Eukaryotic cells are highly heterogeneous. These cells are arranged into different compartments, carrying out separate functions and facilitating biological processes. Proteins are the effector biomolecules targeted to subcellular locations that help fulfill specific tasks in living organisms. Spatial proteomics can help unravel molecularly how protein abundance and localization are altered in cells, which is not feasible in traditional bulk-scale proteomics. To achieve this, our lab has developed a miniaturized sample processing platform called nanoPOTS, reduced separation columns' inner diameter to increase ionization efficiency and concentrate analytes for mass spectrometers and optimized data acquisition modes for increasing proteome coverage in spatial and single-cell proteomics and applying these techniques to studying protein dynamics in various biological samples and conditions.This dissertation details the extension of our techniques to other limited biological samples. We expanded the nanoPOTS sample processing workflow to formalin-fixed, paraffin-embedded tissues (FFPE). By optimizing extraction solvents, times, and temperatures, we obtained the highest proteome coverage in FFPE tissues compared to fresh frozen tissues. Our observations revealed an average of 1312 and 3184 high-confidence master proteins in 50 – 200 µm square cut regions of a 10 µm thick FFPE-preserved mouse liver tissue, achieving 88% of the proteome coverage compared to that obtained from fresh frozen tissues of the equivalent size. We then characterized our fully automated sample preparation and analysis workflow, autoPOTS, for FFPE spatial proteomics. We applied the optimized nanoPOTS sample preparation condition to analyze normal, precancerous, and cancerous lesions of FFPE-preserved pancreatic ductal adenocarcinoma (PDAC) human samples, achieving an average coverage of 3000 proteins from 200 µm squares of each cell type. We identified some highly expressed proteins using differential analysis for cancerous lesions. We also optimized microLIFE, a cellenONE software add-on instrument, to detect and isolate low-input bacteria samples using Escherichia coli (E coli). We collected proteomic data using both Wide Window data Acquisition and Data-Independent Acquisition. On average, we identified 800 and 1300 proteins in WWA and DIA, respectively. We applied microLIFE to identify proteins involved in Salmonella pathogenicity island-I (SPI) impacted by oxygen availability in their growth medium and observed 50% and above average of difference classes of SPI compared with bulk-scale proteomics. This novel software can enable low-input spatial proteomics.

Mass spectrometry imaging (MSI) is a high-throughput technique that in addition to performing protein identification, can capture the spatial localization of proteins within biological tissue. Nevertheless, sample pre-processing and MSI instrumentation limit protein identification capability in MSI compared to more standard tandem mass spectrometry-based proteomics methods. Despite these limitations, the current protein identification approaches used in MSI were originally designed for standard mass spectrometry-based proteomics and do not take advantage of the spatial information acquired in MSI. Herein, I explore the benefit of using the spatial spectral information for protein identification using two objectives. For the first objective, I developed a novel supervised learning spatially-aware protein identification algorithm (SAPID) for mass spectrometry imaging and benchmarked it against ProteinProphet and Percolator, which are state-of-the-art tools for protein identification confidence assessment. I showed that SAPID identifies on average 20% more proteins at <1% false discovery rate compared to the other two algorithms.Furthermore, more proteins are identified when spatial features are used to identify proteins compared to when they are not suggesting their additional benefit. For the second objective, I used SAPID to rescue false positive and false negative protein identifications made by ProteinProphet. By examining a combination of data sampling and learning algorithms, I was able to achieve a good classification performance compared to the baseline given the extremeimbalance in the dataset. Finally, by improving proteome characterization in MSI, our approach will help providing a better understanding of the processes taking place in biological tissues.