ADVANCED TOOLS FOR ISOTOPE PROBING AND MASS SPECTROMETRY IN MICROBIAL PROTEOMICS WITH AERITH AND SIPROS

Abstract

Stable isotope probing (SIP) is a foundational technique in microbial ecology for linking microbial identity and activity to substrate utilization in complex communities. Proteomic SIP extends this idea to the peptide and protein level, where labeled amino acids carry both taxonomic information from their sequence and functional information from the annotation of their parent protein. However, proteomic SIP has historically been limited by three bottlenecks: the high computational cost of searching mass spectra against a protein database over a continuous range of isotopic enrichment levels, the difficulty of separating weakly labeled peptides from unlabeled background in complex metaproteomes, and the lack of accessible tools to visualize and manually validate isotopically labeled peptide-spectrum matches. In this dissertation, I develop a coordinated suite of computational tools that addresses these bottlenecks and extends proteomic SIP from data-dependent acquisition (DDA) on Orbitrap mass spectrometers into the high-throughput narrow-window data-independent acquisition (DIA) regime of the Orbitrap Astral platform. Chapter 2 presents Sipros 4, an upgraded enrichment-resolved database search engine that is more than twenty-fold faster than Sipros 3, accurately recovers the median 13C atom% expected in E. coli standards across 1.07–99 atom% enrichment, and identifies 153 13C-labeled proteins/protein groups in a 13C-methanol soil SIP experiment versus 81, 15, and 13 by Sipros 3, Calisp, and MetaProSIP respectively. Chapter 3 introduces Sipros 5, an end-to-end SIP proteomics platform with universal percentage-labeling supportthat models precursor and fragment isotope envelopes in DDA and DIA data, uses unlabeled negative controls to control the label-specific false discovery rate (FDR), and is paired with a convolutional neural network (CNN)-based WinnowNet rescorer. Sipros 5 produced 2–3× more peptide-spectrum matches (PSMs) than Sipros 4 in DDA and reached approximately fivefold higher PSM yields in 35-minute Astral DIA runs (130,000–330,000 PSMs) compared with 90-minute Orbitrap DDA runs (∼30,000–50,000 PSMs); applied to a 43-day mouse experiment using 15N stable isotope labeling of mammals (SILAM), Sipros 5 quantified protein-turnover kinetics that distinguish microbial proteins (half-life ∼3 days; 90% conversion by ∼10 days) from host proteins (half-life ∼5 days; 90% conversion by ∼17 days). Chapter 4 describes Aerith, an R package that simulates SIP-labeled isotopic fine structures via Monte Carlo and isotopic envelopes via fast Fourier transform (FFT) for arbitrary chemical formulas, simulates fragment-ion envelopes of peptides via residue-level convolution and binomial distributions, scores PSMs with multivariate hypergeometric (MVH), cross-correlation (Xcorr), weighted dot product (WDP), and spectral-entropy scores, and produces publication-quality visualizations through ggplot2 for manual validation across the full 0–100% enrichment range. Together, these tools provide a comprehensive analytical pipeline that improves the sensitivity, accuracy, and interpretability of proteomic and metaproteomic SIP and enables the routine quantitative analysis of substrateassimilation and protein turnover at single-organism and microbiome-community scales.