High-throughput analysis of contrived cocaine mixtures by direct analysis in real time/single quadrupole mass spectrometry and post acquisition chemometric analysis
Horsley, Andrew Blair
MetadataShow full item record
Direct Analysis in Real Time (DART) ionization/mass spectrometry allows for the high throughput analysis of a wide range of materials including but not limited to: solids, liquids, powders, tablets, and plant materials. The ability to detect cocaine was established in a reproducible manner with the use of a DART ionization source (IonSense Inc., Saugus, MA) interfaced to a modified single quadrupole mass spectrometer. Development of a methodology for the detection of cocaine within contrived street quality drug mixtures involved the optimization of the ionization source, sample introduction mechanism, ion guide, and mass analysis parameters. An analytical method was created that utilized ionized helium carrier gas heated to 300°C and an automated sample introduction apparatus consisting of a Linear Rail Enclosure that holds consumable QuickStrip^TM sample cards. Ionized molecules were then fragmented by manipulation of voltage levels within the ion guide to gain more structural information prior to detection by a single quadrupole mass spectrometer. Cocaine was detected by the modified DART/MS analytical platform and gave two peaks within the mass spectrum at m/z 304 and 182. Optimization of in-source fragmentation by manual adjustment of the skimmer focus voltage allowed for the reproducible fragmentation of cocaine and the ability to increase or decrease the amount of fragmentation seen between the two peaks detected for cocaine. With the use of fragmentation, this analytical platform can be classified as a Category A technique as defined by the Scientific Working Group for the Analysis of Seized Drugs. The robust detection of cocaine was demonstrated for reference samples at concentrations as low as 10 ng/μL (50 ng) with high signal abundance greater than ten times the signal to noise ratio. Furthermore, the detection of cocaine at 10 ng/μL was demonstrated for multi component mixtures of up to 14 additional components containing common adulterants and diluents found within street quality samples. In total, 25 common excipients were tested using the same method parameters as optimized for cocaine analysis. Of these 25 excipients tested, five were not detected in positive ion mode (one could be detected in negative ion mode). Of the twenty excipients that could be detected by mass spectrometry, two pairs of excipients (levamisole/tetramisole and creatine/creatinine) could not be differentiated from each other. There were no excipients tested that had equivalent m/z values as those of cocaine. Experimentation into the effects of various excipients at multiple concentrations on the abundance of the two cocaine peaks was performed. Regardless of excipient amount (up to 10 times more concentrated than cocaine) and the number of components (up to 15 total components) the ratio of abundance between the m/z 304 to 182 peaks did not vary greater than 22% relative standard deviation. A match criteria protocol was developed for the ability of an analyst to confirm the presence of cocaine within unknown forensic case samples that have previously tested positive for the presumptive identification of cocaine. The identification of cocaine was based on various factors such as the signal to noise ratio at m/z 304 and 182, the ratio of abundance between those two peaks as well as positive and negative controls. This match criteria protocol was utilized for 25 double blind mock forensic casework samples was performed. Determination for the presence of cocaine within these unknown samples gave an analyst error rate of 0%, with no false positives or false negatives predicted. To further aid human interpretation and identification of compounds within mixtures, the advanced chemometric software, Analyze IQ, was utilized. Development of predictive classification models using a combination of pre-processing steps, principle component analysis and machine learning techniques was achieved. Models were built using 381 unique samples for the purposes of identifying the presence of cocaine within unknown samples. Of all methods available within the Analyze IQ software, the optimization of a model using principle component analysis with support vector machine regression with a radial basis function kernel yielded an initial error rate of 0% for 72 samples tested. Furthermore, of the samples tested against the model, 20 samples were comprised of excipients that were not incorporated into the initial model development process. The inclusion of these samples (10 spiked with cocaine, 10 absent of cocaine), shows that predictive modeling based software can provide an accurate, robust, and evolving approach to the identification of cocaine within sample compositions that have not previously been tested and stored in a database of known reference samples. Predictive modeling has advantages over current mass spectral libraries, which are limited to the identification of pure compounds. To further test the abilities of predictive models, optimized machine learning models were applied to 25 double blind mock forensic casework samples. The predictive modeling error rate was identical to the human interpretation rate for the double blind mock casework samples with a 0% error rate. Using the DART/MS analytical platform, 25 mock forensic casework samples along with positive and negative controls were analyzed and identified for the presence of cocaine within 30 minutes. On the order of 15 to 30 times faster than modern GC/MS and LC/MS methods, the ability to analyze and identify samples faster would allow for an increase in samples being processed on a daily basis and allow for the reduction of case backlogs that currently plague controlled substances sections of forensic science laboratories throughout the United States.