The band positions and intensities were extracted from the spectra of the THCA/THC and CBD powder standards whereas the IR bands from the CBDA cannabinoid were taken from the mentioned reference. These spectra were additionally used and compared with the spectra of the native and decarboxylated flowers and extracts that markedly helped in the band assignments . The most abrupt spectral difference evidenced upon decarboxylation of the flowers is depicted by the increase of the intensity of the bands around 1510, 1430, 1260, 1180, 1040 and 835 cm 1 . The obtained result infer that the emerging bands most likely arise from the vibrations within the THC and CBD units whose content increased upon heating, as the presence of the corresponding acidic forms is majorly reduced. Thus, the medium band at 1510 cm 1occurs from both the CBD and THC and intensity increase upon flower decarboxylation infers increase of their content. In addition, the striking rise of the band at 1424 cm 1 is attributed to the vibrations from the THC molecules since its intensity dominates in the IR spectrum of the pure THC and the CBD lacks this IR absorption band . A similar conclusion is derived for the bands at 1267, 1038 and 835 cm 1 in the decarboxylated flower spectrum that are the strongest band in the THC and either very weak or absent in the CBD counterpart . Furthermore, the bands at 1621 and 1578 cm 1 are blue shifted in the heated samples in comparison to fresh THCA dominant flowers and pure THCA spectrum . This outcome strongly reflects that, upon heating, the THCA conversion to THC might be successfully monitored by infrared spectroscopy even in the complex matrix. Similarly, the recent Raman spectroscopy work for non-invasive and non-destructive differentiation between hemp and cannabis attributed the bands at 1623 and 1295 cm 1 to THCA/THC species assigning the former one to their aromatic ring vibration. On contrary, the heated flowers contracted the intensities of strong 888 and 620 cm 1 bands found in the fresh CBDA rich flowers that are obviously attributed to the CBDA molecules. Namely, these bands are crucial to determine the presence of the CBDA compound appearing as the strongest peaks in the pure sample. Moreover, their absence in the THCA counterpart infers that these bands are analytical signatures to follow the transformation of CBDA into CBD.
It is also worth mentioning that the IR spectra of the decarboxylated flowers made possible to discriminate between CBDA and CBD on one hand and the THCA and THC on the other. Moreover, the spectra of the decarboxylated flowers and the fabricated extracts are practically identical . Thus, the band discussion for the spectrum of the decarboxylated flower is also valid for the bands in the spectrum of the cannabis extract. This approach derived from the IR results for the native and decarboxylated flowers was further con- firmed by the chemometric model encompassing the most discriminating bands at 1580, 1430, 1183 and 1040 cm 1 as predictors for THC concentration .Fourteen MIR spectra acquired from different cannabis grow system extract samples were used to construct the calibration set whereas the remaining 20 spectra were used for the prediction set. The CBD and THC concentrations in the samples varied between 0.38– 39.2% and 18.93–86.99%, respectively . Only the spectral region of 1800–400 cm 1 was used for the statistical modeling because the ATR diamond crystal exhibit two-photon broad IR absorbance bands between 2600 and 1900 cm 1 that reduce the signal-to-noise ratio , and the MIR region above 2600 cm 1 did not contain any specific bands originating from the molecules of interest. The preliminary analysis of the models revealed that Savitskzy-Golay smoothing and the second derivative of the native MIR spectra resulted in bestfit parameters for R2Y, R2X and Q2 . The overlaid spectra from the calibration and prediction set, colored according to the THC and CBD content of the samples are presented in Fig. 4a and b. Considering that CBD and THC were not the sole components in the extract and that the samples originated from different sources , it is logical to assume that some transformation of the spectra will be needed to separate and potentiate the specific bands with quantitative relationship to the components of interest. Therefore, in the further part, only the model based on the above mentioned transformation will be presented and discussed. Three main components were extracted from the PLS model for quantification of THC with 0.934, 0.951 and 0.909 for R2X, R2Y and Q2. The score scatter plot for this model reveals distinctive pattern of the sample variances that can be traced to the concentration of THC in the samples. Indeed, these plots reveal how the X variables are related to each other. In this model, there is a clear trend of distinction among the spectra along the t1 vector related to the THC concentration of the samples, thus pointing out the dominant role of the first component in the overall prediction capability of the model. Furthermore, the VIP plot reveals the spectral regions with the utmost importance in the regression model. The bands with VIP factor larger than 1, are usually considered as important both for explaining the variations in the X matrix as well as to correlate with the Y variables. The spectral regions around 1040, 1425, 1183 and 1577 cm 1 demonstrate the largest VIP factors in the model and should be considered as main predictors for the THC concentration. The latter could be additionally confirmed with the coefficient plot , where the mentioned spectral regions are assigned with the largest regression coefficients. The plotted results of actual versus the predicted THC concentrations are presented in Fig. 5d, and the RMSEE of 4.67% indicates the fit of the observations to the model. The RMSEcv for this model is 5.25% and it is an analogous measure to RMSEE, but estimated using a cross-validation procedure.
The RMSEE and RMSEcv are descriptors for the absolute accuracy of the model, and the values reported here should be expected, taking into consideration the variable origin of the samples , number of samples, the employed concentration range, and the real limits of the analytical technique . Three main components were also extracted from the CBD quantification PLS model with 0.936, 0.991 and 0.972 for R2X, R2Y and Q2. In this model, the score scatter plot also presents a distinctive pattern related to the CBD concentration in the samples that follow a diagonal line between both score vectors , indicating that both the first and second component has significant capabilities for prediction of the Y variable. The VIP plot demonstrated that the same spectral regions of the previous model , with an addition of the region around 880 cm 1 . These regions bear the largest VIP factors, and at the same time, the mentioned spectral ranges are assigned with the largest regression coefficients . The plot of actual versus predicted CBD concentrations reveals a better fit of the points relative to the previous THC model with 1.21 and 2.62% for RMSEE and RMSEcv. The predictive capabilities of both models were evaluated separately on a prediction set of 20 extract samples of various origin and the results are presented in Fig. 6a and b. The root mean square error of prediction, for THC and CBD PLS quantification models were 3.79 and 1.44%, respectively, thus con- firming the previous accuracy descriptors and ruling out the possible bias of the calibration models.The PLS models for the quantification of THC and CBD in decarboxylated Cannabis flowers also employed the second derivatives of the raw MIR spectra with Savitsky-Golay smoothing. The overlaid spectra from the 45 samples used in the calibration and prediction set, colored according to the THC and CBD content of the samples are presented in Fig. 7a and b. As in the previous case, such transformation was chosen as the most optimal regarding the main fit parameters in comparison to the raw, SNV and first order derivative MIR spectra models. Such data transformation is often considered as a necessity in cases where spectra from complex matrices are obtained and the main quantification related bands are overlapped with other components present in the sample. The THC PLS quantification model was build using transformed spectra acquired from 15 decarboxylated samples with various THC content and origin . Five main components were extracted from the model explaining 97.2% of the variations in the X-variables, and producing a high correlation coefficient R2Y = 0.992 with appropriate predictability . The score scatter plot of the second and third component vectors reveal the THC content related pattern, where the THC content increases as a function of the score in both components . The model VIP plot exposes the bands that are related to the THC content in the flowers. The spectral regions around 1620, 1610, 1578, 1425, 1180, 1038, 1010 and 825 cm 1 which are similar to the ones reported in the extract PLS models, are also assigned with large VIP factors. The coefficient plot confirms the previous findings, where the above mentioned spectral regions with the highest VIP factors,marijuana grow system are also assigned with the highest regression coefficients . The RMSEE and RMSEEcv derived from the actual versus predicted THC concentrations plot were 0.43 and 1.53%.
The complexity of the sample , the variability of the samples in regards to their origin and horticultural maturity, as well as the nonuniformity of the plant material and the flower-ATR crystal contact should be considered as main factors that govern the accuracy descriptors. Five main components were extracted for the PLS modeling for CBD content quantification of the Cannabis flowers with R2X = 0.969, R2Y = 0.994 and Q2 = 0.66. As in the previously described model, the score scatter plot of the second and third main components vector presents a distinctive pattern associated with the CBD content in the samples . The bands around 1440, 1185, 1100, 1010, 911, 888, and 826 cm 1 demonstrated the largest VIP factors in the MIR spectra . The mentioned bands were assigned with the highest regression coefficients , thus confirming their association with the CBD content. The actual versus predicted plot of the samples demonstrates a satisfactory level of correlation and the accuracy model parameters RMSEE and RMSEEcv were 0.21 and 1.41%. To evaluate the predictive capability of the above mentioned PLS models for quantification of THC and CBD in decarboxylated Cannabis flowers, a separate prediction set of 30 samples of different origin was employed. The correlation plots of actual versus predicted THC and CBD content are presented in Fig. 9a and b, and the models RMSEP was 2.32% and 1.33% for quantification of THC and CBD, respectively. The RMSEP values are in good agreement with the accuracy descriptors of the models and con- firm the presented predicting capability of the models in a separate independent set of samples.To our knowledge, this study is the first to use a validated questionnaire to assess the association between female sexual function and aspects of cannabis use including frequency, chemovar, and indication. In this survey of more than 400 women, we found a dose response relationship between increased frequency of cannabis use and reduced odds of female sexual dysfunction. In addition, while the increase in index scores was small , increased cannabis use was associated with improved sexual desire, arousal, orgasm, and overall satisfaction as well as overall improved FSFI scores as compared with less frequent users. Older women and those with more comorbidities tended to have more sexual dysfunction. Importantly, our study did not find an association between cannabis chemovar , reason for cannabis use, and female sexual function. As cannabis use has been shown to be associated with increased sexual frequency in the United States, it is possible this may cause positive effects on sexual experiences.7 Much of the research focusing on sexual function and experiences with regard to cannabis began in the 1970s and 1980s. Cannabis’ potential positive effect on female sexual function was noted as early as 1970 by Tart19 who sought to describe the common experiences of cannabis users. He noted in interviews with college students that orgasms are improved, arousal increases, and “sexual feelings are much stronger” leading to more satisfaction.