How do cancer-sniffing dogs sort biological samples? Exploring case-control samples with non-targeted LC-Orbitrap, GC-MS, and immunochemistry methods

Abstract

Early identification of disease onset is regarded as an important factor for successful medical intervention. However, cancer and other long-term latency diseases are rare and may take years to manifest clinically. As such, there are no gold standards with which to immediately validate proposed preclinical screening methodologies. There is evidence that dogs can sort samples reproducibly into yes/no categories based on case-control training, but the basis of their decisions is unknown. Because dogs are sniffing air, the distinguishing chemicals must be either in the gas-phase or attached to aerosols and/or airborne particles. Recent biomonitoring research has shown how to extract and analyze semi- and non-volatile compounds from human breath in exhaled condensates and aerosols. Further research has shown that exhaled aerosols can be directly collected on standard hospital-style olefin polypropylene masks and that these masks can be used as a simple sampling scheme for canine screening. In this article, detailed liquid chromatography-high resolution mass spectrometry (LC-HR-MS) with Orbitrap instrumentation and gas chromatography-mass spectrometry (GC-MS) analyses were performed on two sets of masks sorted by consensus of a four-dog cohort as either cancer or control. Specifically, after sorting by the dogs, sample masks were cut into multiple sections and extracted for LC-MS and GC-MS non-targeted analyses. Extracts were also analyzed for human cytokines, confirming the presence of human aerosol content above levels in blank masks. In preliminary evaluations, 345 and 44 high quality chemical features were detected by LC-MS and GC-MS analyses, respectively. These features were used to develop provisional orthogonal projection to latent structures-discriminant analysis (OPLS-DA) models to determine if the samples classified as cancer (case) or non-cancer (control) by the dogs could be separated into the same groups using analytical instrumentation. While the OPLS-DA model for the LC-HR-MS data was able to separate the two groups with statistical significance, although weak explanatory power, the GC-MS model was not found to be significant. These results suggest that the dogs may rely on the less volatile compounds from breath aerosol that were analyzed by LC-HR-MS than the more volatile compounds observed by GC-MS to sort mask samples into groups. These results provide justification for more expansive studies in the future that aim to characterize specific chemical features, and the role(s) of these features in maintaining homeostatic biological processes.