Comparative genomics of flowering behavior in Cannabis sativa

Mapping of GWAS and QTL markers

In a previous GWAS analysis, Petit et al., (Petit et al., 2020a; Petit et al., 2020b) (Table 1) mapped RADSeq markers to scaffolds of the ‘Purple Kush’ genome (canSat3, GCA_00230575.1). The Purple Kush genome is highly fragmented and consists of 12,836 scaffolds (van Bakel et al., 2011; Hurgobin et al., 2021). To improve on this approach and to unify data from different studies, we identified those scaffolds in the ‘Purple Kush’ genome with SNP markers significantly associated with flowering (LOD scores > 4) (Petit et al., 2020a) and aligned them to the cs10/CBDRx v2 reference genome (annotated from genotype CBDRx:18:580, GCF_900626175.2) using Minimap v2.17 (Li, 2018). Aligned regions containing markers were identified and plotted as separate tracks on a cs10 chromosome karyotype plot using Circos v 0.69-9 (Krzywinski et al., 2009), indicating the association statistic (LOD score) for flowering traits: ‘Beginning of flowering’, ‘Full flowering’, and ‘Length of vegetative phase’ scored at three distinct environments.

A similar approach was used to map USO-31/Carmagnola QTLs described by Woods et al. (2021) to the cs10/CBDRx v2 reference genome (Table 1). Regions containing USO-31/Carmagnola polymorphic SNP markers positioned in the ‘Finola’ genome (GCA_003417725.2) were positioned in the cs10/CBRDx v2 using Minimap 2, as above, to define the endpoints and peaks of the four ‘Days to Maturity’ (DTM.1 through DTM.4) QTLs.

Genomic coordinates were extracted and plotted for the genes identified by Ren et al. (2021) as under selection using cs10/CBDRx2 reference genome protein accessions. Candidate flowering time gene protein sequences reported in the cs10/CBDRx v1 annotation (GCA_900626175.1) that did not correspond to the protein accessions in the cs10/CBDRx v2 annotation were translated and aligned to the cs10/CBDRx v2 genome. Four gene models: evm.model.01.2361 (LD), evm.model.04.2071 (EMF1), evm.TU.01.2503 (FPF) and evm.TU.08.543 (FES1) did not correspond to the reported putative flowering time genes and these were excluded.

Flowering gene identification

Arabidopsis Gene Initiative (AGI) locus codes for 306 ‘flowering time’ and 72 ‘pending flowering time’ protein-encoding gene candidates from Arabidopsis thaliana were obtained from FLOR-ID (Bouché et al., 2016) (accessed on 19 September 2022). Corresponding protein sequences for these A. thaliana genes were obtained from The Arabidopsis Information Resource (TAIR; https://www.arabidopsis.org/). For microRNAs, nucleotide sequences were used. DIAMOND v0.9.24 (Buchfink et al., 2015) was used to compare these A. thaliana sequences to the proteome of C. sativa cs10/CBDRx v2 (GCF_900626175.2) and the best hits with greater than 90% identity were identified as likely orthologs. The longest isoform for each candidate was taken as the corresponding C. sativa cs10/CBDRx v2 ortholog. The microRNA nucleotide sequences for csa-miR156, 159a – b, and 172a – g miRNAs were retrieved via BLASTn analysis of the cs10/CBDRx v2 genome (Das et al., 2015).

To validate this flowering gene identification approach, and to identify additional homologs, we also conducted an Orthofinder analysis (Emms and Kelly, 2015) using the same C. sativa cs10/CBDRx v2 and A. thaliana predicted proteomes. The cs10/CBDRx v2 genome annotation was then further manually examined and additional putative flowering time genes with the keyword annotation ‘flowering’, ‘flower’, ‘time’, ‘circadian’, ‘day’, ‘clock’, and ‘vernalization’ were extracted. Genes were classified using the previously defined categories (Bouché et al., 2016). C. sativa genes with no clear ortholog in A. thaliana were assigned to the category of the most similar A. thaliana protein, based on the Orthogroup analysis using Orthofinder. Locations of the C. sativa flowering time genes in the C. sativa genome were plotted using Circos (Krzywinski et al., 2009).

MADS gene phylogenetic analysis

As the annotation for cs10/CBDRx v2 MADS genes is incomplete, and to resolve the relationships between MADS-domain members, we identified all MADS genes in the cs10/CBDRx v2 genome. An initial search utilized three A. thaliana Type I MADS genes (AT1G01530, AT1G31630, AT5G49490) and three A. thaliana Type II MADS genes (AT1G24260, AT5G23260, AT5G60910) to represent each subgroup of the MADS box gene family (Gramzow and Theißen, 2013). The cs10/CBDRx v2 genome was searched using protein, translated nucleotide and nucleotide BLAST (blastp, tblastx and tblastn) analyses. Duplicate sequences were removed. All C. sativa MADS protein sequences (Supplementary Tables S1, S2) were aligned using Clustal Omega in Geneious Prime 2022.0 (Sievers et al., 2011) and tentatively assigned to clades. Any proteins not containing a complete MADS domain were excluded. An alignment was then generated using Clustal Omega v1.2.4 to assign the CsMADs proteins, including A. thaliana and Vitis vinifera predicted protein sequences (Gramzow and Theißen, 2013), to a clade. The best-fit amino acid substitution model (JTT+R10) was identified using IQ-Tree, and a Maximum Likelihood phylogenetic tree was generated using IQ-TREE 1.6 (Nguyen et al., 2015) (Supplementary Figure S1). A tree of only the Type II sequences was also generated using the aforementioned parameters. Phylogenetic trees were exported to iTOL for visualization (Letunic and Bork, 2007). Details of the accession numbers and clade assignment of CsMADS genes are in Supplementary Table S2.

PEBP gene phylogenetic analysis

To resolve the relationships between PEBPs-domain members, we identified all PEBP-encoding genes in the cs10/CBDRx v2 genome using tblastn and A. thaliana FT (AT1G65480), TFL (AT5G03840), MFT (AT1G18100) and BFT (AT5G62040) protein query sequences. The C. sativa PEBP family protein sequences were aligned with PEBP proteins from A. thaliana, tomato (Solanum lycopersicon; (Cao et al., 2016; Song et al., 2020)), and Chrysanthemum seticuspe ((Oda et al., 2011) using Clustal Omega v1.2.4 (Nguyen et al., 2015). Phylogenetic trees were exported to iTOL for visualization (Letunic and Bork, 2007). Full details of all protein accession numbers are in Supplementary Table S2. For clarity, here we have used the nomenclature suggested by Dowling et al. (2023)

Analysis of protein-protein interactions

Protein sequences for 459 C. sativa cs10/CBDRx v2 flowering time gene candidates were imported into the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database (v.11.5) (Szklarczyk et al., 2020) to generate protein-protein interaction networks. A short-list of 26 proteins from 6 categories of interest was generated using the following parameters: full STRING network, experiments and co-expression data, medium confidence (0.400) (Supplementary Figure S2).

Floral morphology and inflorescence structure

Sexual dimorphism is an important characteristic which has consequences for yield and the chemical composition of C. sativa plants (Welling et al., 2021). C. sativa has nine pairs of homomorphic autosomal chromosomes and a pair of heteromorphic sex chromosomes. Plants are usually diecious with distinct male and female plants (Figures 2A, B), however, plasticity in sexual phenotype can lead to hermaphrodite plants, also known as monecious phenotypes (Moliterni et al., 2004). Male plants (XY) typically flower earlier than female plants (XX) (Bócsa and Karus, 1998; Struik et al., 2000), possibly indicating that there are genes on the Y chromosome that accelerate flowering and/or repressors of flowering on the X chromosome, or that flowering time may be regulated by plant hormones involved in sex differentiation, such as gibberellic acid or ethylene (Galoch, 1978). Male plants produce pollen in hanging inflorescences and female plants produce pistillate flowers in dense clusters, separated by leafy bracts, while the morphology of monecious plants resembles that of female plants prior to the production of male flowers (Moliterni et al., 2004). Monoecious hemp accessions can be classified at flowering by their ratio of developed male to female flowers, which varies by cultivar and environment (Sengbusch, 1952; Faux et al., 2014). In addition to producing separate male and female flowers on a single plant, C. sativa can also produce bisexual flowers (Moliterni et al., 2004). The transition of C. sativa plants from vegetative growth to flowering can be indicated by the formation of undifferentiated primordia in the axils of stipules (protective structures, adjacent to the axillary buds (Heslop-Harrison and Heslop-Harrison, 1969) (Figures 2C, D), and, in some instances, by change of phyllotaxis from opposite to alternate (Bócsa and Karus, 1998) (Figure 3A; Stage 2000). After the appearance of floral primordia, dioecious male plants will form staminate flowers while female plants will develop bracts with no styles, which signifies the development of female flowers (Mediavilla et al., 1998) (Figure 3A; Stage 2200).

Defining the transition from vegetative to inflorescence flowering in C. sativa is complicated by the appearance of solitary flowers (Figure 2D). While a long photoperiod is considered ‘non-inductive’ for C. sativa plants, the development of solitary flowers in shoot internodes demonstrates that these plants are not strictly vegetative (Spitzer-Rimon et al., 2019). For C. sativa plants grown under a long photoperiod, differentiation of the first solitary flowers at the fourth to sixth internodes can occur (Cervantes, 2006). The induction of these solitary flowers is thought to be age-dependent and controlled by internal signals, as opposed to photoperiod (Spitzer-Rimon et al., 2019). Spitzer-Rimon et al. (2022) observed changes in the transcriptomic profile of flowering-related genes among nodes 4, 6, and 7 in female C. sativa seedlings grown under LD conditions. Flowering inducers (such as MOTHER OF FT (MFT), SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1 (SOC1), LEAFY (LFY), and APETALA1 (AP1)) were upregulated while flowering repressors (such as TEMPRANILLO (TEM), TERMINAL FLOWER1 (TFL1), and BROTHER OF FT AND TFL1 (BFT) were downregulated and age-related orthologs (such as SQUAMOSA-PROMOTER BINDING PROTEIN-LIKE (SPL)s, see below) were activated in C. sativa. Given that solitary flowers can develop under both long and short photoperiods, it has been proposed that C. sativa is day-neutral in this aspect of flower-induction (Spitzer-Rimon et al., 2019). Further research directly comparing the timing of induction of solitary flowers in Cannabis plants grown under short and long days is required to determine whether the appearance of solitary flowers is photoperiod independent. It is still unclear as to whether the induction of solitary flowers signifies the end of the vegetative phase, as vegetative growth can continue at the SAM for the period between emergence of solitary flowers and terminal flowering at the shoot apex (Figure 3A; Stages 2201 – 2202).

Inflorescence flowering is marked by changes in the architecture of the shoot apex, which forms a highly branched compound raceme consisting of condensed branchlets and repeating phytomer structures (Figures 2B, 3A; Stage 2202). These phytomer structures consist of an internode, foliage leaf (supported by a petiole), bracts, and solitary flowers (stigma, style, perigonal bract and stipule) (Figures 2D, 3A; Stage 2201) (Spitzer-Rimon et al., 2019). Proliferation of these phytomer structures leads to the development of floral buds (Figure 3B), the main cultivation product of medicinal cannabis (Chandra et al., 2017). The compact nature of inflorescences can vary between genotypes and is affected by environmental stimuli, including light spectrum and intensity (Spitzer-Rimon et al., 2019; Danziger and Bernstein, 2021). While C. sativa is considered a short-day plant, some varieties exhibit photoperiod-independent flowering behavior (Figure 3C; ‘autoflowering’), producing flowers in response to maturity (Gloss, 2015). Similarly, not all plants will form terminal flowers at the apical meristem, even after several months of inflorescence flowering under inductive SD conditions (Spitzer-Rimon et al., 2019). These inconsistencies in flowering behavior indicate that the molecular mechanisms underlying floral initiation and inflorescence structure have a high level of heterogeneity in C. sativa.

The complexity of the morphophysiological characteristics associated with flowering behavior in C. sativa has led to inconsistencies in nomenclature and in the reporting of these traits (Spitzer-Rimon et al., 2019; Petit et al., 2020b; Woods et al., 2021) (Supplementary Table S4). We propose that there are four main events which take place during florogenesis: 1) induction of solitary flowers, typically in the axils of the stipules (Figures 2D, 4A), 2) formation of axillary branches and the transition to higher order branching (Figure 4B), 3) the onset of inflorescence flowering, marked by the formation of flower clusters at the shoot apex and axillary branches (Figure 4C), and finally 4) terminal flowering, when the apical meristem has transitioned to a terminal flower (Figure 4D). Changes in shoot apex architecture and inflorescence flowering can be inducible under short photoperiods and these characteristics appear to be regulated independently of solitary flower formation (Spitzer-Rimon et al., 2019).

Flowering time and phytocannabinoid production

Phytocannabinoid content and yield is known to be highly variable and dependent upon genotype, growth stage, flowering behavior, and cultivation environment. Female C. sativa inflorescences are a rich source of hundreds of specialized metabolites, including phytocannabinoids (Welling et al., 2021; Welling et al., 2022). Phytocannabinoid biosynthesis is concentrated within glandular trichomes (Livingston et al., 2020), present on the perigonal bracts as well as other modified floral leaves within pistillate inflorescences. The capitate stalked trichome is the most abundant trichome morphotype in pistillate inflorescences and these and are principally responsible for the high concentration of phytocannabinoids in C. sativa plants (Livingston et al., 2020).

Many factors are capable of determining phytocannabinoid yield, including plant variety and age, planting density, and light intensity (Backer et al., 2019). Flowering time has a strong effect on phytocannabinoid accumulation, with rapid accumulation occurring in the first 3 weeks of inflorescence flowering (Stack et al., 2021). Importantly, both plant architecture and the accumulation of inflorescence biomass are strongly affected by flowering time (Stack et al., 2021). Comparisons between early and late flowering genotypes also indicate a limited trade-off between floral biomass and phytocannabinoid concentration, with genotypes producing the highest amounts of floral biomass also having the highest phytocannabinoid levels (Stack et al., 2021). This data indicates that the genetic manipulation of flowering pathways could be used as a viable strategy to increase phytocannabinoid yield within C. sativa commercial production systems.

Photoperiod-dependent pathways in C. sativa

The photoperiod-dependent flowering pathway involves light-sensing proteins (phytochromes and cryptochromes) which coordinate with the circadian clock to regulate the expression of the phosphatidylethanolamine-binding protein (PEBP) family, including a sub-family related to the FT protein (Figures 6A, 7A). PEBP members can function both as inducers and inhibitors of flowering. C. sativa is particularly sensitive to photoperiodic changes, with the time to flower reduced in SD conditions (Hall et al., 2012). The PEBP gene family is well represented in C. sativa, with both putative inducers and inhibitors of flowering present (see below) (Figure 7A). The flowering time network of the model species A. thaliana is well-defined with several pathways converging on floral integrator genes (Blümel et al., 2015), including FT, TWIN SISTER OF FT (TSF; (Amasino, 2010)), and SOC1 (Figure 7). FT and its orthologs are synthesized in the leaves of several plant species and encode proteins that function as florigens and anti-florigens, promoting or inhibiting floral initiation at the shoot apex, respectively. A. thaliana possesses five phytochromes: PHYA through PHYE, the signals from which are received by the GIGANTEA-CONSTANS-FT (GI-CO-FT) signaling cascade. Stabilized by PHYA, the nuclear TF CONSTANS (CO) activates transcription of FT (Putterill et al., 1995; Samach et al., 2000). The FT locus produces florigen in the leaves which then travels to the shoot apical meristem to initiate flowering (Corbesier et al., 2007). GI, a circadian clock gene, facilitates the degradation of transcriptional repressors responsible for repressing the expression of CO, indirectly promoting FT (Sawa et al., 2007). CO indirectly upregulates the MADS-box TF gene SOC1, which activates the floral meristem identity gene LEAFY (LFY) to promote flowering (Yoo et al., 2005; Lee et al., 2008). FLOWERING LOCUS C (FLC)–like genes negatively regulate flowering time in the autonomous and vernalization flowering pathways, with elevated levels of FLC resulting in later flowering in A. thaliana (Sheldon et al., 2000). FLOWERING LOCUS D (FLD) codes for the FLD TF, which regulates FLC. FLD facilitates histone demethylation at the FLC locus, deactivating FLC expression and triggering flowering (He et al., 2003; Jiang et al., 2007). The overexpression of TERMINAL FLOWER 1 (TFL1)/CENTRORADIALIS (CEN)-like genes also delays flowering and alters flower architecture in Hevea brasiliensis (Bi et al., 2019) and CENTRORADIALIS (CEN)-like protein 1 (encoded by CET1) is highly expressed in the developing inflorescences of A. thaliana and Antirrhinum (Bradley et al., 1996; Bradley et al., 1997).

CO-like (COL) genes are TFs in pathways associated with growth and development, including the photoperiod-dependent flowering pathway (Figure 6). The COL gene family is known to regulate flowering under both SD and LD conditions, with negative regulators under both photoperiods in rice (Oryza sativa; a facultative SD plant), OsCOL10, OsCOL13 and OsCOL16 as well as Hd1, a promoter of SD dependent flowering that suppresses flowering under LD conditions (Yano et al., 2000; Sheng et al., 2016; Tan et al., 2016; Wu et al., 2017). Overexpression of COL genes in A. thaliana (AtCOL3, AtCOL7 and AtCOL8) delays flowering while the overexpression of AtCOL5 increases the expression of FT to promote flowering (Datta et al., 2006; Hassidim et al., 2009; Takase et al., 2011; Wang et al., 2013). Pan et al. (2021) conducted an analysis of the CONSTANS-like gene family in C. sativa (CsCOL) and identified 13 CsCOL genes (CsCOL1 – CsCOL13), unevenly distributed across 7 chromosomes and primarily located on Chr 10. Ten CsCOL genes were preferentially expressed in the leaves, two in the female flower (CsCOL2 and CsCOL3), and one in the stem (CsCOL13). Most CsCOL genes identified by Pan et al. (2021) exhibited a diurnal oscillation pattern under SD and LD conditions and sequence analysis indicated amino acid differences for CsCOL3 and CsCOL7 among early flowering and late flowering varieties. At peak transcription levels, CsCOL4 and CsCOL11 expression levels were higher in the two early flowering varieties tested, compared to those of the two late flowering varieties. The reverse was true for CsCOL6, CsCOL7, CsCOL9, and CsCOL12. This indicates that there may be multiple CsCOL genes functioning as promoters or suppressers of flowering to regulate flowering time in C. sativa. While gene functions and mechanisms can differ between species, the apparent conservation of GI, CO, and FT in the flowering pathways of many crops (Watanabe et al., 2011), along with the photoperiod-dependent regulation of FT-like expression (Chen et al., 2022) and COL expression in C. sativa suggest that these may be ideal candidates in determining the regulation of flowering time in C. sativa and warrant further investigation.

In soybean (Glycine max), a SD dicot, flowering time is regulated by E genes and JUVENILE (J), also known as GmELF3 (Copley et al., 2018). GmELF3 is orthologous to A. thaliana EARLY FLOWERING3 (ELF3), that encodes a key component of the circadian clock (Lu et al., 2017). E1 is a legume-specific TF and E2, E3, and E4 are orthologous to genes associated with the regulation of flowering time in A. thaliana. E2 (also GmGIGANTEAa) is an ortholog of GIGANTEA (GI), and E3 (GmPHYA3) and E4 (GmPHYA2) are orthologs of PHYA. Under long day conditions, GmPHYA3 and GmPHYA2 promote E1 expression and inhibit GmELF3 expression. E1 up-regulates GmFT4a and down-regulates GmFT2a and GmFT5a, all of which are FT homologs (Xia et al., 2012; Nan et al., 2014; Zhai et al., 2014; Samanfar et al., 2017). GmGIa (a GI homolog) delays flowering under LD conditions by inhibiting GmFT2a (Watanabe et al., 2011). The E1 to E4 loss-of-function alleles result in photoperiod insensitive flowering due to increased FT gene transcript levels (Xu et al., 2013). Under SD conditions, GmELF3 represses E1, releasing the E1 suppression of the GmFT genes, promoting flowering (Xia et al., 2012; Lu et al., 2017). Flowering time variation in soybean is caused, in part, by natural variation in the GmFT gene family (Jiang et al., 2019). C. sativa has two GI (LOC115708742 and LOC115722652), three ELF3 (LOC115703149, LOC115697482 and LOC115707722) and three PHY homologs (PHYA: LOC115719277, PHYB: LOC115721719, and PHYE: LOC115697533) and, as such, these genes may assist in understanding variation in sensitivity to photoperiod in C. sativa (Supplementary Table S1).

Photoperiod affects many aspects of plant development, including the initial elongation of flower stalks, flower initiation (Blümel et al., 2015), meristem termination, bud dormancy and branching. Overexpression of FT homologs induces very early flowering in eudicot plants, such as tomato (Solanum lycopersicum; Lifschitz et al., 2006), and monocot plants, such as rice (Izawa et al., 2002; Kojima et al., 2002). FT and TSF also promote lateral shoot development in A. thaliana, independently of their effect on floral initiation (Hiraoka et al., 2013). Additionally, BRANCHED1/TEOSINTE BRANCHED1-LIKE 1 TF, a key negative regulator of branching in A. thaliana, can inhibit the function of both FT and TSF (Niwa et al., 2013). A similar mechanism exists in C. sativa, given that a short photoperiod promotes intense branching of the inflorescence (Spitzer-Rimon et al., 2019). Research in day-neutral tomato (Solanum lycopersicum) has explored the nature of the relationship between branching and flowering, with late-flowering mutants showing a greater propensity to revert to vegetative functioning in the inflorescence. It has been suggested that there are common mechanisms between the inhibition of vegetative growth in the shoot apical meristem and the number of lateral meristems initiated in the inflorescence (Périlleux et al., 2014). FA (FALSIFLORA) and SINGLE FLOWER TRUSS (SFT) are the tomato orthologs of the A. thaliana LFY and FT genes, respectively (Molinero-Rosales et al., 1999; Lifschitz et al., 2006). Mutants fa and sft exhibit leaf production in the inflorescence (Allen and Sussex, 1996; Molinero-Rosales et al., 1999; Molinero-Rosales et al., 2004) with additive late-flowering phenotypes, indicating that the genes act in parallel pathways (Molinero-Rosales et al., 2004; Thouet et al., 2012). Conversely, FA and SFT are floral promoters, with overexpression of either accelerating flowering (Lifschitz et al., 2006; MacAlister et al., 2012). The early flowering tomato mutant terminating flower (tmf) exhibits a reduction in the number of vegetative phytomers, like that of plants overexpressing FA or SFT (MacAlister et al., 2012). TMF acts upstream of FA and independently of SFT to maintain a vegetative shoot apical meristem. Both FA and LFY are floral meristem identity genes, expressed in leaf primordia before flowering with expression increasing with transition from a shoot apical meristem towards a flowering meristem (Molinero-Rosales et al., 1999; Thouet et al., 2012).

The FAC consists of FT, a 14-3-3 protein, and FD and plays a vital role in promoting flowering in tomato (Pnueli et al., 2001) (Figure 6). SFT interacts with a 14-3-3 protein, in tomato, facilitating the interaction with SELFPRUNING (SP; an ortholog of TERMINAL FLOWER1)-interacting G-BOX (SPGB) to form the FAC (Pnueli et al., 2001). Song et al. (2020) examined the interactions between FTL1, a tomato FT paralog, SPGB and three 14-3-3 isoforms and determined that FTL1 interacts with 14-3-3/2 to form the FAC, with SPGB regulating tomato flowering. Allelic variation in SELF-PRUNING 5G (SP5G), an FT paralog, reduces the LD response and contributes to the loss of day-length-sensitive flowering in tomato (Soyk et al., 2017; Zhang et al., 2018b). FTL1 was induced by SD conditions, as opposed to LD conditions, with transcript levels indicating a strong diurnal oscillation (Song et al., 2020). SFT is a floral inducer but does not respond to day length (Molinero-Rosales et al., 2004; Lifschitz et al., 2006), acting downstream of FTL1 to regulate SD dependent flowering. Disruption of both SP5G and FTL1 function induces day-neutral flowering in tomato, by enhancing or reducing SFT expression under LD or SD conditions (Soyk et al., 2017; Song et al., 2020). SFT induces early flowering in tomato and is conserved in other species (Lifschitz and Eshed, 2006; Lifschitz et al., 2006). In C. sativa, there are nine 14-3-3 and two FD putative homologs present in C. sativa, suggesting the existence of similar pathways (Supplementary Table S1).

Analysis of the expression of FLOWERING LOCUS T-like (FT-like/LOC115697736/FT3) and CET1/LOC115697843/CEN1 in C. sativa accessions from different latitudes shows that wild accessions flowered under both LD and SD conditions and that the cultivated plants only flowered in SDs. FT-like expression was significantly higher in the wild accessions under LD conditions and was positively correlated with the latitude of origin. Cultivated plants showed low FT-like expression under LD conditions, while FT-like expression was high and rapidly followed flowering in all accessions under SD conditions, suggesting that FT-like may promote flowering. The relatively unchanged expression of CET1 across developmental stages has been interpreted by some authors as evidence that flowering behavior is not controlled by autonomous or vernalization pathways and that cultivated C. sativa has adapted to different photoperiods through the regulation of FT-like expression (Chen et al., 2022).

To clarify the relationship between C. sativa PEBP members, we searched the cs10/CBDRx genome and compared the PEBP genes identified to those well characterized PEBP genes from the model plant A. thaliana, tomato, and the SD plant Chrysanthemum seticuspe (Oda et al., 2011). This revealed that there are 12 PEBP family members in C. sativa, with four FT-like (CsFT1 through CsFT4), three closely related to MOTHER OF FT (MFT, CsMFT1 through CsMFT3), two related to TERMINAL FLOWER (TFL) and A. thaliana CENTRORADIALIS (ATC) (CsATC and CsTFL), as well as three BROTHER OF FT (BFT)/CEN genes (Figures 7A, B, Supplementary Table S1). Two of the CsMFT clade genes (CsMFT1/LOC115711426 and CsMFT2/LOC115711470) are almost identical in cs10/CBDRx (Figure 7B), with a five-nucleotide insertion/deletion in the 3’ untranslated region, and two synonymous single nucleotide polymorphisms in the coding region. These two genes are also close together on cs10/CBDRx Chr 3 (NC_044372.1) at 92,271,234 – 92,269,219 bp and 92,136,895 – 92,134,894 bp, respectively. To investigate the possibility that these two annotated genes are incorrectly annotated, perhaps because of heterozygosity-induced assembly errors, we examined the genomes of two other cultivars, Finola and Abacus, and could only detect a single MFT1/2 gene in each case, at the corresponding genomic location. CsFT1 through CsFT4 all have a conserved tyrosine at the Y88 position seen in floral promoting-PEBP proteins (Figure 7B).

The expression of some PEBP family members in C. sativa has been examined in two studies (Chen et al., 2022; Spitzer-Rimon et al., 2022). CsFT3/LOC115697736, also (called FT-like in (Chen et al., 2022)), exhibits increased expression in the first and second apical leaf pairs following the shift from LD to SD conditions in two wild and two cultivated C. sativa strains (Chen et al., 2022). This suggests that the gene may mediate the promotion of flowering in response to a shortening of photoperiod. Six CsPEBP genes were differentially expressed in nodes 4 (vegetative), 6 (vegetative) and 7 (reproductive) (Spitzer-Rimon et al., 2022). The three CsBFT/CEN genes exhibited reduced expression in node 7, compared to nodes 4 and 6, and CsMFT3 showed slightly reduced expression. The CsFT4 gene exhibited increased expression, in node 6, which was unexpected as FT has an amino acid sequence indicative of a floral promoter (Spitzer-Rimon et al., 2022). The expression of CsTFL was also reduced in node 6 and node 7, suggesting it may be involved in the maintenance of vegetative function at the shoot apex in vegetative plants. To further clarify the expression of these genes across the whole C. sativa plant, we examined their expression in a wide variety of tissues using existing RNASeq datasets and found that the relative expression of CsMFT1, CsMFT2, and CsMFT3 was greatest in seed, with CsMFT2 expression reduced in mature Finola flower and Finola root tissues (Figure 7C, Supplementary Table S3).

Temperature-dependent pathways in C. sativa

The vegetative phase is distinguished by a temperature-dependent basic vegetative phase (BVP) and a daylength-dependent photoperiod induced phase (Lisson et al., 2000). In hemp, a base air temperature of ~1°C and a range of 306 – 636°Cd (thermal time) is required for completion of the BVP (Amaducci et al., 2008; Amaducci et al., 2012). The vegetative stage can also be defined by the number of fully developed leaves (Mediavilla et al., 1998) (Figure 2A). While there is little evidence to suggest that C. sativa has vernalization requirements, temperature is known to be a factor affecting the length of the juvenile stage, with reduction in temperature increasing the time to floral initiation and flowering (Supplementary Table S4) (Lisson et al., 2000; Amaducci et al., 2012; Salentijn et al., 2019).

Temperature contributes to the regulation of flowering time through multiple pathways. In A. thaliana, the vernalization pathway controls flowering in response to extended cold periods. The vernalization-related gene VERNALIZATION1 (VRN1) codes for a protein that acts to repress the floral repressor TF, FLC (Figure 6), subsequently allowing the expression of flowering integrator genes (Michaels and Amasino, 1999; Levy et al., 2002). A VRN1 ortholog has also been identified in a hemp QTL for full flowering (Petit et al., 2020a) (Figure 5). Changes in ambient temperature play a key role in the floral induction of A. thaliana under non-inductive SD photoperiods (Balasubramanian et al., 2006; Lee et al., 2007). The type II MADS-box TFs FLOWERING LOCUS M (FLM) and SHORT VEGETATIVE PHASE (SVP) assist in regulating ambient temperature-responsive flowering by repressing the expression of florigen genes (Scortecci et al., 2001; Lee et al., 2007). FLM produces multiple splicing variants including FLM-β and FLM-δ, with overexpression of these resulting in late flowering and early flowering, respectively (Pose et al., 2013). At elevated temperatures, ubiquitin-mediated proteasomal degradation reduces SVP while alternative splicing reduces the abundance of FLM-β but increases the abundance FLM-δ (Jin and Ahn, 2021). SVP was also present in our analysis of protein-protein interactions (Supplementary Figure S2) and has been shown to inhibit floral transition in the A. thaliana autonomous flowering pathway by acting with AGAMOUS-LIKE 24 (AGL24) and AP1 to control floral meristem identity (Gregis et al., 2008).

FLC is central to the flowering regulatory network in A. thaliana and the control of flowering in response to seasonal cues (Madrid et al., 2020). Floral transition is inhibited by FLC binding directly to genes that encode activators of flowering, to repress their transcription (Michaels and Amasino, 1999; Sheldon et al., 1999). FLC targets SOC1 (Hepworth et al., 2002; Helliwell et al., 2006; Searle et al., 2006), which encodes a MADS-domain TF that regulates genes involved in floral transition at the shoot apex (Samach et al., 2000; Immink et al., 2012) and assists with floral transition in non-inductive short days (Moon et al., 2003). SOC1 transcription is activated during vernalization as FLC transcription is repressed (Hepworth et al., 2002; Searle et al., 2006; Deng et al., 2011). FLC binds DNA as heterodimers with other members of the MADS-domain TFs family (de Folter et al., 2005; Li et al., 2008; Gu et al., 2013) and, as such, it is important to consider the specificity of MADS-domain complexes including FLC and partner protein availability when examining FLC function and target-specific regulation (Mateos et al., 2015; Madrid et al., 2020).

There is limited expression data for SOC1 and FLC in C. sativa (Chen et al., 2022) (Figure 6). To clarify the relationship between C. sativa MADS members, we searched the cs10/CBDRx genome and compared the MADS genes identified to those well characterized MADS-box genes from the model plant A. thaliana, and grapevine Vitis vinifera (Figure 8; Supplementary Figure S1, Supplementary Table S2). This identified one FLC-like gene, three SVP-like and three SOC1-like genes suggesting the involvement of these MADS genes in floral transition in C. sativa.

The TF PHYTOCHROME-INTERACTING FACTOR 4 (PIF4) is thought to positively regulate high-temperature-induced flowering by binding to the FT promoter region and increasing FT transcription (Figure 6) (Kumar et al., 2012). PIF4 transcription is regulated by multiple TFs, with TEOSINTE BRANCHED 1/CYCLOIDEA/PCF 5 (TCP5) thought to positively regulate PIF4 transcription in response to warm temperatures. Greater ambient temperature increases PIF4 expression and enhances the accessibility of PIF4, increasing the expression of thermal-responsive genes (Jin and Ahn, 2021). FT binds a membrane phospholipid (phosphatidylglycerol) at low temperatures, restricting mobility. This binding is less preferrable at higher temperatures, allowing FT to travel to the shoot apical meristem and induce flowering. Flowering time is subsequently optimized by the adjustment of florigen (flowering hormone) activity, with cellular membranes sequestering FT by binding the phospholipid, in response to temperature changes (Susila et al., 2021). Similar pathways may exist in C. sativa, where one PIF3 and one PIF5 homolog are present (Supplementary Table S1).