USA). Chromatographic separation was performed using a Phenomenex Kinetex C18 column, 1.7 μm, 150 mm × 2.1 mm (Phenomenex Australia Pty Ltd, Lane Cove, Sydney, NSW, Australia). Mobile phase A was water with 0.1% (v/v) formic acid and mobile phase B was acetonitrile with 0.1% (v/v) formic acid, and the following gradient was used: 10% B, 0–2 min; 10–40% B, 2–3 min; 40% B, 3–5 min; 40–80% B, 5–6 min; 80% B, 6–9 min; 80–90% B, 9–11 min; 90–100% B, 11–12 min; 100% B, 12–15 min; 100–10% B, 15–16 min; and 10% B from 16–20 min.

The mass spectrometer was operated using a HESI interface in positive mode. For single and tandem mass spectrometry (MS and MS²), the orbitrap was the mass analyser used. Pierce FlexMix calibration solution was used to calibrate the mass spectrometer prior to data acquisition, and the internal mass calibrant fluoranthene (Easy-IC) was activated for real-time mass calibration during data acquisition. Xcalibur software (ver. 4.6, ThermoFisher Scientific Inc., see https://www.thermofisher.com/au/en/home/industrial/mass-spectrometry/liquid-chromatography-mass-spectrometry-lc-ms/lc-ms-software/lc-ms-data-acquisition-software.html) was used for data acquisition.

Data processing and statistical analysis

Full MS data were processed using Trace Finder (ver. 5.1, ThermoFisher Scientific Inc., Waltham, MA, USA) for quantitative analysis. Peak identity was determined by a comparison of retention time and MS spectra with reference standards and integration of extracted ion peaks. The concentration of each analyte was determined by interpolation from standard calibration curves generated in MS Excel and results were expressed as average concentration ± standard deviation. Results were plotted using GraphPad Prism (ver. 10.5, GraphPad Software, Boston, MA, USA, see https://www.graphpad.com/features).

The high-resolution accurate mass (HRAM) data were further processed using CD software with the workflow set up for untargeted metabolomics with statistics and detection of unknown compounds using databases. Libraries used were mzCloud (see https://www.mzcloud.org/), ChemSpider (see https://www.chemspider.com/), BioCyc (see https://biocyc.org/) and Metabolika pathways, which is a module within CD for pathway analysis. There were 24,166 compounds detected, grouped by molecular weight and retention time across all the 108 sample extracts analysed, after applying a filter set that excluded peaks in the first and last 0.5 min of the chromatogram and a maximum compound peak area of less than 50,000 (Supplementary Table S2). Within the CD software, the observed m/z values and predicted elemental composition were matched with the online spectral library mzCloud and databases ChemSpider, BioCyc and Metabolika pathways, and the annotation made use of the algorithm ‘mzLogic’ within CD. There were several instances of one annotation for several compounds wherein one compound name was putatively annotated to multiple entries in the list that was misleading and thus required a lot of manual interrogation. For example, there were 67 compounds in the list putatively annotated as CBDA (2) with some annotations based on high scores from spectral library match from mzCloud. This inferred the presence of multiple isomers for CBDA and there were other metabolites in the list that suggested several compounds with molecular isomers.

Differential analysis between chemotype III and chemotype I samples was also carried out in CD software. The significantly downregulated and upregulated metabolites were filtered to only those that had MS² data, and the list was reduced to 114 metabolites (34.5% of downregulated) and 98 metabolites (17% of upregulated) respectively. A mass search (m/z values) for metabolites with MS² data was carried out using the Cannabis Compound Database (CCDB, see https://cannabisdatabase.ca/).³⁷ The putative annotations from CD were compared with the results from CCDB and although there were some compounds that had the same annotation between the two results, there were more annotation hits from the CCDB for compounds that have been reported in cannabis.