Co-expression analysis identifies organ-specific gene regulatory networks responsive to phosphate limitation in hemp (Cannabis sativa L.) Authors: Benjamin Wee Y, Ricarda Jost, James Whelan, Oliver Berkowitz Published: 2026 ABSTRACT AbstractBackgroundCannabis sativa L. is a compelling model species for studying how human selection has shaped plant traits through domestication. Cannabis hemp-type varieties are grown for fibre and seed, whereas drug-types are cultivated for high cannabinoid content in pharmaceutical and recreational use. These contrasting selection criteria have led to divergent physiological and morphological traits, providing a valuable model to investigate underlying molecular and physiological mechanisms. Cultivation of hemp on more marginal soils has favoured a greater capacity to acclimate to nutrient limitations, particularly phosphorus deficiency. However, the mechanisms underlying enhanced phosphorus acclimation in hemp remain poorly understood.ResultsUsing Weighted Gene Co-expression Network Analysis (WGCNA) of RNA-seq data from multiple organs of hemp-type Cannabis grown under contrasting phosphate supplies, we performed a global transcriptome analysis to capture organ-specific and phosphate-responsive gene expression dynamics. This analysis identified twelve co-expression modules capturing coordinated gene expression across organs and phosphate conditions, reflecting processes related to organ function (e.g. root development, fibre production, and flower development) and metabolism (including photosynthesis, nutrient transport, and lipid remodelling), as well as their relationships to physiological traits. Two modules, representing leaf- and root-specific responses, were strongly associated with plant phosphorus status, and cross-comparison with genes under selection during Cannabis domestication showed significant overlap with these modules. Gene regulatory network analysis revealed how gene expression supports prioritisation of resource allocation towards reproductive organs, with the Cannabis homolog of SPX DOMAIN GENE3 (SPX3) identified as a central hub in leaf tissues coordinating the shoot phosphate starvation response. In the root-specific module, root system architecture remodelling was mediated by transcription factors including SCARECROW-LIKE (SCL) family members and a homolog of the C2H2-type transcription factor SENSITIVE TO PROTON RHIZOTOXICITY (STOP). Strigolactones emerged as key hormonal regulators of the root-specific PSR, highlighting their role in phosphate deficiency tolerance in hemp.ConclusionThese findings provide an integrated, organ-specific view of gene networks underlying the phosphate starvation response in hemp and reveal distinct features of nutrient adaptation following domestication. They identify targets for improving nutrient use efficiency in hemp and support comparative studies across Cannabis germplasm.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12870-026-09379-9.Keywords: Hemp, Cannabis sativa L., Phosphorus, Nitrogen, Strigolactones, Gene networks, Co-expression, Nutrient use, Source-sink relations, WGCNA Background Phosphorus (P) is an essential macronutrient and fundamental for plant processes including energy transfer, nucleic acid synthesis, membrane formation, photosynthesis, and reproductive development [1–3]. Consequently, P availability strongly influences crop growth, productivity, and reproductive performance, with plant species and cultivars differing markedly in their P requirements and nutrient-use efficiencies [1, 4]. Changes in nutrient-use efficiency, particularly for nitrogen (N) and P, were major drivers in the domestication of many crops [5, 6]. For example, the domestication of soybean led to increased photosynthetic P-use efficiency and improved acclimation to P-limited conditions [7, 8], whereas in maize loss of an ancestral teosinte allele impacted N assimilation leading to lower seed protein content [9]. Owing to its long domestication history, diverse cultivation practices, and distinct usage-types, Cannabis sativa represents a remarkably valuable model for investigating variation in P-use strategies in crops [10]. Hemp and drug-type cultivars differ substantially in their P requirements and nutrient-use characteristics, reflecting contrasting selection pressures during domestication and cultivation. Hemp-types are generally considered more nutrient efficient and adapted to lower-input environments, whereas drug-type cultivars have largely been selected in high-input cultivation systems with sustained nutrient availability [11–13]. To acclimate to P limitation, plants have evolved a conserved phosphate starvation response (PSR), which coordinates a broad range of physiological and molecular processes, including enhanced Pi uptake and transport, remobilisation of phosphate from older tissues, membrane lipid remodelling, and metabolic adjustments that maintain growth and reproductive development [14, 15]. Emerging evidence in Cannabis further suggests that specifics in the regulation of the PSR may differ between usage-types, highlighting the importance of understanding transcriptional and physiological responses to phosphate availability in Cannabis [11–13]. Recent physiological and gene expression analyses of hemp indicate that this adaptation is associated with efficient internal P recycling and strong source–sink coordination. Under P limitation, nutrients are preferentially remobilised from older source leaves, while reproductive sink tissues maintain P-dependent metabolism and development [12]. These findings suggest that organ-specific regulation of nutrient allocation is a key component of P-use efficiency in hemp. Transcriptomic approaches provide a valuable framework for examining the PSR at the molecular level. Methods such as whole transcriptome sequencing (RNA-seq) generate large and complex datasets which enable the identification of key genes and pathways involved in stress responses across developmental stages and tissue types [12]. When coupled with physiological data, this provides a systems-level understanding of how plants respond to environmental cues. However, the scale and complexity of RNA-seq datasets require careful interpretation to derive meaningful biological insights [16]. Numerous pipelines exist to process RNA-seq data, quantify gene expression, and identify differentially expressed genes (DEGs) through statistical comparison of expression levels [17, 18]. These are often followed by clustering approaches, such as hierarchical or K-means clustering, and gene ontology enrichment analyses to infer functional relevance [19–22]. However, these methods rely on user-defined parameters for gene grouping and provide limited capacity to leverage correlations in transcript abundances for further interpretation and integration with phenotypic traits. Greater insights can be uncovered by extending beyond DEG-centric pipelines, which often prioritize only the most strongly expressed genes, enabling the discovery of more nuanced and concerted transcriptional responses that might otherwise be overlooked [20, 21, 23, 24]. To enable the unbiased analysis of large RNA-seq datasets, gene co-expression network analysis applies the ‘guilt by association’ principle, whereby genes with correlated expression and high connectivity are inferred to share regulatory mechanisms, participate in common pathways, or converge on biological functions under investigation [20, 21, 24, 25]. Weighted Gene Co-expression Network Analysis (WGCNA) is a widely adopted systems biology method that enables the identification of relationships between gene expression patterns, traits, and sample types, supported by robust statistical frameworks [26, 27]. It constructs gene co-expression networks, effectively reducing thousands of genes into an interpretable number of co-expressed groups (modules). Further correlation of these modules to traits of interest allows for isolation of key modules that contain biologically important genes for further downstream analysis, leading to identification of hub genes designated as potential biomarkers and/or key transcription factors for further investigation. It therefore enables the interpretation of complex transcriptomic datasets and supports the identification of actionable biological insights. WGCNA has uncovered key regulatory networks and regulatory genes across many abiotic and biotic stresses—including in a number of studies investigating phosphate use across different model and crop species [28–31] or terpenoid accumulation in Cannabis [32]. A recent study further demonstrated its utility in Cannabis by identifying domestication-associated differences in nutrient use in drug-type cultivars [11]— highlighting its potential for crop improvement and stress resilience. Our previous work characterised the morphological and physiological responses of dual-use hemp-type Cannabis to phosphate (Pi) supply, alongside the expression of a curated set of genes directly involved in P and N metabolism and their regulation, drawing on well-characterized model species such as Arabidopsis and rice [12]. In particular, we identified negative feedback loops between phosphate (Pi; inorganic phosphate) and nitrate signalling that differ from established model systems [12]. However, the global networks underlying these changes in gene expression and their regulatory components were not resolved. Here, we extend this work by employing WGCNA to build gene networks from the full transcriptome and correlating their expression profiles to physiological traits. We identified twelve modules of genes with distinct co-expression patterns across organs and Pi treatments. Subnetworks of highly correlated genes within a leaf-specific module revealed relationships between a key suppressor of the PSR, SPX DOMAIN GENE3 (SPX3), and its downstream targets. Networks associated with the root response to Pi-limitation highlighted a central role for strigolactone signalling. Together, these findings provide a systems-level understanding of the organ-specific PSR in Cannabis and establish a foundation for improving nutrient use strategies in hemp, as well as enabling comparisons with drug-type cultivars. Materials and methods Plant material, treatments and sampling Generation of plant material and biochemical analyses were described in detail previously [12]. Briefly, clonally propagated rooted cuttings of the dual purpose female hemp variety Han NW (MW6-15 [33], original seeds were sourced from the Ecofibre Global Germplasm Collection managed by Southern Cross University, Lismore, Australia) were grown on medium-grade Perlite and established for 13 days under long-day conditions (18-h day/6-h night cycle, 200 µmol m−2 s−1 light intensity) before transfer to flower-inducing short-day conditions (12-h day/12-h night cycle, 600 µmol m−2 s−1 light intensity) under varying P supplies (0, 0.25, 0.5, 1 or 2 mM KH2PO4) for a further 27 days until the early flowering stage. Samples were obtained from different organ types (Top Flower—the terminal inflorescence located at the apex of the main stem, TF; Young Stem, YS; Mature Leaf, ML; Old Leaf, OL; Lateral Root, LR). Determination of key metabolites (nitrate, phosphate, sulfate, protein, starch, chlorophyll, anthocyanin) and total elemental analysis were also described previously [12]. RNA-seq analyses Quantification of gene expression determined by RNA-seq for TF, YS, ML, OL and LR organs was described previously and all data were sourced from Wee Y et al. (2025) [12]. Genes with TPM > 10 per organ type across treatments were retained to identify differentially expressed genes (DEGs) using the “sleuth” package, with cut-offs for false-discovery-rate (FDR) < 0.05 and expression change greater than two-fold (|log2FC|> 1). The function prcomp was used to perform principal component analysis in R (4.3.2) on the raw transcriptomic dataset (TPM). The relationship was visualized using the function autoplot in base R. WGCNA and module detection analyses The R package “WGCNA” [26] was used to build a weighted gene correlation network (WGCN) using TPM values as input in R. To construct the WGCN, a β-soft power threshold (SFT) of 15 was used (R2 = 0.80) (Supplementary Figure S1), using an unsigned network and modified parameters (minModSize = 200; maxBlokSize = 21,000; mergingThresh = 0.25). This resulted in 12 modules with distinct module eigengene (the first principal component of each module) profiles for organs and treatments. Based on module membership and trait significance, gene expression profiles were correlated with quantified metabolites, gene ontology (GO) terms, and positively selected genes in domesticated Cannabis usage-types relative to basal types, based on whole genome resequencing of 110 Cannabis accessions [34]. For each module, GO term over-representation analysis was performed using the “clusterProfiler “ package, utilizing the simplify function with the “Rel” operation to remove redundancy [35]. GO annotations for the analysis were based on the cs10 genome assembly (GCA_900626175.2) and obtained from NCBI. Fisher’s exact test was performed using the fisher.test function in base R to assess whether differentially expressed genes were non-randomly distributed across co-expression modules among positively selected Cannabis genes [34]. A 2 × 12 contingency table (DE vs. non-DE by module) was constructed, and p-values were estimated via Monte Carlo simulation (1,000,000 iterations, simulate.p.value = TRUE, B = 1e6). These genes were further evaluated for their enrichment in specific co-expression modules. For each module, a 2 × 2 contingency table was constructed, comparing module membership of Cannabis genes and the background WGCNA input gene dataset. P-values were considered significant at p < 0.05. The results were visualised using the package “ggstatsplot” [36], representing the proportion of differentially expressed genes as a percentage of total genes in each respective module. Network construction Networks were constructed from the shortlisted “red” and “greenyellow” modules by exporting the topological overlap matrix (TOM) of module genes using the exportNetworkToCytoscape function from the WGCNA package. A soft threshold of 0.2 was applied to retain only meaningful edge connections, reflecting co-expression strength between gene pairs. The resulting node (genes in the exported network) and edge (gene–gene connections) lists were imported into Cytoscape to construct gene regulatory networks for each module [37]. To improve interpretability of the generated networks, module-specific filtering strategies were applied. Due to the large size of the ‘red’ module, a bait-gene approach was implemented to extract first-neighbour sub-networks focused on the relevant PSR genes of interest. The small “greenyellow” module was processed without bait-based filtering. For both modules, genes lacking annotation, not differentially expressed, or with low module membership (kME < 0.74) were excluded. To provide further biological relevance, gene families with known involvement in the PSR in Arabidopsis were matched to putative orthologs in the cs10 reference genome and verified via local BLASTp searches against Arabidopsis proteins annotated in TAIR (https://www.Arabidopsis.org/), using a cut-off of E < 10–50 and selecting matches with the lowest E-values [38, 39]. Once networks were constructed, hub genes were further characterised by calculating betweenness centrality (BC) for each gene within each network via the “CytoNCA” plugin, which quantifies the importance of a gene as an intermediary in network connectivity [40]. Heatmap construction The ComplexHeatmap package in R was used to generate heatmaps [41]. Heatmaps were constructed using z-scored (row-scaled) data to visualise the normalised distribution of expression across organ types in response to the Pi treatments for each gene or metabolite. Results Principal component analysis reveals phosphate-driven, organ-specific transcriptome responses In our previous work, we profiled transcriptional changes across five organs—top flower (TF, the final flower developed from the apical meristem), young stem (YS), mature leaf (ML), old leaf (OL), and lateral roots (LR) – of a dual-purpose hemp cultivar under four Pi supply levels (0, 0.25, 1 and 2 mM KH2PO4). This revealed expression changes in a curated set of genes involved in P and N assimilation pathways and their regulation across organs, and their relationship to physiological responses [12]. Here, we extend this analysis to the whole-transcriptome level. Principal component analysis (PCA) showed that Pi supply was the main driver of RNA-seq sample segregation (Fig. 1A; PC1). Samples from plants supplied with sufficient to excess Pi (1–2 mM) clustered together across organ types, whereas organ-specific segregation became more pronounced under Pi limitation (0–0.25 mM). In LR and OL, samples under Pi limitation (0.25 mM) were clearly separated from those under complete Pi withdrawal (0 mM), whereas in ML these conditions clustered together. By contrast, sink (TF) and transport (YS) organs showed distinct separation of Pi withdrawal samples from other treatments, with Pi-limited samples more closely aligned with Pi-sufficient conditions. Principal component analysis reveals phosphate-driven, organ-specific transcriptome responses. Principal component analysis (PCA) of RNA-seq–derived transcriptomic profiles across organs under contrasting phosphate (Pi) supply. Samples separate primarily according to Pi availability along PC1, while variation along PC2 is consistent with differences between biological replicates. Abbreviations: LR, lateral roots; OL, old leaf; ML, mature leaf; YS, young stem; TF, top flower Overall, these transcriptomic patterns aligned with our previous findings on differential expression of P/N-related genes under Pi deficiency [12]. WGCNA identifies organ-specific and Pi-responsive gene co-expression modules To identify groups of co-expressed genes and assess their biological relevance, we performed weighted gene co-expression network analysis (WGCNA). After removal of lowly expressed genes (TPM < 10 across organ types per treatment), 18,665 genes were retained for analysis [26]. In agreement with previous results [12], hierarchical clustering identified one outlier sample (OL under Pi withdrawal), which was removed prior to network construction (Fig. 2A). WGCNA subsequently defined 12 modules of co-expressed genes (Fig. 2B, Supplementary Table S2). The module eigengenes, the first principal component for a given module, were calculated as proxies for the average expression of genes across organs and treatments (Fig. 2C). This separated modules with organ-specific but Pi-insensitive expression of genes (black, blue, brown, pink, purple, turquoise) from those with organ-specific and Pi-responsive expression (red, greenyellow, magenta, green and yellow). The latter likely contain hub genes associated with the phosphate starvation response (PSR). WGCNA identifies organ-specific and phosphate-responsive gene co-expression modules. A Hierarchical clustering of samples identified the OL sample under Pi withdrawal (P0) as an outlier, which was removed prior to network construction. The red line indicates the cut height used to define clusters (cut height = 5000). B Gene dendrogram showing the 12 co-expression modules identified based on the topological overlap matrix, with modules indicated by colour. The heatmap below shows gene significance, representing the correlation between individual gene expression profiles and measured physiological traits [26]. C Module eigengenes across organs (LR, lateral root; OL, old leaf; ML, mature leaf; YS, young stem; TF, top flower) and phosphate (Pi) treatments (0, 0.25, 1.0, 2.0 mM), highlighting organ-specific and Pi-responsive gene expression patterns. Module membership and associated genes are provided in Supplementary Table S2 To assess module functionality, gene ontology (GO) over-representation analysis was performed to identify biological processes associated with genes within each module (Fig. 3; Supplementary Table S3). Among organ-specific modules, the root-specific blue module was enriched for reactive oxygen species (ROS)- and ubiquitination-related GO terms, with the gene list including many peroxidases and E3 ubiquitin ligases needed for root growth [42]. The brown module, characterised by YS-specific expression, was enriched for GO terms such as cell-wall and lignin synthesis, including several genes encoding subunits of cellulose synthase or cellulose-modifying enzymes, likely driving cell elongation and expansion for fibre production [43]. The flower-specific turquoise module was enriched for developmental processes such as cell-cycle and nucleic acid transport, the latter associated with nuclear RNA export, consistent with rapid cell division in this organ. GO term enrichment of WGCNA gene co-expression modules. Bubble plot showing enriched Gene Ontology (GO) terms after over-representation analysis across the 12 co-expression modules identified by WGCNA. Dot colour represents statistical significance (false discovery rate; − log10 FDR) of enrichment, and dot size indicates the gene ratio (proportion of module genes associated with a given GO term). The number of genes in each module is shown in brackets. Modules associated with organ-specific phosphate starvation responses (PSR) are highlighted in colour. Full GO term annotations and associated genes are provided in Supplementary Table S3 For the modules showing P-responsive gene expression in leaves, the yellow module was enriched for photosynthesis-related GO terms, while the magenta module was enriched for mRNA and protein transport-related GO terms (Fig. 3), suggesting down-regulation of these processes under severe P-limitation (Fig. 2C). The leaf-specific red and root-specific greenyellow modules showed the strongest up-regulation of associated genes by P-limitation (Fig. 2C). Correspondingly, enriched GO terms for the red module included ‘cellular responses to Pi starvation’, ‘glycolipid biosynthesis’, and ‘flavonoid biosynthesis’. Underlying genes are classical markers of the PSR such as SPX DOMAIN GENE (SPX) or MONOGALACTOSYL DIACYLGLYCEROL SYNTHASE (MGD) [14, 15]. In the root-specific greenyellow module, enriched GO terms included ‘inorganic anion transport’, with transcripts encoding multiple high-affinity phosphate and sulfate transporters as well as Pi exporter PHOSPHATE1 (PHO1). Additional enrichment for reactive oxygen species (ROS) detoxification was observed, including transcripts encoding peroxidases associated with root growth [44] (Supplementary Table S3). Interestingly, genes associated with the GO term ‘nodulation’ were also enriched, despite Cannabis not forming symbiotic nodules. Closer inspection revealed that these genes encode transcription factors involved in ethylene and strigolactone signalling pathways, suggesting that these genes coordinate developmental and hormonal signalling pathways in hemp that influence root growth responses to Pi supply [45]. Co-expression modules correlate with metabolite profiles across Pi treatments A hallmark of WGCNA is its ability to relate gene modules to quantified traits using MEs as proxies of gene expression, enabling identification of statistically significant relationships. Normalisation of metabolite concentrations showed their distribution across tissues and suggested coordinated allocation to protect sink organs under varying Pi supply (Fig. 4A) [12]. Pi preferentially accumulated in TF when compared to other organ types, indicating targeted allocation of resources to this sink organ. Consistently elevated Pi levels in YS are in line with its role in transport to floral organs. Maintaining Pi flux and assimilation into organic P (Po) compounds, such as RNA, likely supports critical processes associated with flowering irrespective of external Pi availability [46]. This also translated into consistently higher levels of total N, particularly protein and RNA, indicating that developmental processes are maintained in floral organs even under Pi stress [47]. Metabolite distribution across organs and correlation with gene co-expression modules. A Distribution of metabolites across organs of Cannabis plants under varying phosphate (Pi) supply. Metabolite concentrations were normalised by z-scoring for comparison (data from [12]). Metabolites not determined in YS are indicated in grey. LR, lateral root; OL, old leaf, ML, mature leaf, YS, young stem, TF, top flower, chlorophyll (total chlorophyll); Pi (phosphate); Po (organic P). B Correlation between module eigengenes derived from WGCNA and normalised tissue concentrations of Pi and Po. Pearson correlation coefficients (r) are shown, with corresponding p-values in brackets. Negative correlations for the red and greenyellow modules reflect increased gene expression coinciding with decreased Pi or Po concentrations under Pi deficiency. The complete set of module–trait correlations is shown in Supplementary Figure S2 Several metabolites showed organ-specific accumulation under Pi limitation, including anthocyanins, total carbon (C), and multiple macro- (K, nitrate) and micronutrients (Fe, Mn, Zn). While anthocyanin accumulation in P-limited leaves was expected, starch accumulated with increasing Pi supply. Sulfate accumulated in roots irrespective of Pi supply, but under Pi limitation was increasingly allocated to shoots, especially YS and TF, indicating a critical role in Pi-stress mediated responses such as lipid remodelling [48]. Correlation analysis between module eigengenes and metabolite profiles was performed to identify associations between gene expression and metabolite abundance for each module (Fig. 4B, Supplementary Figure S2). Focusing on the Pi and organic P (Po) fractions, the brown, purple, and turquoise modules showed the strongest positive correlations with tissue Pi and Po concentrations, reflecting higher expression of genes within these modules on Pi-accumulating organs such as TF and YS (Fig. 2C). By contrast, the black module showed the strongest negative correlation with Pi and Po, reflecting high expression of its constituent genes in leaves, where Pi and Po accumulation was comparatively lower (Fig. 2C, Fig. 4A). While weaker, greenyellow and red modules also showed negative correlation to organ Pi and Po concentration (r = −0.39, p < 0.1 and r = −0.41, p < 0.1; respectively). Negative correlation was expected given these genes were upregulated in P-limited LR and OL/ML, respectively, to help the plant cope with low Pi availability. Together, WGCNA identified coordinated patterns of gene expression across organs and Pi treatments. Notably, the leaf- and root-specific modules (red and greenyellow, respectively) showed consistent negative correlations with Pi status and supply, highlighting their relevance to the phosphate starvation response (PSR) in hemp. Domestication-related genes are enriched in Pi-responsive modules To further characterise module function, genes previously identified as being under positive selection in domesticated compared to basal Cannabis [34] were cross-referenced with genes assigned to the 12 modules (Supplementary Table S4). The leaf-specific red module and root-specific greenyellow module were significantly enriched for domestication-related genes (Fisher’s exact test; red: p < 0.0001; greenyellow: p = 0.04), indicating that these modules capture genes associated with both the phosphate starvation response (PSR) and human selection (Fig. 5). Inspection of genes within these modules identified SPX3, AGAMOUS-LIKE MADS-BOX PROTEIN66, and PURPLE ACID PHOSPHATASE2 in the red module, and AMINO ACID PERMEASE1, NODULATION SIGNALLING PROTEIN2, and PEROXIDASE72 in the greenyellow module. Notably, a high proportion (> 70%) of domestication-related genes in these modules were also differentially expressed, in contrast to other modules where non-differentially expressed genes predominated (Fig. 5). Distribution of domestication-related genes across WGCNA modules. Domestication-related genes identified by Ren et al. (2021) were mapped to co-expression modules. P-values indicate enrichment of these genes within each module (Fisher’s exact test), with statistically significant values (p < 0.05) shown in bold; n denotes the number of domestication-related genes per module. Bar plots show the proportion of differentially expressed genes (DEGs) among domestication-related genes within each module. The red and greenyellow modules exhibit the highest proportion of DEGs. Representative phosphate starvation response (PSR) genes within modules are indicated on the right. Full gene distributions are provided in Supplementary Table S4. Gene abbreviations: AGAMOUS-LIKE MADS-BOX PROTEIN66, AGL66; AMINO ACID PERMEASE1, AAP1; NITRATE REDUCTASE, NIA; PROTEIN NRT1/PTR FAMILY4.5, NPF4.5; NODULATION SIGNALLING PROTEIN2, NSP2; SPX-DOMAIN CONTAINING PROTEIN3, SPX3; PURPLE ACID PHOSPHOTASE2, PAP2; PEROXIDASE72, POX72 These findings suggested that genes within the root- and leaf-specific modules, and their coordinated regulation, were important for the domestication of Cannabis. As these modules exhibited the strongest transcriptional responses to Pi limitation, defining the phosphate starvation response (PSR) in roots and leaves, respectively, we focused subsequent gene regulatory network (GRN) analyses on these modules. Regulation of PSR in hemp protects sinks from nutrient imbalances Due to the large number of genes in the leaf-specific module (1473 genes), a ‘bait gene’ approach was applied using SPX3 (LOC115797978), encoding a central negative regulator of PSR and also a domestication-related gene [34, 49]. SPX3 was highly responsive to P limitation across hemp organs with up to 30-fold induction (Fig. 6, Supplemental Tables S5,S6). Filtering for functionally annotated first neighbours of SPX3 yielded a highly connected network comprising 86 genes with 2844 edges. Betweenness centrality (BC), defined as the frequency with which a node lies on the shortest paths between other nodes in the network, highlighted genes acting as key intermediates in coordinating network activity (Fig. 6A). The identified genes are associated with PSR-related processes, including Pi scavenging, transport and signalling, lipid remodelling, glycolytic Pi bypass, and P/N signalling, as well as broader processes such as phytohormone signalling and organ growth regulation (Fig. 6; Supplementary Table S6). Gene regulatory network centred on SPX3 in the leaf-specific module. A Co-expression network of SPX3 first neighbours in the leaf-specific (red) module, filtered for functionally annotated and differentially expressed genes (DEGs). Edges represent co-expression relationships, with weights > 0.4 shown in bold (all edges ≥ 0.2). Node size reflects betweenness centrality (BC), a measure of how central a gene is within the network, reflecting its potential role in linking multiple processes. Node colours denote functional categories as indicated (see Supplementary Table S6). B Expression profiles of network genes across organs (LR, lateral root; OL, old leaf; ML, mature leaf; YS, young stem; TF, top flower) under varying phosphate (Pi) supply. Genes are grouped by functional category as in (A). Values represent TPM-normalised expression after z-scoring Among the top hub genes identified was another SPX homolog (SPX1, LOC115725567), confirming the SPX protein family as central regulators of the PSR in hemp leaves. SPX proteins regulate the PSR through sequestration of the master regulator PHOSPHATE RESPONSE1 (PHR1), thereby restricting binding of this transcription factor to PSR gene promoters containing the P1BS motif [14]. Their conserved up-regulation across species is thought to enable rapid suppression of the PSR upon Pi resupply [50]. Both SPX isoforms were identified in a comparison of two contrasting drug-type chemovars [11]. Among the first neighbours of SPX3 in the network are a PHR1-like family member (PHL11, LOC115702127) and multiple targets of PHR1 identified in Arabidopsis, including genes encoding purple acid phosphatases (PAPs), ribonucleases (RNases) and phosphate transporters (PHTs) [3]. Members of the INOSITOL 1,3,4-TRISPHOSPHATE 5/6-KINASE (ITPK) family were also present, which regulate inositol polyphosphate levels controlling the SPX–PHR1 interaction [51], together suggesting that the SPX-PHR1 mediated pathway is preserved in hemp. In addition, PHOSPHATE TRANSPORTER TRAFFIC FACILITATOR1 (PHF1), a regulator of phosphate transporter localisation, is also among the first neighbours of SPX3 [52]. Genes involved in the release of Pi via phospholipid hydrolysis (PLDζ1) and their replacement by galactolipids (MGDs, DGDs) and sulfolipids (UGP3, SQD1/2) were also part of the network [53, 54]. One of the top hub genes was PHOSPHATIDATE PHOSPHATASE2 (PAH2) which is critical for the dephosphorylation of phosphatidic acid to diacylglycerol [55]. A homolog of the transcription factor NITRATE-INDUCIBLE, GARP-TYPE TRANSCRIPTIONAL REPRESSOR 1 (NIGT1) was part of the SPX3 network (Fig. 6), suggesting a co-ordination of P and N homeostasis. In Arabidopsis, NIGT1 suppresses nitrate acquisition and SPX1/2-mediated sequestration of PHR1 under P-limitation [56]. Two NRT1 family members (NPF6.4, NPF5.2) were identified as first neighbours in the network (Fig. 6), suggesting remobilisation of nitrate from hemp leaves (Fig. 4). In addition, CALCINEURIN B-LIKE PROTEIN1 (CBL1) and two CBL-INTERACTING PROTEIN KINASES (CIPK12 and CIPK32) were part of the SPX3 network, indicating potential formation of CBL–CIPK complexes that regulate nitrate transporter activity [57]. Pi limitation is known to alter phytohormone production, sensitivity, and transport, including abscisic acid (ABA), ethylene and gibberellin (GA) [58], which were represented in the SPX3 network by several associated genes (Fig. 6A). Mitogen-activated protein kinase (MAPK) cascades, which regulate downstream signalling, were also present, including two MAPKKK18 homologs and a downstream MAPK (MAPK2) [59]. ABA-related components included PP2C-27, encoding a PP2C-type protein phosphatase, and its interaction partner SERINE/THREONINE-PROTEIN KINASE 2 A [60]. Ethylene signalling was represented by the biosynthetic enzyme METHYLTHIOADENOSINE NUCLEOSIDASE2 (MTN2) and the receptor REVERSION-TO-ETHYLENE SENSITIVITY1 HOMOLOG (RTH). Gibberellin signalling was represented by the nuclear receptor GIBBERELLIN-INSENSITIVE DWARF1B (GID1B) [61–64]. Pi-stress alters developmental progression, which was reflected in the SPX3 network of the leaf-specific module by genes associated with flower development. These included transcription factors such as HEADING DATE REPRESSOR 1 (HDR1) and CRC DOMAIN-CONTAINING PROTEIN (TSO1) (Fig. 6B). HDR1 acts as a suppressor of flowering in rice and its increased expression in hemp suggests a role in reducing flower initiation under P limitation [65]. In Arabidopsis, TSO1 balances cell proliferation [66] – its transcriptional up-regulation in hemp may therefore limit inflorescence growth when P becomes limiting. Regulation of growth was further supported by the presence of the E3 ubiquitin ligase BIG BROTHER-RELATED (BB), a top hub gene that negatively regulates organ size and promotes senescence in Arabidopsis [67]. The metabolic plasticity of glycolysis, via the glycolytic bypass, is a key acclimation strategy to maintain energy conversion in photosynthetic tissues when P is limiting [3]. The WGCNA identified several enzymes of these pathways among the hub genes of the leaf-specific module network: four SOLUBLE INORGANIC PYROPHOSPHATASES1 (PPA1), two PHOSPHOENOLPYRUVATE CARBOXYLASE KINASE1 (PPCK1) genes, as well as a ATP-DEPENDENT 6-PHOSPHOFRUCTOKINASE5 (PFK5), PYRUVATE KINASE (PK) and GLUCOSE-6-PHOSPHATE 1-EPIMERASE (G6PE) gene (Fig. 6). PPCK1, a key component of the Pi-independent glycolytic bypass, supports carbon flux under Pi limitation by enhancing phosphoenolpyruvate carboxylase (PEPC) activity [68]. Co-induction of PFK5 and PK suggests that, alongside activation of Pi-saving bypasses, certain ATP-dependent glycolytic steps remain active, potentially to balance metabolic demand using the remaining Pi [68]. Similarly, G6PE, an enzyme involved in carbohydrate interconversion in yeast, may contribute to glycolytic flexibility by enabling alternative hexose utilisation pathways to sustain metabolism in sink organs [57]. In summary, analysis of the leaf-specific module revealed a conserved regulatory network centred around the domestication-related gene SPX3. Up-regulation of these hub genes under Pi limitation highlights their role in integrating metabolic processes with P-dependent growth and development, likely through regulation of source–sink relationships. PSR in roots involves the regulation of the strigolactone pathway For the greenyellow module, comprising 280 root-specific and P limitation-induced genes (Fig. 2C), a GRN was constructed from 49 differentially expressed and functionally annotated genes. The resulting network exhibited high connectivity, with 737 edges, more than half of which (378) had weights > 0.4, highlighting the strong co-expression structure underlying this module (Fig. 7A, Supplementary Table S7). Gene network of the root-specific module. A Co-expression network of genes in the root-specific (greenyellow) module, filtered for functionally annotated and differentially expressed genes (DEGs). No single dominant hub gene was identified; instead, genes are primarily associated with root architecture modulation and strigolactone biosynthesis. Edges represent co-expression relationships, with weights > 0.4 shown in bold (all edges ≥ 0.2). Node size reflects betweenness centrality (BC), a measure of how central a gene is within the network, reflecting its potential role in linking multiple processes. Node colours denote functional categories as indicated (see Supplementary Table S7). B Expression profiles of network genes across organs (LR, lateral root; OL, old leaf; ML, mature leaf; YS, young stem; TF, top flower) under varying phosphate (Pi) supply. Genes are grouped by functional category as in (A). Values represent TPM-normalised expression after z-scoring Although relatively few genes in the network were directly associated with Pi homeostasis, these formed key drivers of Pi uptake and root-to-shoot translocation. This group included three homologs of the high-affinity PHT1;4 transporter family (LOC115721968, LOC115722577, LOC115721895), together with VACUOLAR PHOSPHATE TRANSPORTER 1 (VPT1/PHT5; LOC115708335) and PHOSPHATE1 (PHO1; LOC115699756), the latter mediating Pi loading into xylem vessels [69–72] (Fig. 7A,B). Upregulation of LOC115710836, encoding a member of the BRUTUS (BTS) family of E3 ubiquitin ligases that target iron-responsive transcription factors as well as PHR1, points to reduced Fe uptake to maintain balance with Pi availability [73]. Coordination of ion and metal transport is further supported by the induction of the iron-deficiency responsive IRON-REGULATED PROTEIN2 (IREG2; LOC115697119) and HIGH AFFINITY K + TRANSPORTER5 (HAK5), both present within the network (Fig. 7) [74]. In addition, two genes encoding putative nitrate transporters, NPF4.5 and NPF8.1 (LOC115699885, LOC115701357), were identified, consistent with the observed increase in nitrate accumulation in P-limited roots (Fig. 4A). This contrasted with model species such as Arabidopsis and rice, where root nitrate uptake is typically suppressed under Pi limitation via the PHR1–NIGT1 regulatory module [75]. Under Pi stress, and in parallel with the upregulation of high-affinity PHT1 transporters to enhance Pi acquisition, primary root growth is inhibited while lateral root and root hair proliferation increase root surface area [31, 76]. This shift is associated with reduced cell division and elongation, alongside structural modifications affecting cell wall composition and thickness [77, 78]. Reflecting these responses, the top hub gene CASPARIAN STRIP MEMBRANE PROTEIN1 (CASP1), together with many of its first neighbours, is linked to root growth and cell wall organisation (Fig. 7A). These include PROTEIN TRICHOME BIREFRINGENCE-LIKE 1 (TBR1), CASPARIAN STRIP MEMBRANE PROTEIN5-LIKE (CASP5), COBRA-LIKE PROTEIN4 (COBL4), and NON-CLASSICAL ARABINOGALACTAN PROTEIN 30-LIKE (AGP30). Notably, three GRAS-type transcription factors of the SCARECROW-LIKE (SCL) family, as well as a homolog of the C2H2 transcription factor SENSITIVE TO PROTON RHIZOTOXICITY (STOP), were identified within the network. SCL genes are key regulators of root development, but also function in the shoot, which may explain the distinct expression profiles observed for the two SCL isoforms (Fig. 7B) [79]. STOP1/2 inhibit primary root growth under Pi stress by restricting root cell elongation through peroxidase-dependent cell wall stiffening [80]. In Arabidopsis, Pi deficiency additionally promotes ammonium uptake and subsequent accumulation of STOP1, triggering malate excretion to mobilise Pi from the soil [81]. In line with this regulatory mechanism, multiple peroxidases were present in the module, including PEROXIDASE 12, 27, 57, 60 and 72 (POX12, POX27, POX57, POX60, POX72), supporting conservation of the STOP1/2-mediated response. Together, the upregulation of STOP within this network suggests a role in coordinating P/N balance with root growth dynamics. The most prominent feature of this root-specific and Pi-responsive network was the presence of multiple genes associated with the strigolactone pathway (Figs. 7A, 8). Strigolactones are key integrators of nutrient homeostasis and plant development, with levels typically increasing under N and/or phosphate deficiency across diverse species, including rice, tomato, pea, Medicago and Arabidopsis [82]. The network included genes encoding enzymes involved in strigolactone biosynthesis from carotenoid precursors, including PHYTOENE SYNTHASE (PSY), as well as the subsequent conversion of β-carotene into active strigolactone via CAROTENOID CLEAVAGE DIOXYGENASE 7 and 8 (CCD7/8) and CARLACTONOIC ACID METHYLTRANSFERASE (CLAMT). In addition, genes involved in strigolactone turnover, such as CARBOXYLESTERASE 15 (CXE15), were also present. These genes showed strong induction under Pi limitation, with expression changes ranging from 2- to 60-fold (Figs. 7, 8) [83]. The network contained three homologs of the transcription factors NODULATION SIGNALING PATHWAY1 (NSP1) and NSP2 which are known to regulate strigolactone biosynthesis [84]. In rice, NSP1/2 act downstream of PHR2 to enhance strigolactone production by inducing expression of DWARF27 (D27) under P limitation [45]. P-responsive expression of two hemp NSP1/2 homologs (LOC115702418, LOC115712444) was limited to the root, although LOC115702418 showed some response in the stem (YS, Fig. 8). Expression of strigolactone biosynthesis pathway genes in the root-specific module. Schematic representation of the strigolactone biosynthesis and degradation pathway. Transcriptional regulators NSP1/2 regulate D27 expression to promote strigolactone synthesis under phosphate (Pi) limitation. Heatmaps show expression of pathway genes assigned to the root-specific (greenyellow) module across organs (LR, lateral root; OL, old leaf; ML, mature leaf; YS, young stem; TF, top flower) under varying Pi supply. Values represent TPM-normalised expression after z-scoring. Gene names are colour-coded according to module membership (blue, greenyellow, red, turquoise). Gene abbreviations: PHYTOENE SYNTHASE 2, PSY2; DWARF27, D27; NODULATION SIGNALING PATHWAY1 and 2, NSP1/2; CAROTENOID CLEAVAGE DIOXYGENASE7 and 8, CCD7/8; CYP, CYTOCHROME P450 subfamilies, CYP711A and CYP722C; CARLACTONOIC ACID METHYLTRANSFERASE, CLAMT; LATERAL BRANCHING OXIDOREDUCTASE, LBO; CARBOXYLESTERASE15, CXE15 Together, these results indicate that in hemp, strigolactones coordinate root responses to Pi limitation and likely contribute to the regulation of shoot branching, as described for Arabidopsis and rice [85, 86]. The predominantly root-specific expression of strigolactone-associated regulators further supports local control of strigolactone biosynthesis under Pi stress. Discussion Pi availability is a key determinant of plant productivity, driving coordinated metabolic and cellular responses to limitation [87]. Integrating phenotypic, metabolic, and transcriptomic analyses enables identification of local and whole-plant mechanisms that improve Pi acquisition and utilisation [88–91]. Our systems-level analysis identified two distinct organ-specific modules linked to Pi status, revealing hub genes and regulatory pathways central to the PSR in hemp. While the PSR is centred on the conserved master regulator PHR1 and its interaction with SPX proteins across plant species, the broader transcriptional networks underlying these responses are context-dependent and shaped by species-specific traits and resource allocation strategies [49, 92–96]. This study represents the first gene regulatory network analysis of Pi starvation in Cannabis, providing a systems-level perspective for understanding transcriptional coordination of Pi adaptation in this crop. Coordinated PSR prioritizes sink stability in hemp Coordinated regulation of nutrient homeostasis under Pi limitation is a hallmark of the PSR [14, 15]. In source leaf tissues, a co-expression network of genes upregulated under Pi limitation highlighted SPX3 as a central hub (Fig. 6). Together with NIGT1.2 and the PHR1 homolog PHL11 (LOC115702127), both closely associated with SPX3, this forms a canonical regulatory node consistent with that observed across plant species [49, 50, 95, 97]. In line with the negative relationship between gene expression in this network and plant P status (Fig. 4), expression patterns in source leaves suggest that hemp prioritises Pi remobilisation towards floral organs by maintaining a strong PSR (Fig. 6B). This resembles strategies observed in crops with high sink strength such as maize and rice [98–100], and may also reflect domestication-driven selection, as SPX3 was identified as a locus under positive selection in Cannabis [34]. Maintaining carbon flux into sink floral organs, supported by the co-expression network and metabolite profiles (Figs. 4A, 6), has been linked to high P-use efficiency in other crops. In hemp, this suggests prioritisation of reproductive resilience over conservation of Pi in source tissues, further supported by the restriction of the negative organ size regulator BB to source leaves (Fig. 6B). These altered developmental cues under nutrient limitation in hemp also likely extend to floral morphogenesis [69, 101]. In some species, Pi limitation delays flowering to allow additional Pi accumulation prior to the transition to reproductive growth [102]. PHO1, a key regulator of root-to-shoot Pi transport, also influences flowering time in response to Pi nutrition in Arabidopsis and rice [69], where its mutation delays flowering via an AGAMOUS-like transcription factor and the floral repressor SHORT VEGETATIVE PHASE (SVP). In hemp, however, no delay in flower initiation was observed. While PHO1 was identified in root-associated gene sets, constitutive expression was maintained in sink tissues, particularly in TF (Fig. 7). Although the mechanisms linking Pi status to floral development remain unclear, the upregulation of floral meristem regulators TSO1 and HDR1 in leaves suggests that flowering cues in hemp are directly responsive to Pi availability. Root specific PSR and the role for the strigolactone pathways in hemp Local Pi perception in roots drives transcriptional reprogramming that enhances Pi acquisition through coordinated changes in root architecture, transporter activity, and mobilisation of external organophosphates [30, 103]. These responses include primary root inhibition alongside increased lateral root and root hair proliferation, as well as the formation of specialised root structures [44, 76, 104, 105]. In hemp, a root-associated co-expression network identified genes highly responsive to Pi supply, with central hubs linked to phytohormone signalling, root architecture modification, and nutrient transport (Fig. 7). These expression patterns suggest a robust Pi foraging strategy that enhance resilience to Pi-deficient soils [13]. Parallels of this development can be drawn with the evolutionary strategies of some Australian Proteaceae and white lupin which are adapted to extremely Pi-deficient soils [105–107]. Comparable strategies have evolved in species adapted to low-P environments, such as Australian Proteaceae and white lupin, which develop specialised root systems and attenuated feedback regulation of Pi uptake to maximise acquisition [108]. A key component of this adaptive root response in hemp involves strigolactone biosynthesis, with network analysis indicating strong Pi-dependent regulation of core strigolactone pathway genes. Strigolactones are established regulators of root development under nutrient stress and of shoot branching [109–111], and their induction in hemp suggests a central role in coordinating growth with nutrient availability, as observed in other species such as tomato [112, 113]. Notably, strigolactone-associated genes, rather than auxin-related genes, dominated transcriptional responses in this root-associated co-expression network. While auxin is a major driver of root responses to Pi limitation in model species [114–119], strigolactone-mediated modulation of auxin transport and signalling often occurs independently of local transcriptional changes [120]. The prominence of strigolactone-related genes may also reflect domestication, as the strigolactone biosynthesis regulator NSP2 was identified as a target of positive selection in Cannabis [34]. In addition to root responses, Pi-responsive expression of strigolactone pathway genes in shoots suggests a role in regulating shoot architecture (Fig. 8), consistent with observations in rice but contrasting with legumes where NSP1/2 primarily regulate nodulation [45, 121–123]. Similar to the functional role of NSP1/2, DWARF14 (D14) is a strigolactone receptor regulating shoot architecture. In rice under Pi deficiency, strigolactones promote the degradation of SPX4 via formation of a protein complex containing D14, and d14 mutants show increased tiller numbers as compared to wild types [124, 125]. Consistent with this integration, D14 was also identified as a target of human selection in hemp [34], and reduced shoot branching under low Pi conditions was observed in our previous work [12]. Together, these findings support a conserved PHR–SPX nutrient sensing system in hemp that interfaces with strigolactone signalling to coordinate root and shoot development in response to nutrient status. Unlike hemp, which was historically grown and selected under field conditions characterized by lower nutrient availability, drug-type cannabis has been intensively bred for high-input, and more controlled environments [13]. Modern crop varieties selected under high fertiliser input often lose traits advantageous for survival in nutrient-poor soils [126, 127]. In many crops, this ‘genetic erosion’ is supported by the superior Pi uptake of traditional cultivars or their wild ancestors compared to modern high-yielding cultivars under Pi limiting conditions [128, 129]. A similar divergence may exist between hemp and drug-types, with hemp potentially retaining ancestral adaptations to low-Pi environments. Further comparative analyses of hemp- and drug-type Cannabis varieties will be particularly interesting to identify selective traits [11]. Conclusions The integration of transcriptomic and metabolic data in this study provides a deeper understanding of the organ-specific regulatory networks governing Pi homeostasis in hemp. These findings identify candidate targets for improving P use efficiency, including SPX–PHR regulatory hubs in leaves and regulators of root architecture and strigolactone biosynthesis linked to Pi acquisition. From an applied perspective, these insights highlight targets for selecting or developing nutrient-efficient genotypes in Cannabis and other crops [130]. Supplementary Information Supplementary Material 1: Supplementary Figure S1 Determining soft-threshold for WGCNA. (A) Scale-free topology plot and (B) Mean connectivity plots were generated to determine the optimal soft threshold (SFT) for network construction. The plots are used to determine the power value that ensures a scale-free topology for co-expression networks, balancing network robustness and sparsity for downstream analysis. A SFT of 15 was chosen for downstream network creation. Supplementary Material 2: Supplementary Figure S2. Correlations of Module Eigengenes and quantified metabolites. Correlation of Module Eigengenes (ME) derived from WGCNA and normalised concentrations of quantified metabolites. Given are the Pearson correlation coefficients (r) with the correspondingp-values in brackets. The former is also visualized via colour scale for ease of interpretation. Trait data is derived from [12]. Supplementary Material 3: Supplementary Table S1: Number of up- and downregulated DEGs per organ type and treatment. In addition, number of DEGs specific to one organ per treatment are given. DEGs relative to the control treatment of 1 mM Pi have been filtered out for lowly expressed genes (TPM <10 across all organ-types per treatment) and adj. p-value<0.05, |log2FC| > 1). Supplementary Table S2: Summary of WGCNA analysis for module detection. The cs10 genome LOC gene ID is provided, alongside annotations, WGCNA module allocation, associated kME (module membership) values and an indication if the gene was differentially expressed or not (DE / notDE). Supplementary Table S3: Results of GO term enrichment analysis. For each gene their associated module membership, the enriched GO term(s), GO term description, gene ratio (module genes associated with GO term / total number of genes in module), background ratio (GO term associated genes / total number of cs10 genes with GO annotation) and statistical metrics are indicated. ID = GO term identifier, description = GO term description, BgRatio = background ratio, p.adjust = adjusted p-value, FDR = false discovery rate, count = number of genes within the module associated with a given GO term. Supplementary Table S4: Overlap of module genes with domestication-related genes. Genes were taken from Ren et al. (2021)'s large scale whole-genome resequencing of 110 Cannabis accessions from worldwide origins. Given are the cs10 gene LOC ID, protein ID, annotation, module membership / colour, their selection signatures. Table refers to the Supplementary Table number in Ren et al. (2021). The last column indicates if the gene was differentially expressed (DE) or not (nonDE) in comparison to the 1 mM Pi control treatment or whether transcript abundance for a given candidate was below the expression cutoff of TPM < 10. Candidates highlighted in Fig. 5 are highlighted in bold. Supplementary Table S5: Summary of differentially expressed genes. Given are the log2-fold changes for statistically significant genes (FDR < 0.05) across organ types in comparison to the 1 mM Pi control treatment. LOC IDs are derived from the cs10 genome annotation and represent NCBI RefSeq gene IDs, alongside gene annotation, and expression changes across organ types and treatments. LR, lateral root; OL, old leaf; ML, mature leaf; YS, young stem; TF, top flower. Abbreviated organ types are followed by numbers indicating Pi treatments (_0, _0.25, _2 = 0, 0.25, and 2.0 mM Pi). Supplementary Table S6: Summary of red module genes captured in the network. Shown are LOC IDs are derived from the cs10 genome annotation and represent NCBI RefSeq gene IDs, alongside annotation, betweenness connectivity (BC) scores, assigned functional category, kME (module membership) value, as well as individual Pearson correlation coefficients (r) with normalised organic P and Pi concentrations. Supplementary Table S7: Summary of greenyellow module genes captured in the network. Shown are LOC IDs are derived from the cs10 genome annotation and represent NCBI RefSeq gene IDs, alongside annotation, betweenness connectivity (BC) scores, assigned functional category, kME (module membership) value, as well as individual Pearson correlation coefficients (r) with normalised organic P and Pi concentrations. Acknowledgements The authors would like to thank Stefanie Kabitz, Dr. Filippa Brugliera (both formerly Cann Group Pty. Limited) and colleagues within the ARC Industrial Transformation Research Hub for Medicinal Agriculture for useful discussions and feedback. Special thanks to Dr. Bhavna Hurgobin for initial guidance using OrthoFinder and Dr Matthew Welling for providing the Han NW seeds. Authors’ contributions BWY.: Conceptualization; Methodology; Investigation; Data curation; Formal analysis; Visualization; Writing – original draft; Writing – review & editing. RJ: Conceptualization; Supervision, Data curation; Formal analysis; Writing– original draft; Writing – review & editing. JW: Conceptualization; Supervision; Writing – review & editing; Funding acquisition. OB: Supervision; Conceptualization; Data curation; Formal analysis; Writing –original draft; Writing – review & editing. Funding This work was funded by the Australian Research Council Research Hub for Medicinal Agriculture (grant ID IH180100006) and the Australian Research Council Research Hub for Protected Cropping (grant ID IH240100024). Data availability RNA-seq read data are deposited at the NCBI SRA database under project ID PRJNA1278883. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1: Supplementary Figure S1 Determining soft-threshold for WGCNA. (A) Scale-free topology plot and (B) Mean connectivity plots were generated to determine the optimal soft threshold (SFT) for network construction. The plots are used to determine the power value that ensures a scale-free topology for co-expression networks, balancing network robustness and sparsity for downstream analysis. A SFT of 15 was chosen for downstream network creation. Supplementary Material 2: Supplementary Figure S2. Correlations of Module Eigengenes and quantified metabolites. Correlation of Module Eigengenes (ME) derived from WGCNA and normalised concentrations of quantified metabolites. Given are the Pearson correlation coefficients (r) with the correspondingp-values in brackets. The former is also visualized via colour scale for ease of interpretation. Trait data is derived from [12]. Supplementary Material 3: Supplementary Table S1: Number of up- and downregulated DEGs per organ type and treatment. In addition, number of DEGs specific to one organ per treatment are given. DEGs relative to the control treatment of 1 mM Pi have been filtered out for lowly expressed genes (TPM <10 across all organ-types per treatment) and adj. p-value<0.05, |log2FC| > 1). Supplementary Table S2: Summary of WGCNA analysis for module detection. The cs10 genome LOC gene ID is provided, alongside annotations, WGCNA module allocation, associated kME (module membership) values and an indication if the gene was differentially expressed or not (DE / notDE). Supplementary Table S3: Results of GO term enrichment analysis. For each gene their associated module membership, the enriched GO term(s), GO term description, gene ratio (module genes associated with GO term / total number of genes in module), background ratio (GO term associated genes / total number of cs10 genes with GO annotation) and statistical metrics are indicated. ID = GO term identifier, description = GO term description, BgRatio = background ratio, p.adjust = adjusted p-value, FDR = false discovery rate, count = number of genes within the module associated with a given GO term. Supplementary Table S4: Overlap of module genes with domestication-related genes. Genes were taken from Ren et al. (2021)'s large scale whole-genome resequencing of 110 Cannabis accessions from worldwide origins. Given are the cs10 gene LOC ID, protein ID, annotation, module membership / colour, their selection signatures. Table refers to the Supplementary Table number in Ren et al. (2021). The last column indicates if the gene was differentially expressed (DE) or not (nonDE) in comparison to the 1 mM Pi control treatment or whether transcript abundance for a given candidate was below the expression cutoff of TPM < 10. Candidates highlighted in Fig. 5 are highlighted in bold. Supplementary Table S5: Summary of differentially expressed genes. Given are the log2-fold changes for statistically significant genes (FDR < 0.05) across organ types in comparison to the 1 mM Pi control treatment. LOC IDs are derived from the cs10 genome annotation and represent NCBI RefSeq gene IDs, alongside gene annotation, and expression changes across organ types and treatments. LR, lateral root; OL, old leaf; ML, mature leaf; YS, young stem; TF, top flower. Abbreviated organ types are followed by numbers indicating Pi treatments (_0, _0.25, _2 = 0, 0.25, and 2.0 mM Pi). Supplementary Table S6: Summary of red module genes captured in the network. Shown are LOC IDs are derived from the cs10 genome annotation and represent NCBI RefSeq gene IDs, alongside annotation, betweenness connectivity (BC) scores, assigned functional category, kME (module membership) value, as well as individual Pearson correlation coefficients (r) with normalised organic P and Pi concentrations. Supplementary Table S7: Summary of greenyellow module genes captured in the network. Shown are LOC IDs are derived from the cs10 genome annotation and represent NCBI RefSeq gene IDs, alongside annotation, betweenness connectivity (BC) scores, assigned functional category, kME (module membership) value, as well as individual Pearson correlation coefficients (r) with normalised organic P and Pi concentrations. Data Availability Statement RNA-seq read data are deposited at the NCBI SRA database under project ID PRJNA1278883.