Analysis of the 10 major cannabinoids in cannabis together with the internal standard, norgestrel, and the surrogate, PPMB, with detection at 227 nm is shown in Figure 1. Complete resolution of the 10 cannabinoids in the calibration standard was obtained except for the CBDA/CBGA pair, which was 95% resolved (Fig. 1A). This resolution was adequate for the quantitation of the two cannabinoids. Norgestrel and PPMB were well resolved from the cannabinoids. The analysis of a supercritical fluid CO₂ extract of the NIDA cannabis material is shown in Figure 1B. The sample had not been subjected to decarboxylation, and therefore showed significant amounts of CBDA and THCA. Importantly, no interferences were observed that would confound the analysis of the cannabinoids, surrogate, or internal standard.

Calibration curves for the 10 cannabinoids and PPMB showed excellent linearity, with coefficients of determination (R²) for each analyte exceeding 0.9995 (Table 1). Analytical precision was determined by analysis of 8 replicates of samples spiked with the 10 cannabinoids and PPMB, each at 1.8 μg/mL. Coefficients of variation (%CV) from these analyses ranged from 0.87% to 5.27%. LODs and LOQs for the cannabinoids (Table 1) indicated that the sensitivity of the method was adequate for the determination of cannabinoids in medical marijuana products. Norgestrel and PPMB were initially investigated for use as the internal standard and surrogate respectively based on favorable chromatographic properties and molar absorptivity at 227 nm. Norgestrel and PPMB were found to be quite stable during analytical procedures and under normal storage conditions. Storage of the stock solutions of norgestrel and PPMB for 70 h at ambient temperature or 1 month at 4°C resulted in <1% change in their concentrations. The efficacy of norgestrel as an internal standard is evidenced by the linearity of the calibration curves, which are based on analyte/internal standard area ratios, and by subsequent accuracy, precision, reproducibility, and stability studies of the method. PPMB showed comparable values for calibration curve R², LOD, and LOQ to those of the cannabinoids (Table 1). Further studies of the accuracy, precision, reproducibility, and stability of the method verified the suitability of PPMB as the surrogate.

The HPLC-PDA method was validated using sample matrices that were comparable to those that were to be encountered in the NYS Medical Marijuana Program. Initial matrix validation studies involved spiking hemp oil into sesame oil and the ethanol-propylene glycol tincture and comparing the analytical recoveries with those of the original hemp oil. The hemp oil contained 19.1% (wt/wt) CBD, but also contained measurable amounts of CBDV, CBDA, CBG, Δ⁹-THC, and CBC. Analytical recoveries were ≥98.2% for each of the six cannabinoids derived from the hemp oil spiked into sesame oil and the tincture solution (data not shown). Commercial MCT oil was also investigated as a matrix, since this material is used as an excipient for some products formulated into gel capsules and other products. Results from the analysis of low- (1.67 μg/mL), mid- (5.0 μg/mL), and high-level (45 μg/mL) spikes of cannabinoids and PPMB into MCT oil are shown in Table 2. At each of the three levels, measured cannabinoid concentrations were all within ±10% of the spike levels, indicating no significant matrix effect.

Following method validation and certification by ELAP,¹⁴ the method was used to analyze medical marijuana products from the NYS ROs. From January 2016 through April 2018, over 1300 medical marijuana product samples including tinctures, oils, capsules, and vaporizer cartridge oils were assayed. While potencies on product labels referred only to, on a mass basis, total Δ⁹-THC (Δ⁹-THC + THCA) and total CBD (CBD + CBDA), complete profiles of the 10 cannabinoids shown in Figure 1 were determined. Over the 2-year period reported here, the accuracy and precision of the method remained unchanged as determined by QC data from each analytical batch. Figure 2 shows the compilation of QC data for the analysis of Δ⁹-THC and CBD. The spike recoveries were determined to be within ±10% of the theoretical value for both THC and CBD. There was no trend in recovery during the analysis period. Similar results were obtained for the other cannabinoids (data not shown).

The high precision of the method was also maintained over the 2-year period. The %CV for the total Δ⁹-THC and total CBD normalized by the percentage of all lots tested during 2016–2017 shows remarkable intralot reproducibility (Fig. 3). The average percentage of each cannabinoid was determined from 5–14 samples per final lot and multiplied by the gross dose weight to report as weight percentages. When the cannabinoid percentage was less than the LOQ, the value of “0” was used in the total cannabinoid calculation. The %CV of each lot was calculated and binned by increasing percent variation. Most of the product lots had a %CV of <3%. The percentage of lots progressively decreased with increasing %CV, reflecting the normal statistical distribution of the combined contributions of method error and product inhomogeneity. The cumulative total Δ⁹-THC and total CBD values increase similarly, indicating no method or product reproducibility bias by the major cannabinoid. Overall, the analytical method performs well for both total Δ⁹-THC and total CBD potency determinations across many product types and concurrently demonstrates the homogeneity of the product samples for various medical marijuana products available in NYS.

A detailed breakdown of analytical precision by product type (Fig. 4) shows similar reproducibility, but with some matrix-specific exceptions. Although many product types are available in NYS, they can be separated into four major categories: tinctures, oils/solutions/drops, capsules/tablets, and vaporizer oils. Final product potency data from all ROs were sorted into these four categories. The product type was observed to influence the reproducibility in terms of %CV. The method reproducibility is predicted to be constant, while individual product homogeneity likely has greater variability across the product types.

The ethanol-based tinctures were found to have the greatest intralot reproducibility, followed closely by the oil/solution/drops products. The slight difference between the total Δ⁹-THC and total CBD in the <2 and 2–3%CV bins may be due to the varying numbers and types of tinctures. The capsule/tablet products, which contained both oil-based and powder-based formulations, had distinctly larger and broader %CV distribution than the oral liquid-based products, but certainly not what would be considered high variation. The vaporizer oils showed the highest variation. Several sources of both method error and inhomogeneity could contribute to this observation. Oil samples are removed from the vaporizer pens/cartridges using a syringe or spatula. The vaporizer oils are the most concentrated, require the smallest dose, and are highly viscous; therefore, increased weighing and dilution errors occur in their analysis using this and similar potency methods. Inhomogeneity due to vaporizer cartridge packaging is also possible. The viscous vaporizer oils are in contact with surfaces of the cartridge materials, including metals, glass, plastics, wicks, and seals. Additionally, liquid exposed to air inside the cartridge is subject to oxidation and evaporation of the more volatile fractions of natural terpenes and excipients. This combination of factors, and the large variation in product formulations and vaporizer devices among the ROs, may lead to larger variations in potency measurements than for other forms.

An additional requirement of the NYS Medical Marijuana Program is the determination of stabilities of total Δ⁹-THC and total CBD in the products. Overall, product stability was found to be excellent. The mean stabilities for each product type at 30 and 60 days after opening were ≥98.8%. Unopened product stabilities at 120, 240, and 360 days postpackaging exceeded 83% for total Δ⁹-THC and total CBD in all product types (Table 3).