Correlations among morphological and biochemical traits in high‐cannabidiol hemp ( Cannabis sativa L.)

2.2. Field preparation and maintenance

Trials were planted at two Cornell University field sites: Geneva, NY (McCarthy Farm: 42.9, −77.0) and Ithaca, NY (Bluegrass Lane Turf and Ornamental Farm: 42.5, −76.5). Weather station data can be found at http://newa.cornell.edu (Last accessed on 11/22/2022). Each site was cultivated and raised beds formed with drip irrigation and black plastic mulch prepared every 1.83 m on center. Fertilizer (19‐19‐19, Phelps Supply Inc., Phelps, NY) equivalent to 95.3 kg N ha⁻¹ was spread under the plastic mulch in Geneva and was broadcast pre‐planting in Ithaca. Landscape fabric was deployed to control weeds in alleys between rows.

Each cultivar was planted in five‐plant plots in a replicated, randomized complete block design with four replicate blocks at each site. Seedlings and rooted cuttings were transplanted into raised beds on June 11, 2020 (Geneva, NY) and June 12, 2020 (Ithaca, NY). Plants were spaced 1.23 m apart within rows. After transplanting, the plots were irrigated using in‐bed drip irrigation as needed throughout the season to maintain optimal soil moisture (>.27 m³ m⁻³). HOBOnet RXW‐SMD‐10HS soil moisture sensors (Onset, Bourne, MA) were installed at a depth of 10 cm and used to assess when irrigation was necessary. Fertilizer (Jack’s 12‐4‐16 Hydro FeED RO, 11.3 kg per application) was included in the irrigation on two occasions in early and late July.

2.3. In‐season measurements

The heights of the middle three plants of each five‐plant plot were measured weekly after transplant until there was no change in height for two consecutive weeks. All plants were surveyed weekly to determine the onset of flowering following the protocol described by Carlson et al. (2021), where “terminal flowering” describes distinct clusters of pistillate flowers observed at shoot apices (Figure S1). Plants that produced staminate flowers were promptly removed from the field to maintain unpollinated female inflorescences.

For cannabinoid quantification, inflorescence samples of the 32 cultivars were collected from a single plant in every plot starting 1 week after terminal flowering and again at 3 and 5 weeks after terminal flowering. The top 10 cm of the apical inflorescence—including bracts, leaves, stems, and floral tissue—was sampled for the time series. The first plant in the plot was sampled for the 1‐week timepoint, the fifth for the 3‐week timepoint, and the second for the 5‐week timepoint. Shoot tip samples were dried in a climate‐controlled room (~30% RH and below 33°C) and then milled to a fine powder in a Ninja® Pro blender (SharkNinja, Needham, MA). Milled samples were stored at 4°C prior to high pressure liquid chromatography (HPLC) analysis. For each sample, 50 mg of dried, milled tissue was mixed with 1.5‐mL ethanol by high‐speed shaking at room temperature with a Tissuelyser (Qiagen), and filtered through a SINGLE StEP PTFE Filter Vial (Thomson). The resultant liquid was directly subjected to HPLC analysis (Dionex UltiMate 3000; Thermo Fisher) with biphenyl‐4‐carboxylic acid (BPCA) as an internal standard, using a Phenomenex Kinetex 2.6‐μm Polar 100‐Å column 150 × 4.6 mm heated at 35°C. Samples were injected and eluted at 1.2 mL min⁻¹ over a 10‐min gradient, from 80% acetonitrile, .1% formic acid, to 90% acetonitrile, .1% formic acid, followed by a 2 min isocratic step. Absorbance was measured at 214 nm. For cannabinoids with concentrations greater than the upper limit of the standard curve, a 20× dilution of the ethanol extract was prepared and run using the same protocol. The following cannabinoids were quantified for each sample: tetrahydrocannabinolic acid (THCA), Δ⁹‐tetrahydrocannabinol (THC), cannabidiolic acid (CBDA), CBD, cannabichromenic acid (CBCA), cannabichromene (CBC), cannabigerolic acid (CBGA), cannabigerol (CBG), cannabinol (CBN), tetrahydrocannabivarin (THCV), tetrahydrocannabivarinic acid (THCVA), cannabidivarin (CBDV), cannabidivarinic acid (CBDVA), cannabicyclol (CBL), cannabicyclolic acid (CBLA), and Δ⁸‐tetrahydrocannabinol (Δ⁸‐THC) (Dataset S2). To control for variation in decarboxylation of acid‐form cannabinoids and cultivar‐dependent variation in relative proportions of different cannabinoids, the statistical analysis was performed on the sum of total potential cannabinoid percentages for all measured cannabinoids. See tab. 3 in Stack et al. (2021) for formulas used to calculate total potential percentages for individual cannabinoids (total potential THC, total potential CBD, etc.).

2.4. Time‐of‐harvest and post‐harvest measurements

At harvest, 5 weeks after terminal flowering, canopy architecture traits of the third plant in each plot were measured following the method described by Carlson et al. (2021), including height, maximum canopy diameter (MCD), and the height at the maximum canopy diameter (MCDH). Following the measurements, the stems were cut at soil level and the total wet biomass of each plant in a plot was measured. The third plant in each five‐plant plot was partitioned into five sections (S1 through S5) based on the length of the main stem (Figure 1). Sections were subsequently air‐dried in a greenhouse with industrial fans.

FIGURE 1

Schematic of (a) a hemp plant in the field, (b) the branching structure of the hemp plant, (c) the canopy‐level plant architecture measurements taken prior to harvest, and (d) the sectioning scheme used to partition the plant into five sections.

After drying, basal stem diameter, total dry biomass—including stems, leaves, and inflorescence material—and dry stripped inflorescence biomass were measured for each section. The full list of traits measured with relevant formulas can be found in Table S1. Cannabinoids were quantified by HPLC following the protocol above using a subsample of the homogenized stripped biomass for each section (Dataset S3).

3.2. Biomass distribution among sections is cultivar‐dependent

There was an effect of cultivar on total stripped inflorescence biomass production (F[31,200] = 3.33, p < .001) (Figure 3a). Further, the proportion of total stripped biomass produced in each section of the canopy varied by cultivar (Figure 3b). On one extreme, over 75% of the total stripped biomass produced by GVA‐H‐19‐1077‐008 was in S5, the section closest to the base of the plant. In contrast, less than 30% of the total biomass produced by ‘Umpqua’ was in S5. K‐means clustering analysis of the proportions of dry stripped biomass in each section separated the cultivars into four clusters (Figure 3b). The clusters corresponded almost perfectly with the variation in the proportion of stripped biomass in S5.

FIGURE 3

Bar plots of stripped biomass produced by 32 high‐cannabidiol (CBD) hemp cultivars partitioned into five sections where S1 is branches derived from the top fifth of the main stem, S2 those from the second to top, and so on. Colors indicate the mean stripped biomass per plant (a) and the proportion of stripped biomass (b) from each of the five sections. Clusters in (b) are cultivar‐level k‐means clusters based on the proportions of dry biomass produced in each section. Letters indicate statistically significant differences in stripped biomass production among cultivars based on a post hoc Tukey’s HSD test.

3.3. Variation in plant architecture is correlated with biomass distribution

There was substantial variation in plant architecture among and within the 32 cultivars evaluated with plant height and MCD ranging from 61 to 249 and 55 to 265 cm, respectively (Figure 4 and Dataset S1). The ratio of the MCD to height was positively correlated with the proportion of dry biomass in S5 at the level of cultivar (R ² = .47, F[1,30] = 26.1, p < .001) and plot (R ² = .29, F[1,228] = 91.35, p < .001) (Figure 5), and negatively correlated with the log of the ratio between stripped biomass in S4 and S5 at the level of cultivar (R ² = .44, F[1,30] = 23.37, p < .001) and plot (R ² = .20, F[1,228] = 55.28, p < .001) (Figure 5). There was a significant relationship between bicone volume and harvest index at the level of cultivar (R ² = .57, F[2,29] = 55.28, p < .001) and plot (R ² = .39, F[2,227] = 73.11, p < .001) (Figure 4). The proportion of dry biomass in S5 was negatively correlated with the log of the amount of stripped inflorescence biomass per unit area at the level of cultivar (R ² = .33, F[1,30] = 14.56, p < .001) and plot (R ² = .09, F[1,228] = 23.81, p < .001) (Figure 5).

FIGURE 4

Variation in plant architecture within and among high‐cannabinoid hemp cultivars. (a) Plot of maximum canopy diameter (MCD) versus height where points are cultivar means and error bars are the standard error of the mean. Dashed lines indicate MCD to height ratios of .5, 1, and 2. (b) Second‐degree polynomial regression of harvest index on bicone volume where points are cultivar means and error bars are the standard error of the mean. (c) Average kite models by cultivar contrasted from MCD, height, and the height of the maximum canopy diameter (MCDH). Colors in all panels indicate the k‐means cluster to which that a cultivar belongs based on the proportions of dry biomass in each biomass section.

FIGURE 5

Correlations between (a) maximum canopy diameter (MCD):Height ratio and proportion of dry biomass in S5, (b) MCD:Height ratio and log of the ratio of stripped biomass produced in S4 and S5, and (c) proportion of dry biomass in S5 and log of stripped biomass per unit area. Each point and set of error bars, depicting the standard error of the mean, represents one cultivar. Colors indicate cultivar‐level k‐means clusters based on the proportions of stripped inflorescence biomass produced in each section. Regression lines and confidence intervals are cultivar‐level generalized linear models.

3.4. Intra‐plant variation in cannabinoid concentration is cultivar‐dependent

When modeling the concentration of total potential cannabinoids sampled from the biomass sections using a mixed‐effects model, there were significant main effects of cultivar (F[31,502.12] = 16.23, p < .001) and biomass section (F[1,988.23] = 434.14, p < .001), and a significant interaction between cultivar and biomass section (F[31,988.22] = 5.24, p < .001). After correcting for multiple testing for the 32 mixed‐effects model F‐tests (α = .0016), 17 of the cultivars had statistically significant variation in total cannabinoid concentration among plant sections (Figure 6). For those cultivars with significant variation among sections, the cannabinoid concentration in S1 was always significantly greater than the concentration in S5.

FIGURE 6

Boxplots showing the distribution of total potential cannabinoid concentration measured in post‐harvest stripped biomass samples for plants partitioned into five sections where S1 represents branches derived from the top fifth of the main stem, S2 those from the second to top, and so on. (a) Boxplot of all samples from all cultivars. (b) Thirty‐two cultivar‐level boxplots ordered by the slope of the regression predicting total potential cannabinoids using biomass section. P‐values reported are for mixed models predicting total potential cannabinoid concentration where biomass section was a fixed effect and the plot that the sample was from was a random effect. Asterisks indicate a significant effect of biomass section after correcting for multiple testing and letters indicate statistically significant differences between sections of a cultivar based on a post hoc Tukey’s HSD test.

4.2. Distribution of inflorescence biomass within the plant canopy and relationships with plant architecture

As inflorescence biomass is the primary product of interest when producing high‐cannabinoid C. sativa, the position and distribution of this biomass throughout the canopy is critical for growers. Based on the data presented here, growers should prefer plants that produce a greater proportion of biomass in the upper canopy. Plants that had a greater proportion of biomass in S5 produced less inflorescence biomass per unit area (Figure 5c), and plants with a large bicone volume also had a significantly lower harvest index, potentially due to the additional stem biomass needed to support the large basal branches (Figure 4b).

Moving toward a high‐density production system, where individual plants are smaller but production per unit area is greater, has been effective in increasing yields across dozens of cropping systems (Assefa et al., 2018; Duvick, 2005; Postma et al., 2021). Through tandem improvement of cultural practices for high‐density cultivation, as well as development of cultivars selected for high‐density production, cultivators have been able to increase yields and production efficiency. While the bulk of this discussion focuses on outdoor production systems, similar principles govern increasing production density in controlled environment systems (Amundson et al., 2012; Rodriguez et al., 2007). Selection strategies toward improved high‐cannabinoid cultivars for indoor production should consider optimization of biomass distribution for high‐density cultivation systems, specifically by reducing the need for labor‐intensive manual pruning and training.

4.3. Cultivar‐dependent intra‐plant variation in cannabinoid concentration

In contrast with previous studies of intra‐plant variation in cannabinoid concentration (Bernstein et al., 2019; Crispim Massuela et al., 2022; Danziger & Bernstein, 2021a, 2021b; Richins et al., 2018), many of the cultivars evaluated in this study had little variance in cannabinoid concentration across the canopy. This incongruence could be attributed to many factors, including the different genotypes used in the studies, variation in sampling methodology, and differences in cultivation environment. Nevertheless, all cultivars with significant variation among sections followed the same general pattern described in the literature, with the greatest concentrations of cannabinoids accumulating in the upper‐most part of the canopy.

The degree of variation in total cannabinoid concentration throughout the canopy is particularly important to growers in the context of regulatory compliance. Because THCA and CBDA are synthesized in predictable ratios by the CBDA synthase enzyme in chemotype III C. sativa (Stack et al., 2021; Toth et al., 2020; Toth et al., 2021; Yang et al., 2020; Zirpel et al., 2018), remediation of non‐compliant floral biomass by blending it with plant tissue containing lower THC concentration is effective primarily due to differences in the total concentration of cannabinoids among plant sections. If there is not substantial variation in cannabinoid concentration in stripped inflorescence biomass by section, dilution with inflorescence biomass from lower sections of the canopy would not reduce the total THC concentration sufficiently to bring it below the compliance threshold.

4.4. Correlations with intra‐plant variation in cannabinoid concentration

The factors underpinning the variation, or lack of variation, in cannabinoid concentration throughout the canopy must be understood to develop efficient production systems. This dataset complements previous work investigating the impact of plant architecture manipulation on intra‐plant variation in cannabinoid accumulation (Crispim Massuela et al., 2022; Danziger & Bernstein, 2021a, 2021b). Specifically, given that manipulation of plant architecture and canopy density can cause local changes in cannabinoid concentration, does existing variation in plant architecture among cultivars correlate with intra‐plant variation in cannabinoid content? Interestingly, there was a significant interaction between canopy area and dry canopy density in predicting decrease in cannabinoid concentration. Plants with a small canopy area, independent of canopy density, showed very little reduction in cannabinoid concentration, while plants with a greater canopy area exhibited canopy density‐dependent decreases in cannabinoid concentration. This lends support to hypotheses that factors including light penetration and density of non‐inflorescence leaves influence intra‐plant variation in cannabinoid concentration. In many species, variation in light penetration has been shown to contribute to intra‐plant variation in secondary metabolites, including: flavonoids (Del Valle et al., 2018), anthocyanins (González‐Talice et al., 2013), and tannins (Mole et al., 1988). Recently, Danziger and Bernstein (2022) found that increased shading as a result of greater planting density decreased cannabinoid concentrations in axillary inflorescences from the bottom of the plant. Additional work examining light penetration into the canopy, photosynthetic activity throughout the canopy, and sub‐canopy lighting will provide a better understanding of the degree to which variability in light intensity modulates variation in cannabinoid concentration.

Stepwise regression modeling indicated that stripped inflorescence biomass produced in S4 was the best single predictor of intra‐plant variation cannabinoid content, such that plants with greater amounts of stripped inflorescence biomass in S4 had a greater decrease in cannabinoid concentration moving down the plant. The next most important predictors were the amount of biomass in S3, the proportion of S4 that comprised stripped inflorescence biomass, and the kite angle ratio. Of the nine predictors in the final model, six were biomass measurements, while the other three were plant architecture measurements. Though it was not selected as a predictor in the final model, the relationships between the selected plant architecture and biomass measurements with canopy density suggest that variation in micro‐environments throughout the plant canopy drives variation in cannabinoid concentration. Plants that produce less stripped inflorescence biomass and have a lower relative MCDH could have a more homogenous and open canopy environment (light, temperature, humidity, airflow, etc.) and thus a more homogeneous cannabinoid profile throughout the canopy.

Further studies are needed to characterize the interactions between genotype, environment, planting density, and manipulation of plant architecture, specifically contrasting genotypes with and without innate intra‐plant variation in cannabinoid concentration.

We are grateful to the research teams from the L. Smart, C. Smart, Viands, Rose, and Bergstrom labs for their excellent technical assistance on this project, especially Deanna Gentner, Stephen Snyder, Allison DeSario, Teagan Zingg, Savanna Shelnutt, Emily McFadden, Lauren Carlson, Colin Day, Garrett Giles, Dr. Heather Grab, Nicholas Kaczmar, Jesse Chavez, Ryan Crawford, Jason Schiller, Johanna Gertin, and the farm crews at Cornell AgriTech and the CUAES campus area farms. We are also grateful to Stem Holdings Agri, Sunrise Genetics, Phytonyx, Industrial Seed Innovations, Castetter Sustainability Group, Ryes Creek, Front Range Biosciences, Charlotte’s Web, Blue Forest Farms, Wessels’ Farm, and Boring Hemp for generously contributing cultivars.