Abstract
Understanding the stability and degradation pathways of cannabinoids during storage is crucial for both medicinal and forensic applications of Cannabis sativa L. This study investigated the long-term storage effects on cannabinoid profiles in Cannabis sativa L. (C. sativa) resin. 150 resin samples, seized by the Royal Gendarmerie of Morocco and stored for long periods of up to 8 years, were analyzed. Cannabinoid percentages were quantified using gas chromatography with a flame ionization detector (GC/FID). Fresh, unstored cannabis samples initially exhibited the greatest overall extraction yield (37.9 ± 1.45%) and were characterized by high Δ9-tetrahydrocannabinol (THC) concentrations (35.16 ± 3.87%), alongside lower levels of cannabidiol (CBD) (3.92 ± 0.40%) and cannabinol (CBN) (0.87 ± 0.35%). Interestingly, after 2 years of storage, a significant decrease in THC content was observed 2.74 ± 0,33%, accompanied by a corresponding increase in CBD (6.71 ± 0.77%) and CBN (6.94 ± 0.83%) concentrations. Principal component analysis (PCA) revealed strong correlations between storage time, yield of extraction, and cannabinoid profiles, and enabled the classification of samples based on storage duration and cannabinoids content. Polynomial and multiple regression were developed to predict the relationship between cannabinoid content and storage time. The prediction results indicate that the cubic model exhibit superior predictive performance compared to linear and quadratic alternatives in prediction cannabinoids in terms of storage time. Further, a multiple linear regression model with a hight R2 = 0.99 was developed to predict the storage time based on cannabinoid levels. These findings provide valuable insights into the long-term stability of cannabinoids in C. sativa resin and hold potential implications for forensic applications, particularly in cases involving the analysis of aged cannabis samples.
Supplementary Information
The online version contains supplementary material available at 10.1038/s41598-025-03888-7.
Keywords: Cannabis sativa L. resin, Cannabinoids, Long-term storage, Principal component analysis, Modeling
Subject terms: Plant sciences, Chemistry, Mathematics and computing
Introduction
Cannabis is a genus of plants in the Cannabaceae family, comprising three sub-species: C. sativa L., Cannabis indica Lam., and Cannabis ruderalis J. C. sativa L. (commonly known as marijuana) originates from Central Asia, specifically India and China1. By the early sixteenth century, its cultivation had spread across the Middle East, Europe, and South America2. C. sativa is produced in about 60 countries, and its principal sources are in the countries of Southwest Asia and North Africa3,4. The production and cultivation of industrial varieties of cannabis in Morocco have a significant socio-economic impact. Until recently, cannabis cultivation in Morocco was prohibited. However, in June 2021, a significant shift in policy occurred when the Moroccan government approved draft law N°13–21. This legislation marked a turning point by legalizing the cultivation and production of cannabis for specific purposes, including medical, pharmaceutical, and industrial applications5.
Cannabis produces a unique class of terpenophenolic compounds known as cannabinoids, with over 130 identified to date6. Of these, Δ9-tetrahydrocannabinol is the most biologically active. Cannabinoids are synthesized and stored in glandular trichomes, which are distributed across the majority of the aerial parts of the Cannabis plant, with the highest concentration typically found in the unfertilized female flowers7. THC is the primary psychoactive constituent and is responsible for mood- and consciousness-altering effects8. Cannabinoids are being actively studied for potential therapeutic benefits in conditions such as cancer, Alzheimer’s disease, Huntington’s disease, Parkinson’s disease, epilepsy, and cardiovascular disorders9–11. Most medicinal effects of cannabinoids are due to the action of cannabinoids other than THC, such as CBD and CBN. CBD exhibits various potential therapeutic effects, including anti-epileptic, anti-dystonic, anti-emetic, and anti-inflammatory properties, and is recognized for its potential neuroprotective actions12,13. CBN provides analgesic, antibacterial, and antifungal activities14.
Cannabis resin is a sticky substance produced by the glandular trichomes of the Cannabis plant (C. sativa or Cannabis indica), containing cannabinoids like THC and CBD, terpenes, and other phytochemicals that contribute to its psychoactive and therapeutic properties15. It is harvested through methods such as dry sifting, ice water extraction, or solvent extraction, where trichomes are separated and processed into concentrated forms like hashish or Cannabis oil. The resin is then purified and cured to enhance its quality, with proper storage ensuring its stability16.
The cannabinoid profile in C. sativa is highly variable and influenced by several factors, including cultivation conditions (e.g., light, temperature, and nutrient availability)17, the location within the plant (e.g., leaves, flowers, or trichomes), and genetic differences between varieties18. Consequently, the cannabinoid concentrations in the produced resin are not uniform, which underscores the importance of considering these factors when studying long-term cannabinoid transformation and storage effects. Despite cannabinoids extensive and increasing use, data on the variability of most cannabinoids in Cannabis resin during a long-term storage remains scarce. Traditionally, the analysis of cannabinoids has relied on gas chromatography (GC) or liquid chromatography (LC) techniques coupled with mass spectrometry (MS) as the detection method. Alternatively, flame-ionization detection (FID) may be used for GC analyses, or an ultra-violet diode array detector (UV-DAD) for LC. While these methods are accurate and selective, they can also be costly and time-consuming19. A predictive model that estimates cannabinoid levels based solely on storage duration would be invaluable for optimizing inventory management, establishing standardized storage guidelines, and reducing the need for extensive chemical analysis. On the other hand, Predicting the storage life of Cannabis resin through cannabinoid content analysis offers valuable insights for forensic science, quality control, and regulatory compliance20. As THC degrades to CBN over time, the THC/CBN ratio serves as a useful indicator of storage duration21. This predictive capability aids law enforcement in tracing illicit drug networks and estimating seizure timelines22. For the pharmaceutical and nutraceutical industries, it ensures product potency and safety throughout the supply chain23.
This study aimed to thoroughly examine the impact of long-term storage on the chemical composition and the extraction yield of Cannabis resin. In this context, the Forensic Institute of the Royal Gendarmerie Laboratory of Morocco analyzed 150 chemical profiles from samples stored for periods extending to 8 years. To our knowledge, this is the first study to examine Cannabis samples stored for such an extensive duration, providing unique insights into the long-term stability and degradation of cannabinoids. A range of statistical tools were used, including two-way ANOVA with post hoc tests to determine the statistical significance of storage duration on cannabinoid composition, Principal Component Analysis to identify correlations among cannabinoids, storage time, and yield of extraction and for samples classification, polynomial and multiple linear regression to develop predictive models of the relationship between cannabinoid and storage time. By conducting this multifaceted analysis, we aimed to bridge a significant gap in the literature regarding long-term cannabinoid stability and degradation pathways.
Materials and methods
Sample origin, storage conditions, and preparation
The 150 resin samples analyzed in this study were seized by the Royal Gendarmerie of Morocco during law enforcement operations in 2015, originating primarily from the Rif Mountains. Post-seizure, samples were stored as forensic evidence in secure facilities under standard protocols, involving placement in opaque polyethylene bags or cardboard boxes, kept in dark, dry rooms at ambient temperature (20–25 °C). The resin, produced via traditional dry-sieving methods, was compressed into blocks of varying sizes (5–500 g) and forms (rectangular or irregular slabs). These storage conditions reflect typical real-world forensic archiving practices, providing ecologically valid insights into long-term cannabinoid degradation. From the same batch of samples, 30 random samples were taken every 2 years, forming the five groups listed in Table 1 (A, B, C, D, and E), and analyzed by GC/MS. To ensure representative sampling, every samples’ group were collected from multiple spatial points within each block (outer surface, inner core) and homogenized prior to analysis.
Table 1.
Samples of C. sativa resin according to storage time.
| Samples | Storage time |
|---|---|
| A | Non-stored |
| B | 2 years |
| C | 4 years |
| D | 6 years |
| E | 8 years |
This grouping allowed for a systematic evaluation of cannabinoid degradation trends over time, ensuring robust analysis and validation of the predictive models.
Extraction procedures and sample preparation
Ultrasound assisted extraction (UAE) was performed by a 50% ethanolic solvent (ethanol:water, 1:1, v/v) using an adjustable ultrasonic bath with a power of 200 W and a frequency of 35 kHz. The 50% ethanol solvent was selected for UAE based on its proven efficacy in extracting cannabinoids from C. sativa24. Research indicates that ethanol not only enhances the extraction yield of cannabinoids but also helps preserve the chemical integrity of these compounds during extraction, making it a preferred solvent in many studies24. After grinding and homogenization, 100 mg of powder was extracted with 20 ml of the ethanol/water solvent, already containing 0.25 g/L of n-Nonadecane as an internal standard. The experiments were performed using increasing extraction times of 5, 12.5, and 20 min. After ultrasonic extraction, 1 μL was analyzed by GC/FID. Each experiment was carried out in triplicate. For yield of extraction calculation, the extracts were filtered under vacuum and the solvents were evaporated using a rotary evaporator. The formula (Eq. 1) was used to determine the yield of extraction on a dry weight basis:
![]() |
1 |
W2 is the weight of the extract and the container, W1 is the weight of the container alone, and W0 is the weight of the initial dried sample. This value represents the total mass of the crude extract obtained after solvent evaporation, expressed as a percentage of the initial dry weight (100 mg) of the C. sativa resin powder used for extraction.
Gas chromatography/flame ionization detector analysis
Gas chromatography with flame ionization detector (GC/FID) analyses were performed on an Agilent 6890N instrument equipped with an autosampler, a split and splitless injector, and a flame ionization detector. The separation was accomplished using an Agilent capillary column, type 5% phenyl 95% methylsiloxane HP-5 (30 m × 0.32 mm × 0.25 μm), with helium and liquid air as carrier gas. A 22-min oven temperature program was adopted: 60 °C isothermals for 2 min, 15 °C/min ramp, and 280 °C isothermals for 5 min. Splitless injection was heated to 270 °C, while the FID detector was heated to 280 °C.The injection was performed with an automatic injector. After five washes, the microsyringe injected 1 µL of sample into the GC column for separation, with detection by flame ionization (FID). All samples were chemically analyzed for THC, CBD, and CBN content. Component identification was confirmed by comparing the retention time of the standard and test samples. All the experiments were carried out in triplicate. The appropriate method for the quantification of THC, CBD and CBN cannabinoids in Cannabis and its derivatives with GC/FID is a validated method based on an intra-laboratory study. According to the results found, CBN used for validation as a stable compound acts in FID in the same way as THC and CBD. The CBN/THC and CBN/CBD ratios are very close to 1 according to these results; CBN is suitable for the calibration of other cannabinoids. The calibration graphs produced were linear in the range of 0.1–16% and with a correlation coefficient of 0.999. The limit of detection (LD) and the limit of quantification (LOQ) are respectively of the order of 0.02% and 0.08%, the repeatability and reproducibility have proven to be sufficient and acceptable. The expanded relative uncertainty was estimated at 5% (K = 2). The GC/FID method quantified total cannabinoid content (acidic + neutral forms), as high-temperature injection and column conditions induce decarboxylation of acidic precursors (e.g., THCA, CBDA) into their neutral counterparts (THC, CBD). This approach precludes separate quantification of acidic and neutral forms, a limitation inherent to gas chromatography without prior derivatization. The GC/FID method’s suitability was confirmed via rigorous intra-laboratory validation, which is performed periodically every 6 months. Linearity was excellent (R2 ≥ 0.9998) across the 0.1–16% (w/w) calibration range (7 levels). Accuracy met pre-defined criteria, assessed via bias from five intermediate precision runs. These runs (different days/operators) demonstrated strong inter-day precision (%RSD ≤ 1.7% for levels ≥ 0.5%), ensuring ongoing reliability for comparing results across the study’s 8-year timeframe.
Chemometric analysis
Principal component analysis was utilized to identify the correlation between variables and to determine the similarity between individuals based on the studied variables, percentage of cannabinoids, storage time and the yield of extraction. Linear, quadratic and cubic regression were used as a predictive method that allows us to create a suitable model by finding the relationship between various independent (storage time) and dependent variables (percentages of cannabinoids.). Six parameters were employed to assess and compare the performance of the different regression models: Akaike Information Criterion (AICc), Bayesian Information Criterion (BIC), sum of squared errors (SSE), mean squared error (MSE), and root mean squared error (RMSE), and the coefficient of determination R225. AICc and BIC are model selection criteria that estimate prediction error while balancing goodness-of-fit with model complexity. SSE quantifies the total variability around the fitted regression line, while MSE and RMSE measure average error magnitudes. Finally, R2 represents the proportion of variance in the dependent variable explained by the independent variable, serving as a measure of overall model fit. PCA and polynomial regression were performed using JMP Pro 17 software.
Statistical analysis
Two way Anova and Tukey post hoc test were used to evaluate the effects of storage and extraction duration on the percentage of the three main cannabinoids and the yield of extraction. The two-way Anova tests were performed using GraphPad prism 9 software. Regression fitting was carried out using JMP 17 PRO.
Results and discussion
Effect of storage and extraction time on yield of extraction and cannabinoids
To track the changes in the four parameters, extraction yield, THC, CBD, and CBN over time, the average values of these parameters were calculated for each storage duration. These average values were then compared across the different storage periods using ANOVA analysis supplemented by Tukey’s post-hoc test. The extraction yield and cannabinoid profile of C. sativa resin exhibited significant variations correlated with sample storage duration. The non-stored sample (A) yielded the highest resin content at 37.9 ± 1.45%, while the oldest sample (E), stored for 8 years, produced the lowest extraction yield at 5.10 ± 0.84%. Cannabinoid concentrations also showed marked differences across storage periods. THC content was highest in the fresh sample at 35.16 ± 3.87% and lowest in the 08-year stored sample at 0.44 ± 0.2%. CBD content peaked in the 2-year stored sample at 6.71 ± 0.77% and reached its minimum in the 08-year stored sample at 1.76 ± 0.31%. Similarly, CBN content was maximum in the 2-year stored sample at 6.94 ± 0.83% and minimum in the 08-year stored sample at 2.94 ± 0.51%. The two-way analysis of variance revealed that storage time significantly influenced several key parameters: extraction yield, THC, CBD, and CBN contents, all expressed as percentages. In contrast, neither extraction time nor the interaction between storage time and extraction time demonstrated statistically significant effects on these parameters. The Tukey’s test results presented in Fig. 1a provide a detailed analysis of the extraction yield across the different storage durations. The statistical analysis revealed that the extraction yield for each storage period was significantly different from the yield of extraction obtained for the other storage durations. This finding suggests that the rate of resin extraction decreases in a stepwise manner as the storage time increases. As shown in Fig. 1b, 2 years storage time caused the proportion of THC to decrease dramatically. Furthermore, The Tukey test results indicate significant statistical differences between the THC of freshly processed samples and those stored for 8 years. In contrast, the yields of extraction of samples stored for the other three durations showed no statistically significant differences from each other. Figure 1C shows that CBD levels increase after 2 years of storage, reaching a mean value of 6.71 ± 0.77%, and then decrease for the remainder of the storage time. Moreover, Tukey’s test reveals statistically significant differences among the extraction yield obtained from samples stored for 2, 4, and 8 years. In contrast, no statistically significant difference was observed between the CBD content of freshly processed samples and those stored for 6 years. Figure 1d illustrates the temporal evolution of CBN levels across different storage durations. The concentration of CBN is lowest in freshly processed samples, then increases progressively, reaching its peak in samples stored for 2 years (6.94 ± 0.83%). Following this maximum, CBN levels exhibit a declining trend for subsequent storage periods of 4, 6, and 8 years. Tukey’s post-hoc test revealed that the cannabinol (CBN) concentrations observed in samples stored for 2 and 4 years were not statistically different from each other. However, these concentrations were significantly different from those observed in samples from other storage periods.
Fig. 1.
The variation of C. sativa resin extraction yield and cannabinoids according to storage time and the time. (a): extraction yield, (b): amount of THC, (c): amount of CBD, (d): amount of CBN. Different letters indicate a statistically different means (P > 0.05) based on the Tukey test.
This study implied that storage time significantly impacted the extraction yield and cannabinoids (P < 0.05). However, the extraction time had no significant effect on yield of extraction and cannabinoids. Our results confirm those reported by McPartland and Russoindicating that the oxidative degradation of THC to cannabinol CBN was more marked with prolonged storage26. In addition, previous investigations showed that storage time significantly impacted THC (degradation) and CBN (formation) content27,28. In this way, Sannikovaet al. noted that the amount of THC decreased with increased CBN levels during storage27. However Trofin et al. reported that THC and CBD content decreased (degradation), while CBN content increased (formation) over the storage period28. Regarding the time effect, it was found that 1-year storage significantly affected cannabinoids contents29–31. While our results indicate no significant effect of extraction time, previous studies reported that the extraction time significantly affected the cannabinoid amount32,33. This finding could encourage the future pharmaceutical exploitation of C. sativa resin because of its high CBD and CBN content and low THC content by storing Cannabis resin.
The cannabinoids profile of C. sativa resin according to storage time
Figure 2 presents the GC/FID chromatographic profiles of representative samples from groups A, B, C, D, and E, illustrating the dynamic changes in cannabinoid composition over extended storage periods. The chromatograms display peaks corresponding to the internal standard (IS) n-Nonadecane, followed by CBD, THC, and CBN, with each peak area reflecting the relative concentration of these cannabinoids. As shown in Fig. 2a, the fresh sample (Group A) exhibits a high THC content (37.06%), with significantly lower levels of CBD (4.92%) and CBN (0.85%). This initial profile is consistent with the typical cannabinoid composition of freshly harvested Cannabis resin, where THC predominates due to its biosynthesis as THCA, which decarboxylates to THC during drying and storage34,35.
Fig. 2.
Representative GC-FID chromatograms showing the effect of storage time on the cannabinoid profile of C. sativa resin. Panels correspond to storage duration: (a) Sample A (not stored), (b) Sample B (2 years), (c) Sample D (6 years), (d) Sample E (8 years). IS: internal standard. Identified peaks: IS (Internal Standard, n-Nonadecane), CBD (Cannabidiol), THC (Δ9-Tetrahydrocannabinol), CBN (Cannabinol).
After 2 years of storage (Group B), a marked transformation in the cannabinoid profile is observed, characterized by a substantial decrease in THC content (2.67%) and corresponding increases in CBD (6.84%) and CBN (7.12%) levels (Fig. 2b). This trend aligns with the well-documented degradation pathways of THC, which undergoes oxidation to form CBN, a process accelerated by prolonged storage and exposure to environmental factors such as light and temperature20,21,28.
The trend of THC degradation and CBN formation persists in samples stored for 4 years (Group C), which exhibit further reductions in THC (2.24%) and increases in CBN (6.51%) and CBD (5.58%) (Fig. 2c). This continued decline in THC content, accompanied by the accumulation of CBN, underscores the time-dependent nature of cannabinoid degradation. Similar patterns are observed in samples stored for 6 and 8 years (Groups D and E, respectively), where THC levels drop to negligible amounts, and CBD and CBN concentrations also decline, albeit at a slower rate (Fig. 2d,e). This overall reduction in cannabinoid content over extended storage periods highlights the progressive degradation of cannabinoids, likely due to oxidative and photochemical processes, as well as the potential evaporation of volatile compounds28,35. The degradation of THC to CBN follows first-order kinetics, as demonstrated by Zamengo et al.21, who reported that nearly 100% of THC degrades within 4 years under suboptimal storage conditions. This rapid degradation is consistent with our findings, where THC levels drop significantly after 2 years of storage. The study by Ross and ElSohly36 further supports this observation, showing a 16.6% loss of THC after 1 year and a 41.4% loss after 4 years at room temperature. These findings emphasize the importance of storage conditions in preserving cannabinoid stability, as lower temperatures and darkness have been shown to mitigate THC degradation28,35.
The formation of CBN as a degradation product of THC is a key indicator of Cannabis aging, as highlighted by multiple studies21,23. CBN, while less psychoactive than THC, possesses distinct pharmacological properties, including sedative and anti-inflammatory effects, which may influence the therapeutic profile of aged Cannabis products37. The increase in CBN levels in ou study, particularly after 2 years of storage, aligns with these findings and underscores the role of oxidative processes in cannabinoid degradation.
The observed discrepancy between THC degradation and CBN/CBD accumulation can be attributed to multiple factors: (1) THC degrades into unmeasured products such as Δ⁸-THC, hydroxylated derivatives, and polymeric compounds38; (2) volatilization of terpenes and adsorption of cannabinoids to packaging materials may reduce recoverable mass22,35; (3) analytical limitations (GC/FID) precluded quantification of acidic precursors (THCA, CBDA) and non-volatile degradation byproducts31,34.
The inspection of the chromatograms in Fig. 2 clearly demonstrates increasing chemical complexity over time. Besides the target analytes and internal standard, numerous unidentified peaks emerge and grow, especially in samples stored for extended durations (Fig. 2b–e). These likely correspond to various cannabinoid isomers, oxidation products, and other degradation compounds. Although their structural elucidation was outside the defined scope of this GC/FID-based quantitative study, further characterization using methods like GC–MS would be valuable for a more complete mapping of the degradation processes in stored C. sativa resin.
Principal component analysis
Principal component analysis (PCA) was performed on the data in supplementary table to identify similar individuals (A, B, C, D, and E) and correlations between variables (storage time, cannabinoids, and yield of extraction). This powerful chemometric tool allowed for the explicit representation of individuals and variables, showing the various profiles among samples based on storage times, solvents, and times of extractions. The individuals (A, B, C, D, and E) were represented by 150 samples of C. sativa resin. The variables were represented by the storage time, the average yield of extraction, and the cannabinoid percentage in the C. sativa resin. The Kaiser–Meyer–Olkin (KMO) index and the Bartlett test were calculated to test the feasibility of the PCA. The KMO index showed a value of 0.6, which is greater than 0.5, and the Bartlett test indicated that the test is significant (P < 0.05), as a result, the PCA is feasible.
The quality of representation of the variables indicated that most of them were well represented by the two first components (Table 2). The first component explained 64.54% of the variables, while the second explained (34.03%.). As a result, these two first components accounted for 98.57% of the total information in this dataset. The storage time, yield of extraction, and THC variables were well represented in the first component (PC1), while CBD and CBN were well represented explained by the second component (PC2).
Table 2.
The variable representation quality on the two first components, eigenvalues, percentages, and cumulated percentages of explained variability of the two components.
| Components | Storage time | Yield | CBD | THC | CBN | Eigenvalues | Percentages of explained variability (%) |
Cumulated percentages (%) |
|---|---|---|---|---|---|---|---|---|
| PC1a | − 0.91 | 0.98 | 0.06 | 0.99 | − 0.66 | 3.22 | 64.54 | 64.54 |
| PC2 | − 0.38 | 0.19 | 0.99 | − 0.09 | 0.73 | 1.70 | 34.03 | 98.57 |
aPrincipal component.
The loading plot (Fig. 3) and correlation heat map (Fig. 4) reveal several noteworthy relationships among the variables. All observed correlations are statistically significant, with the exception of the relationship between THC and CBD. A strong positive correlation exists between THC and yield of extraction (r = 0.96), as well as between CBN and CBD (r = 0.67). Although a direct biochemical link between CBD and CBN formation is not implied, the positive correlation observed in our longitudinal dataset is likely attributable to the fact that both ‘Total CBD’ and CBN levels showed a concurrent increasing trend over the extended storage durations investigated in this study.
Fig. 3.
Loading graph illustrated the projection of variables on the first and second principal components.
Fig. 4.
Heat map of the correlations between yield of extraction, cannabinoids amount and storage time.
Conversely, strong negative correlations are evident between storage time and both yield of extraction (r = − 0.96) and THC content (r = − 0.87). These findings confirm that shorter storage periods are associated with higher yield of extraction and THC content. Additionally, a negative correlation between THC and CBN (r = − 0.72) indicates an inverse relationship between these cannabinoids. This correlation likely results from the gradual transformation of THC into CBN during storage, explaining why increased the extraction yield and THC content correspond with decreased CBN levels. The degradation of THC to CBN was previously reported. In this way, Repka et al. investigated THC stability within polymeric matrices at various temperatures. Their findings revealed that THC degradation led to CBN formation, with the extent of conversion varying based on temperature. At 120 °C and 160 °C, CBN accounted for 9.0% and 7.8% of the total degraded THC, respectively. Notably, when the temperature was raised to 200 °C, the conversion rate of THC to CBN increased significantly, reaching 29.1%39. In the same way, Turner and ElSohly suggested a potential degradation pathway for THC’s conversion to CBN. Their proposed mechanism involves the formation of intermediate compounds, specifically epoxy and hydroxylated derivatives, during the transformation process40.
The principal component analysis (PCA) biplot in Fig. 5 reveals the distribution of individuals and variables across the first two principal components. This biplot shows five distinct clusters, each corresponding to one of the five storage periods. This clustering indicates that the individuals within each group share similar characteristics. The first principal component captures the majority of the variance in the non-stored samples, which are represented by Cluster A. This suggests that the first component is more strongly associated with yield of extraction and THC parameters. In contrast, the remaining four clusters, representing the stored samples, are better aligned with the second principal component. This implies that the second component is more closely related to the CBD and CBN variables.
Fig. 5.
Biplot represented the projection of variables and individuals on the first and second principal components; (A): Non-stored, (B): 2 years, (C): 4 years, (D): 6 years, (E): 8 years.
The PCA biplot in Fig. 5 further reveals that the group of non-stored samples (Cluster A) is the richest in terms of yield of extraction and THC content. In contrast, the groups on the left side of the biplot represent the stored samples (Clusters B, C, D, and E). Clusters B and C are distinguished by their relatively higher levels of CBN and CBD. Clusters D and E, on the other hand, exhibit a different profile—they have low THC amounts, medium levels of CBN and CBD. This analysis of the biplot helps elucidate the relationships between the storage time, cannabinoid profiles, and yield for the Cannabis samples under study. Several research found that the strong negative correlation between THC and CBN content was significantly reduced over time21,41. However, other authors found no direct correlation between CBN formation and THC degradation22,42.
Prediction of cannabinoids in terms of storage time
Prediction by polynomial regression
To predict cannabinoid levels as a function of storage time, we employed three polynomial regression models (linear, quadratic, and cubic) and evaluated their performance using K-fold cross-validation (k = 5). The dataset of 150 samples (Supplementary table) was divided into subsets for model fitting (120 samples) and evaluation (30 samples). Each model was fitted to the estimation subset, and its predictive accuracy was assessed on the evaluation subset. The polynomial regression method predicts variable y (dependent variable) based on predictive x variables (independent variable). The storage year was the predictor x, while THC, CBN, and CBD content were the dependent variables. The purpose of this method was to establish models that would allow us to predict the storage time from the percentages of THC, CBN, and CBD.
Model recapitulation
This section aims to create a mathematical model using linear, quadratic and cubic regression to predict the percentages of THC, CBN, and CBD based on the storage time.
Figure 6 illustrates the changes in THC, CBD, and CBN levels over storage time, comparing linear (red), quadratic (green), and cubic (blue) model fits. The analysis of variance (Table 3) revealed significant regression effects for linear, quadratic, and cubic models across all three dependent variables: THC, CBD, and CBN (p < 0.0001). While the linear and quadratic models exhibited significant lack of fit (p < 0.05), the cubic models for all three variables demonstrated negligible lack of fit (p > 0.05). These results suggest that the cubic model offers a better fit to the data compared to both the linear and quadratic models. An advanced comparative analysis of linear, quadratic, and cubic regression models for predicting cannabinoid degradation during storage revealed the cubic model as the most suitable across all three investigated cannabinoids (Table 4). This model consistently yielded the lowest AICc, BIC, SSE, MSE, and RMSE values while demonstrating the highest R2 values.
Fig. 6.
The evolution of THC, CBD and CBN levels over the storage period, comparing linear (red), quadratic (green) and cubic (blue) model fits.
Table 3.
Analysis of variance for the three regression models.
| Source | DF | Sum of Squares | Mean Square | F Ratio | Prob > F | ||
|---|---|---|---|---|---|---|---|
| THC | Linear fit | Model | 1 | 20.422.56 | 20,422.60 | 432.68 | < 0.0001* |
| Error | 148 | 6985.64 | 47.20 | ||||
| Lack Of Fit | 3 | 6545.73 | 2181.91 | 719.18 | < 0.0001 | ||
| Pure Error | 145 | 439.91 | 3.03 | ||||
| C. Total | 149 | 27,408.21 | |||||
| Polynomilal fit degree = 2 | Model | 2 | 26,385.37 | 13,192.70 | 1896.03 | < 0.0001* | |
| Error | 147 | 1022.83 | 7.00 | ||||
| Lack Of Fit | 2 | 582.92 | 291.46 | 96.07 | < 0.0001 | ||
| Pure Error | 145 | 439.91 | 3.03 | ||||
| C. Total | 149 | 27,408.21 | |||||
| Polynomilal fit degree = 3 | Model | 3 | 26,956.53 | 8985.51 | 2904.50 | < 0.0001* | |
| Error | 146 | 451.67 | 3.09 | ||||
| Lack Of Fit | 1 | 11.76 | 11.76 | 3.88 | 0.05 | ||
| Pure Error | 145 | 439.91 | 3.03 | ||||
| C. Total | 149 | 27,408.21 | |||||
| CBD | Linear fit | Model | 1 | 80.57 | 80.57 | 30.60 | < 0.0001* |
| Error | 148 | 389.75 | 2.63 | ||||
| Lack Of Fit | 3 | 356.67 | 118.89 | 521.01 | < 0.0001 | ||
| Pure Error | 145 | 33.09 | 0.23 | ||||
| C. Total | 149 | 470.33 | |||||
| Polynomilal fit degree = 2 | Model | 2 | 425.43 | 212.72 | 696.52 | < 0.0001* | |
| Error | 147 | 44.89 | 0.31 | ||||
| Lack Of Fit | 2 | 11.81 | 5.90 | 25.87 | < 0.0001 | ||
| Pure Error | 145 | 33.09 | 0.23 | ||||
| C. Total | 149 | 470.33 | |||||
| Polynomilal fit degree = 3 | Model | 3 | 436.87 | 145.63 | 635.59 | < 0.0001* | |
| Error | 146 | 33.45 | 0.23 | ||||
| Lack Of Fit | 1 | 0.36 | 0.36 | 1.60 | 0.21 | ||
| Pure Error | 145 | 33.09 | 0.23 | ||||
| C. Total | 149 | 470.33 | |||||
| CBN | Linear fit | Model | 1 | 96.54 | 96.54 | 19.50 | < 0.0001* |
| Error | 148 | 732.61 | 4.95 | ||||
| Lack Of Fit | 3 | 668.22 | 222.74 | 501.56 | < 0.0001 | ||
| Pure Error | 145 | 64.39 | 0.44 | ||||
| C. Total | 149 | 829.15 | |||||
| Polynomilal fit degree = 2 | Model | 2 | 756.23 | 378.12 | 762.26 | < 0.0001* | |
| Error | 147 | 72.92 | 0.50 | ||||
| Lack Of Fit | 2 | 8.53 | 4.26 | 9.60 | 0.0001 | ||
| Pure Error | 145 | 64.39 | 0.44 | ||||
| C. Total | 149 | 829.15 | |||||
| Polynomilal fit degree = 3 | Model | 3 | 763.14 | 254.38 | 562.61 | < 0.0001* | |
| Error | 146 | 66.01 | 0.45 | ||||
| Lack Of Fit | 1 | 1.62 | 1.62 | 3.65 | 0.06 | ||
| Pure Error | 145 | 64.39 | 0.44 | ||||
| C. Total | 149 | 829.15 |
Statistically significant at P < 0.05.
Table 4.
The parameters of the quality prediction of THC, CBD and CBN for the linear, quadratic and cubic models for fitting and validation datasets.
| Regression fitting | Data subset | AICc | BIC | SSE | MSE | RMSE | R2 | |
|---|---|---|---|---|---|---|---|---|
| THC | Linear fit | Fitting | 1007.99 | 1016.86 | 6985.64 | 47.20 | 6.87 | 0.75 |
| Validation | 201.42 | 204.70 | 1148.89 | 41.03 | 6.41 | 0.76 | ||
| Polynomilal fit degree = 2 | Fitting | 721.91 | 733.68 | 1022.83 | 6.96 | 2.64 | 0.96 | |
| Validation | 133.10 | 137.11 | 107.77 | 3.99 | 2.00 | 0.98 | ||
| Polynomilal fit degree = 3 | Fitting | 601.45 | 616.08 | 451.67 | 3.09 | 1.76 | 0.98 | |
| Validation | 47.097 | 51.603 | 5.565 | 0.214 | 0.463 | 0.999 | ||
| CBD | Linear fit | Fitting | 575.08 | 583.95 | 389.75 | 2.63 | 1.62 | 0.17 |
| Validation | 117.27 | 120.55 | 69.51 | 2.48 | 1.58 | 0.17 | ||
| Polynomilal fit degree = 2 | Fitting | 253.01 | 264.77 | 44.89 | 0.31 | 0.55 | 0.90 | |
| Validation | 34.53 | 38.53 | 4.03 | 0.15 | 0.39 | 0.95 | ||
| Polynomilal fit degree = 3 | Fitting | 211.02 | 225.65 | 33.45 | 0.23 | 0.48 | 0.93 | |
| Validation | 17.30 | 21.81 | 2.06 | 0.08 | 0.28 | 0.98 | ||
| CBN | Linear fit | Fitting | 669.74 | 678.61 | 732.61 | 4.95 | 2.22 | 0.12 |
| Validation | 132.36 | 135.64 | 114.95 | 4.11 | 2.03 | 0.11 | ||
| Polynomilal fit degree = 2 | Fitting | 325.76 | 337.53 | 72.92 | 0.50 | 0.70 | 0.91 | |
| Validation | 19.09 | 23.10 | 2.41 | 0.09 | 0.30 | 0.98 | ||
| Polynomilal fit degree = 3 | Fitting | 312.98 | 327.62 | 66.01 | 0.45 | 0.67 | 0.92 | |
| Validation | 13.10 | 17.60 | 1.79 | 0.07 | 0.26 | 0.99 |
BIC Bayesian information criterion, AIC corrected Akaike information criterion, SSE sum of squared errors, MSE mean squared error, RMSE Root mean squared error, R2 coefficient of determination.
The linear, quadratic, and cubic polynomial regression models were evaluated to predict THC degradation over storage time. The linear model exhibited limited accuracy, with an R2 of 0.75–0.76 and high RMSE (6.41–6.87 for fitting and validation subsets, respectively), reflecting poor alignment with the non-linear degradation kinetics. The quadratic model significantly outperformed the linear model, achieving an R2 of 0.96–0.98 and lower RMSE (2.00–2.64) for fitting and validation subsets, respectively, though slight overfitting was observed. The cubic model demonstrated near-perfect predictive performance, with validation R2 = 0.999 and RMSE = 0.463, alongside the lowest AICc (601.45) and BIC (616.08), indicating optimal balance between model complexity and accuracy. These results underscore the cubic model’s superiority in capturing the non-linear degradation dynamics of THC, making it the most reliable tool for estimating storage duration in forensic and pharmaceutical applications and highlighting the complexity of the degradation process beyond simpler linear or quadratic relationships. Potential overfitting was specifically addressed using the AICc and BIC model selection criteria, which penalize model complexity. By selecting the cubic model based on its minimum AICc and BIC values, we ensured that its superior fit sufficiently justified its increased number of parameters compared to simpler models, thus favoring parsimony and mitigating the risk of overfitting. The use of a cubic polynomial reflects its empirical effectiveness in describing the complex degradation kinetics observed, which likely involve multi-phasic processes or changing matrix effects over the extended storage period not adequately captured by simpler linear or quadratic models. These finding are confirmed by the Fig. 7 showing the curve of actual values as a function of predicted ones for the prediction of THC (A), CBD (B) and CBN (C) obtained by the cubic model for fitting and validation datasets.
Fig. 7.
The curve of actual values as a function of predicted ones for the prediction of THC (A), CBD (B) and CBN (C) obtained by the cubic model for fitting and validation datasets.
In the same way, the linear, quadratic, and cubic polynomial regression models were evaluated to predict CBD variation over storage time. The linear model performed poorly, with low explanatory power (R2 = 0.17 for both fitting and validation subsets) and high error (RMSE = 1.62–1.58), indicating a weak linear relationship. The quadratic model significantly improved accuracy, achieving a fitting R2 = 0.90 (RMSE = 0.55) and validation R2 = 0.95 (RMSE = 0.39), capturing non-linear trends effectively. The cubic model further enhanced performance, with fitting R2 = 0.93 (RMSE = 0.48) and near-perfect validation metrics (R2 = 0.98, RMSE = 0.28), alongside the lowest AICc (211.02) and BIC (225.65). These results highlight the cubic model’s superior ability to describe CBD dynamics, balancing complexity and predictive accuracy, and offering critical insights for applications requiring precise monitoring of cannabinoid stability, such as pharmaceutical quality control or forensic investigations.
Concerning CBN, the linear model exhibited poor performance, with low explanatory power (R2 = 0.12 and 0.11 for fitting and validation subsets, respectively) and high error (RMSE = 2.222 and 2.03), indicating a negligible linear relationship. The quadratic model significantly improved accuracy, achieving a fitting R2 = 0.91 (RMSE = 0.70) and validation R2 = 0.98 (RMSE = 0.30), demonstrating strong non-linear alignment. The cubic model further optimized performance, with fitting R2 = 0.92 (RMSE = 0.67) and near-perfect validation metrics (R2 = 0.99, RMSE = 0.26), alongside the lowest AICc (312.98) and BIC (327.62). These results confirm the cubic model’s superiority in capturing the complex dynamics of CBN formation, offering exceptional precision for applications such as forensic aging estimates or pharmaceutical stability assessments, where accurate prediction of oxidative degradation is critical.
Coefficients estimation of the cubic model
Table 5 summarizes the estimated coefficients, t-student values, and observed probability (p-values) for each coefficient of the cubic models.
Table 5.
Estimated regression coefficients for the for the linear, quadratic and cubic models.
| Term | Coefficient | Estimate | t Ratio | Prob >|t| | |
|---|---|---|---|---|---|
| THC | Intercept | a | 3.30 | 3.460 | 0.001* |
| Storage time | b | − 0.10 | − 1.830 | 0.0070* | |
| (Storage time)2 | c | 0.055 | 21.080 | < 0.0001* | |
| (Storage time)3 | d | − 0.005 | − 13.590 | < 0.0001* | |
| CBD | Intercept | a | 10.35 | 39.84 | < 0.0001* |
| Storage time | b | − 0.28 | − 17.58 | < 0.0001* | |
| (Storage time)2 | c | − 0.0157 | − 22.15 | < 0.0001* | |
| (Storage time)3 | d | 0.0007 | 7.07 | < 0.0001* | |
| CBN | Intercept | a | 8.72 | 23.9 | < 0.0001* |
| Storage time | b | − 0.13 | − 5.67 | < 0.0001* | |
| (Storage time)2 | c | − 0.024 | − 23.99 | < 0.0001* | |
| (Storage time)3 | d | 0.0005 | 3.91 | 0.0001* |
*Statistically significant at P < 0.05.
The results showed that all coefficients a, b, c and d had a significant effect (P < 0.05) on the three dependent variables THC, CBN, and CBD. The 3 degree regression models relating the storage time to the percentages of THC, CBN, and CBD is represented by the following equations:
![]() |
2 |
![]() |
3 |
![]() |
4 |
Previous research by Zamengo et al.21 explored the use of predictive models to estimate the storage time (t) of Cannabis samples. In this study, the authors proposed a method to approximate the age of Cannabis material stored under different conditions, using experimental measurements of THC and CBN concentrations. The study introduces a simpler empirical equation to model the kinetic behavior of the CBN/THC ratio over time. In contrast to this previous work, our study focused on predicting cannabinoid content as a function of storage time. These findings represent a promising advancement in estimating cannabinoid levels based on storage duration. However, it is important to emphasize that the developed models are specifically tailored to the storage conditions outlined in our study.
Prediction of storage time in terms of cannabinoids
Prediction by multiple linear regression
In a complementary analysis, we aimed to model storage time as a function of cannabinoid content (THC, CBD, and CBN) using the same dataset. This inverse approach seeks to predict the storage duration of Cannabis resin based on its chemical composition, which has significant implications for forensic and quality control applications. We employed multiple linear regression (MLR) with three models: a main-effects model (M1), a model including interaction terms (M2), and a model incorporating both interaction and quadratic terms (M3). The dataset of 150 samples (Supplementary table) was divided into subsets for model fitting (120 samples) and evaluation (30 samples) using K-fold cross-validation (k = 5). Each model was fitted to the estimation subset, and its predictive performance was assessed on the evaluation subset.
Models recapitulation
This section focuses on developing mathematical models to predict storage time based on cannabinoid ratios. We evaluated three models: one with main terms only (M1), another with main and interaction terms (M2), and a third with main, interaction, and quadratic terms (M3). All three models showed significant regression effects (p < 0.0001). The models details in Table 6 show that the model M1 demonstrated strong performance with fitting R2 = 0.96 (RMSE = 1.97) and validation R2 = 0.99 (RMSE = 1.16), indicating a robust linear relationship. M2 improved accuracy further, achieving fitting R2 = 0.98 (RMSE = 1.30) and validation R2 = 0.99 (RMSE = 0.98), highlighting the importance of interaction effects. The most complex model, M3, exhibited near-perfect predictive capability, with fitting R2 = 0.99 (RMSE = 0.90) and validation R2 = 0.99 (RMSE = 0.312), alongside the lowest AICc (39.84) and BIC (40.58). These results underscore M3’s superiority in capturing the intricate, non-linear relationships between cannabinoid content and storage time, offering unparalleled precision for forensic and pharmaceutical applications requiring reliable age estimation of Cannabis resin. These finding are supported by the curves of actual values as a function of predicted ones for the three models M1 (A), M2 (B) and M3 (C) for fitting and validation datasets (Fig. 8) showing a more perfect strait line for the M3 model for both fitting and validation datasets.
Table 6.
The parameters of the quality prediction of storage time in terms of cannabinoids for M1, M2 and M3 models for fitting and validation datasets.
| Regression fitting | Data subset | AICc | BIC | SSE | MSE | RMSE | R-Square |
|---|---|---|---|---|---|---|---|
| M1: Model with main terms | Fitting | 635.16 | 649.80 | 565.52 | 3.87 | 1.97 | 0.96 |
| Validation | 102.41 | 106.91 | 40.58 | 1.35 | 1.16 | 0.99 | |
| M2: Model with main and interaction terms | Fitting | 515.48 | 538.55 | 248.63 | 1.69 | 1.3 | 0.97 |
| Validation | 98.52 | 102.87 | 28.53 | 0.95 | 0.98 | 0.98 | |
| M3: Model with main, interaction and quadratic terms | Fitting | 409.87 | 441.07 | 115.1 | 0.81 | 0.9 | 0.99 |
| Validation | 39.837 | 40.584 | 2.928 | 0.098 | 0.312 | 0.99 |
BIC Bayesian information criterion, AIC corrected Akaike information criterion, SSE sum of squared errors, MSE mean squared error, RMSE root mean squared error, R2 coefficient of determination.
Fig. 8.
The curve of actual values as a function of predicted ones for the three models M1 (A), M2 (B) and M3 (C) for fitting and validation datasets.
Coefficients estimation of the M3 model
Table 7 presents a comprehensive summary of the M3 models, detailing the estimated coefficients, t-student values, and observed probabilities (p-values) for each coefficient.
Table 7.
Estimated regression coefficients for the for the linear, quadratic and cubic models.
| Term | Coefficient | Estimate | Std Error | t Ratio | Prob >|t| |
|---|---|---|---|---|---|
| Intercept | β0 | 26.85 | 1.50 | 17.86 | < 0.0001* |
| %THC | β1 | − 1.00 | 0.16 | − 6.3 | < 0.0001* |
| %CBD | β2 | − 4.69 | 0.45 | − 10.35 | < 0.0001* |
| %CBN | β3 | 2.73 | 0.24 | 11.36 | < 0.0001* |
| %THC*%CBD | β12 | 0.51 | 0.09 | 5.83 | < 0.0001* |
| %THC*%CBN | β13 | − 0.45 | 0.04 | − 11.4 | < 0.0001* |
| %CBD*%CBN | β23 | 4.86 | 0.42 | 11.49 | < 0.0001* |
| %THC*%THC | β12 | 0.00 | 0.01 | 0.37 | 0.7104 |
| %CBD*%CBD | β22 | − 2.97 | 0.29 | − 10.13 | < 0.0001* |
| %CBN*%CBN | β32 | − 1.72 | 0.14 | − 12.08 | < 0.0001* |
*Statistically significant at P < 0.05.
The analysis reveals that all coefficients demonstrate a statistically significant effect (P < 0.05), with the notable exception of the quadratic coefficient β12. This exception suggests that while most factors in the model contribute significantly to predicting storage time, the quadratic interaction between certain variables may not be as influential. The M3 models, which relate storage time to the percentages of THC, CBN, and CBD, can be expressed through the following equations:
![]() |
5 |
The prediction of storage time for cannabinoids such as THC, CBD, and CBN is crucial for both forensic analysis and product quality control. Previous studies on predicting Cannabis resin storage time using cannabinoid levels have shown varying degrees of success. In this way, A mono-exponential function can model THC degradation and CBN formation, allowing predictions of storage time based on current cannabinoid concentrations21. In another study, The CBN to THC concentration ratio serves as an effective indicator of the age of stored marijuana, with significant THC loss correlating with increased storage time36. The current study’s multiple linear regression model, incorporating multiple cannabinoid levels, has achieved a remarkably high R2 of 0.99 for both fitting and validation datasets, suggesting a significant improvement in predictive power. This advance could have major implications for forensic applications and quality control in the Cannabis industry. However, it’s important to consider factors such as sample range, storage conditions43, and the potential for overfitting when interpreting these results. Further validation with diverse sample sets and under various storage conditions would be valuable to ensure the model’s robustness and generalizability, addressing challenges in accurately dating Cannabis samples.
Conclusion
This study provides valuable insights into the long-term stability and degradation patterns of cannabinoids in C. sativa L. resin. Our findings demonstrate that prolonged storage significantly impacts cannabinoid profiles, characterized by a notable decrease in THC content and a concurrent increase in CBD and CBN concentrations. This shift in cannabinoid composition is particularly relevant for both medicinal and forensic applications. The increase in CBD and CBN over time suggests that storage could be strategically employed to enhance the potential therapeutic value of C. sativa L. by naturally promoting a higher concentration of these non-psychoactive cannabinoids, known for their potential anti-inflammatory and analgesic properties. From a forensic perspective, understanding these time-dependent changes in cannabinoid profiles is crucial for accurate interpretation of evidence. The developed predictive models, offer a valuable tool for estimating the relationship between storage duration and cannabinoid content, potentially aiding forensic investigations. Further research investigating the specific mechanisms underlying these cannabinoid transformations, as well as the influence of different storage conditions, will be crucial to fully leverage these findings for tailored applications in medicine and law enforcement.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Author contributions
N.F.: Conceptualization, Methodology, Validation, Writing-Original draft preparation; M.F.: Methodology, Investigation, Software, Writing-Original draft preparation; H.S.: Supervision, Methodology, Investigation; A.E.B.: Methodology, Investigation, Validation; T.B.: Methodology, Investigation, Validation; S.A.: Methodology, Investigation, A.B.: Methodology, Investigation, Validation; A.F.: Conceptualization, Methodology, Validation Supervision, review & editing.
Data availability
Data Availability: The data used for the multivariate and predictive analyses, are available in the Electronic Supplementary Material.
Declarations
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.McPartland, J. Cannabis: The plant, its evolution, and its genetics—with an emphasis on Italy. Rend. Lincei31, 939–948 (2020). [Google Scholar]
- 2.Chandra, S., Lata, H. & ElSohly, M. A. Cannabis sativa L. botany and biotechnology. 10.1007/978-3-319-54564-6.
- 3.Stambouli, H., El Bouri, A., Bellimam, M. A., Bouayoun, T. & El Karni, N. Cultivation of Cannabissativa L. in northern Morocco. Bull. Narc.57, 79–118 (2005). [PubMed] [Google Scholar]
- 4.Stambouli, H., El Bouri, A. & Bouayoun, T. Évolution de la teneur en δ9-THC dans les saisies de résines de cannabis au Maroc de 2005 à 2014. Toxicol. Anal. Clin.28, 146–152 (2016). [Google Scholar]
- 5.Fadil, M. et al. Enhancement of ultrasonic-assisted extraction and antioxidant potential of phenolic compounds from Moroccan Cannabis sativa: A green intelligent approach integrating optimal mixture process design and artificial neural networks. Sustain. Chem. Pharm.37, 101359 (2024). [Google Scholar]
- 6.Lange, B. M. & Zager, J. J. Comprehensive inventory of cannabinoids in Cannabissativa L.: Can we connect genotype and chemotype?. Phytochem. Rev.21(4), 1273–1313 (2021). [Google Scholar]
- 7.Bernstein, N., Gorelick, J. & Koch, S. Interplay between chemistry and morphology in medical cannabis (Cannabissativa L.). Ind. Crops Prod.129, 185–194 (2019). [Google Scholar]
- 8.Babson, K. A., Sottile, J. & Morabito, D. Cannabis, cannabinoids, and sleep: A review of the literature. Curr. Psychiatry Rep.10.1007/s11920-017-0775-9 (2017). [DOI] [PubMed] [Google Scholar]
- 9.Maurya, N. & Velmurugan, B. K. Therapeutic applications of cannabinoids. Chem. Biol. Interact.293, 77–88 (2018). [DOI] [PubMed] [Google Scholar]
- 10.Pellati, F. et al. Cannabissativa L. and nonpsychoactive cannabinoids: Their chemistry and role against oxidative stress, inflammation, and cancer. BioMed Res. Int.2018, 1691428 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Pisanti, S. et al. Cannabidiol: State of the art and new challenges for therapeutic applications. Pharmacol. Ther.175, 133–150 (2017). [DOI] [PubMed] [Google Scholar]
- 12.Grotenhermen, F. & Müller-Vahl, K. Das therapeutische Potenzial von Cannabis und Cannabinoiden. Dtsch. Arztebl. Int.109, 495–501 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Russo, E. B. Cannabinoids in the management of difficult to treat pain. Ther. Clin. Risk Manag.4, 245–259 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Citti, C., Braghiroli, D., Vandelli, M. A. & Cannazza, G. Pharmaceutical and biomedical analysis of cannabinoids: A critical review. J. Pharm. Biomed. Anal.147, 565–579 (2018). [DOI] [PubMed] [Google Scholar]
- 15.ElSohly, M. & Gul, W. Constituents of Cannabissativa. in Handbook of Cannabis, vol. 3(1093), 187–188 (2014).
- 16.Happyana, N. & Kayser, O. Monitoring metabolite profiles of Cannabissativa L. trichomes during flowering period using 1H NMR-based metabolomics and real-time PCR. Planta Med.82, 1217–1223 (2016). [DOI] [PubMed] [Google Scholar]
- 17.Saloner, A. & Bernstein, N. Effect of potassium (K) supply on cannabinoids, terpenoids and plant function in medical cannabis. Agronomy12, 1242 (2022). [Google Scholar]
- 18.Danziger, N. & Bernstein, N. Shape matters: Plant architecture affects chemical uniformity in large-size medical cannabis plants. Plants10, 1834 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Elkins, A. C., Deseo, M. A., Rochfort, S., Ezernieks, V. & Spangenberg, G. Development of a validated method for the qualitative and quantitative analysis of cannabinoids in plant biomass and medicinal cannabis resin extracts obtained by super-critical fluid extraction. J. Chromatogr. B1109, 76–83 (2019). [DOI] [PubMed] [Google Scholar]
- 20.Trofin, I. G., Vlad, C. C., Noja, V. V. & Dabija, G. Identification and characterization of special types of herbal cannabis. UPB Sci. Bull. Ser. B Chem. Mater. Sci.74, 119–130 (2012). [Google Scholar]
- 21.Zamengo, L. et al. The role of time and storage conditions on the composition of hashish and marijuana samples: A 4-year study. Forens. Sci. Int.298, 131–137 (2019). [DOI] [PubMed] [Google Scholar]
- 22.Lindholst, C. Long term stability of cannabis resin and cannabis extracts. Aust. J. Forens. Sci.42, 181–190 (2010). [Google Scholar]
- 23.Peschel, W. Quality control of traditional cannabis tinctures: Pattern, markers, and stability. Sci. Pharm.84, 567–584 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Olejar, K. J. et al. Ultrasonic-Assisted extraction of cannabidiolic acid from cannabis biomass. J. Vis. Exp.10.3791/63076 (2022). [DOI] [PubMed] [Google Scholar]
- 25.Pham, H. A new criterion for model selection. Mathematics7, 1215 (2019). [Google Scholar]
- 26.McPartland, J. M. & Russo, E. B. Cannabis and cannabis extracts. J. Cannabis Ther.1, 103–132 (2001). [Google Scholar]
- 27.Sannikova, N. The effect of storage conditions on cannabinoid stability. Ascens. Sci. Inc.15, 1–7 (2020). [Google Scholar]
- 28.Trofin, I. G., Dabija, G., Váireanu, D. I. & Filipescu, L. The influence of long-term storage conditions on the stability of cannabinoids derived from cannabis resin. Rev. Chim.63, 422–427 (2012). [Google Scholar]
- 29.Pacifici, R. et al. Evaluation of long-term stability of cannabinoids in standardized preparations of cannabis flowering tops and cannabis oil by ultra-high-performance liquid chromatography tandem mass spectrometry. Clin. Chem. Lab. Med.56, e94–e96 (2018). [DOI] [PubMed] [Google Scholar]
- 30.Lewis, R., Ward, S., Johnson, R. & Burns, D. T. Distribution of the principal cannabinoids within bars of compressed cannabis resin. Anal. Chim. Acta538, 399–405 (2005). [Google Scholar]
- 31.Lazarjani, M. P., Young, O., Kebede, L. & Seyfoddin, A. Processing and extraction methods of medicinal cannabis: A narrative review. J. Cannabis Res.10.1186/s42238-021-00087-9 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Cai, C. et al. Green extraction of cannabidiol from industrial hemp (Cannabissativa L.) using deep eutectic solvents coupled with further enrichment and recovery by macroporous resin. J. Mol. Liq.287, 110957 (2019). [Google Scholar]
- 33.Chang, C. W., Yen, C. C., Wu, M. T., Hsu, M. C. & Wu, T. Y. Microwave-assisted extraction of cannabinoids in hemp nut using response surface methodology: Optimization and comparative study. Molecules22, 1894 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Berman, P. et al. A new ESI-LC/MS approach for comprehensive metabolic profiling of phytocannabinoids in Cannabis. Sci. Rep.8, 1–15 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Milay, L., Berman, P., Shapira, A., Guberman, O. & Meiri, D. Metabolic profiling of cannabis secondary metabolites for evaluation of optimal postharvest storage conditions. Front. Plant Sci.11, 583605 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Ross, S. A. & ElSohly, M. A. CBN and D9-THC concentration ratio as an indicator of the age of stored marijuana samples*. Boletín estupefacientes (1997).
- 37.Grotenhermen, F. Pharmacokinetics and pharmacodynamics of cannabinoids. Clin. Pharmacokinet.42, 327–360 (2012). [DOI] [PubMed] [Google Scholar]
- 38.Raharjo, T. J. & Verpoorte, R. Methods for the analysis of cannabinoids in biological materials: A review. Phytochem. Anal.15, 79–94 (2004). [DOI] [PubMed] [Google Scholar]
- 39.Repka, M. A., Elsohly, M. A., Munjal, M. & Ross, S. A. Temperature stability and bioadhesive properties of Δ9-tetrahydrocannabinol Incorporated hydroxypropylcellulose polymer matrix systems. Drug Dev. Ind. Pharm.32, 21–32 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Turner, C. E. & Elsohly, M. A. Constituents of Cannabissativa L. XVI A possible decomposition pathway of Δ9-tetrahydrocannabinol to cannabinol. J. Heterocycl. Chem.16, 1667–1668 (1979). [Google Scholar]
- 41.Grafström, K., Andersson, K., Pettersson, N., Dalgaard, J. & Dunne, S. J. Effects of long term storage on secondary metabolite profiles of cannabis resin. Forens. Sci. Int.301, 331–340 (2019). [DOI] [PubMed] [Google Scholar]
- 42.Carbone, M. et al. Chemical characterisation of oxidative degradation products of Δ9-THC. Tetrahedron66, 9497–9501 (2010). [Google Scholar]
- 43.Sharma, P., Murthy, P. & Bharath, M. M. S. Chemistry, metabolism, and toxicology of cannabis: Clinical implications. Iran. J. Psychiatry7, 149–156 (2012). [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data Availability: The data used for the multivariate and predictive analyses, are available in the Electronic Supplementary Material.













