In recent years, the push from society to employ medicinal cannabis in the treatment of intractable and debilitating illnesses has been met by some governments legalizing research into medicinal cannabis [3,4,5]. Research in this area has escalated but the understanding of the complex composition of cannabis extract is still rudimentary [17]. In one such program, we seek to profile and measure terpenes and terpenoids in medicinal cannabis inflorescence through GCMS to inform genome-wide association studies (GWAS) and other aspects of medicinal cannabis accelerated precision breeding for the development of novel designer strains. The method should be:

• Comprehensive—detecting the highest possible number of individual compounds.
• Quantitative—reliably providing absolute quantitation for as many compounds as possible.
• High-throughput—processing the maximum number of samples possible in as short a time as possible, from as little biomass as possible.

For our purposes, absolute quantitation, while desirable, is not strictly a prerequisite, as meaningful correlations to genomics data can be drawn from relative quantities, i.e., analyte-internal standard response ratios. Similarly, baseline separation, defined as R > 1.5 and generally a requirement for validated methods, is not our priority. As we are expecting close to 1000 offspring from 2–3 individual crossing events, resulting in 5000 potential samples (including replication), increasing sample throughput over existing methods was a major consideration.

We evaluated a range of sample extraction and introduction strategies, two headspace techniques (static headspace and solid-phase microextraction (SPME)) and solvent extraction. SPME was evaluated for its ease of use and comparably high sensitivity [28]. It has been used extensively for volatile profiling in cannabis and cannabis products [29,30], and has been reviewed [23]; however, to our knowledge, none of these methods were validated for quantitation. Most of our preliminary identification work was carried out using SPME, and the method reliably provided very strong signals for all monoterpenes and earlier eluting sesquiterpenes. Higher boiling sesquiterpenes were present, but their signals were comparatively weaker, most likely due to insufficient volatility at the extraction temperature or insufficient fiber incubation. Higher extraction temperatures (> 100 °C) may aid in volatilizing these higher boiling compounds, but will most likely result in a loss of more volatile monoterpenes from the fiber and longer exposure times (60–120 min, e.g., [31]) would reduce sample throughput.

Higher extraction temperatures (145 °C) were employed for static headspace (SHS) analysis. This methodology is generally regarded as providing superior quantitation compared to SPME and has been described previously for the quantification of terpenes in cannabis [32,33]. We found that, for our purposes, it lacked sensitivity for all sesquiterpenes, despite the higher extraction temperature.

The third sample extraction and introduction trialed was liquid extraction with an organic solvent. Solvents tested were hexane, ethyl acetate, and isopropanol. While all performed equally well in terms of their extraction characteristics, isopropanol and, to a lesser extent, ethyl acetate caused peak fronting, particularly for early eluting monoterpenes (Figure 3). Hexane also afforded clear, lighter colored extracts which, in our experience, translates to lower amounts of non-target compounds and nonvolatile contaminants, resulting in higher instrument uptime and reliability. Hexane was therefore chosen for subsequent experiments. Comparing chromatograms for the three extraction strategies, it became apparent that only solvent extraction was able to provide a full representation of all relevant mono- and sesquiterpenes and -terpenoids (Figure 1). In terms of throughput, headspace techniques would be advantageous, as they require minimal sample preparation (weighing); however, the hexane extract strategy still allows for well over 100 samples to be processed within a work day, equivalent to several days’ worth of instrument time. It was therefore decided to use solvent extraction with hexane for subsequent experiments.

Next, two additional columns were evaluated for their chromatographic characteristics. The initial method development described above was carried out on a DB-5 non-polar column. Other columns evaluated from our inventory were DB-17 and VF-35, both of intermediate polarity. A 5 °C/min gradient was chosen as a trade-off between shortest possible run time and best possible resolution. The runtime achieved is shorter than that reported by Giese and colleagues using a 3.5 °C gradient [24], while also resolving five additional, later eluting sesquiterpenes, including the validation compounds guaiol and bisabolol. Furthermore, by employing column backflushing after elution of bisabolol, we can prevent cannabinoids, column bleed, and other high-boiling compounds from entering the MS, reducing contamination of the source and therefore the need for more frequent source cleaning. Using a DB-17 column may also marginally reduce the runtime, as demonstrated in Figure 2 but would mean a trade off in terms of compound identification, as this column type is very rarely used to calculate retention times indices. Initial tests also showed it did not offer significant improvements in terms of resolution over the DB-5, which affords baseline or near-baseline separation for the validation compounds (Table 1). Notably, the resolution values here were calculated using extracted ion chromatograms (EICs) from a combination of standard and sample runs, which we believe is a more accurate representation of actual cannabis samples than presenting resolution data on standard runs alone [25,34]. The p-cymene, limonene, and eucalyptol eluted in close proximity in the total ion chromatogram (TIC), but are nonetheless resolved due to their unique quantifier ions. Furthermore, only limonene was detected in significant quantities in the available biomass. Based on the samples used for determining precision, the amount of eucalyptol and β-phellandrene was approximately 1.5% that of limonene each, while that of p-cymene was below the detection limit (cp Table 4).

A total of 49 distinct individual compounds were detected using this method. Of these, 23 were classified as monoterpenes and -terpenoids, the remainder as sesquiterpenes and -terpenoids. Compounds were assigned putative identities based on spectral library matching and retention time indexing. As far as possible, these predictions were confirmed using authentic standards. While monoterpene standards are readily available commercially, sesquiterpene standards have proven more difficult to obtain, an issue also encountered in other studies [6,32,35]. With the standards available to us, we were able to conclusively identify 18 out of the 23 monoterpenes, and 8 out of 26 sesquiterpenes, as summarized in Table 1; for a further 2 monoterpenes (β-phellandrene and 4thujanol) and 3 sesquiterpenes (bergamotene isomer 3, (−)-α selinene and δ-guaiene), we found matching retention time indices in the NIST webbook. The α-thujene was initially identified as α-phellandrene, based on spectral library matching, but did not match the retention time of an authentic α-phellandrene standard, and was assigned its current identity based on the RI published by Ascrizzi and colleagues [29].

The method was validated for 22 compounds, as listed in Table 5, for specificity (resolution) linearity, quantitation and detection limits, accuracy (recovery), and precision. Nerolidol was removed from the validation early due to poor sensitivity. Detection and quantitation limits for the remaining 21 compounds ranged from 0.017 (LOD) and 0.052 μg/mL (LOQ) for β-pinene to 1.144 and 3.468 μg/mL for bisabolol. The majority of compounds can be detected at less than 0.1 μg/mL in solution, equivalent to 2 ppm in biomass (w/w dry weight), and quantitated at less than 0.2 μg/mL, equivalent to 4 ppm in biomass. All LODs/LOQs were well below what is generally considered pharmacologically relevant (500 ppm) [18], and the responses were linear for the ranges tested. Using an MS/MS approach, as advocated by Shapira and colleagues [32] instead of acquiring scan data, was considered initially and may provide lower detection limits but also prevent the re-examination of data for previously “unexpected” compounds.

Accuracy was determined by spiking cannabis samples with three concentrations of the validation terpenes within the linear range and adjusting for endogenous levels. Recoveries mostly fall in the acceptable 80–120% range, with the exception of isopulegol (61.8% high; 66.2% mid), βcis caryophyllene (62.0% high; 72.9% mid), and caryophyllene oxide (69% high; 76.3% mid). Interestingly, the values for the low spike concentration for these compounds do fall within the 80–120% range. The precision of the method as determined by the %RSD for eight sample preparations, was generally between 1% and 10%, with the exception of α-terpinene (16.81% Analyst 1; 7.86% Analyst 2) and eucalyptol (20.86% Analyst 1; 13.55% Analyst 2). Four validation compounds—3-carene, p-cymene, ocimene, and nerolidol—were below detection limits entirely in the cannabis material and therefore not part of the precision determination.

Given the large concentration range of terpenes in cannabis, we consider these values acceptable for our purpose. The validation parameters obtained with this method are comparable to previously published studies that use much longer gradient times and that are therefore less suitable for high-throughput terpene profiling [25,32]. As Shapira and colleagues note [32], the actual accuracy and precision can be influenced by minor variations in sample preparation and variation in endogenous levels. Consequently, we saw the largest outlier in precision in two of the lowest level compounds.