Enhancing Computational Tools for Ion Mobility-Mass Spectrometry-Based Untargeted Workflows (Cascadia Proteomics Symposium 17)

Ion mobility spectrometry (IMS) is a rapid and highly reproducible molecular-shape separation technique. While IMS has shown great utility when coupled with MS for analysis of complex samples, methods for processing the complex data generated have lagged behind. In fact, the incorporation of this extra IMS separation dimension requires upgrades and optimization of existing computational pipelines and the development of new algorithmic strategies to fully exploit the advantages of the technology. Here, we investigate MS pre-processing algorithms to extend feature detection and quantification performance, as well as integration of IMS collisional cross section (CCS) libraries into data analysis tools for molecular characterization. These strategies were applied for untargeted analyzes of biofluid samples to evaluate changes in endogenous metabolites and xenobiotics. IMS-MS data files were acquired by an Agilent 6560 Ion Mobility Q-TOF MS system. A software tool for IMS data analysis and processing was developed (C#) to apply multidimensional smoothing, saturation correction and generate new raw data files. Intensity values in each frame (an IMS cycle) were smoothed first in the drift dimension followed by smoothing of the chromatographic dimension considering neighboring frames. High signal intensity values that reached beyond detector capacity were identified by the characteristic flat profile at the apex of saturated peaks. Agilent Mass Profiler was used for feature extraction and sample alignment. A CCS library was created from small molecule standards. Features were annotated using Agilent ID Browser. Statistical analyses of results were performed in R. The developed software tools were integrated to process IMS-MS data from urine samples previously analyzed using a solid-phase extraction method. To minimize the effects of low ion statistics (e.g., jagged profiles), several smoothing kernels were evaluated: Gaussian, Savitzky-Golay, moving average and weighted moving average. Among those, moving average smoothing provided the best results for retrieving low-abundance features and merging at least 40% more of the features with split profiles. Smoothing increased by a factor of 2.5 the number of high quality features (quality score ≥ 80; a 0−100 scale considering signal-to-noise, number of isotopic ions and m/z stability). More specifically, 1332 features were found in at least 20 of the 96 analyzed samples, compared to 360 features found without smoothing. These improvements were also reflected by a 3-fold increase of the number of features having abundances with less than 20% coefficient of variation. A filtering strategy was incorporated in the smoothing heuristic to reduce background noise, consequently decreasing (by half, on average) file size, memory usage and processing time for feature detection and alignment. Furthermore, saturation correction improved quantification and mass accuracy of analytes with high intensity signals. We therefore observed that the implemented strategies enhanced the dynamic range of measurement: smoothing towards the lower end and saturation correction towards the higher end of abundances.

These data quality improvements also allowed us to compute more accurate CCS values from the IM-MS measurements, which in combination with CCS libraries can help to discriminate the feature of interest. For instance, our CCS library increased the identification confidence of creatinine for a detected feature (114.0656 m/z, 122.69 CCS), distinguishing among 6 hits (0.01 mass tolerance) from the METLIN database. Work is in progress to compare and examine in detail the benefits of incorporating predicted and experimental CCS libraries in the metabolite identification workflow.