Identification of Novel Psychoactive Substances in Complex Mixtures Using a Design of Experiment Guided Approach and Handheld Raman Spectroscopy

The screening of NPS in street samples has been proven to be problematic in recent years due to their fast appearance, the unavailability of adequate reference materials and their complexity. The aim of this project focussed on the use of handheld Raman spectroscopic technique in the identification of NPS in samples of street-like complexity. Additionally, it mainly endeavoured at the application of a model in NPS screening and its ability to predict NPS responses in complex mixtures. In fact, chemometrics namely a mixtures’ design of experiments approach was used to propose a set of 26 NPS samples of varying amount of NPS (5F-PB-22, N-Me-2-AI and phenibut) and of most common cutting agents/adulterants (benzocaine, caffeine, creatine, and sodium glutamate) to represent maximum variability of a five components mixture. Each mixture was analysed via a Rigaku Progeny handheld Raman spectroscopy device using its in-built algorithms namely wavelet and Rigaku mixture. Matching results obtained from these initial studies were evaluated and were re-implemented in the design of experiment for the generation of a model used to predict NPS responses for proposed test samples. Initial analysis of all 26 vials on a NPS library (99 reference materials) using wavelet displayed good NPS detection notably in samples of high concentrations (above 80 mg) with matching values between 0.59-0.98. Benzocaine and caffeine showed major influence on NPS identification in samples of increased complexity. This was mainly observed in samples of low NPS concentration (10-24 mg) where only 5F-PB-22 displayed detection between 0.10-0.20. However, this was mostly due to the abundant detection of analogues and/or structurally similar substances which were absent within the sample but were similar in Raman spectra to excipients. This was mostly the case with benzocaine derivatives such as dimethocaine and procaine. For comparison purposes, alternative sub-libraries were created containing only five references, an NPS among all three and all cutting agents and adulterants, hence three sub-libraries created. Consequently, analysis of the same 26 samples in similar conditions against sub-libraries exhibited 100% NPS identification for all NPS using wavelet. The influence of analogues was recurrent using Rigaku mixture algorithm against full NPS library. Yet, the algorithm displayed better matching results with an increased matching range between 0.79-0.99 in presence of above 80 mg of NPS. This was due to its ability to match samples spectra to multiples matches for each result. Thus, the values obtained could not be assigned to a match in particular. With wavelet being able to correlate to single matches, each value corresponded to the match obtained hence wavelet results were carried forward for the design of the model. Analysis from the model showed promising results in its capability to predict matching responses for pre-determined sample composition. Screened against the full NPS library, 15% of the test samples (only 5F-PB-22 samples) displayed NPS detection as well as matching values correlating to the predictions. While sub-library analysis showed improvement in detecting NPS in all test vials (100%) as noticed in the initial studies, it mainly highlighted the ability of the model to consistently predict within range of the obtained matching values in 86% of proposed samples. These outcomes regarding the complexity of samples could be of help to forensic scientists and border control officers in understanding how the use of cutting agents and adulterants happen to hinder detection of NPS by correlating to other substances of similar spectral profile. More importantly, it confirms the use of a model in NPS screening could further show promise in predicting NPS responses depending on the samples composition which would be useful in tackling complexity issue faced in street samples