Toward a Cannabis Terroir: Untargeted Metabolomic Profiling of Authentic Samples Using Gas Chromatography–High‐Resolution Mass Spectrometry (GC‐HRMS) and Liquid Chromatography–High‐Resolution Tandem Mass Spectrometry (LC‐HRMS/MS)

ABSTRACT

Cannabis sativa L. constituents, such as cannabinoids, terpenes, flavonoids, and other secondary metabolites, determine the plant's (medicinal) effects and properties in a complex interplay, a phenomenon known as the entourage effect. However, environmental influences like cultivation method, soil, light, and climate might also influence the plant's chemical composition—and thus its therapeutic profile. Much like in viticulture, the concept of a “cannabis terroir” might play an important role in determining the plant's chemical phenotype. The aim of this study was therefore to make these complex properties analytically accessible and develop a comprehensive metabolomics workflow using gas chromatography–high‐resolution mass spectrometry (GC‐HRMS) and liquid chromatography–high‐resolution tandem mass spectrometry (LC‐HRMS/MS) in positive and negative ionization mode, applying HILIC and reversed phase chromatography to assess multiple chemical classes. Data processing and statistical analysis were done in MS‐DIAL and MetaboAnalyst, respectively. The method was applied to 35 CBD‐type cannabis flowers grown under different environmental conditions, and compounds belonging to various chemical classes were successfully detected. Principal component analysis revealed distinct clustering of the samples, and key discriminative features were identified, including cannabinoids, terpenes such as β‐caryophyllene and α‐humulene, cuticular alkanes (e.g., pentacosane and nonacosane), and polar compounds such as choline and trigonelline. The markers enabled a discrimination of samples not only by chemical phenotype but also by cultivation environment, supporting the emerging concept of a cannabis terroir. In conclusion, this study introduces an analytical framework for the comprehensive chemical profiling of cannabis employing GC‐HRMS and LC‐HRMS analysis and advanced statistical techniques.

This study developed a comprehensive metabolomics workflow using GC‐HRMS and LC‐HRMS/MS and analyzed 35 CBD‐type cannabis flowers grown under varying conditions. Key discriminative compounds enabled classification by chemical phenotype and cultivation environment. The results showed that environmental factors influence chemical profiles, supporting the concept of a “cannabis terroir.”

1. Introduction

Cannabis sativa has been used for both medical and recreational purposes for centuries. It is one of the most popular psychoactive drugs worldwide, and its applications in medicine are growing, for example, for treating anorexia, nausea, pain, or spasticity, also due to changing legislations in various countries [1, 2, 3]. Within the species, there are large differences in the plant ingredients, which is why the plants are classified into different types. One of the most common classifications for the cannabis plant is currently based on the phenotype (chemotypes or chemovars), which is oriented around the tetrahydrocannabinol (THC) or cannabidiol (CBD) content of the plant and is typically divided into five groups [4]:

• Type I, when THC is predominant (drug‐type, THC > 0.3% and CBD < 0.5%);

• Type II, when the THC and CBD content is balanced (intermediate‐type, generally with CBD predominant);

• Type III, when CBD is predominant (THC < 1%);

• Type IV, when cannabigerol (CBG) is predominant;

• Type V, when there is no detectable amount of phytocannabinoids.

However, beyond the two cannabinoids commonly used for classification, over 100 additional cannabinoids have been identified, such as cannabichromenic acid (CBCA), cannabigerolic acid (CBGA), cannabidivarinic acid (CBDVA), and tetrahydrocannabivarinic acid (THCVA), along with their respective decarboxylated forms [5, 6]. In addition, the cannabis plant contains > 100 terpenes, dozens of flavonoids, and various minor compounds such as stilbenes, lignans, phytosterols, alkaloids, and amides [6, 7]. These constituents shape the character of the plant and are collectively responsible for the pharmacological profile.

Because of the use of cannabis in various fields such as medicine, food industry, cosmetics, and structural engineering, new varieties are created by breeders all over the world, resulting in an overwhelming number of chemical phenotypes (“strains”) on the market [4, 8]. Environmental and cultivation factors such as soil type, climate, altitude, and microbial communities influence the chemical and sensory profile even further, resulting in the need for a more complex classification. This is also true regarding recent cannabis legalization in various countries, or the amendment to the Swiss Federal Narcotics Act allowing pilot trials that involve the dispensing of Type I or II cannabis for nonmedicinal purposes since 2022. This means that currently, there are cannabis samples on the market belonging to the same type that are considered either legal or illegal. Therefore, a higher degree of differentiation of cannabis samples within the current classification gains importance not only in the fields of medicine and pharmacology but also for forensic sciences and jurisdiction. Beyond the increasing diversity of cannabis products on the market, a more detailed understanding of how environmental factors influence the plant's chemical composition is urgently needed. Drawing parallels from viticulture, the establishment of analytical platforms capable of mapping terroir‐driven chemical variability is essential in medical, forensic, and regulatory contexts.

In consequence, highly sophisticated analytical methods covering a broad range of constituents (metabolites) of the cannabis plant at once are of great interest. Metabolomics summarizes the investigation of mainly endogenous metabolites (small molecules with a molecular weight of generally < 1000 Da) in an organism and thereby represents the end of the “omics” cascade preceded by genomics, transcriptomics, and proteomics. Given the significant influence of environmental factors on the metabolome, it represents the omics science closest related to the phenotype [9]. In the context of forensic toxicology, metabolic studies in humans or animals typically aim for qualitative and/or quantitative investigation of metabolite changes caused by a certain stimulus, such as a particular disease or a drug intake (also referred to as toxicometabolomics) in order to find a diagnostic or analytical marker. In contrast, plants' metabolomic studies usually aim to identify new compounds or classify chemical phenotypes, for example, in the frame of testing pharmacological activities [10]. Analysis of the metabolome can be generally done by two different principles: untargeted and targeted metabolomics. Briefly, targeted designs will compare changes by a hypothesis‐driven approach for initially selected, limited numbers of compounds, usually in a quantitative way. Untargeted metabolomics, on the other hand, tries to cover theoretically all metabolite information within a sample without preselection, therefore including also unknown and unexpected compounds. It is mainly characterized by comprehensive data acquisition followed by extensive data processing and statistical evaluation in a relative or semiquantitative manner [9]. This might also represent a promising approach for further classification and differentiation of cannabis phenotypes [10]. The metabolome of the cannabis plant has only been investigated superficially to date. Hazekamp et al. used a targeted gas chromatography coupled to mass spectrometry (GC‐MS) approach and compared cannabis plants obtained from multiple sources and were able to identify the cannabis constituents that defined the samples into distinct chemotype groups. The study indicated the usefulness of a PCA approach for chemotaxonomic classification of cannabis varieties [11]. In 2016, the same group showed the effectiveness of a targeted metabolomics approach for chemotaxonomic mapping of cannabis varieties for medical use [12]. Cerrato et al. achieved a more detailed classification by a liquid chromatography coupled to high‐resolution mass spectrometry (LC‐HRMS) untargeted metabolomics approach for the detection of over a hundred phytocannabinoids [4]. This approach clearly indicated several new subgroups within the traditional classifications, which arise from a unique composition of the minor phytocannabinoids. The existence of these subgroups was stated to be of critical importance for evaluating the pharmacological effects of cannabis. Other researchers also used untargeted LC‐HRMS methods to differentiate between cannabis varieties or geographic regions [13, 14, 15]. Stupak et al. tested the use of sample derivatization prior to GC‐HRMS analysis for cannabis metabolome investigations [16]. All these studies show that HRMS techniques are a valuable tool to achieve a detailed classification of cannabis samples.

Therefore, this study aimed to develop an analytical method that allows the characterization of cannabis samples based on their metabolome, considering not only major plant components such as THC and/or CBD but also minor and even unknown components. The method will be used to assign specific fingerprints to 35 cannabis varieties belonging to the same type (Type III due to the legal status in Switzerland), making it possible to differentiate and classify cannabis samples beyond the common chemical phenotypes. This might improve knowledge and evaluation of the pharmacological effects of cannabis and establish knowledge about the terroir of cannabis plants—an environmental imprint reflected in the plant's chemical profile—similar to that observed in viticulture, which could be used to differentiate, for example, between legal and illegal plant material, or to investigate sensory attributes.

2. Materials and Methods

2.1. Chemicals and Reagents

Cannabis Terpenes Standards #1 and #2 were purchased from Restek GmbH (Bad Homburg v. d. Höhe, Germany). A C7–C40 saturated alkanes mixture, Supelco Terpene Mix A and B, trigonelline, choline, THC, CBD, other cannabinoids, and the internal standards trimipramine and propofol were obtained from either Sigma‐Aldrich (Buchs, Switzerland) or Cayman Chemical (Ann Arbor, MI, USA). Water was purified using a Purelab Ultra Millipore filtration unit (Labtech, Villmergen, Switzerland). All other chemicals were of the highest available purity and were purchased from Merck (Zug, Switzerland).

In total, 35 cannabis samples from 19 different producers were collected in January 2022 (Table S1). Of these, 10 were cultivated outdoors, 17 indoors, and 8 under greenhouse conditions. All samples were classified as CBD‐type varieties (Type III) with declared CBD contents ranging from 5% to 17% and THC contents below 1% and therefore legal in Switzerland.

2.2. Sample Preparation

Three inflorescences per cannabis variety were homogenized. From each homogenized sample, 100 mg was extracted in 5 mL of ethanol by ultrasonication for 10 min. Negative controls (n = 6) were prepared by ultrasonication of 5 mL of ethanol under identical conditions. For untargeted GC‐HRMS analysis, 20 μL of the extract was diluted with 980 μL of methanol containing 1‐μg/mL propofol and trimipramine as internal standards. For LC‐HRMS analysis, 50 μL of extract was diluted with 950 μL of methanol containing 1‐μg/L creatinine‐d3 and 2‐μg/L THC‐d3 as internal standards. All samples were prepared in triplicate. In order to enable subsequent feature filtering, a pooled sample and three row‐diluted pooled samples were prepared. The pooled sample was generated by combining equal volumes of all sample extracts. The row‐diluted pooled samples were prepared at concentrations of 50%, 100%, and 150% relative to the original pooled extract dilution. All samples were stored at −20°C until analysis.

2.3. GC‐HRMS Analysis

The samples were transferred to a GC autosampler (Gerstel MultiPurposeSampler MPS; Gerstel, Mülheim, Germany), controlled by Maestro software (version 1.4.40.1; Gerstel). Analysis was performed on a TRACE 1300 GC system (Thermo Scientific, Bremen, Germany) coupled to a Q Exactive GC Orbitrap mass spectrometer (Thermo Scientific, Bremen, Germany). Chromatographic separation was carried out on a TraceGOLD TG‐5SilMS capillary column (30 m × 0.25 mm i.d., 0.25‐μm film thickness; Thermo Scientific) with helium as the carrier gas at a constant flow rate of 1 mL/min. The inlet and transfer line temperatures were set to 250°C. The oven temperature program was as follows: initial temperature 50°C (1 min hold), ramped at 12°C/min to 320°C, followed by a final hold at 320°C for 4 min, resulting in a total runtime of 27.5 min. The injection volume was 1 μL with a split ratio of 5:1. The HRMS was operated in electron ionization (EI) mode at 70 eV, with the ion source temperature set to 230°C. Data acquisition was performed in full‐scan mode across a mass range of m/z 40–500 at a resolving power of 60,000 (FWHM at m/z 200), starting after a filament delay of 2.5 min. Lock mass correction was applied using background column bleed ions (m/z 207.03240, 225.04290, and 381.05110) for real‐time recalibration of potential m/z shifts. The system was controlled by Xcalibur software (version 4.0; Thermo Scientific). Samples were analyzed in randomized order. A pooled sample was injected every sixth sample to monitor system stability. Pooled dilution samples were analyzed four times, evenly distributed across the batch. The standard deviation of the retention time and the peak areas of the internal standards was assessed to evaluate method reproducibility.

2.4. LC‐HRMS Analysis

All samples were analyzed four times on a Shimadzu LC 40 XD3 system (Shimadzu, Muttenz, Switzerland) coupled to a Sciex X500R QTOF mass spectrometer (Sciex, Darmstadt, Germany).

Two injections were performed using a reversed‐phase column (XSelect HSS T3 RP‐C18, 150 mm × 2.1 mm, 2.5 μm particle size; Waters, Baden‐Dättwil, Switzerland) with water (A) and methanol (B), both containing 0.1% formic acid, as mobile phases. The gradient elution started at 10% B (0.5 min hold), increased to 100% B over 11.5 min, followed by a 2.5‐min hold at 100% B and re‐equilibration to starting conditions for 1.5 min. Two additional injections were performed using a hydrophilic interaction chromatography (HILIC) column (SeQuant ZIC‐HILIC, 150 mm × 2.1 mm, 3.5 μm particle size; Merck, Darmstadt, Germany) with water (A) and acetonitrile (C), both containing 0.1% formic acid, as mobile phases. The gradient began at 95% C (1 min hold), decreased to 40% C over 10 min, further down to 10% C within 2 min, held for 1 min, followed by a 2‐min re‐equilibration to initial conditions. The column oven temperature was set to 40°C for all runs. The injection volume was 1 μL, and the flow rate was 0.5 mL/min for both methods. Extensive needle and capillary washing with methanol, acetonitrile, and isopropanol was performed between injections to prevent carryover.

The mass spectrometer was operated in both positive and negative electrospray ionization (ESI) mode for each chromatographic method. A full‐scan TOF‐MS experiment was carried out in the range of m/z 50–800 with an accumulation time of 0.1 s, a declustering potential (DP) of 100 V with a ±20 V spread, and a collision energy (ce) of 5 V. This was followed by information‐dependent acquisition (IDA) of TOF‐MS/MS spectra. IDA settings included a maximum of seven candidate ions per cycle (intensity threshold: > 100 cps), dynamic background subtraction, and exclusion of previously selected ions for 4 s after three occurrences. MS/MS spectra were acquired in the m/z range of 50–800 with an accumulation time of 50 ms, a DP of 100 V ± 20 V, and a ce of 35 V ± 15 V. Automated mass calibration was performed after every tenth sample. The standard deviation of the retention time and the peak areas of the internal standards was assessed to evaluate method reproducibility.

2.5. GC‐HRMS Data Preprocessing

After conversion of the data files to .abf with the Reifycs Analysis Base File Converter v1.3, the data were preprocessed in MS‐DIAL v. 4.9.221218 applying the following parameters: mass range, 40–500 Da; retention time, 2.5–27.5 min; sigma window value, 0.5; minimum peak height, 2000; mass slice width and mass accuracy for centroiding, 0.025 Da; and smoothing method: linear weighted moving average, smoothing Level 3 scans, average peak width 10 scans, and retention time tolerance 0.05 min. All other parameters were set to default.

2.6. LC‐HRMS Data Preprocessing

After conversion of the data files to .mzML in ProteoWizard v3.0, the data were preprocessed in MS‐DIAL v. 4.9.221218 applying the following parameters: MS1 tolerance, 0.02 Da; MS2 tolerance, 0.025 Da; mass range, 50–800 Da; retention time, 0–16 min; minimum peak height, 1000; mass slice width, 0.03 Da; and smoothing method: linear weighted moving average, smoothing Level 2 scans, minimum peak width 5 scans, sigma window value 0.5, MS/MS abundance cut off 1000, alignment retention time, and MS1 tolerance 0.1 min and 0.015 Da. All other parameters were set to default.

2.7. Data Processing and Statistical Analysis

The MS‐DIAL raw data (peak areas) were exported, weight‐corrected, and filtered in the nPYc‐Toolbox [17] according to the following criteria: The relative standard deviation (RSD) of a feature in the pooled samples had to be ≤ 25%; the Pearson correlation in the pooled row diluted samples had to exceed 0.7; and the RSD of a feature in the actual samples had to be at least 20% higher than in the pooled samples. If multiple features represented the same analyte, the features with lower abundance were manually removed from the dataset. Further data processing was done in Metaboanalyst 5.0 [18]. Probabilistic quotient normalization (PQN) was performed using the pooled samples. Prior to statistical analysis, log transformation and Pareto scaling were applied. LC features obtained in positive and negative ionization mode were combined for multivariate analysis. Statistical analyses included analysis of variance (ANOVA, p < 0.05) and Tukey's HSD post hoc test, as well as principal component analysis (PCA), partial least squares discriminant analysis (PLS‐DA), and sparse PLS‐DA (sPLS‐DA). Model quality was evaluated by five‐fold cross‐validation (CV) and permutation testing to assess the risk of overfitting. GC‐HRMS features of interest were identified by comparison with certified reference standards, the NIST library v2.3, and the publicly available MS‐DIAL library “all records with Kovats RI,” which contains EI‐MS spectra and Kovat's retention indices for 9062 unique compounds. Kovat's indices were estimated by comparison to certified reference material consisting of C7–C40 saturated alkanes. LC‐HRMS features of interest were identified using certified reference materials and the software SIRIUS 6.0.7, a Java‐based tool for the structural elucidation of small molecules from tandem mass spectrometry data [19]. Identification were scored in Group 1 (identified against a reference standard), Group 2 (putatively annotated compounds, for example, matched to available databases), Group 3 (putatively characterized compound classes), and Group 4 (unknown), respectively, according to the suggestions of the metabolomics standard initiative [20].

3. Results

3.1. Feature Selection and Data Reduction

Peak picking and alignment across GC‐HRMS and LC‐HRMS datasets yielded 510 features for GC‐HRMS and 11,873 features for LC‐HRMS, combining positive and negative ionization modes from both HILIC and reversed‐phase chromatography. Specifically, LC‐HRMS generated 2453 features (HILIC positive), 2652 (HILIC negative), 4003 (RP positive), and 2765 (RP negative). After applying quality control filtering in the nPYc‐Toolbox and manual curation, 113 GC‐HRMS features, 88 LC‐HRMS features in positive mode, and 108 in negative mode remained for statistical analysis. Of the retained LC‐HRMS features, only a small number were derived from HILIC chromatography (six positive and four negative), with the majority originating from reversed‐phase separation. Evaluation of the internal standards showed stable retention times in all methods with standard deviations of < 0.02 min and acceptable RSDs of the peak areas of < 25%.

3.2. GC‐HRMS

PCA revealed distinct clustering of the cannabis varieties based on their metabolic profiles (Figure 1). In the PLS‐DA analysis based on cultivation conditions, samples grown under greenhouse and indoor conditions showed partial overlap but were clearly separated from those grown outdoors (Figure 2A). When the dataset was reduced to include only the most relevant features, separation between greenhouse‐ and indoor‐grown varieties also became apparent (Figure 2B). One‐way ANOVA identified 77 features with significant differences between cultivation conditions. The 10 most important discriminative features in the PLS‐DA or sPLS‐DA analysis of cultivation conditions (top 10 compounds based on the VIP score) included eight terpenes (β‐caryophyllene, α‐humulene, caryophyllene oxide, β‐ocimene, terpinolene, and three unidentified terpenes), three alkanes (pentacosane, heptacosane, and nonacosane), one unidentified cannabinoid, and one additional unidentified compound (Table 1). For the differentiation of the 35 cannabis varieties, the 10 most important features in the PLS‐DA model (top 10 in VIP score) were all terpenes: terpinolene, β‐ocimene, 3‐carene, γ‐terpinene, α‐bisabolol, three unidentified monoterpenes, and two unidentified sesquiterpenes (Table 1). Cross‐validation and permutation testing results are presented in the supporting information (Figure S1) and did not indicate overfitting. The mass spectra of the most important features are provided in Figures S02–S20.

FIGURE 1.

PCA plot of the 35 CBD‐type varieties S01–S35 analyzed with GC‐HRMS. S01–S10 were cultivated outdoors, S11–S27 were cultivated indoors, and S28–S35 were cultivated in a greenhouse.

FIGURE 2.

PLS‐DA (A) and sparse PLS‐DA (B) plots of the growing conditions greenhouse (GH; red), indoor (ID; green), and outdoor (OD; blue) analyzed with GC‐HRMS.

TABLE 1.

The 10 most important GC‐MS features in PLS‐DA and sPLS‐DA separating the varieties (V) and the growing conditions (G) indoors (ID), outdoors (OD), and greenhouse (GH) including retention time (RT), mass‐to‐charge ratio (m/z), the level of identification according to the MSI [20], the top 10 group regarding variable importance in projection (VIP), and the Tukey's HSD post hoc test results.

RT, min	m/z	Analyte name	Level of identification	Top 10 in VIP scores	Tukey's HSD
6.50	93.0700	3‐carene	1	V (PLS‐DA)	ID‐GH; OD‐ID
6.83	91.0543	beta‐ocimene	1	V (PLS‐DA); G (PLS‐DA)	ID‐GH; OD‐ID
7.03	91.0543	gamma‐terpinene	1	V (PLS‐DA)	ID‐GH; OD‐ID
7.39	91.0543	terpinolene	1	V (PLS‐DA); G (PLS‐DA)	ID‐GH; OD‐ID
8.13	79.0543	monoterpene	3	V (PLS‐DA)	ID‐GH
8.59	69.0336	monoterpene	3	V (PLS‐DA); G (PLS‐DA)	ID‐GH; OD‐ID
8.63	111.0800	monoterpene	3	V (PLS‐DA); G (PLS‐DA)	ID‐GH; OD‐ID
11.49	91.0543	beta‐caryophyllene	1	G (sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
11.87	93.0699	Alpha‐humulene	1	G (sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
12.39	139.1118	sesquiterpene	2	G (sPLS‐DA)	OD‐GH; OD‐ID
13.19	91.0543	caryophyllene oxide	1	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
13.46	96.0570	not identified	4	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
13.71	111.0805	sesquiterpene	3	V (PLS‐DA)	ID‐GH; OD‐ID
14.02	111.0805	sesquiterpene	3	V (PLS‐DA)	ID‐GH; OD‐ID
14.07	119.0856	alpha‐bisabolol	1	V (PLS‐DA)	OD‐GH; OD‐ID
19.18	189.0910	cannabinoid	3	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
20.34	57.0699	pentacosane	1	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
21.61	57.0699	heptacosane	1	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
22.81	57.0699	nonacosane	1	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID

3.3. LC‐HRMS

PCA showed distinct clustering of the cannabis varieties based on their LC‐HRMS profiles (Figure 3). The PLS‐DA plots based on cultivation conditions showed partial overlap between indoor, greenhouse, and outdoor samples (Figure 4). Sparse PLS‐DA (sPLS‐DA) did not improve group separation (data not shown). One‐way ANOVA identified 50 negatively and 53 positively ionized features with statistically significant differences between cultivation types. The 10 most important discriminative features in PLS‐DA or sPLS‐DA (top 10 compounds regarding VIP score) for the varieties and cultivation conditions are listed in Table 2 (positive ionization) and Table 3 (negative ionization). In negative ionization mode, CBD, Δ9‐THC, CBNA, two additional cannabinoids (Level 3), a flavonoid, a terpene glycoside (Level 3), and another terpene derivative (Level 3) were identified among the top features. The terpene glycoside was detected twice, once in HILIC and once in reversed‐phase negative mode. In positive mode, CBD, THC, choline, and trigonelline (Level 1); three putative cannabinoids (Level 3); and one flavonoid (Level 3) were identified. Choline was identified twice in positive mode—once in HILIC and once in reversed‐phase analysis. The flavonoid detected in positive mode was also among the top 15 features in negative ionization. Cross‐validation and permutation testing results are presented in Figure S1 and did not indicate overfitting. The mass spectra of the most important features are provided in Figures S21–S36 and S37–S51 for positive and negative ionization, respectively.

FIGURE 3.

PCA plot of the 35 CBD‐type varieties S01–S35 analyzed with LC‐HRMS in (A) negative mode and (B) positive mode.

FIGURE 4.

PLS‐DA plots of the growing conditions greenhouse (GH; red), indoor (ID; green), and outdoor (OD; blue) analyzed with LC‐HRMS in (A) negative mode and (B) positive mode.

TABLE 2.

The 10 most important positive ionized LC‐MS features in PLS‐DA and sPLS‐DA separating the varieties (V) and the growing conditions (G) indoors (ID), outdoors (OD), and greenhouse (GH) including retention time (RT), mass‐to‐charge ratio (m/z), chromatographic mode, the level of identification according to the MSI [20], the top 10 group regarding variable importance in projection (VIP), and the Tukey's HSD post hoc test results.

RT, min	m/z	Analytical mode	Analyte name	Level of identification	Top 10 in VIP scores	Tukey's HSD
0.68	104.1073	RP positive	choline	1	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
2.06	146.1177	HILIC positive	not identified	4	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
5.21	104.1074	HILIC positive	choline	1	G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
6.06	138.0554	HILIC positive	trigonelline	1	G (sPLS‐DA)	ID‐GH; OD‐GH
10.97	359.2226	RP positive	not identified	4	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
10.97	381.2034	RP positive	not identified	4	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
11.16	359.2219	RP positive	cannabinoid	3	G (sPLS‐DA)	ID‐GH; OD‐ID
11.19	333.2429	RP positive	cannabinoid	3	V (PLS‐DA); G (PLS‐DA	ID‐GH; OD‐ID
11.80	315.2310	RP positive	CBD	1	V (PLS‐DA); G (PLS‐DA	ID‐GH; OD‐ID
11.80	337.2145	RP positive	not identified	4	V (PLS‐DA)	ID‐GH; OD‐ID
11.80	415.1586	RP positive	not identified	4	V (PLS‐DA); G (PLS‐DA	ID‐GH; OD‐ID
11.90	401.2331	RP positive	not identified	4	G (sPLS‐DA)	ID‐GH; OD‐ID
12.29	407.1856	RP positive	flavonoide	3	V (PLS‐DA); G (PLS‐DA	ID‐GH; OD‐ID
12.37	317.2485	RP positive	not identified	3	G (sPLS‐DA)	OD‐GH; OD‐ID
12.57	315.2323	RP positive	d9‐THC	1	V (PLS‐DA)	ID‐GH; OD‐ID
12.77	315.2323	RP positive	cannabinoid	3	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID

TABLE 3.

The 10 most important negative ionized LC‐MS features in PLS‐DA and sPLS‐DA separating the varieties (V) and the growing conditions (G) indoors (ID), outdoors (OD), and greenhouse (GH) including retention time (RT), mass‐to‐charge ratio (m/z), chromatographic mode, the level of identification according to the MSI [20], the top 10 group regarding variable importance in projection (VIP), and the Tukey's HSD post hoc test results.

RT, min	m/z	Analytical mode	Analyte name	Level of identification	Top 10 in VIP scores	Tukey's HSD
2.06	549.2342	HILIC negative	not Identified	4	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	OD‐GH; OD‐ID
2.94	551.2502	HILIC negative	terpene glycoside	3	V (PLS‐DA); G (PLS‐DA)	OD‐GH; OD‐ID
3.99	713.3019	HILIC negative	not identified	4	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
7.50	551.2499	RP negative	terpene glycoside	3	G (PLS‐DA and sPLS‐DA)	OD‐GH; OD‐ID
9.00	245.1181	RP negative	terpene derivative	3	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
10.12	389.1970	RP negative	not identified	4	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
10.31	373.2020	RP negative	not identified	4	G (sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
10.98	357.2068	RP negative	cannabinoid	3	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
11.20	331.2282	RP negative	not identified	4	V (PLS‐DA)	ID‐GH; OD‐ID
11.24	357.2069	RP negative	not identified	4	G (sPLS‐DA)	ID‐GH; OD‐ID
11.69	343.1908	RP negative	cannabinoid	3	G (sPLS‐DA)	ID‐GH; OD‐GH; OD‐ID
11.81	313.2169	RP negative	CBD	1	V (PLS‐DA); G (PLS‐DA)	ID‐GH; OD‐ID
12.29	405.1706	RP negative	flavonoide	3	G (PLS‐DA, top 15)	ID‐GH; OD‐ID
12.77	313.2175	RP negative	THC	1	V (PLS‐DA); G (PLS‐DA and sPLS‐DA)	ID‐GH; OD‐ID
13.05	353.1758	RP negative	CBNA	1	G (sPLS‐DA)	ID‐GH; OD‐GH

4. Discussion

This study aimed to establish a sophisticated analytical platform for the comprehensive metabolomic profiling of cannabis flowers and to apply it to authentic samples. Both GC‐HRMS and LC‐HRMS were employed for that purpose and were successfully applied to distinguish cannabis samples from different growing conditions. GC‐HRMS provided the highest resolution in terms of chromatographic separation and mass spectrometry. In addition, the results were achieved by only one injection—in contrast to LC‐HRMS, where the results of four injections were combined. However, the extension of the GC‐MS workflow applying prior derivatization might expand the variety of detectable analytes [16]. The most important discriminative features in the PLS‐DA or sPLS‐DA analysis of cultivation conditions using GC‐HRMS included β‐caryophyllene, α‐humulene, caryophyllene oxide, β‐ocimene, and terpinolene, which are mainly synthesized in the glandular trichomes of the plant as defense compounds [21]. Among the top 10 features, β‐caryophyllene is one of the most abundant terpenes and serves as a major discriminator in both PLS‐DA and sPLS‐DA analyses. Environmental stress (e.g., UV exposure, pathogens, or drought) might induce an upregulation of its biosynthesis, which is of major interest because of its bioactive properties as a selective CB2 receptor agonist [22].

Besides terpenoid compounds, aliphatic hydrocarbons such as pentacosane, heptacosane, and nonacosane were accessible by GC‐HRMS and were shown to be an important chemical group for the discrimination of phenotypes. These compounds are part of the cuticula, a wax layer covering the surface of leaves, stems, and flowers, and showed differences in the growth conditions. This finding was also observed by Tipple et al., who found significant differences between the concentrations of especially nonacosane in enclosed versus open‐field growth settings [23]. The accessibility of alkanes offers another dimension for the discrimination of cannabis phenotypes and their taxonomy. These compounds are not known to have a pharmacological effect; they are included in cannabis essential oil and might, for example, influence the pharmacokinetics of cannabinoids by influencing their absorption.

LC‐HRMS accessed the major cannabinoids whose biosynthesis is under genetic regulation but is also influenced by environmental factors. Minor cannabinoids were successfully detected, contributing to separating the phenotypes. However, identification of unknown LC‐MS features was more challenging compared with the GC‐MS features, for example, due to the better comparability of GC‐MS spectra in library searches. Noncannabinoid polar compounds such as choline, trigonelline, and flavonoids were identified as important features for phenotype separation and contributed to the overall profile of the plant. Trigonelline and choline were found to significantly increase in concentration with increasing harvesting time point of the cannabis plant by Spano et al. [24]. A comparison of the two LC techniques, HILIC and RP, revealed that RP was more suitable for detecting compounds with significant differences between varieties than HILIC, with a total of 196 versus only 10 features, respectively. In addition, among the six important HILIC features, two compounds (choline and the terpene glycoside) were also identified among the top 10 RP features. Therefore, omission of HILIC experiments would be an option to reduce workload without missing relevant data. However, identifying several identical compounds with different techniques among the top 10 important features demonstrated that data acquisition, preprocessing, and statistical analysis were successful and plausible within each technique.

A comparison of the growing conditions separated the outdoor samples from those grown indoors and in greenhouses (Figures 2 and 4). However, the samples differed not only in the growing conditions but also probably in other (unknown) characteristics, such as genetics, age, or storage conditions. The influence of these characteristics on the result cannot be excluded and could be investigated in another experiment. Nevertheless, these results show that the presented workflow might be used to investigate the origin of cannabis samples and provide further, more detailed differentiation compared with the current classification in only five chemotypes. This will be helpful in forensic investigations or for cannabis traders to verify the sellers' declarations. Among the 35 samples, the two varieties “Cookies Kush” (S07 and S30) and “CPure Fedtonic” (S06 and S33) were represented twice, but grown under different conditions, once outdoors (S07 and S06) and once in a greenhouse (S30 and S33). The samples of these two varieties were clearly separated in the PCA plots of the GC‐HRMS data (Figure 1) and the negative LC‐HRMS data (Figure 3A). This shows that differences in samples that do not have genetic origins can be detected. This might be of interest for pharmaceutical quality control, for example, to monitor batch‐to‐batch differences or the influence of storage conditions, as for example, stability issues do not only concern cannabinoids but also other relevant plant constituents such as terpenes [25]. Our data showed again that environmental conditions influence the plant's chemical composition, similar to wine. This might lead to characteristic chemical profiles correlated to defined geographic origin or cultivation practice and might be used for product traceability, quality, safety, and consumer information.

The incomplete identification of some detected features and the absence of genetic data limited the application of the analytical platform. Research, including genotyping and untargeted metabolomics with the established analytical methods, is of great importance and could be part of future studies applying even more detailed structural elucidation, such as NMR. Furthermore, a transformation of the analytical platform to a targeted workflow, which is more effective regarding workload and therefore cost, should be investigated. However, the results supported the concept of a cannabis terroir, which was successfully analyzed by a platform using GC‐HRMS and LC‐HRMS/MS.

The established analytical workflow provides a versatile and transferable platform for the comprehensive chemical profiling of cannabis plants. In the future, it could be applied to broader sample sets—including Type I and II cannabis—as well as controlled cultivation trials to systematically study the impact of individual environmental factors. Additionally, the current untargeted setup could be adapted into a targeted platform focusing on key marker compounds identified in this study, thereby reducing analytical workload and improving throughput for routine applications such as batch control, product verification, or forensic casework.

5. Conclusion

An analytical platform using GC‐HRMS and LC‐HRMS for untargeted metabolomic profiling of cannabis flowers was successfully established and applied to authentic samples under different growing conditions. Metabolomic fingerprints, including major and minor cannabinoids, terpenes, aliphatic hydrocarbons, flavonoids, and other secondary metabolites, enabled the separation of samples. Key discriminative compounds included bioactive terpenes such as β‐caryophyllene, as well as cuticular alkanes, which may serve as novel markers for cultivation‐related adaptation. The results further support the emerging concept of a cannabis terroir—an environmental imprint reflected in the plant's chemical profile—similar to that observed in viticulture. This underlines the need for comprehensive profiling approaches beyond classical THC/CBD quantification to ensure product standardization, traceability, and authenticity. Future studies integrating genetic and sensory data will be instrumental in deepening our understanding of the complex relationships between genotype, environment, and metabolite expression in cannabis.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1 Numbers, growing conditions (OD = outdoor, ID = indoor, GH = greenhouse), producers, and names of the cannabis varieties.

Figure S1 Cross‐validation (fivefold CV) and permutation testing results of the PLS‐DA plots comparing the growing conditions with (A) GC‐HRMS, (B) LC‐HRMS in negative mode, and (C) LC‐HRMS in positive mode.

Figure S2 Mass spectrum in the QC pool sample of 3‐carene.

Figure S3 Mass spectrum in the QC pool sample of beta‐ocimene.

Figure S4 Mass spectrum in the QC pool sample of gamma‐terpinene.

Figure S5 Mass spectrum in the QC pool sample of terpinolene.

Figure S6 Mass spectrum in the QC pool sample of the monoterpene with RT 8.13 and m/z 79.0543.

Figure S7 Mass spectrum in the QC pool sample of the monoterpene with RT 8.59 and m/z 69.0336.

Figure S8 Mass spectrum in the QC pool sample of the monoterpene with RT 8.63 min and m/z 111.0800.

Figure S9 Mass spectrum in the QC pool sample of beta‐caryophyllene.

Figure S10 Mass spectrum in the QC pool sample of alpha‐humulene.

Figure S11 Mass spectrum in the QC pool sample of the sesquiterpene with RT 12.39 min and m/z 139.1118.

Figure S12 Mass spectrum in the QC pool sample of caryophyllene oxide.

Figure S13 Mass spectrum in the QC pool sample of the unidentified feature with RT 13.46 min and m/z 96.0570.

Figure S14 Mass spectrum in the QC pool sample of the sesquiterpene with RT 13.71 and m/z 111.0805.

Figure S15 Mass spectrum in the QC pool sample of the sesquiterpene with RT 14.02 and m/z 111.0805.

Figure S16 Mass spectrum in the QC pool sample of alpha‐bisabolol.

Figure S17 Mass spectrum in the QC pool sample of the cannabinoid with RT 19.18 min and m/z 189.0910.

Figure S18 Mass spectrum in the QC pool sample of pentacosane.

Figure S19 Mass spectrum in the QC pool sample of heptacosane.

Figure S20 Mass spectrum in the QC pool sample of pentacosane.

Figure S21 Mass spectrum in the QC pool sample of choline (RP).

Figure S22 Mass spectrum in the QC pool sample of the unidentified feature with RT 2.06 min and m/z 146.1177.

Figure S23 Mass spectrum in the QC pool sample of choline (HILIC).

Figure S24 Mass spectrum in the QC pool sample of trigonelline.

Figure S25 Mass spectrum in the QC pool sample of the unidentified feature at 10.97 min with m/z 359.226.

Figure S26 Mass spectrum in the QC pool sample of the unidentified feature at RT 10.97 with m/z 381.2034.

Figure S27 Mass spectrum in the QC pool sample of the cannabinoid at 11.16 min with m/z 359.2219.

Figure S28 Mass spectrum in the QC pool sample of the cannabinoid at 11.19 min with m/z 333.2429.

Figure S29 Mass spectrum in the QC pool sample of CBD.

Figure S30 Mass spectrum in the QC pool sample of the unidentified feature at 11.80 min with m/z 337.2145.

Figure S31 Mass spectrum in the QC pool sample of the unidentified feature at RT 11.80 min with m/z 415.1586.

Figure S32 Mass spectrum in the QC pool sample of the unidentified feature at RT 11.90 min with m/z 401.2331.

Figure S33 Mass spectrum in the QC pool sample of the flavonoide at 12.29 min with m/z 407.1856.

Figure S34 Mass spectrum in the QC pool sample of the unidentified feature at 12.37 min with m/Z 317.2485.

Figure S35 Mass spectrum in the QC pool sample of d9‐THC.

Figure S36 Mass spectrum in the QC pool sample of the cannabinoid at RT 12.77 min with m/z 315.2323.

Figure S37 Mass spectrum in the QC pool sample of the unidentified feature at RT 2.06 min (HILIC) with m/z 549.2342.

Figure S38 Mass spectrum in the QC pool sample of the terpene glycoside at RT 2.94 min (HILIC) with m/z 551.2502.

Figure S39 Mass spectrum in the QC pool sample of the unidentified feature at RT 3.99 min (HILIC) with m/z 713.3019.

Figure S40 Mass spectrum in the QC pool sample of the terpene glycoside at 7.50 min (RP) and m/z 551.2499.

Figure S41 Mass spectrum in the QC pool sample of the terpene derivative at RT 9.00 min (RP) and m/z 245.1181.

Figure S42 Mass spectrum in the QC pool sample of the unidentified feature at RT 10.12 min (RP) and m/z 389.1970.

Figure S43 Mass spectrum in the QC pool sample of the unidentified feature at 10.31 min (RP) and m/z 373.2020.

Figure S44 Mass spectrum in the QC pool sample of the cannabinoid at RT 10.98 min (RP) and m/z 357.2068.

Figure S45 Mass spectrum in the QC pool sample of the unidentified feature at 11.20 min with m/z 331.2282.

Figure S46 Mass spectrum in the QC pool sample of the unidentified feature at RT 11.24 min and m/z 357.2069.

Figure S47 Mass spectrum in the QC pool sample of the cannabinoid at RT 11.69 min (RP) and m/z 343.1908.

Figure S48 Mass spectrum in the QC pool sample of CBD.

Figure S49 Mass spectrum in the QC pool sample of the flavonoid at RT 12.29 min and m/z 405.1706.

Figure S50 Mass spectrum in the QC pool sample of THC.

Figure S51 Mass spectrum in the QC pool sample of CBNA.

Poetzsch S., Poetzsch M., Kraemer T., and Steuer A., “Toward a Cannabis Terroir: Untargeted Metabolomic Profiling of Authentic Samples Using Gas Chromatography–High‐Resolution Mass Spectrometry (GC‐HRMS) and Liquid Chromatography–High‐Resolution Tandem Mass Spectrometry (LC‐HRMS/MS),” Drug Testing and Analysis 17, no. 10 (2025): 2086–2095, 10.1002/dta.3922.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.