We found that the Raman spectrum of hemp is dominated by vibrational bands that could be assigned to cellulose, carotenoids and lignin, Fig. 1, Table 1.¹³ At the same time, spectra collected from GC, TCC and TS clearly demonstrate presence of THCA. Specifically, we observed vibrational bands at 780, 1295, 1623, and 1666 cm⁻¹ in the spectra of GC, TCC and TS that can be assigned to THCA. We also observed a change in the intensities of bands at 916, 993, 1000, 1084, 1155, 1185, 1267, 1321, and 1525 cm⁻¹ in the spectra collected from GC, TCC and TS relative to the intensities of these bands in the Raman spectrum of hemp. However, the change in intensities of these vibrational bands cannot be used for confirmatory diagnostics of THCA presence because these bands have also been observed in the spectrum of hemp and can be assigned to carbohydrates and cellulose. Nevertheless, the change in intensities of these vibrational bands suggests about structural differences between hemp and cannabis plants. For instance, we found that vibrational bands at 993, 1000 and 1525 cm⁻¹, which can be assigned to carotenoids, are more intense in the spectrum collected from hemp relative to the intensity of this band in the spectra collected from cannabis species. This result indicate that hemp have higher carotenoid content comparing to cannabis plants. Similarly, we found that intensities of 946, 1084 and 1155 cm⁻¹ bands are stronger in the spectrum of hemp comparing to cannabis plants. This finding suggests that amount of cellulose is higher in hemp comparing to cannabis species.¹³

To further prove our expectation that RS can be used for highly accurate differentiation between hemp and cannabis, we used OPLS-DA analysis. The final model, containing one predictive component, 2 orthogonal components and 1001 (701–1700 cm⁻¹) out of 1651 original wavenumbers, was used to generate the misclassification table (Table 2) and the loadings plot (Fig. 2).

The first predictive component (PC) (Fig. 2) explain 94% of the variation between classes. Absolute intensities in the loading spectrum are proportional to the percentage of the total variation between classes explained by each wavenumber. The model identified the peak at 781 cm⁻¹, which could be assigned to THCA (Table 1), cellulose and lignin peaks at ∼925 cm⁻¹ and the bands at 1260–1320 cm⁻¹, which correspond to both THCA and cellulose. Also, the model identified peaks at 1440 cm⁻¹, which could be assigned to aliphatic vibrations and bands at 1623–1660 cm⁻¹, which originate from THCA, to be the strongest spectral markers of cannabis, which supports the conclusions of our qualitative spectral analysis above. The model explained 48% of the variation (R2X) in the spectra and correctly assigned all 86 spectra to their classes (Table 2, Fig. S1†). This indicates that coupling of OPLS-DA with RS allows for a 100% accurate differentiation between cannabis and hemp.

Next, we asked a question whether RS can be used for quantitative prediction of the THCA content in cannabis, as well as identification of cannabis variety. HPLC analyses of TCC, GC and TS samples allowed to determine the THCA content which was found to be 10.31%, 6.12% and 4.05%, respectively. Our results demonstrate that intensity of the 1623 cm⁻¹ band directly correlates with the amount of the THCA content in the cannabis, Fig. 1. We used analysis of variance (ANOVA) to visualize the correlation of intensity of 1623 cm⁻¹ band with the content of THCA in the analyzed plant material, Fig. 3. These results suggest that RS can be used for quantitative prediction of the THCA content in intact plant materials. However, more experimental work is needed at this point to determine the accuracy and the range of THCA prediction. This work is currently in progress in our laboratory.

Utilization of OPLS-DA allowed for quantitative differentiation between three different classes on cannabis, Table 3. The final model, containing 2 predictive components, 2 orthogonal components and 1001 (701–1700 cm⁻¹) out of 1651 original wavenumbers, was used to generate the misclassification table (Table 3) and the loadings plot (Fig. S2†). The first two predictive components (PC) explain 42% and 36% of the variation between classes respectively, which collectively accounts for 78% of the total class-to-class variation. Absolute intensities in the loadings spectra are proportional to the percentage of the total variation between classes explained by each wave-number within each component. The model identified the peak at 781 cm⁻¹ (PC1), cellulose and lignin peaks at 925 cm⁻¹ (PC1), the bands at 1250–1340 cm⁻¹ (PC1), 1440 cm⁻¹ (PC1) and the region of 1580–1670 cm⁻¹ (PC1) to be the strongest spectral markers representing the three cannabis species, which supports the conclusions of our qualitative spectral analysis above. The model explained 48% of the variation (R2X) in the spectra and 78% (R2Y) of the variation between the classes. Furthermore, the model correctly assigned 84 out of 86 spectra to their classes (Table 3). Our results demonstrate that coupling of OPLS-DA with RS allows for a ∼97% accurate identification of the cannabis variety.

It should be noted that in plants THC is present in carboxylated form of THC, known as THCA. Upon decarboxylation, which can be induced by thermal heating, THCA is converted to THC. Spectroscopically, the carboxyl group of THCA is evident from the vibration at 1691 cm⁻¹, which was observed in our spectra. One could expect that other vibrational bands that were assigned to THCA would originate from THC since decarboxylation is the only structural transformation in the molecule upon THCA to THC conversion. Specifically, our spectroscopic analysis was based on 1623 cm⁻¹ band, which originate from aromatic moiety present in both THCA and THC. Therefore, we can speculate that RS allows to predict the amount of THC in the analyzed sample without necessary oxidation of THCA to THC.