Profiling of seized cannabis samples using gas chromatography, statistical methodology and multivariate data analysis

Worldwide, cannabis is the most commonly seized and used drug, in which Δ9tetrahydrocannabinol is the main psychoactive component that causes the mind-altering effects. In Europe, cannabis is mainly cultivated indoors using high-tech equipment to mimic the best environmental conditions. By doing so, it is possible to produce multiple growths throughout the year.
Nowadays, the current market contains many different cannabis products ranging from marijuana and hashish to edibles, CBD products and semi-synthetic or synthetic cannabinoids. This dissertation focused on the analysis of the herbal material.
The analysis of controlled drugs is one of the major domains of interest in forensic laboratories. For cannabis analysis, the main focus is to identify and quantify certain cannabinoids taking into account the legal framwork, but many other forensic purposes may be the goal as well. A specific application is illicit drug profiling where the aim is to find or confirm links between seizures, which can be a very useful law enforcement tool during police investigations. When investigating judges ask to perform illicit drug profiling, usually a comparison between different cases is needed. Profiling cocaine, heroin and amphetamine-type stimulants has already been successful. In this thesis, a method for cannabis profiling was developed combining statistical methodology with multivariate data analysis. Finding a suitable method to compare different seizures is challenging because marijuana is a natural product of which the chemical composition depends on many environmental factors, causing heterogeneity in the plant specialized metabolites, also within one plantation.
Gas chromatography (GC) is the most frequently applied analytical technique in illicit drug profiling and was also used in our research to analyze the seized cannabis samples. First, target-based GC chemical profiles containing eight cannabinoids were used, where the intra (within)- and inter (between)-plantation variabilities were studied and evaluated by similarity analysis. Several data pretreatment techniques were investigated for their discriminating ability between different locations. This was done by determining linked-thresholds and by applying prognostic error rates, i.e. false negatives and false positives (FPs), as decision-making parameters with the goal to find a preprocessing method that allowed the best distinction between seized cannabis samples. This workflow was used for all studies conducted in this thesis. Initially, promising results were found, aiming at obtaining low FPs, which is important when the results need to be used as additional evidence in court. However, this first data set only had a limited number of intra-samples per plantation where it is very likely that the model did not contain all the information about the possible variation in different cultivation sites.
Subsequently, a larger data set composed of entire GC fingerprints that comprised all components in the sample, was applied to compare the herbal material. This fingerprint approach is new in illicit drug profiling, where usually only selected compounds are used. The same methodology as explained above was used on the entire profiles. An additional data pretreatment step with herbal fingerprints, i.e. alignment, is necessary because small retention time shifts occur between the different chromatographic runs. Even though there was a large decrease in FPs after selecting the best data preprocessing, the obtained FP rate was still too high.
Finally, the same large data set as in the second part was studied, but only using the eight previously selected cannabinoids as profile. In addition, an intra-variability study was carried out to investigate whether the initial model was based on representative intra-information. The latter is important to take into account when developing a profiling method, especially for defining decision limits. This was initially not the case and a considerably larger intra-location variability was observed. As a result, an optimized model was established, providing an improvement of FPs. However, the FP rate is still not at a level that was acceptable in court.
To conclude, by applying proper data preprocessing, a considerable improvement of the FPs was established in each case study, but the FP rate never reached values below 10%, which was the initial imposed target value. However, this research did allow identifying some problems that can occur when profiling natural products, where suggestions about the possible next steps were discussed to further reduce the number of FPs.