GC-MS–Informed Aroma Profiling and Sensory Lexicon Development for Cannabis Inflorescence

Decades of prohibition have constrained scientific understanding of cannabis quality, leaving THC potency as a misleading proxy for consumer value and contributing to public health risks. Increasing evidence indicates that pleasant aroma, rather than cannabinoid concentration, is the primary determinant of perceived quality. However, the chemical basis of Cannabis aroma in whole inflorescence remains poorly defined.

Research conducted by the Department of Food Science and Technology of Oregon State University (Corvallis, Oregon) and the Legacy Research Institute of Dow Neurobiology (Portland, Oregon) addressed this gap by developing a standardized aroma lexicon and associated sensory methodology, supported by gas chromatography–mass spectrometry (GC-MS) analysis of terpenes and terpenoids plus gas chromatography-pulsed flame photometric detector (GC-PFPD) analysis of volatile sulfur compounds. Chemical–sensory associations were evaluated to identify contributors to key aromatic attributes, and aroma similarities between type I (high THC, low CBD; cannabis) and type III (low THC, high CBD; hemp) Cannabis were explored. Together, these results advance a potency-independent, analytically grounded framework for defining Cannabis inflorescence quality.

LCGC International spoke to Thomas H. Shellhammer, Nor’Wester professor of Fermentation Science and professor of Food Science at Oregon State University, about this work, and the paper (1) that resulted from it.

What led your team to use gas chromatography–mass spectrometry (GC–MS) for volatile profiling, and how did you determine it was the most appropriate analytical technique for capturing the aromatic diversity of intact Cannabis inflorescences?

We selected gas chromatography–mass spectrometry (GC–MS) because it is the most established and widely validated analytical approach for profiling volatile aroma compounds in complex plant matrices, including terpenes and other low-molecular-weight volatiles known to contribute to aroma. GC–MS provides both chromatographic separation and structural information, which is essential when working with a chemically diverse and incompletely characterized system like Cannabis aroma.

A key practical factor in our decision was regulatory. Because drug-type Cannabis (high-THC) remains a Schedule I substance under U.S. federal law, my laboratory cannot legally handle it for chemical analysis without a DEA Schedule I license. Rather than redesign the study around that constraint, we partnered with an external, accredited analytical laboratory that routinely conducts cannabis volatile analyses and operates fully within regulatory compliance. This allowed us to include representative commercial samples while maintaining institutional compliance.

Importantly, the laboratory employed a previously published, industry-standard GC–MS method, rather than a bespoke or experimental approach. That was a deliberate choice. Using a validated method ensured the data were reproducible, comparable with existing literature, and reflective of how Cannabis volatiles are currently characterized in both research and commercial testing contexts. Our goal was not to develop a new analytical technique, but to ask whether commonly measured chemical profiles meaningfully explain sensory diversity in intact Cannabis inflorescences.

Finally, GC–MS was appropriate for the scientific question because it captures a broad spectrum of volatile compounds present in the headspace of uncombusted flower, which aligns directly with how aroma is perceived orthonasally by consumers. While no single technique can fully capture the sensory complexity of aroma, GC–MS remains a strong tool for comprehensive volatile profiling and provided a robust chemical framework for characterizing Cannabis aroma diversity

Could you describe how you prepared the intact inflorescence samples for GC–MS analysis—particularly your approach to preserving volatile integrity and preventing compound loss or transformation during handling?

Because our objective was to characterize the aroma of intact, uncombusted Cannabis inflorescence as it would be experienced by consumers, sample handling was designed to preserve volatile integrity while minimizing physical or chemical disruption of the flower.

We followed handling and storage practices recommended by the cannabis industry and analytical laboratories that routinely work with volatile aroma compounds. After drying and trimming, intact inflorescences were placed into high-barrier flexible pouches that were flushed with inert gas (nitrogen) prior to sealing. This approach reduces oxygen exposure, thereby limiting oxidative degradation or transformation of aroma-active compounds during storage. Samples were then stored under refrigerated conditions until analysis.

We intentionally avoided frozen storage. While freezing can reduce volatilization losses, it also risks damaging glandular trichomes through mechanical fracture, which can alter the physical integrity of the inflorescence and potentially bias volatile release during analysis. Because trichomes are the primary reservoirs for many aroma-active compounds in cannabis, preserving their structural integrity was a priority.

Importantly, we did not grind, homogenize, or otherwise disrupt the samples prior to analysis. Maintaining intact inflorescences helped ensure that volatile release during GC–MS analysis reflected the native aroma profile rather than artifacts introduced by aggressive sample preparation. Overall, this handling strategy was designed to balance chemical stability with physical preservation, allowing us to measure aroma-relevant volatiles in a form that closely represents how cannabis flower is stored, handled, and smelled in real-world settings

The study quantified both terpenes and volatile sulfur compounds (VSCs). What specific GC–MS parameters or column choices allowed you to resolve low-abundance, high-impact sulfur compounds alongside abundant terpenes?

(This question answered by Dr. Michael Qian, Shellhammer's colleague in the Department of Food Science and Technology, Oregon State University)

Volatile sulfur compounds (VSC) of hemp only were analyzed in Dr. Michael Qian’s flavor chemistry lab at OSU. A sulfur specific detector (pulsed flame photometric detector, PFPD) was used for VSC detection, a polar, thick film of column (DB-FFAP column, 30 m × 0.32 mm I.D.,1 µm film thickness, Agilent Technologies) was used for separation because it can retain highly volatile sulfur compound and give a sharp peak. Otherwise, a thick non-polar column (such as DB-sulfur column) can be used. Since we are using a sulfur-specific detector for the analysis, the coeluting terpenes do not give signals and thus do not interfere with the VSC detection and qauntification. For further information, please see our previous publications (2-4).

Two approaches were used for the analysis of VSC in hemp. Dimethylsulfide (DMS) was analyzed using static headspace GC-PFPD due to its high content in the sample. For this analysis, the hemp flowers without stems were placed in a sample vial, and 5 mL 5% ethanol in deionized water was added. The added ethanol improved the extraction efficiency of analytes from the solid matrix; it also provided a consistent vapor pressure (such as buffering system) so the volatility of DMS was not dramatically impacted by the other volatile terpenes. In addition, an ethyl methyl sulfide (EMS) was added as internal standard for DMS quantitative analysis. EMS was used as internal standard because it has the same behavior as the DMS regardless the volatile composition. The vial was then sealed and incubated before the headspace was taken for injection. Standard curve of DMS was built for quantitation. Please see our additional publication on the subject (5,6).

For the analysis of other VSCs (beyond DMS), a solid phase microextraction technique was used for volatile extraction due to their extreme low abundance.

How did GC–MS outputs inform or challenge your sensory lexicon, and what compounds beyond terpenes appeared most influential?

The GC–MS data were critical not because they confirmed a simple chemistry–aroma relationship, but because they clearly demonstrated its limitations. One of the most important outcomes of the study was showing that terpene composition alone did not reliably predict how Cannabis inflorescences were perceived sensorially. While GC–MS revealed clear chemical clustering based on terpene profiles, those clusters aligned poorly with the sensory groupings generated using the aroma lexicon.

This result directly challenged a dominant assumption in both the scientific literature and the cannabis industry that terpene profiles are sufficient proxies for aroma character. In our data, only one terpene, terpinolene, showed a consistent association with specific sensory descriptors, namely citrus and chemical. Even well-known compounds such as d-limonene, which is widely assumed to drive citrus aroma, did not correlate with citrus perception in intact inflorescences. This mismatch reinforced the need for a sensory lexicon that is grounded in perception rather than inferred from chemistry alone.

Rather than undermining the lexicon, the GC–MS results validated its purpose. The sensory lexicon captured meaningful and reproducible differences among samples that were not explainable by terpene chemistry, highlighting that aroma perception emerges from complex mixtures rather than single compound drivers. In that sense, the chemistry data strengthened our confidence that the lexicon was capturing real perceptual variation rather than simply restating known chemical differences.

Beyond terpenes, volatile sulfur compounds (VSCs) emerged as particularly noteworthy, despite being present at much lower concentrations. Using sulfur-selective detection, we identified more than 40 sulfur-containing compounds in type III samples, including dimethyl sulfide, methional, and dimethyl trisulfide, as well as several tentatively identified or potentially novel compounds. Although these compounds did not strongly predict sensory clusters in a statistical sense, which was likely due to sample size and the complexity of odor interactions, since their known sensory potency suggests they may contribute disproportionately to attributes such as skunky, savory, tropical, or fermented notes.

Overall, the GC–MS data shifted the interpretation of Cannabis aroma away from a terpene-centric view and toward a more holistic understanding. They underscored that aroma is driven by interactions among multiple classes of volatile compounds, including sulfur compounds, and likely esters and aldehydes not fully captured here, and thus reinforcing the conclusion that sensory evaluation remains essential for defining Cannabis aroma quality and diversity.

How was the GC–MS dataset integrated with sensory data?

The GC–MS dataset was integrated with the sensory data using a series of exploratory and predictive multivariate statistical techniques designed to evaluate whether chemical fingerprints aligned with perceptual aroma descriptors. Our goal was not to force a model fit, but to test how well commonly used chemical data could explain sensory outcomes.

We first applied principal component analysis (PCA) to the terpene dataset and Correspondence Analysis to the sensory dataset. PCA was used to visualize patterns of variation, identify dominant contributors within each data type, and assess whether samples that clustered together chemically also clustered together sensorially. While terpene PCA revealed clear chemical structure and repeatable clusters, those groupings showed little correspondence with sensory-based clusters derived from aroma descriptor frequencies.

To more directly integrate chemistry and sensory data, we employed multifactor analysis (MFA) and partial least squares regression (PLSR). These approaches are commonly used in flavor science to relate instrumental measurements to sensory perception. MFA allowed us to examine shared variance between terpene chemistry and sensory attributes in a combined analytical space, while PLSR was used to test whether chemical variables could predict sensory descriptors.

In practice, both MFA and PLSR showed poor overall model fit. The number of explanatory variables was large relative to the number of samples, and the relationships between individual compounds and sensory attributes were weak or inconsistent. This outcome was informative rather than disappointing. It demonstrated that, for intact Cannabis inflorescence, terpene chemistry alone does not strongly predict sensory aroma profiles.

Nevertheless, these analyses consistently highlighted one notable relationship. Terpinolene showed a repeatable association with citrus and chemical aroma descriptors across multiple statistical approaches. Beyond that, most sensory variation was not well explained by the measured chemical variables.

Overall, integrating GC–MS data with sensory data using multivariate statistics reinforced a central conclusion of the study. Chemical composition and sensory perception occupy related but largely independent dimensions, and sensory evaluation remains essential for capturing aroma differences that are not evident from chemical fingerprints alone

Looking ahead, how do you see GC-based approaches evolving for Cannabis aroma research — for example, through two-dimensional GC×GC, olfactometry (GC–O), or coupling GC data with machine learning for predictive aroma modeling?

Looking ahead, GC-based approaches will remain central to advancing Cannabis aroma research, but their greatest impact will come from expanding beyond conventional one-dimensional GC–MS and integrating more perception-focused and data-driven tools.

From the outset, GC–olfactometry (GC–O) was identified as an ideal next step for this work. GC–O allows human assessors to directly evaluate odor-active compounds as they elute from the chromatograph, making it one of the most powerful tools for identifying aroma-impact compounds rather than simply abundant compounds. This is especially important for Cannabis, where highly potent odorants can exert strong sensory effects at extremely low concentrations. While GC–O was part of the original study design, it could not be implemented due to budget constraints. That limitation reinforces, rather than diminishes, its importance as a future direction.

Comprehensive two-dimensional gas chromatography (GC×GC) also holds substantial promise. Cannabis inflorescence contains a chemically dense and highly co-eluting mixture of volatiles, and GC×GC offers dramatically improved separation capacity compared to conventional GC. This technique would be particularly valuable for resolving trace sulfur compounds, oxygenated volatiles, and other minor constituents that may be masked in one-dimensional separations but contribute disproportionately to aroma perception. As with GC–O, the primary barrier is not scientific merit but cost and instrument access.

Another important frontier is the integration of GC data with machine learning and advanced statistical modeling. Larger, more diverse datasets that combine comprehensive chemical profiling with well-curated sensory data could enable models that move beyond simple linear relationships. Machine learning approaches may be better suited to capturing the nonlinear, interactive nature of odor perception, where aroma character emerges from mixtures rather than individual compounds. However, such models will only be as good as the data that feed them. High-quality sensory data and chemically comprehensive inputs are prerequisites for meaningful prediction.

Taken together, the future of Cannabis aroma research lies in combining richer GC-based analytical tools with rigorous sensory science. Techniques such as GC–O and GC×GC can help identify which compounds truly matter perceptually, while computational approaches can help interpret their combined effects. Achieving this will require sustained investment, but the payoff is a more accurate, scientifically grounded understanding of Cannabis aroma that moves beyond terpene-centric assumptions and toward aroma quality as it is experienced.

References

1. Isaacson, S. E.; Wilson-Poe, A. R.; Ye, T. et al. Beyond Potency: A Proposed Lexicon for Sensory Differentiation of Cannabis sativa L. Aroma. PLoS One 2025, 20 (10), e0335125. DOI: 10.1371/journal.pone.0335125
2. Fang, Y.; Qian, M. C. Sensitive Quantification of Sulfur Compounds in Wine by Headspace Solid-Phase Microextraction Technique.J. Chromatogr. A 2005, 1180, 177-185. DOI: 10.1016/j.chroma.2005.05.024
3. Vazquez-Landaverde,P.: Torres, J. A.; Qian, M. C.Quantification of Trace Volatile Sulfur Compounds in Milk by Solid-Phase Micro-Extraction and Gas Chromatography-Pulsed Flame Photometric Detection. J. Dairy Sci. 2006, 89, 2919-2927. DOI: 10.3168/jds.S0022-0302(06)72564-4
4. He, J.;Zhou, Q.; Peck, J. et al. The Effect of Wine Closures on Volatile Sulfur and Other Compounds During Post-Bottle Ageing. Flavour Fragr. J. 2013, 28 (2). 118–128. DOI: 10.1002/ffj.3137
5. Davis, P. M.;Qian, M. C. Effect of Ethanol on the Adsorption of Volatile Sulfur Compounds on Solid Phase Micro-Extraction Fiber Coatings and the Implication for Analysis in Wine. Molecules 2019, 24, 3392. DOI: 10.3390/molecules24183392
6. Davis, P. M.;Qian, M. C. Effect of Wine Matrix Composition on the Quantification of Volatile Sulfur Compounds by Headspace Solid-Phase Microextraction-Gas Chromatography-Pulsed Flame Photometric Detection. Molecules 2019, 24, 3320. DOI: DOI: 10.3390/molecules24183320