Authors
Shabnam Salimi, Wentao Zhao, Maria Partida-Aguilar, Elle Harwood, Hayley Purcell, Daniel Raftery

Institutions
University of Washington, Department of Anesthesiology and Pain Medicine, North West Metabolomic Research Center, Mitochondrial Metabolism Center

Abstract
Doxorubicin (DOX) is an effective chemotherapeutic agent whose anti-tumor effects are mediated through DNA damage and apoptosis induction in cancer cells. However, DOX also exerts substantial off-target toxicity that contributes to accelerated aging phenotypes, mitochondrial dysfunction, metabolic dysregulation, and reduced healthspan. We hypothesized that DOX induces both short- and long-term metabolomic alterations in skeletal muscle myoblasts and that spermidine or rapamycin may mitigate these metabolic perturbations. Mouse skeletal muscle myoblasts were treated with 0.2 µM doxorubicin, followed by evaluation of metabolomic changes at short-term (3 days) and long-term (10 days) time points using untargeted Q-TOF metabolomics with SERRF normalization and machine learning–based analyses. We additionally evaluated the effects of spermidine and rapamycin treatment on cell viability and metabolomic remodeling following DOX exposure. DOX treatment induced marked alterations in pathways related to NAD+ metabolism, glutamate metabolism, coenzyme A metabolism, oxidative stress, inflammation, vitamin B6 metabolism, pyrimidine and purine metabolism, and mitochondrial function. Short-term exposure demonstrated depletion of NAD+ metabolites, perturbations in nitrogen metabolism, and reductions in vitamin E–related metabolites and biopterin-associated pathways. Long-term exposure further revealed inflammatory prostaglandin signaling and persistent mitochondrial metabolic dysfunction. Spermidine and rapamycin partially reversed several DOX-associated metabolic abnormalities. Both interventions altered pathways related to glutathione metabolism, folate metabolism, amino acid metabolism, mitochondrial energetics, and inflammatory signaling. Methionine isotope tracing experiments demonstrated increased methionine cycle activity, elevated S-adenosylmethionine (SAM), and increased decarboxylated SAM (dc-SAM) following spermidine and rapamycin treatment, suggesting potential epigenetic regulation through altered DNA methyltransferase activity. These findings demonstrate that DOX induces profound metabolic remodeling associated with aging-related hallmarks, including mitochondrial dysfunction, oxidative stress, and altered one-carbon metabolism. Spermidine and rapamycin may represent promising strategies to mitigate chemotherapy-induced metabolic and epigenetic dysfunction. Future studies integrating DNA methylation analyses will further define the molecular mechanisms underlying these protective effects.

Authors
Emma Timmins-Schiffman, Karen Junge, Katarina Abrahamsson, Michael Riffle, Ruben Shtrestha, Brook Nunn

Institutions
University of Washington (ETS, KJ, MR, and BN); University of Gothenburg (KA); Bruker (RS)

Abstract
Despite its austere landscape, Antarctica is home to abundant life, some of which is found in the seasonal sea ice that doubles the size of the continent during the austral winter months. Marine microbes are caught in this massive ocean freezing event and metabolically active communities develop in sea ice brine channels. These sea ice microbes have the genomic potential to participate in a variety of ecologically important nutrient cycling and biogeochemical processes, including the bromocarbon cycle. Bromocarbons are highly concentrated in polar sea ice and some forms, when volatilized, can chemically deplete ozone. However, studies of these microbes are hindered by problems such as field site access, the impermanence of the habitat, a lack of available sequencing resources, the low abundance of microbes in this system, and a lack of knowledge of the variability of species and metabolic processes over space and time. We have developed a method to leverage de novo sequencing of high coverage Bruker timsTOF Ultra2 metaproteomics data from Antarctic sea ice, snow and seawater, combined with a limited sequenced metagenome. Our pipeline allowed for the identification of over 41,000 peptides across a diverse assemblage of ice- and snow-bound microbes, revealing metabolic processes associated with important environmental variables, such as temperature, salinity, and bromocarbon concentration. These discoveries reveal the roles of microbes in an extreme environment that is under threat from global ocean warming.

Authors
Claire E. Elbon1, Gabrielle Chebli2, Emma TImmins-Schiffman1, Micheal Riffle1, Bo Wen1, William Noble1, Brook L. Nunn1

Institutions
1. University of Washington, Seattle, WA. 2. Georgia Institute of Technology, Atlanta, GA.

Abstract
Harmful algal blooms threaten coastal ecosystems, fisheries, shellfish harvests, and public health across Washington State, yet current monitoring approaches largely detect blooms after biomass or toxin accumulation has already occurred. Currently, blooms are primarily defined in “initiation” and “termination” phases, which are determined by rate changes in chlorophyll in the water column. A major unresolved challenge is determining the full suite of bloom phases; including why, when, and how they initiate, terminate, and the duration of a bloom. Understanding molecular signals at the protein-level of the co-associated bacterial and viral community provides a unique opportunity to define bloom phase transitions, particularly whether bloom decline reflects viral lysis, algal physiological collapse, dormancy, or bacterial restructuring of bloom-derived organic matter. Here, I use a high-frequency multi-omic time series collected every four hours over 22 days in East Sound, Washington, to connect microbial community state changes with active protein-level function during harmful algal bloom succession. Initial analysis of 133 bacterial and viral (1-0.22 micron filter) metagenomic time points revealed that the microbial community organizes into four distinct ecological states across the bloom time series. These states do not simply mirror chlorophyll fluorescence, suggesting that microbial and viral succession captures underlying bloom-phase structure that is not visible from biomass alone. I use these metagenome-defined ecological states as the framework for interpreting high-resolution DIA metaproteomics across the full time series. Metaproteomic samples were generated using a novel high-throughput, low biomass S-Trap 96 well plate preparation method and run using 20-minute gradients on an Orbitrap Astral Mass Spectrometer optimized for our complex environmental samples. For each metaproteomic sample, I will search DIA spectra against a time-matched metagenome through a collaborative workflow that integrates metagenomic and metaproteomic analyses. The resulting data are processed using Carafe3 (a prototype developed by Bo Wen) which reduces metagenome-derived protein databases by retaining only nonredundant proteotypic peptides, enabling deep peptide detection and protein-level interpretation across microbial and viral communities with minimal desktop GPU computing. By aligning peptide abundance patterns with the four metagenomically defined ecological states, I will identify the functions that distinguish bloom initiation, maintenance, decline, and post-bloom restructuring. Specifically, I will evaluate viral infection through structural and replication proteins, stress and mortality through oxidative-stress, proteolytic signatures, dormancy or persistence through cell-wall remodeling and stress-tolerance proteins, and processing of bloom-derived biomass through carbohydrate degradation, peptide degradation, respiration, and nutrient-scavenging pathways. This phase-guided metaproteomic framework will test whether bloom transitions are marked by reproducible peptide-level signatures and whether specific molecular processes precede visible bloom decline. Together, this work moves harmful algal bloom monitoring beyond reactionary water sampling toward a mechanistic, protein-level understanding of microbial ecosystem transitions. By using metagenomic ecological states to guide interpretation of high-resolution DIA metaproteomics, this project establishes metaproteomics as a powerful tool for resolving active microbial and viral controls on coastal bloom dynamics.

Authors
Julien Margaret A. Dagan, Alesi R. Escobedo, Anran Yu, Miklos Guttman

Institutions
University of Washington

Abstract
Hydrogen Deuterium Exchange-Mass Spectrometry (HDX-MS) is routinely employed for probing protein structures and conformational dynamics. However, its reproducibility is limited by back-exchange after following labeling, which complicates exchange rate measurements and data interpretation. While most back-exchange occurs in solution or during ionization, some studies have also observed deuterium loss post-ionization within the MS. We hypothesize that ambient water is entering the MS and contributing to deuterium loss. Here, we quantitatively track deuterium loss using EDTA and P1 peptide to systematically evaluate water vapor-driven back exchange in both negative and positive modes across five widely used MS platforms: Thermo Orbitrap Tribrid (Ascend and Eclipse), Waters Synapt G2-Si, Thermo Finnegan LTQ, and Bruker timsTOF fleX. This study aims to define the magnitude of water vapor effects and highlight platform-dependent differences.

Authors
Anais Gentilhomme1,  Chris Hsu1, Bo Wen1, Justin A. Sanders1, William S. Noble1, Laura Zelter2, Michael Riffle1, Susan Q. Lang3, William J. Brazelton4, and Brook L. Nunn1

Institutions
1University of Washington (Department of Genome Sciences) 2Florida State University (Department of Chemistry & Biochemistry) 3Woods Hold Oceanographic Institution (Department of Geol. and Geophys. and Marine Chem. and Geochem.) 4Blue Marble Space Institute of Science (Seattle, WA)

Abstract
We present current efforts in an ongoing investigation to develop and evaluate low-input metaproteomic workflows tailored to Lost City Hydrothermal Field (LCHF) chimneys, aiming to identify expressed microbial functions. LCHF is an ultramafic, serpentinizing submarine hydrothermal vent system characterized by high pH (9-11), temperatures of ~100℃ in venting fluids, and elevated concentrations of formate, methane, and sulfide that can support C1-metabolizing organisms (Kelley et al. 2005, Brazelton et al., 2006). These characteristics make LCHF an ideal study site and analog for serpentinizing systems predicted to be on icy ocean worlds such as Europa and Enceladus. Metaproteomics allow us to both detect and measure microbial functional responses to environmental conditions. Metagenomic datasets for LCHF chimneys and venting fluids are available (Alian et al., 2025, Brazelton et al., 2022), but no metaproteomic studies have been reported. The goal of this study is in two phases, to first identify chemosynthetic-associated peptides within these chimneys and the second is to then apply a targeted proteomics approach on 1.5 km of LCHF uncharacterized rock core to detect peptides that are relevant for life detection in the subsurface of ocean worlds with serpentinizing hydrothermal systems. Samples from five different active LCHF chimneys and three carbonate veins were sampled during the 2018 ROV Jason Lost City expedition (i.e. Brazelton et al., 2022, Moore et al., 2021). These samples are not only composed of a complex matrix but also have low cell counts ( < 2E76 cells/g of chimney rock) and therefore have inherently low protein concentration (20-100ng total protein). For phase one metaproteomic investigation, one of the chimneys with the highest amount of predicted biomass, Sombrero, was selected for methods and search development and prepared in technical triplicate. The protein fraction was isolated, and the cellular biomass solubilized by a combination of mechanical and chemical disruption using an in-house EDTA/SDS solution. Proteins were purified using High Recovery S-trap to isolate and wash the proteins embedded in the complex matrix, followed by trypsin digestion to peptides. Then digested peptides were analyzed using an Orbitrap Astral mass spectrometer coupled to a Vanquish Neo UHPLC system and the data were acquired using a data dependent acquisition (DDA). It was searched against metagenome-derived protein database that was chimney-specific and contained the top 100k most abundant contigs using Comet and Percolator (Peptide Spectral Match (PSM) False Discovery Rate < 0.01) where results were visualized and analyzed in limelight and R (4.3.2) (Chambers et al., 2012, Eng et al., 2013, Käll et al., 2007, Riffle et al., 2025a,b). Across these technical replicates, peptide-spectrum matches were detected under strict false discovery control of 0.01, though peptide recovery was low, which is consistent with the low biomass of this system. Despite these challenges, we detected about ~1400 LCFH- associated unique peptide sequences across samples. In addition, we are currently comparing our metagenomic based proteomic identification findings with an agnostic de novo approach to identifying peptides without relying on a reference database. This approach is especially relevant for life detection applications as it presents the evaluation of peptide level biomarkers in environments where genomic context may be absent.

Authors
Abanoub Mikhael, Gillian Leach and David Goodlett

Institutions
UVic Genome BC Proteomics Centre, Victoria, BC, Canada

Abstract
Identifying post-translational modifications (PTMs) in proteomics remains challenging because conventional database-search workflows rely on predefined modification lists, sites or numbers. As a result, modified peptides may remain unidentified when a modification is not included in the search parameters, or when tandem MS fragmentation patterns are incomplete or insufficient for confident assignment. Here, we introduce Library-Assisted Mass Defect Analysis (LAMDA), a mass-defect-guided framework that uses curated structural libraries (e.g., modification lists and protein sequences) to define a peptide’s chemically plausible modification search space. In addition to accurate mass, LAMDA constrains the search space using experimental conditions such as protease rules, protein sequence information, and a modification library. Across glycosylation, phosphorylation, acetylation, and DSSO-based cross-linking, LAMDA generated hypotheses that enabled rapid validation with existing tandem mass spectra and importantly recovered assignments missed by conventional database searches. These results establish LAMDA as a general, discovery-oriented workflow that bridges accurate-mass analysis, chemical reasoning, and tandem MS confirmation. Although demonstrated here primarily for peptide modifications, the framework is readily adaptable to other analyte classes when appropriate structural libraries are available. Overall, LAMDA enables an objective, discovery-driven analysis by allowing the data to define the search space.

Authors
Reta Birhanu Kitata1*, Zhangyang Xu1, Nadia Bayou2, Rui Zhao3, William B Christler1, Matthew J Gaffrey1, Karl K Weitz1, Mara Serena Serafini2, Elisabetta Molteni2, Vladislav A Petyuk1, Massimo Cristofanilli2, Tao Liu1, Carolina Reduzzi2, Tujin Shi1*

Institutions
1Biological Sciences Division, Pacific Northwest National Laboratory, Richland, WA 99354, USA 2Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine, New York, NY 10021, USA 3Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, Richland, WA 99354, USA

Abstract
With advanced mass spectrometry (MS)-based proteomics, deep proteome coverage can be achieved from bulk cells. However, such bulk measurement obscures cell to cell heterogeneity, precluding proteome profiling of single cells and small numbers of cells. Recent advances in MS-based single-cell proteomics (SCP) have revolutionized the SCP field for comprehensive characterization of cellular heterogeneity. We and others have recently developed multiple convenient SCP methods (e.g., surfactant-assisted one-pot single-cell sample preparation termed SOP) based on standard 96-, 384-PCR well plates or low-bind PCR tube.

However, they have different types of shortcomings. Herein we report an easily adoptable improved Surfactant-assisted One-Pot (iSOP) method enabling robust low-volume single-cell processing on 384-well plate for high throughput SCP.

iSOP-MS capitalizes on processing of single cells at an easily manageable volume of ~3 µL, which can be fully automated for single cells. With a commonly accessible LC-MS platform, iSOP-MS quantified ~1,200–1,800 protein groups from single HeLa or MCF7 cells. Application of iSOP-MS to two neuroblastoma cell lines has enabled reliable identification of an average of ~1,700 and ~2,050 protein groups from single BE2-C and SK-N-SH cells, respectively, and precise characterization of cellular heterogeneity. Overall, iSOP-MS provides a robust and convenient platform for routine, cost-effective, quantitative SCP analysis.

Authors
Tim S. Veth1, Emmajay Sutherland1, David Bergen2, Rafael D. Melani2, Christopher Mullen2, and Nicholas M. Riley1*

Institutions
1Department of Chemistry, University of Washington, Seattle, WA, USA 2Thermo Fisher Scientific, San Jose, CA, USA

Abstract
Glycan heterogeneity is a fundamental property of glycoproteins, but the high degree of glycosite-level heterogeneity leads to technical challenges in measuring glycoproteoforms. A holistic understanding of glycan modification states is critical to translating glycoproteome regulation to biological function, but bottom-up glycoproteomics cannot recapitulate the full ensemble of glycoproteoforms from glycopeptide measurements alone. Promising efforts to profile masses of intact glycoproteins have recently explored data-independent acquisition (DIA) coupled with proton-transfer charge reduction (PTCR) or electron-capture-induced charge reduction mass spectrometry (MS). While valuable for generating broad glycoproteoform mass distributions, these approaches have remained limited in their ability to generate discrete glycoproteoform mass measurements, largely because they rely on low-resolving power measurements and deconvolution that does not account for isotopic information. To this end, we developed a DIA-PTCR workflow that couples high-resolving power (Rp ~240,000 at m/z 200) tandem mass spectra with an open-source processing suite to define glycoproteoform populations within 20 ppm mass accuracy thresholds. We demonstrate the glycoproteoform characterization capabilities of this platform using a collection of glycoproteins with well-described translational interests (EpCAM, TIGIT, CD40, PDL1, and CD24). We showcase how intact glycoproteoform masses acquired using our high-resolving power DIA-PTCR (hRp-DIA-PTCR) approach can be integrated with bottom-up intact glycoproteomics and Direct-Mass Technology (i.e., Orbitrap-based charge-detection MS) acquisitions to inform structural and biological insights. Altogether, our hRp-DIA-PTCR method extends the current capabilities of intact glycoprotein analyses by enabling robust characterization of isotopically resolved proteoforms and facilitating deep biological interpretation of glycosylation heterogeneity. Our open-source informatics platform includes a GUI-based tool called PTsliCR to clean PTCR scans directly from DIA-PTCR raw files and a deconvolution R package called IsoTrac, both of which are freely available on GitHub at https://github.com/riley-research.

Authors
Matthew D. Berg, Alexis Chang, Kyle Hess, Ricard A. Rodriguez-Mias, Judit Villén

Institutions
Department of Genome Sciences, University of Washington, Seattle, WA, USA

Abstract
DNA sequencing has identified millions of natural genetic variants that alter protein sequence. However, determining the functional impact of these variants remains challenging. Traditional mutagenesis approaches are not scalable for millions of variants and high-throughput approaches such as deep mutational scanning are limited to investigating one protein per experiment. To address this, our lab has developed Miro – a high-throughput proteomic approach to functionally annotate the impact of missense mutations across entire proteomes. In this approach, stochastic errors in protein synthesis are induced to create amino acid substitutions throughout all expressed proteins within a cell. Biochemical selections that probe general protein properties like solubility, thermal stability, ligand binding, protein-protein interactions and post-translational modifications are then applied to the collection of protein variants. After selection, variants are quantified by mass spectrometry to determine the functional impact of each mutation on the measured property. Here, we harness a collection of tRNAs that we engineered to mis-incorporate alanine at non-alanine codons and determine the impact of over 50,000 alanine substitutions on the thermal stability of more than 2000 yeast proteins. We find impactful substitutions are enriched at protein-protein interfaces and uncover both known and novel residues that facilitate these interactions. Our work establishes proteome-scale alanine scanning as a powerful approach for mapping mutational sensitivity and for uncovering the structural and functional determinants of protein thermal stability.

Authors
Lauren Fields1, Bo Wen1, Chris Hsu1, Deanna L. Plubell2, Philip M. Remes2, Lilian R. Heil2, Michael J. MacCoss1

Institutions
1Department of Genome Sciences, University of Washington, Seattle, Washington, 2ThermoFisher Scientific, San Jose, CA, United States

Abstract
Introduction Despite the widespread adoption of data-independent acquisition (DIA), linear ion trap (LIT) instruments have been largely overlooked in favor of high resolution accurate mass (HRAM) instruments. LITs have several strengths, including fast scan speed and sensitivity, alongside practical benefits of being cost effective. While challenges remain in adapting unit-resolution instrumentation to DIA proteomics, we show that the acquisition speed of the Thermo Stellar MS, combined with emerging computational strategies, can minimize limitations of instrument mass accuracy. We demonstrate a stepwise approach to improve DIA performance by constraining search space, correcting local space charge effects, and refining retention time (RT) predictions, achieving improved precursor and protein detections. These results highlight the potential of unit-resolution platforms for DIA proteomics. Methods Tryptic peptides from HeLa cell lysates were analyzed by the Thermo Scientific Orbitrap Astral and Stellar MS using identical LC and analytical column setups to compare analyzer-specific performance. Gas-phase–fractionated DIA acquisitions spanning sequential 100 m/z windows were collected to generate spectrum libraries. A machine learning strategy was developed to correct local space charge effects and improve mass measurement accuracy of Stellar data. Performance was benchmarked across DIA-NN (v.1.8.0; 2.2.0), XCorrDIA, and EncycopeDIA.

DIA-NN v.2.2.0 was optimized for subsequent analyses using MS/MS-only searching with a filtered FASTA database at 1% FDR. Fragment ion intensity predictions were refined through Carafe (v.2) to account for Stellar-specific fragmentation patterns and run-specific RT. Preliminary Data or Plenary Speakers Abstract To adapt DIA frameworks originally designed around HRAM to unit-resolution instrumentation, we implemented a peptide-centric analytical strategy. We benchmarked multiple analysis engines (e.g., DIA-NN, EncyclopeDIA, XCorrDIA), across Stellar datasets, evaluating their reliance on mass accuracy versus alternative measured peptide properties. DIA-NN demonstrated robustness to reduced mass accuracy, leveraging other features (e.g., improved retention time from Carafe) to drive peptide identification. Performance was further improved by restricting analysis to MS/MS-only, eliminating unnecessary dependence on MS1 information, inherently limited on LIT platforms, producing a 74% increase in precursors detected relative to analyses performed with MS1 data. We hypothesized that a high-resolution reference proteome could bridge the performance gap between unit-resolution and HRAM platforms by constraining the peptide search space to experimentally observable candidates. By limiting the query space to peptides detected on the Orbitrap Astral, we reduced the search space while retaining sensitivity for observed peptides. This search space reduction proved impactful, increasing precursor detections by an additional 33% on the Stellar. To improve the mass measurement accuracy, we implemented a lightweight machine learning model designed to correct local space-charge effects and other systematic errors. This model improved the root mean square error by 27%. We addressed fragmentation and RT model mismatches by fine tuning the AlphaPeptDeep-based prediction model using Carafe. This enabled fragment ion predictions that more accurately reflect the behavior of the Stellar MS. Additionally, the RT was improved specific to the column and HPLC setup. These computational refinements yielded an additional 22% increase in precursor identifications, altogether lending an over 3-fold improvement in precursors compared to starting conditions, expanding detections from 18,755 to 59,201 precursors. Ongoing work extends this framework toward quantitative benchmarking, comparing beam-type and resonance CID fragmentation. Collectively, these results demonstrate a practical and scalable pathway for high-throughput DIA proteomics on a nominal linear ion trap instrument. Novel Aspect This work democratizes DIA proteomics, achieving results via a nominal mass linear ion trap, minimizing the gap with HRAM instruments.

Authors
Doudou Yu1,10, Kristine A. Tsantilas1,2,10, Michael Riffle1, Christine C. Wu1, Bo Wen1, Caitlin S. Latimer3, Gary A. Churchill4, Stephanie McGrath5,6,7, Julie A. Moreno5,7,8, William S. Noble1,9*, Michael J. MacCoss1*

Institutions
1Department of Genome Sciences, University of Washington, Seattle, WA, USA. 2Translational Science and Therapeutics Division, Fred Hutchinson Cancer Center, Seattle, WA, USA 3Department of Laboratory Medicine and Pathology, University of Washington, Seattle, WA, USA 4The Jackson Laboratory, Bar Harbor, ME, USA 5Brain Research Center, Colorado State University, Fort Collins, CO, USA 6Department of Clinical Sciences, College of Veterinary Medicine and Biomedical Sciences, Colorado State University, Fort Collins, CO, USA 7Center for Healthy Aging, Colorado State University, Fort Collins, CO, USA 8Department of Environmental & Radiological Health Sciences, College of Veterinary Medicine and Biomedical Sciences, Colorado State University, Fort Collins, CO, USA 9Department of Computer Science, University of Washington, Seattle, WA, USA 10These authors contributed equally

Abstract
Aging is a major risk factor for disease across species. Blood plasma, a minimally invasive biopsy of the whole body, offers a window into age-associated conditions such as neurodegeneration. Yet whether and how plasma proteomic signatures of aging are conserved across species remains understudied, for example when comparing inbred laboratory mice with genetically diverse companion dogs that share the environments and health-care systems of their human owners. Here we quantified the plasma extracellular vesicle (EV) proteome across the lifespan in 86 mice and 200 dogs, using Mag-Net, a species-agnostic EV-enrichment method coupled with data-independent acquisition mass spectrometry. Overcoming the dynamic-range challenge of the plasma proteome, Mag-Net measured 5,675 proteins from 63,125 precursors in mice and 6,848 proteins from 87,020 precursors in dogs. Aging clocks built from these plasma EV features accurately predicted chronological age within each species (correlation R = 0.92 in mice and R = 0.83 in dogs) and dementia in dogs, with top features implicating the inflammatory response. Notably, 38 protein features used to predict age in dogs were conserved across both species and have been repeatedly implicated in the aging human plasma proteome. Furthermore, we built a combined dog-mouse clock from shared protein features that predicted age in both species. Together, these results demonstrate that easily collected plasma EV proteomics can identify interpretable, conserved protein signatures of aging, including brain-aging signatures detectable in plasma that predict dementia in dogs. These conserved signatures may serve as cross-species biomarkers for evaluating aging interventions, streamlining a translational pipeline from laboratory mice to companion dogs and, ultimately, to humans.

Authors
Kayla A. Markuson, Chinmayee Deshpande, Nicholas M. Riley

Institutions
University of Washington

Abstract
Mass-spectrometry based glycoproteomics relies heavily on enrichment methods to isolate modified peptides from complex samples. However, no single enrichment strategy has been universally accepted as a standard in the field. The use of diverse chemistries, such as strong anion exchange and hydrophobic interaction liquid chromatography, as well as different enrichment platforms (e.g., solid-phase extraction), directly impacts reproducibility and the reliability of downstream data. Introducing automation to glycopeptide enrichment has the potential to improve sensitivity, selectivity, and reproducibility in glycoproteomics workflows used across the field. Additionally, automation decreases analyst hands-on time, allowing for high-throughput studies. Here we demonstrate an automated glycopeptide enrichment approach using strong-anion exchange magnetic beads. Experiments used cell lysates from human immortalized myelogenous leukemia K562 cells. Proteins were digested with trypsin and desalted with Strata-X SPE cartridges (Phenomenex). Glycopeptides were enriched with Oasis-MAX SPE cartridges (Waters) and strong-anion exchange magnetic beads (ReSyn BioSciences) and. Data were collected on an Orbitrap Ascend Tribrid MS (Thermo Fisher Scientific) coupled to a Vanquish Neo UHPLC system. Samples were separated online with a 25 cm, 75 µm ID Aurora Series Gen3 LC column packed with 1.6 µm C18 particles (IonOpticks).

Data acquisition used Autonomous Dissociation-type Selection utilizing a real-time library search to perform stepped HCD MS/MS for N-glycopeptides and EThcD for O-glycopeptides. Automated experiments were performed on the KingFisher Flex (Thermo Scientific). Data were processed using MSFragger-Glyco and GlyCounter.

Standard glycopeptide enrichment has historically involved solid-phase extraction (SPE) cartridges. However, the analyst time required for these workflows limits scalability, and the SPE format potentially limits sensitivity and selectivity. Current solid phase extraction methods, including SAX and HILIC, require upwards of 90 minutes to enrich up to twelve samples. In contrast, an automated platform using magnetic beads instead of SPE packing material can process 96-well plates and require ~20 minutes of analyst time for set up. Using strong anion exchange bead chemistry coupled to magnetic beads (ReSyn Bioscienes), we have successfully enriched glycopeptides from standard protein tryptic-based digests. The flexibility of an automated workflow enables simultaneous analysis of multiple variables, including bead chemistry, sample type, and load, wash, and elution conditions. Under optimized conditions, we observe an increase in the percentage of MS/MS spectra containing glycan-specific ions from 75 to 83%. By comparison, SPE methods designed to reduce analyst time, such as Oasis MAX, yielded a lower proportion of glycan-specific MS/MS spectra (71%). Our preliminary data also shows that this automated strategy maintains, and even improves upon, reproducibility achieved in slower SPE-based methods that are state-of-the-art for the field.

Additionally, bead-based enrichment eliminates additional SPE preparation steps, including cartridge activation and conditioning, thereby reducing workflow complexity and solvent usage. Enrichment performance was evaluated across a broad range of starting material (25–500 µg), confirming robust performance at higher peptide loads. The automated platform also accommodates bead volumes ranging from 5 µL to 100 µL per well without unintended bead loss during processing. When combined with our previously published automated dissociation-type selection mass spectrometry method optimized for glycopeptide analysis, this workflow yields a higher proportion of spectra containing glyco-associated ions. Overall, automated digestion and enrichment using platforms such as the KingFisher Flex System enable high-throughput, precise, and reproducible sample handling.

Authors
Christopher J. Weir, Anna B. Lin, Trung Nguyen, Matthew F. Bush

Institutions
University of Washington

Abstract
Introduction Temperature is often used as a way to modulate protein stability and control rates of reactions. Previous methods for incorporating temperature control into ESI have required long equilibration times and unwieldy hardware. Programmed-temperature electrospray ionization (ptESI) is a technique that modulates sample temperature as a function of time and analyzes the sample in real time using native MS. We are using ptESI to monitor processes including protein folding, the effects of ligands on protein stability, and to One example is the ability to monitor disulfide reduction in real time. To take these experiments to the next level, we have developed (1) a new ptESI source that takes advantage of 3D-printing and off the shelf hardware and (2) software to generate and visualize XML-based method files, control and track experiment progress in real time, and create SQLite files that encode the data and metadata for downstream integration with MS data.

Methods The ptESI source has a low thermal mass and modulates the temperature of liquid samples as a function of time according to user-defined temperature profiles. The source was positioned in front of a Waters Cyclic IMS system to enable the real-time characterization of ions using a variety of MS-based experiments. Ribonuclease A was chosen as a test case due to its four native disulfide bonds and strong thermal stability. Samples of ribonuclease A were prepared in aqueous 200 mM ammonium acetate at pH 7 with 5 mM dithiothreitol (DTT).

Control of the ptESI source is accomplished using a 3D-printed box that provides all necessary power, communication, and water cooling. It uses consumer-grade computer components and other commonly available hardware for a relatively low-cost solution. A codebase and GUI have been developed with AI assistance to provide easy methods for creating temperature profiles with XML sequences and gathering relevant experiment metadata and storing it in a stable SQLite format.

Results The ptESI source was characterized and it was found that liquid samples could be heated or cooled with high fidelity at rates exceeding ±60 °C⸱min–1. Additional experiments have shown that a wide variety of temperature profiles can be used for protein stability measurements.

Samples of ribonuclease A containing DTT were run through a temperature profile that went from 30 to 90 °C at 30 °C⸱min–1, during which time the sample was continuously analyzed by nESI-MS. The initial mass spectrum was consistent with the expected mass of the disulfide-intact protein; that spectrum persisted until ~90 °C and then rapidly began to change. The sudden change suggests that all four native disulfide bonds were reduced and the rate of that reaction may be strongly linked to protein structure and the accessibility of the reducing agent to the disulfide bonds. This process may be especially cooperative, i.e., the loss of the first disulfide bond preferentially favors the unfolded state and further increases the accessibility of the remaining disulfide bonds. Future work will refine the source and design and continue preliminary experiments applying real-time disulfide reduction to therapeutic antibodies of interest.

Authors
Kiran Iyer, Rosalynn Molden, Erin Weisenhorn, Hannah Townsend, Luis Fernandez-Ruiz, Shannon Hayes

Institutions
Just Evotec Biologics, Protein Metrics LLC

Abstract
Glycans can significantly influence the stability, activity, and immunogenicity of biologics, making their analysis crucial for assessing the quality of antibody therapeutics, especially biosimilars. Presented here is a tool to streamline mass spectrometry-based glycan analysis automated glycan grouping with software. From released glycans to peptide-level multi-attribute method (MAM) workflows, automated glycan grouping and intuitive visualizations can transform complex datasets into actionable insights. Work presented here shows the utility of glycan grouping in bridging data from multiple analytical methods such as MAM and released glycan analysis. Additionally, glycan grouping was used to compare day of culture samples from different bioreactors using mass spectrometry (MS)

Authors
Anna B. Lin, Christopher J. Weir, Trung Nguyen, Matthew F. Bush

Institutions
University of Washington

Abstract
Introduction: The measurement and characterization of protein stability is critical to many aspects of biopharmaceutics. In the context of quality control, understanding protein stability is essential for ensuring the consistency and integrity of manufactured therapeutics. In biopharmaceutical research, this same understanding is critical to probing many life-threatening diseases associated with protein misfolding. One such example is transthyretin (TTR), a homotetrameric protein implicated in the development of familial amyloid polyneuropathy (FAP) and senile systemic amyloidosis (SSA). Transthyretin can dissociate into individual monomers that are prone to misfolding and aggregation, leading to the formation of fibrils associated with these diseases. To date, more than 150 TTR mutations have been identified each with their own stability profile. Among these, the V30M variant is one of the most recognized, exhibiting accelerated amyloid formation relative to the wild-type protein. A promising strategy to combat this process involves the use of pharmacological chaperones, which work by shifting the thermodynamic landscape of protein folding while also providing kinetic facilitation of folding the desired structure. However, pharmacological chaperones have only been successfully developed for a handful of diseases, including tafamidis for TTR amyloidosis. The small number of approved pharmacological chaperones across all diseases emphasizes how challenging it is to identify molecules with these properties, highlighting the need for novel technology capable of enabling measurements that can identify viable candidates across many proteins. To meet this need, the Bush Lab has developed programmed-temperature electrospray ionization (ptESI), a method that modulates the temperature of liquid samples across varying gradients while enabling real-time mass spectrometry analysis. In this work, ptESI is used to characterize the kinetic stability of transthyretin, and tafamidis is applied to provide preliminary validation of ptESI as a ligand screening tool, with the broader goal of extending this approach toward identifying pharmacological chaperones for other protein misfolding diseases.

Methods: Wild-type and V30M transthyretin (3 uM tetramer) solutions are prepared in a blend of 200 mM ammonium acetate and 200 mM acetic acid for a final pH of 4. Samples are buffer exchanged and loaded into the ptESI source using nano-ESI borosilicate capillaries. Temperature programs spanning 40 °C to 75 °C are used to modulate the temperatures of solutions while MS data is continuously acquired using a Waters Cyclic IM-MS system.

Preliminary Data: Under acidic conditions (pH 4) and thermal stress, wild-type TTR exhibits partial tetramer dissociation at ~75 °C, reaching a plateau over a 10 minute hold at that temperature. Under identical conditions, V30M undergoes complete and significantly faster dissociation into its monomer subunits. In preliminary analysis, an exponential decay is fitted to each replicate for both species. Individual decay rate constants are used to determine a mean apparent decay rate constant and direct comparisons of the decay rates between the two species confirm that the mutation significantly destabilizes the tetramer and complete dissociation of V30M into subsequent monomer subunits is observed prior to the onset of WT-TTR dissociation. However, all samples appear to undergo subtle behavioral changes as a function of time after sample preparation, strongly suggesting that the samples are not at equilibrium. For both WT and V30M, this is reflected in differences in the initial delay period preceding dissociation as well as a shift in decay rates during dissociation. Despite the variance that this contributes to the data there is a significant difference between the stability of WT and V30M.

To validate ptESI as a tool for pharmacological chaperone screening, tafamidis is introduced to evaluate its effect on TTR thermal dissociation. When the tafamidis bound complex is subjected to identical conditions as before, a significant stabilizing effect is observed as expected. Tafamidis bound WT-TTR remains intact until a point at which dissociation begins at a significantly reduced rate relative to apo WT-TTR. Preliminary data has shown a similar stabilizing effect of tafamidis on V30M transthyretin as well. Ongoing work aims to continue characterizing this temperature dependent behavior and begin a screening of a small library of ligands.

Authors
Gina M Many 1 , Hasmik Keshishian 2 , Gregory Smith 3 , Natalie M Clark 2 , Gayatri Iyer 4 , Patrick Hart 2 , Malene E Lindholm 5 , Samuel Montalvo 5 , Zidong Zhang 3 , Christopher Jin 6 , James A Sanford 1 , Steven A Carr 2 , Joshua N Adkins 1 7 , D R Mani 2 , Sue C Bodine 8 , Scott Trappe 9 , Joseph A Houmard 10 , Nicolas Musi 11 , Kim M Huffman 12 , William E Kraus 12 , Lauren M Sparks 13 , Anna E Thalacker-Mercer 14 , Stuart C Sealfon 3 , Ashley Y Xia 15 , Daniel H Katz 5 , Christopher B Newgard 12 , Charles F Burant 4 , Paul M Coen 13 , Bret H Goodpaster 13 ; MoTrPAC Study Group

Institutions
1 Biological Sciences Division, Pacific Northwest National Laboratory, Richland, WA. 2 Proteomics Platform, Broad Institute of MIT and Harvard, Cambridge, MA. 3 Icahn School of Medicine at Mount Sinai, New York, NY. 4 Department of Internal Medicine, University of Michigan, Ann Arbor, MI. 5 Department of Medicine, Stanford University, Stanford, CA. 6 Department of Genetics, Stanford University, Stanford, CA. 7 Oregon Health and Science University, Portland, OR. 8 Oklahoma Medical Research Foundation, Oklahoma, University of Oklahoma, Oklahoma City, OK. 9 Human Performance Laboratory, Ball State University, Muncie, IN. 10 Department of Kinesiology, Human Performance Laboratory, East Carolina University, Greenville, NC. 11 Cedars Sinai Medical Center, Los Angeles, CA. 12 Duke University School of Medicine, Durham, NC. 13 AdventHealth Orlando, Translational Research Institute, Orlando, FL. 14 The University of Alabama at Birmingham, Birmingham, AL. 15 National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), National Institutes of Health.

Abstract
The Molecular Transducers of Physical Activity Consortium (MoTrPAC) was established to systematically characterize the molecular basis of the health benefits of exercise. Here, we present the integrative, multi-omics response of human skeletal muscle to acute endurance (EE) and resistance (RE) exercise. Distinct temporal responses were observed, with changes in ATAC-seq, phosphoproteome, and metabolome occurring before changes in the transcriptome and proteome. These distinct temporal multi-omic dynamics were used to identify transcriptional regulatory hubs converging around MEF2A and NFIC regulation of autophagy, angiogenesis and metabolism. Further, early RE-specific phosphoproteome signatures counteracted epigenetic modifications and downregulated transcripts involved in protein turnover. Additional findings include suppression of HIPK2/3 kinase signatures linked to the acute exercise regulation of sarcomeric proteins TTN, NEB, ANKRD2 and LMOD2. Our data demonstrate distinct temporal regulation across the multi-omic landscape of human skeletal muscle, with EE and RE eliciting common and unique molecular signatures.

Authors
Andrea I. Gutierrez, Lillian T. Tatka, Gaelle Mercenne, Julia Robbins, Daniele Canzani, Anastasiya Prymolenna, Sebastian J. Paez, William E. Fondrie, Lindsay K. Pino, Alexander J. Federation

Institutions
Talus Bioscience

Abstract
Introduction (120 words) STAT6 is a transcription factor (TF) central to immune regulation and implicated in cancer progression, making it a highly attractive therapeutic target. Upon activation, STAT6 translocates to the nucleus and binds chromatin to regulate genes involved in immune signaling, tumor microenvironment modulation, and disease progression. Despite its biological relevance, STAT6 remains a challenging drug target, as transcription factors generally lack conventional druggable binding sites, and limited successful small molecule approaches have succeeded. To address the challenges associated with drugging STAT6, we developed a high-throughput screening approach leveraging subcellular fractionation, quantitative chromatin proteomics, and machine learning-powered in silico compound screening to identify small molecule STAT6 inhibitors.

Methods (120 words) We adapted regulome profiling, our semi-automated high-throughput screening platform, to quantify changes in the chromatin-bound proteome, including STAT6, following compound treatment. Briefly, cells are first stimulated with IL-4 for 30 minutes to induce STAT6 nuclear translocation and chromatin binding. Compound treatment and stimulation is performed using an Echo acoustic dispenser, followed by subcellular fractionation to enrich DNA-bound proteins found in the chromatin fraction. Chromatin proteins are digested via SP3, and resulting peptides are quantified using DIA mass spectrometry on a timsTOF Ultra II. Data analysis is performed using DIA-NN version 1.8.1 on the quantms Nextflow pipeline.

Preliminary data (300 words) We first qualified the STAT6 target on our mass spectrometry-based regulome profiling platform. Specifically, we evaluated cell models, stimulation conditions, and treatment durations that maximize quantitative accuracy and reproducibility for chromatin-associated STAT6. We found that stimulating cells for 30 minutes with IL-4 was required to achieve abundant, reproducible, and quantifiable detection of STAT6, a critical optimization step required for reliable screening.

Second, we trained a machine learning algorithm on millions of protein-compound interactions and leveraged it to prioritize a set of candidate inhibitors from a library of 12,000 covalent compounds, reducing the experimental space of the primary screen. We empirically screened the selected small molecules in the STAT6 cell-based system and have thus far identified 25 preliminary hits that significantly inhibit the function of STAT6 chromatin-binding. Traditional drug screening efforts yield 0.1 to 1% hit rates on average, where our platform has achieved a 5% hit rate for this difficult target. Our target-agnostic screening platform captures compound-induced changes across the broader chromatin-associated proteome, providing insight into selectivity, potential off-target effects, and pathway-level consequences of STAT6 modulation. This systems-level view enables early assessment of both efficacy and specificity, critical considerations for advancing compounds toward therapeutic development. The identified hits are currently undergoing validation through orthogonal cell-based assays to confirm STAT6 target engagement and assess downstream functional effects on immune and cancer-related signaling pathways. The scalability and throughput of regulome profiling positions this approach as a powerful strategy for screening against difficult TF targets implicated in cancer and immune dysregulation. Ongoing validation efforts will prioritize lead compounds for advancement into a STAT6-focused drug discovery campaign.

Novel aspect (20 words) A high-throughput chromatin proteomics platform combining machine learning compound prediction and DIA-LCMS to drug challenging targets such as STAT6.

Authors
Michal Maes, Casey Propst, Alex Zelter, Winnie Wen, Michael MacCoss, Nina Isoherranen

Institutions
Department of Pharmaceutics, Department of Genome Sciences, University of Washington

Abstract
Identification of the enzymes involved in the metabolism of a drug candidate is critical in drug development to predict drug-drug interactions, interindividual variability and impact of disease states and pharmacogenetics on drug disposition. Despite their prevalence in drug metabolism, the enzymes carrying out hydrolysis reactions remain poorly characterized. While many drugs are known to undergo hydrolysis in vivo, the specific enzymes that contribute to amide and ester hydrolysis remain poorly characterized. This project focuses on characterizing in vitro hydrolase activity with the ultimate goal of identifying major drug hydrolase enzymes in human tissues and determining the relative importance of these enzymes in drug clearance. As the liver is one of the main metabolic organs, we prepared subcellular fractions from individual liver donors from the UW human liver bank. These were tested for hydrolase activity with compounds known to undergo hydrolytic metabolism in vivo. As proof of concept, we show that hydrolysis of Z-Arg-Arg-AMC, which is expected to be lysosomal, happens exclusively in the fractions containing cytosolic proteins. Interestingly, formation of the inactive metabolite of clopidogrel, clopidogrel carboxylic acid, was observed in both the cytosolic and the microsomal fractions although CES1, the enzyme suggested to be responsible for this reaction, is localized to the ER lumen. The abundance of CES1 was quantified using DIA-based proteomic analysis. CES1 abundance showed a significant correlation with clopidogrel hydrolysis activity in microsomes. In contrast, clopidogrel hydrolysis activity in the cytosolic fraction did not correlate with CES1 abundance. These findings suggest that, in addition to CES1, another as-yet-unidentified cytosolic enzyme is also involved in clopidogrel hydrolysis. As hydrolysis pathways play essential roles in numerous biochemical processes across diverse tissues, we investigated the proteomic expression of hydrolase enzymes in a variety of human tissues. As adipose tissue showed a high abundance of CES1, we tested whether clopidogrel hydrolysis activity was present in human omental and subcutaneous adipose tissue. The hydrolysis activity appeared different between the omental and subcutaneous adipose depots, but the data suggest additional enzyme contributions beyond CES1 in adipose tissue as well. Further analysis of correlations between enzymatic activity assays and the full proteome data in individual donors and different tissues may allow identification of unknown hydrolase enzymes contributing to drug metabolism. These findings highlight the power of combining DIA-derived protein abundance with activity assays to identify drug metabolizing hydrolase enzymes.

Authors
Conor P Herlihy, Elijah Biletch, Lidan Li, Olivia Weissenfels, Keriann M. Backus, Brian J Beliveau, Devin K Schweppe

Institutions
University of Washington, University of California at Los Angeles

Abstract
Regulating nuclear structure is paramount to carrying out fundamental cellular processes and preventing onset of disease. Much of that maintenance is mediated through regulating the diverse protein, DNA, and RNA composition of nuclear compartments. In spite of this critical function, the core molecular composition of many nuclear compartments are still not fully understood. Deciphering de novo multiomic environments of specific proteins and nucleic acid loci in an unbiased and comprehensive manner is challenging with the currently available approaches. To overcome this challenge, we developed a proximity labeling platform using photosensitizer-dependent oxidation and capture by amine (POCA). Our approach enables target specific proteome identification using photosensitizer conjugated oligonucleotides or antibodies and proximity labeling followed by affinity purification and mass spectrometry. Thus, developing the first oligo-directed photocatalytic labeling approach. To better understand the molecular composition of nuclear speckles, nucleoli, and constitutive heterochromatin, we leveraged POCA targeting their defining molecular components. Employing complementary approaches has revealed the diverse molecular composition of these compartments. For example, targeting both SON and MALAT1 at nuclear speckles shows significant enrichment of 24 proteins exclusive to Malat1, 226 proteins exclusive to SON, and 56 shared proteins. Additionally, targeting both NPM1 and ITS1 has elucidated local proteomes unique to the periphery and the core of the nucleolus. Together these findings establish the flexibility of our method for quantitative proximity labeling proteomics targeting both nucleic acids and proteins to define sub-compartment composition and better understand nuclear structure.

Authors
Jinyu Liu, Alex Zelter, Aprajita Yadav, Michael Riffle, Lindsay C. Czuba, Yue Wen, Michael MacCoss, Katya Rubinow, Nina Isoherranen

Institutions
University of Washington

Abstract
Adipose tissue serves as an energy reservoir and as an active endocrine organ, playing a central role in metabolic homeostasis. Subcutaneous (SC) and omental (OM) adipose depots are functionally and histologically distinct, with OM adipose tissue more strongly associated with insulin resistance and metabolic disease. Retinoids, including retinol and its bioactive metabolite, all-trans retinoic acid (atRA), are essential signaling molecules that regulate lipid metabolism, immune function, and cellular differentiation. Animal models have suggested that retinoid homeostasis is altered in obesity. Yet retinoid concentrations in human adipose tissue depots have not been previously measured, and the relationship between tissue retinoids and the local proteome remains uncharacterized. We hypothesized that adipose tissue retinoid homeostasis is regulated locally in an autocrine manner, and that depot-specific retinoid concentrations correlate with both the tissue proteome and systemic metabolic markers.

Paired SC and OM adipose biopsies were collected from 31 non-diabetic participants (20 female, 11 male; age 25–65 years; BMI 21–56 kg/m²) during elective surgeries. Tissue concentrations of retinol, all-trans retinoic acid, and 13-cis retinoic acid were quantified by LC-MS/MS. Proteomic profiles were generated using LC-MS/MS-based data-independent acquisition (DIA). Depot-specific differences in retinoid concentrations and proteomic signatures were identified, and correlations between retinoid concentrations and proteomic and clinical variables (BMI, insulin, adiponectin, and leptin) were tested.

Retinoid concentrations differed significantly between SC and OM depots, and the relationship between tissue and circulating retinoid levels was depot dependent. Consistent with these differences, ALDH1A2, a key retinoic acid-synthesizing enzyme, was also differentially expressed between depots. Tissue or serum retinoid levels were significantly correlated with BMI, leptin, and adiponectin. Proteomic analyses identified retinoid-metabolizing enzymes, including members of the ADH, ALDH, and AOX families, as significantly correlated with tissue retinoid concentrations across donors. Notably, ADH7 was associated with multiple retinoid species in both depots. The retinoid-correlated protein networks were distinct across depots and retinoid species, indicating depot-specific regulatory patterns. Finally, a predictive model integrating retinoid metabolism-related proteins and circulating retinoid levels was developed for estimating tissue retinoid concentrations.

This study provides the first characterization of depot-specific retinoid profiles in human adipose tissue, linking them to the local proteome and identifying the enzymatic basis for inter-depot differences. Correlations between tissue retinoid levels, adipokines, and metabolic markers further support a functional connection between retinoid signaling and lipid metabolism in adipose tissue.

Authors
Kayleigh Voos1*, Gayatri Iyer2*, Anne Marie Weitzel3, Gina Many1, Laurie Goodyear4, Sue Bodine5, Karyn Esser6, Joshua Adkins1, Charles Burant3 and the MoTrPAC Study Group

Institutions
1 Pacific Northwest National Laboratory, Richland, WA. 2 Department of Computational Medicine and Bioinformatics, University of Michigan, Ann Arbor, MI. 3 Department of Internal Medicine , University of Michigan, Ann Arbor, MI. 4 Joslin Diabetes Center, Harvard Medical School, Boston, MA. 5 Oklahoma Medical Research Foundation, Oklahoma City, OK. 6 University of Florida, Gainesville, FL.

Abstract
Skeletal muscle plays a central role in generating force for movement and mediating adaptive responses to exercise. In this study, we examined the temporal molecular response of rat skeletal muscle to a single bout of acute endurance exercise across multiple molecular layers, from immediately post-exercise to 48 hours post-exercise. We observed that the phosphoproteome and epigenome exhibited the most robust early responses to exercise. Notably, several molecular links were identified between phosphoproteomic changes and epigenomic regulation, suggesting a coordinated interplay between these layers. Additionally, peak-to-gene correlation analyses revealed associations between epigenetic changes and corresponding alterations in transcriptional programs related to protein degradation and folding, transcription factor activity, and aerobic respiration. Multi-omics factor analysis (MOFA) across molecular layers further revealed a subset of exercise-responsive and sex-dependent modules with clear temporal dynamics. These included proteins and genes associated with mitochondrial respiration, cytoskeletal remodeling, sarcomere assembly, RNA metabolism, and inflammatory responses. Furthermore, temporal clustering of metabolomic, phosphoproteomic, and transcriptomic data revealed early, middle, and late molecular programs responsive to an acute exercise bout. Together, these data provide a time-resolved molecular characterization of the skeletal muscle response to acute exercise in untrained female and male rats. These findings contextualize early molecular responses that precede muscle adaptation to training and enhance our understanding of muscle biology on the minute-to-hour timescale following a single bout of endurance exercise.

Authors
Gayatri Mishra1*, Anna Berim1, David Roger Gang

Institutions
1Institute of Biological Chemistry, The Washington State University, Pullman, WA 99164, USA

Abstract
Popping beans are derived by crossing indigenous South American nuña beans with temperate-adapted common bush beans. Popping beans are a unique group of Phaseolus vulgaris L. that expand upon heating, similar to popcorn. This emerging crop is attracting interest because of its unique popping trait, making it well-suited for nutritious, protein- and fibre-rich snack foods. Despite their culinary and nutritional potential, the biochemical basis of this distinctive popping behaviour remains poorly understood. Here, we present the first integrated proteomic and metabolomic analysis of diverse popping bean accessions and compare them with commercial common bean accessions to identify protein and metabolite signatures associated with the popping trait. Whole seeds from six popping bean and six commercial bean accessions were powdered and analysed using liquid chromatography coupled with tandem mass spectrometry (LC–MS/MS) with electrospray ionisation (ESI). A total of 1,878 proteins were identified across all accessions. Comparative proteomic analyses revealed distinct protein signatures between popping and commercial beans. KEGG pathway enrichment analysis identified glycan degradation, valine/leucine/isoleucine biosynthesis, pantothenate and CoA biosynthesis, and porphyrin metabolism as major enriched pathways in popping beans, whereas commercial bean accessions were characterised by enrichment of proteins associated with endoplasmic reticulum protein processing, including molecular chaperones and protein-folding proteins. Metabolomic profiling further differentiated the two bean groups. Popping beans exhibited higher abundances of raffinose, stachyose, flavonoids (quercetin, myricetin, epicatechin, and astragalin), and organic acids. In contrast, commercial bean accessions were enriched in amino acid-related metabolites, including histidine, ornithine, serine, and urocanic acid, together with several phenolic acid derivatives. Integration of proteomic and metabolomic datasets revealed distinct biochemical signatures that differentiate popping and commercial bean seeds.

This work provides new insights into the biochemical basis of the popping trait and identifies distinct protein and metabolite signatures associated with popping beans. These findings enhance our understanding of the nutritional composition of popping beans and support their potential development as a novel, nutritious food crop and source of value-added food products.

Keywords: Emerging crop; LC–ESI–MS/MS; Metabolomics; Nutritional composition; Popping trait; Proteomics; Seed proteins; Storage proteins.

Authors
Kristine A. Tsantilas, Jeffrey R. Whiteaker, Myung Sik Jeong, Rose C. Pletcher, Jeremy Hoog, Lei Zhao, Hong Wang, Max Short, Ian Hagemann, Regine M. Schoenherr, Uliana J. Voytovich, Richard G. Ivey, ChenWei Lin, Foluso Ademuyiwa, Jingqin Luo, Cynthia X. Ma, Amanda G. Paulovich

Institutions
1. Fred Hutchinson Cancer Center, Seattle, WA 2. Washington University in St. Louis, St. Louis, MO

Abstract
Antibody-drug conjugates (ADCs) are monoclonal or bispecific antibody-based therapeutics with an increasing variety of cellular antigen targets that can deliver a cytotoxic “payload” to tumors. As the number of available ADC antigens has increased, the need to match patients with potential therapies has grown. Classically, this required quantifying target proteins separately with immunohistochemistry. We sought to remove this bottleneck by applying a multiplexed immuno-PRM assay (“ADC-PRM panel”) in patient selection by measuring 7 ADC target antigens. Furthermore, we expanded the panel to include characterization of a DNA-damage payload target and the pharmacodynamic response associated with successful payload delivery.

The ADC-PRM assay quantifies 128 peptides derived from 64 proteins. The workflow entails isotope dilution with cleavable isotopic standards coupled with peptide immunoaffinity enrichment of endogenous and internal standard peptides with 106 antibodies. LC-MS analysis of 256 precursors in a single 32.5-minute gradient was performed using an Evosep ONE (Whisper Zoom SPD40) coupled to a PRM method on an Orbitrap Astral mass spectrometer (Thermo). Basic performance metrics of the ADC-PRM panel were characterized in our CLIA-CAP environment in two matrices: LCL57 cell line exposed to 10GY of radiation (LCL57-10GY) and pooled frozen breast tumor lysate from two individuals. The assay was applied to a retrospective cohort of frozen breast cancer patient biopsies treated with HER2-directed antibody therapy.

The panel quantifies 7 tumor antigens currently targeted using monoclonal or bispecific ADCs (B7H3, B7H4, EGFR, FOLR1, HER2, HER3, PDL-1) including 2 that are clinically approved (HER2, FOLR1). Downstream, we measure a common DNA-damage agent payload target (TOP1), the pharmacodynamic response (DNA-damage response, “DDR”), and additional relevant downstream pathways including RAS signaling, RTK response, and proteins involved in the regulation of cell cycle, cell death, survival, and proliferation.

In preliminary studies, we utilized LCL57-10GY and frozen breast tumor lysate to evaluate linearity and protein input requirements of the assay. In a peptide response curve, 72% of the analytes the panel measures have an estimated lower limit of quantification (LLOQ) of 100 fmol or lower. We could detect endogenous signal from 104/128 peptides (77%) in 250 μg of frozen breast tumor. We found most of the detected analytes were robust down to 25 μg of input where 98/128 peptides were detected - including 10/35 phosphopeptides. This is crucial for sample-limited patient biopsies.

We applied this ADC-PRM assay to a cohort of 47 retrospective samples from the PALTAN trial – a human clinical trial of breast cancer patients on a neoadjuvant treatment including trastuzumab, palbociclib, and letrozole. This included samples collected from patients at baseline prior to treatment, during treatment Cycle 1 on Day 15, and at surgery after completing 4 treatment cycles where each treatment cycle was 28 days. Preliminary results showed that we can quantify proteins associated with the 3 therapeutics in the study. We measured the trastuzumab antigen (HER2) in all the samples and observed a reduction in the estrogen receptor (ESR1) after treatment with letrozole. We quantified proteins downstream of CDK4/6 (CDKN1A, CCNE1, and RB1) and known DDR targets (H2AX, NBN, NUMA1) that are induced by DNA damage caused by palbociclib.

Authors
Kayla A. Markuson, Chinmayee Deshpande, Nicholas M. Riley

Institutions
University of Washington

Abstract
Mass-spectrometry based glycoproteomics relies heavily on enrichment methods to isolate modified peptides from complex samples. However, no single enrichment strategy has been universally accepted as a standard in the field. The use of diverse chemistries, such as strong anion exchange and hydrophobic interaction liquid chromatography, as well as different enrichment platforms (e.g., solid-phase extraction), directly impacts reproducibility and the reliability of downstream data. Introducing automation to glycopeptide enrichment has the potential to improve sensitivity, selectivity, and reproducibility in glycoproteomics workflows used across the field. Additionally, automation decreases analyst hands-on time, allowing for high-throughput studies. Here we demonstrate an automated glycopeptide enrichment approach using strong-anion exchange magnetic beads. Experiments used cell lysates from human immortalized myelogenous leukemia K562 cells. Proteins were digested with trypsin and desalted with Strata-X SPE cartridges (Phenomenex). Glycopeptides were enriched with Oasis-MAX SPE cartridges (Waters) and strong-anion exchange magnetic beads (ReSyn BioSciences) and. Data were collected on an Orbitrap Ascend Tribrid MS (Thermo Fisher Scientific) coupled to a Vanquish Neo UHPLC system. Samples were separated online with a 25 cm, 75 µm ID Aurora Series Gen3 LC column packed with 1.6 µm C18 particles (IonOpticks).

Data acquisition used Autonomous Dissociation-type Selection utilizing a real-time library search to perform stepped HCD MS/MS for N-glycopeptides and EThcD for O-glycopeptides. Automated experiments were performed on the KingFisher Flex (Thermo Scientific). Data were processed using MSFragger-Glyco and GlyCounter.

Standard glycopeptide enrichment has historically involved solid-phase extraction (SPE) cartridges. However, the analyst time required for these workflows limits scalability, and the SPE format potentially limits sensitivity and selectivity. Current solid phase extraction methods, including SAX and HILIC, require upwards of 90 minutes to enrich up to twelve samples. In contrast, an automated platform using magnetic beads instead of SPE packing material can process 96-well plates and require ~20 minutes of analyst time for set up. Using strong anion exchange bead chemistry coupled to magnetic beads (ReSyn Bioscienes), we have successfully enriched glycopeptides from standard protein tryptic-based digests. The flexibility of an automated workflow enables simultaneous analysis of multiple variables, including bead chemistry, sample type, and load, wash, and elution conditions. Under optimized conditions, we observe an increase in the percentage of MS/MS spectra containing glycan-specific ions from 75 to 83%. By comparison, SPE methods designed to reduce analyst time, such as Oasis MAX, yielded a lower proportion of glycan-specific MS/MS spectra (71%). Our preliminary data also shows that this automated strategy maintains, and even improves upon, reproducibility achieved in slower SPE-based methods that are state-of-the-art for the field.

Additionally, bead-based enrichment eliminates additional SPE preparation steps, including cartridge activation and conditioning, thereby reducing workflow complexity and solvent usage. Enrichment performance was evaluated across a broad range of starting material (25–500 µg), confirming robust performance at higher peptide loads. The automated platform also accommodates bead volumes ranging from 5 µL to 100 µL per well without unintended bead loss during processing. When combined with our previously published automated dissociation-type selection mass spectrometry method optimized for glycopeptide analysis, this workflow yields a higher proportion of spectra containing glyco-associated ions. Overall, automated digestion and enrichment using platforms such as the KingFisher Flex System enable high-throughput, precise, and reproducible sample handling.

Authors
Robert W. Seymour1, Adelle Priest1, Christian Garrard2, Ben Turley2, JD Logan2, Josh Brown2, John C. Price2, Samuel H. Payne1

Institutions
1) Biology Department, Brigham Young University, 2) Chemistry Department, Brigham Young University

Abstract
Understanding protein folding dynamics and how they change allows us to unlock greater insight into underlying disease processes and normal functioning. However, discovering the folding dynamics using high-resolution structure determination is slow and unfeasible at proteome scale. Mass spectrometry can be used to identify changes in folding states and therefore exposes a unique window into how experimental perturbations affect the stability of all proteins at once. The Iodination Protein Stability Assay (IPSA) probes stability by identifying solvent accessible YMCH residues across a chemical denaturant gradient. By applying IPSA across experimental conditions, we can identify shifts in stability due to factors such as drug treatment or genetic mutations. Although the biochemical assay is robust, it lacks a rigorous statistical method to determine whether differences in stability are meaningful. We implemented both a frequentist and a Bayesian method to perform the statistical analysis of the IPSA data. This parallel implementation allows for comparison between the standard and an underutilized, but promising, method. We found that the Bayesian method reports 6 times more robust changes compared to the frequentist method at the same confidence/credible cutoff. This statistical pipeline allows for the identification of more stability shifts than previously possible, extending the usefulness of these assays and promoting more discoveries.

Authors
Jose Humberto Giraldez Chavez, Joshua Hunsaker, Caleb Coons, Ryan T. Kelly, Samuel H. Payne

Institutions
Brigham Young University

Abstract
In recent years, the depth of single-cell proteomics has dramatically increased, with contemporary studies identifying upwards of 5,000 to 9,000 proteins per cell. Bioinformatics methods for assessing the statistical rigor of peptide/protein identification ensure that the reported data has a known false-discovery rate. However, robust methods to assess the accuracy of quantitative measurements currently do not exist. Because single-cell proteomics is dominated by ultra-low abundance peptide signals, we need an objective and rigorous metric to assert quantitative accuracy. Using a unique ground truth dataset, we developed a machine learning model that discriminates between accurate and erroneous peptide quantification. This model was trained on a single-cell aliquot Human, Yeast, E. coli proteome mixture dataset with predefined changes in abundance between a pair of runs. By comparing expected versus observed ratios, we generated labels for accurate and erroneous quantification. Our model predicts that a substantial fraction of single-cell proteins are not accurately quantified. Filtering these poorly quantified peptides can dramatically improve protein abundance estimates and overall dataset quantitative quality.

Authors
Michael Riffle, Alex Zelter, Brendan X MacLean, Michael J MacCoss

Institutions
University of Washington

Abstract
Exploratory data analysis is where scientific insight is created and where it is most quietly compromised. Working interactively with an AI assistant, a scientist can generate observations that vanish after the session, run many data-dependent tests whose cumulative false-discovery burden goes unrecorded, and produce results that cannot be easily regenerated from their inputs. We present the Findings Workflow, a formalized human–AI collaboration, encoded as a repository of agent definitions, skills, and conventions, that addresses these failures by making the finding the durable, first-class unit of analysis. As a scientist and an AI agent explore a dataset together, every substantive insight is automatically captured as a structured, numbered finding that pins its data version, code, parameters, and environment for exact regeneration; records its caveats and status; links into a curated finding graph; and is promoted to "validated" only after independent, blinded re-derivation. The system ships with baked-in scripts for many common data-exploration and statistical analyses, and with opinionated, principled rules for proper statistics, validation, script-writing, and research. Dedicated subagents conduct and track research efficiently, building a reusable, fact-checked knowledge base. Correctness is foundational: data-loading fidelity is rigorously tested and verified before any analysis, since a silent read error is a common-mode failure that defeats downstream validation. Reporting rules and agents then turn validated findings and research into sharable research reports that support downstream manuscript preparation. Grounded in proteomics but discipline-agnostic, it converts the implicit hazards of AI-assisted exploration into explicit, auditable safeguards, reframing the AI from a transient analyst into a curator of trustworthy scientific knowledge.

Authors
Christopher J. Weir, Anna B. Lin, Trung Nguyen, Matthew F. Bush

Institutions
University of Washington

Abstract
Introduction Temperature is often used as a way to modulate protein stability and control rates of reactions. Previous methods for incorporating temperature control into ESI have required long equilibration times and unwieldy hardware. Programmed-temperature electrospray ionization (ptESI) is a technique that modulates sample temperature as a function of time and analyzes the sample in real time using native MS. We are using ptESI to monitor processes including protein folding, the effects of ligands on protein stability, and to One example is the ability to monitor disulfide reduction in real time. To take these experiments to the next level, we have developed (1) a new ptESI source that takes advantage of 3D-printing and off the shelf hardware and (2) software to generate and visualize XML-based method files, control and track experiment progress in real time, and create SQLite files that encode the data and metadata for downstream integration with MS data.

Methods The ptESI source has a low thermal mass and modulates the temperature of liquid samples as a function of time according to user-defined temperature profiles. The source was positioned in front of a Waters Cyclic IMS system to enable the real-time characterization of ions using a variety of MS-based experiments. Ribonuclease A was chosen as a test case due to its four native disulfide bonds and strong thermal stability. Samples of ribonuclease A were prepared in aqueous 200 mM ammonium acetate at pH 7 with 5 mM dithiothreitol (DTT).

Control of the ptESI source is accomplished using a 3D-printed box that provides all necessary power, communication, and water cooling. It uses consumer-grade computer components and other commonly available hardware for a relatively low-cost solution. A codebase and GUI have been developed with AI assistance to provide easy methods for creating temperature profiles with XML sequences and gathering relevant experiment metadata and storing it in a stable SQLite format.

Results The ptESI source was characterized and it was found that liquid samples could be heated or cooled with high fidelity at rates exceeding ±60 °C⸱min–1. Additional experiments have shown that a wide variety of temperature profiles can be used for protein stability measurements.

Samples of ribonuclease A containing DTT were run through a temperature profile that went from 30 to 90 °C at 30 °C⸱min–1, during which time the sample was continuously analyzed by nESI-MS. The initial mass spectrum was consistent with the expected mass of the disulfide-intact protein; that spectrum persisted until ~90 °C and then rapidly began to change. The sudden change suggests that all four native disulfide bonds were reduced and the rate of that reaction may be strongly linked to protein structure and the accessibility of the reducing agent to the disulfide bonds. This process may be especially cooperative, i.e., the loss of the first disulfide bond preferentially favors the unfolded state and further increases the accessibility of the remaining disulfide bonds. Future work will refine the source and design and continue preliminary experiments applying real-time disulfide reduction to therapeutic antibodies of interest.

Authors
Michael MacCoss, Bo Wen, Nick Shulman, Mike Riffle, Matt Chambers, Brendan MacLean

Institutions
University of Washington, Department of Genome Sciences

Abstract
Open-source software for data-independent acquisition (DIA) proteomics has fallen behind closed and commercial tools, which now deliver faster searches, greater detection numbers, and fewer missing values. This gap matters for reproducibility, method development, and equitable access, since reproducible science depends on inspectable scoring and laboratories should not need a per-seat license to analyze their proteomics data. We asked whether a principal investigator who understands the underlying algorithms, but has written little code in the past fifteen years, could build a state-of-the-art DIA search tool using agentic AI coding tools (Claude Code). The result is Osprey, an open-source peptide-centric DIA search tool that complements Carafe and Skyline. Osprey parses common inputs (mzML, DIA-NN TSV, blib), calibrates retention time and m/z by LOESS regression, scores peptides with 21 spectrum and chromatogram features, detects peaks with a modified continuous wavelet transform, controls FDR at the peptide, precursor, and protein level, and performs a cross-run reconciliation step that forces a consensus retention time and eliminates missing values. Predicted fragment ion abundances, retention time, and ion mobility from Carafe are essential to the scoring. FDR control was validated with FDRBench. On Astral HeLa DIA files, Osprey runs faster than DIA-NN while detecting a closely overlapping set of peptides. The tool has since been ported to C# within ProteoWizard (OspreySharp), with substantially improved performance and identical results. Development requiring a clear plan and a verification strategy at every step (unit tests, plotted intermediate outputs, and cross-tool comparison). We will describe how Osprey was evaluated and demonstrate its performance on data from multiple instrument platforms with both accurate and nominal mass accuracy. Planned features include the use of ion mobility to improve selectivity and localization-specific scoring for post-translational modifications.

Authors
Conor P Herlihy, Elijah Biletch, Lidan Li, Olivia Weissenfels, Keriann M. Backus, Brian J Beliveau, Devin K Schweppe

Institutions
University of Washington, University of California at Los Angeles

Abstract
Regulating nuclear structure is paramount to carrying out fundamental cellular processes and preventing onset of disease. Much of that maintenance is mediated through regulating the diverse protein, DNA, and RNA composition of nuclear compartments. In spite of this critical function, the core molecular composition of many nuclear compartments are still not fully understood. Deciphering de novo multiomic environments of specific proteins and nucleic acid loci in an unbiased and comprehensive manner is challenging with the currently available approaches. To overcome this challenge, we developed a proximity labeling platform using photosensitizer-dependent oxidation and capture by amine (POCA). Our approach enables target specific proteome identification using photosensitizer conjugated oligonucleotides or antibodies and proximity labeling followed by affinity purification and mass spectrometry. Thus, developing the first oligo-directed photocatalytic labeling approach. To better understand the molecular composition of nuclear speckles, nucleoli, and constitutive heterochromatin, we leveraged POCA targeting their defining molecular components. Employing complementary approaches has revealed the diverse molecular composition of these compartments. For example, targeting both SON and MALAT1 at nuclear speckles shows significant enrichment of 24 proteins exclusive to Malat1, 226 proteins exclusive to SON, and 56 shared proteins. Additionally, targeting both NPM1 and ITS1 has elucidated local proteomes unique to the periphery and the core of the nucleolus. Together these findings establish the flexibility of our method for quantitative proximity labeling proteomics targeting both nucleic acids and proteins to define sub-compartment composition and better understand nuclear structure.

Authors
Lauren Fields1, Bo Wen1, Chris Hsu1, Deanna L. Plubell2, Philip M. Remes2, Lilian R. Heil2, Michael J. MacCoss1

Institutions
1Department of Genome Sciences, University of Washington, Seattle, Washington, 2ThermoFisher Scientific, San Jose, CA, United States

Abstract
Introduction Despite the widespread adoption of data-independent acquisition (DIA), linear ion trap (LIT) instruments have been largely overlooked in favor of high resolution accurate mass (HRAM) instruments. LITs have several strengths, including fast scan speed and sensitivity, alongside practical benefits of being cost effective. While challenges remain in adapting unit-resolution instrumentation to DIA proteomics, we show that the acquisition speed of the Thermo Stellar MS, combined with emerging computational strategies, can minimize limitations of instrument mass accuracy. We demonstrate a stepwise approach to improve DIA performance by constraining search space, correcting local space charge effects, and refining retention time (RT) predictions, achieving improved precursor and protein detections. These results highlight the potential of unit-resolution platforms for DIA proteomics. Methods Tryptic peptides from HeLa cell lysates were analyzed by the Thermo Scientific Orbitrap Astral and Stellar MS using identical LC and analytical column setups to compare analyzer-specific performance. Gas-phase–fractionated DIA acquisitions spanning sequential 100 m/z windows were collected to generate spectrum libraries. A machine learning strategy was developed to correct local space charge effects and improve mass measurement accuracy of Stellar data. Performance was benchmarked across DIA-NN (v.1.8.0; 2.2.0), XCorrDIA, and EncycopeDIA.

DIA-NN v.2.2.0 was optimized for subsequent analyses using MS/MS-only searching with a filtered FASTA database at 1% FDR. Fragment ion intensity predictions were refined through Carafe (v.2) to account for Stellar-specific fragmentation patterns and run-specific RT. Preliminary Data or Plenary Speakers Abstract To adapt DIA frameworks originally designed around HRAM to unit-resolution instrumentation, we implemented a peptide-centric analytical strategy. We benchmarked multiple analysis engines (e.g., DIA-NN, EncyclopeDIA, XCorrDIA), across Stellar datasets, evaluating their reliance on mass accuracy versus alternative measured peptide properties. DIA-NN demonstrated robustness to reduced mass accuracy, leveraging other features (e.g., improved retention time from Carafe) to drive peptide identification. Performance was further improved by restricting analysis to MS/MS-only, eliminating unnecessary dependence on MS1 information, inherently limited on LIT platforms, producing a 74% increase in precursors detected relative to analyses performed with MS1 data. We hypothesized that a high-resolution reference proteome could bridge the performance gap between unit-resolution and HRAM platforms by constraining the peptide search space to experimentally observable candidates. By limiting the query space to peptides detected on the Orbitrap Astral, we reduced the search space while retaining sensitivity for observed peptides. This search space reduction proved impactful, increasing precursor detections by an additional 33% on the Stellar. To improve the mass measurement accuracy, we implemented a lightweight machine learning model designed to correct local space-charge effects and other systematic errors. This model improved the root mean square error by 27%. We addressed fragmentation and RT model mismatches by fine tuning the AlphaPeptDeep-based prediction model using Carafe. This enabled fragment ion predictions that more accurately reflect the behavior of the Stellar MS. Additionally, the RT was improved specific to the column and HPLC setup. These computational refinements yielded an additional 22% increase in precursor identifications, altogether lending an over 3-fold improvement in precursors compared to starting conditions, expanding detections from 18,755 to 59,201 precursors. Ongoing work extends this framework toward quantitative benchmarking, comparing beam-type and resonance CID fragmentation. Collectively, these results demonstrate a practical and scalable pathway for high-throughput DIA proteomics on a nominal linear ion trap instrument. Novel Aspect This work democratizes DIA proteomics, achieving results via a nominal mass linear ion trap, minimizing the gap with HRAM instruments.

Authors
Alex Zelter[1], Ditte Iversen[2], Tore B Stage[2], Michael J. MacCoss[1], Nina Isoherranen[3]

Institutions
Departments of [1]Genome Sciences, [3]Pharmaceutics University of Washington, Seattle, WA, USA; [2]Clinical Pharmacology, Pharmacy and Environmental Medicine, Department of Public Health, University of Southern Denmark, Odense, Denmark

Abstract
Covalent modification of proteins by drugs and reactive metabolites creates a molecular record of chemical exposure and contributes to drug efficacy, toxicity, and immune-mediated adverse reactions. Despite decades of research identifying drug-protein adducts, the in vivo kinetics governing their formation, accumulation, and persistence in humans remain poorly understood due to the analytical challenges of detecting site-specific protein modifications in clinical samples.

Here, we developed a longitudinal clinical adductomics strategy to quantify the temporal dynamics of dicloxacillin-derived protein modifications in plasma collected during a controlled human drug-drug interaction study. We used discovery-based mass spectrometry to identify dicloxacillin-modified peptides in plasma proteins. We then developed a targeted proteomic assay capable of quantifying site-resolved drug adducts across plasma proteins in clinical samples. We applied this assay to samples collected at ten timepoints following dosing across three study days spanning one month of thrice-daily dicloxacillin administration, providing direct measurement of drug-protein adduct kinetics in humans at the amino acid level. Dicloxacillin adducts exhibited rapid formation after dosing, progressive accumulation during repeated exposure, and persistence beyond circulating drug elimination. Individual modification sites displayed distinct kinetic profiles, revealing site-specific adduct formation and elimination. Longitudinal analysis demonstrated substantial interindividual variability in the presence of adducted peptides despite controlled dosing. Unexpectedly, despite the approximately three-week half-life of circulating albumin, some of the dicloxacillin-albumin adducts declined substantially within hours after dosing, closely tracking predicted plasma dicloxacillin pharmacokinetics. These results establish clinical adductomics as a strategy for time-resolved measurement of covalent drug-protein modifications in humans. By capturing the formation, persistence, and dynamic remodeling of drug adducts, this approach provides a framework for understanding the role of protein adduct dynamics in drug response, adverse reactions, and personalized pharmacology.

Authors
Justin Sanders, William Noble, Uri Keich

Institutions
University of Washington department of Genome Science , University of Sydney Department of Statistics

Abstract
Introduction Recently, significant advances in de novo peptide sequencing have been made by using deep learning models trained on massive datasets of labeled mass spectra. However, a major roadblock limiting the adoption of these tools is a lack of rigorous false discovery rate (FDR) control for de novo predictions. When peptides are assigned to acquired mass spectra via database search, target-decoy competition is the standard method for estimating FDR. However, no analogous FDR-controlling procedure exists for the case of de novo peptide sequencing. Accordingly, we propose a method for controlling the FDR for de novo sequencing results which is applicable specifically in the setting where some, but not all, peptides in the sample come from a known sequence database. Most notably, our proposed procedure is strictly more powerful than database search, yielding the same set of discoveries among database peptides as a database search procedure of choice, along with an additional set of FDR-controlled external discoveries obtained via de novo sequencing.

Methods We consider a setup where the goal is to identify an additional set of de novo discoveries on only the set of spectra not identified by database search. In brief, our proposed procedure works by using correctly predicted reference peptides as an empirical estimate of the score distribution for correct de novo predictions. The distribution of scores assigned to non-reference de novo peptides is then treated as a mixture distribution of scores from correct de novo peptides, drawn from the empirical correct score distribution, and scores from incorrect peptides, drawn from some unknown null distribution. Mixture parameters are inferred based on weak assumptions on the distribution of de novo sequencing scores, allowing for direct estimation of the FDR on de novo discoveries.

Results We test our procedure on a collection of publicly available mass spectrometry runs from eight different experiments in four species. To evaluate the performance of our FDR control, we simulate the presence of true de novo peptides by holding out portions of the reference database. We then obtain a conservative estimate of the false discovery proportion (FDP) by calling all de novo discoveries from the held out portion of the reference correct, and all other discoveries incorrect. Based on this experiment, we evaluate both the validity and the power of our FDR control when the held out portion of the database ranges from ~5% all the way up to ~90%. This experiment offers a realistic simulation of the full range of common de novo applications, from metaproteomics and antibody sequencing, where a large portion of peptides are not in the reference, to immunopeptidomics and studies of alternate translation, where a tiny fraction of peptides are expected to be de novo. We find that FDR is well controlled in all settings. We then apply our method to three downstream applications: sequencing a monoclonal antibody, analyzing a cave bear sample, and exploring psyllid endosymbionts. In all three settings our FDR control procedure offers valuable statistical confidence to de novo analysis.

Authors
Jinyu Liu, Alex Zelter, Aprajita Yadav, Michael Riffle, Lindsay C. Czuba, Yue Wen, Michael MacCoss, Katya Rubinow, Nina Isoherranen

Institutions
University of Washington

Abstract
Adipose tissue serves as an energy reservoir and as an active endocrine organ, playing a central role in metabolic homeostasis. Subcutaneous (SC) and omental (OM) adipose depots are functionally and histologically distinct, with OM adipose tissue more strongly associated with insulin resistance and metabolic disease. Retinoids, including retinol and its bioactive metabolite, all-trans retinoic acid (atRA), are essential signaling molecules that regulate lipid metabolism, immune function, and cellular differentiation. Animal models have suggested that retinoid homeostasis is altered in obesity. Yet retinoid concentrations in human adipose tissue depots have not been previously measured, and the relationship between tissue retinoids and the local proteome remains uncharacterized. We hypothesized that adipose tissue retinoid homeostasis is regulated locally in an autocrine manner, and that depot-specific retinoid concentrations correlate with both the tissue proteome and systemic metabolic markers.

Paired SC and OM adipose biopsies were collected from 31 non-diabetic participants (20 female, 11 male; age 25–65 years; BMI 21–56 kg/m²) during elective surgeries. Tissue concentrations of retinol, all-trans retinoic acid, and 13-cis retinoic acid were quantified by LC-MS/MS. Proteomic profiles were generated using LC-MS/MS-based data-independent acquisition (DIA). Depot-specific differences in retinoid concentrations and proteomic signatures were identified, and correlations between retinoid concentrations and proteomic and clinical variables (BMI, insulin, adiponectin, and leptin) were tested.

Retinoid concentrations differed significantly between SC and OM depots, and the relationship between tissue and circulating retinoid levels was depot dependent. Consistent with these differences, ALDH1A2, a key retinoic acid-synthesizing enzyme, was also differentially expressed between depots. Tissue or serum retinoid levels were significantly correlated with BMI, leptin, and adiponectin. Proteomic analyses identified retinoid-metabolizing enzymes, including members of the ADH, ALDH, and AOX families, as significantly correlated with tissue retinoid concentrations across donors. Notably, ADH7 was associated with multiple retinoid species in both depots. The retinoid-correlated protein networks were distinct across depots and retinoid species, indicating depot-specific regulatory patterns. Finally, a predictive model integrating retinoid metabolism-related proteins and circulating retinoid levels was developed for estimating tissue retinoid concentrations.

This study provides the first characterization of depot-specific retinoid profiles in human adipose tissue, linking them to the local proteome and identifying the enzymatic basis for inter-depot differences. Correlations between tissue retinoid levels, adipokines, and metabolic markers further support a functional connection between retinoid signaling and lipid metabolism in adipose tissue.

Authors
Parker S. Reyes, Gabriel B. Wilson, Ivin C. Tait, Caleb T. Coons, Hsien-Jung L. Lin, Ryan T. Kelly, Samuel H. Payne

Institutions
Parker S. Reyes, Gabriel B. Wilson, Ivin C. Tait, Caleb T. Coons, Hsien-Jung L. Lin, Ryan T. Kelly, Samuel H. Payne

Abstract
Recent advances in LC-MS proteomics pose new constraints for existing pipelines and softwares. High-throughput LC-MS allows for increased statistical scale, enabling broader experimental complexity such as longitudinal and multi-condition projects. MSConnect 2.0 is a vendor-independent data and analysis management system designed to handle the current and future demands of high throughput proteomics.

MSConnect2.0 is the next iteration from our legacy software, MSConnect. Updates to the architecture were taken to make this tool a scalable, all-in-one software to aid researchers across the entire experimental process. We reimagined MSConnect2.0 to operate at the project level, redesigning all related schemas to aggregate data around projects down to sample level metadata. We also implemented a series of analytical modules. We developed a sequence run file generator that captures all sample/project level metadata and stores that metadata in the relational database. We also have the QC and system suitability module, that captures longitudinal data from HYE and PRTC.

These updates to our MSConnect2.0 platform provide the necessary features and architectural foundation to keep pace with high throughput LC-MS proteomics. These efforts in building a software that can grow with the community in its aid to further biological significance.

Authors
Doudou Yu1,10, Kristine A. Tsantilas1,2,10, Michael Riffle1, Christine C. Wu1, Bo Wen1, Caitlin S. Latimer3, Gary A. Churchill4, Stephanie McGrath5,6,7, Julie A. Moreno5,7,8, William S. Noble1,9*, Michael J. MacCoss1*

Institutions
1Department of Genome Sciences, University of Washington, Seattle, WA, USA. 2Translational Science and Therapeutics Division, Fred Hutchinson Cancer Center, Seattle, WA, USA 3Department of Laboratory Medicine and Pathology, University of Washington, Seattle, WA, USA 4The Jackson Laboratory, Bar Harbor, ME, USA 5Brain Research Center, Colorado State University, Fort Collins, CO, USA 6Department of Clinical Sciences, College of Veterinary Medicine and Biomedical Sciences, Colorado State University, Fort Collins, CO, USA 7Center for Healthy Aging, Colorado State University, Fort Collins, CO, USA 8Department of Environmental & Radiological Health Sciences, College of Veterinary Medicine and Biomedical Sciences, Colorado State University, Fort Collins, CO, USA 9Department of Computer Science, University of Washington, Seattle, WA, USA 10These authors contributed equally

Abstract
Aging is a major risk factor for disease across species. Blood plasma, a minimally invasive biopsy of the whole body, offers a window into age-associated conditions such as neurodegeneration. Yet whether and how plasma proteomic signatures of aging are conserved across species remains understudied, for example when comparing inbred laboratory mice with genetically diverse companion dogs that share the environments and health-care systems of their human owners. Here we quantified the plasma extracellular vesicle (EV) proteome across the lifespan in 86 mice and 200 dogs, using Mag-Net, a species-agnostic EV-enrichment method coupled with data-independent acquisition mass spectrometry. Overcoming the dynamic-range challenge of the plasma proteome, Mag-Net measured 5,675 proteins from 63,125 precursors in mice and 6,848 proteins from 87,020 precursors in dogs. Aging clocks built from these plasma EV features accurately predicted chronological age within each species (correlation R = 0.92 in mice and R = 0.83 in dogs) and dementia in dogs, with top features implicating the inflammatory response. Notably, 38 protein features used to predict age in dogs were conserved across both species and have been repeatedly implicated in the aging human plasma proteome. Furthermore, we built a combined dog-mouse clock from shared protein features that predicted age in both species. Together, these results demonstrate that easily collected plasma EV proteomics can identify interpretable, conserved protein signatures of aging, including brain-aging signatures detectable in plasma that predict dementia in dogs. These conserved signatures may serve as cross-species biomarkers for evaluating aging interventions, streamlining a translational pipeline from laboratory mice to companion dogs and, ultimately, to humans.

Authors
David Ross Hall1, Brianna I. Gonzalez1, Athena A. Schepmoes1, Thomas L. Filmore2, Tyler A. Sagendorf1, Bennett Drucker1, Karin D. Rodland3, J. Charles A. Lacson4,5, Clifton L. Dalgard6,7, Isaac Brownell8, Jerry S.H. Lee4,9,10,11,12, Craig D. Shriver4, Elaine S. Keung13, Liesl S. Grenier14, Vladislav A. Petyuk1, and Tao Liu1,*

Institutions
1 Biological Sciences Division, Pacific Northwest National Laboratory, Richland, WA, USA 2 Environmental Molecular Sciences Division, Pacific Northwest National Laboratory, Richland, WA, USA 3 Department of Cell, Developmental, and Cancer Biology, Oregon Health & Science University, Portland, OR, USA 4 Murtha Cancer Center Research Program, Department of Surgery, Uniformed Services University of the Health Sciences, Bethesda, MD, USA 5 The Henry M. Jackson Foundation for the Advancement of Military Medicine, Inc., Bethesda, MD, USA 6 Department of Anatomy, Physiology and Genetics, Uniformed Services University of the Health Sciences, Bethesda, MD, USA 7 The American Genome Center, Uniformed Services University of the Health Sciences, Bethesda, MD, USA 8 Dermatology Branch, National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Insti-tutes of Health, Bethesda, MD, USA 9 Ellison Medical Institute, Los Angeles, CA, USA 10 Department of Chemical Engineering and Material Sciences, Viterbi School of Engineering, University of Southern California, Los Angeles, CA, USA 11 Dornsife College of Letters, Arts and Sciences, University of Southern California, Los Angeles, CA, USA 12 Department of Medicine, Keck School of Medicine, University of Southern California, Los Angeles, CA, USA 13 Department of Pathology and Laboratory Medicine, Walter Reed National Military Medical Center, Be-thesda, MD, USA 14 Department of Dermatology, Walter Reed National Military Medical Center, Bethesda, MD, USA * Corresponding author: tao.liu@pnnl.gov

Abstract
Early detection determines melanoma outcomes, yet current screening relies on visual inspection that misses molecular changes preceding clinical diagnosis. To identify serum protein signatures of melanoma development, we leveraged the Department of Defense Serum Repository, analyzing 390 longitudinal serum samples from 73 melanoma cases and matched controls across four timepoints (4 and 2 years prior, diagnosis, and 2 years post-diagnosis) using data-independent acquisition mass spectrometry with Seer Proteograph nanoparticle enrichment. We quantified 3,364 proteins and applied machine-learning strategies combining cross-sectional case-control comparisons with longitudinal tracking of within-individual changes. Cross-sectional analysis at diagnosis achieved AUC 0.823 (95% CI: 0.723-0.923), identifying eight consensus features selected in >50% of cross-validation folds: PRSS1, GSK3B, FAM20C, CES1, CCL14, EPHA10, LMAN2, and ITIH1. These signatures, spanning proteases, immune activation, and extracellular matrix re-modeling, demonstrate proof-of-concept for serum-based melanoma detection and provide candidate biomarkers for validation.

Authors
Yuhui Hong1, Isha Gokhale1, Justin Sanders2, Gwenneth Straub1, Wout Bittremieux3, and William Stafford Noble1,2

Institutions
1 Department of Genome Sciences, University of Washington 2 Paul G. Allen School of Computer Science and Engineering, University of Washington 3 Department of Computer Science, University of Antwerp

Abstract
In a proteomics tandem mass spectrometry experiment, each observed fragmentation spectrum ideally corresponds to a single peptide sequence. In practice, however, a significant proportion of observed spectra are chimeric, meaning that they were generated by a mixture of two or more peptides. The de novo sequencing task aims to infer the amino acid sequence of the peptide responsible for generating a given spectrum, but standard models assume a single peptide per spectrum and therefore cannot account for chimeras. We have modified the Casanovo de novo sequencing model to detect chimeric spectra. We introduced a separator token to delineate the sequences in a chimeric spectrum, and we modified Casanovo's loss function to minimize loss across different sequence orderings during training. The precursor filter was removed after decoding, since one sequence in a chimeric spectrum typically exhibits an extremely large precursor mass difference. To train and evaluate the model, we assembled a mixed single-sequence and chimeric spectra dataset spanning ten species, each drawn from a separate PRIDE repository. For each species we selected the data-dependent acquisition project with the most qualifying raw files, totaling 3,132 runs. Spectra were searched with MSFragger in DDA+ mode against species-specific target-decoy databases and re-scored with Percolator, treating each peptide of a chimeric spectrum as a separately scored peptide-spectrum match (PSM). After discarding spectra with more than two sequences, keeping at most two spectra per distinct peptide or peptide pair, and limiting each peptide to ten chimeric spectra, the dataset comprised 5,650,819 spectra from 1,784,919 peptides, 53.3% of them chimeric. We partitioned the data at the sequence level using a greedy algorithm that eliminates peptide overlap between splits. We evaluate models by treating each spectrum and its chimera as separate PSMs and computing average precision (AP) as the area under the precision-coverage curve. On a held-out test set, our chimeric model raised average precision from 0.54 to 0.67 at the peptide level and from 0.75 to 0.86 at the amino acid level relative to a non-chimeric baseline. A two-pass alternative, which predicts one peptide, removes its matched peaks, and sequences the residual spectrum, gave essentially no improvement (peptide-level AP 0.54), indicating that jointly modeling chimeras is substantially more effective than sequential peak subtraction.

Authors
Michael R. Hoopmann, Jimmy K. Eng, Devin K. Schweppe

Institutions
University of Washington

Abstract
Real-time search (RTS) is the adaptation of sequence identification algorithms for use during proteomics-focused mass spectrometer data acquisition. RTS is used to drive automated decision making processes for powerfully adaptive instrument methods. RTS algorithms must operate efficiently while considering hurdles, such as differential modifications, that create exponential increases in computation time. Here we present Telescope, a lightweight database search algorithm and sandbox, designed to develop and test methods for RTS. Telescope accepts spectral data streams in real time, and performs large proteome-wide searches, including PTM analysis, in multithreaded fashion to keep pace with today’s fastest instruments. Using Telescope provides a foundation for developing more sophisticated RTS applications and advancing the capabilities of real time mass spectrometry.

Authors
Kiran Iyer, Rosalynn Molden, Erin Weisenhorn, Hannah Townsend, Luis Fernandez-Ruiz, Shannon Hayes

Institutions
Just Evotec Biologics, Protein Metrics LLC

Abstract
Glycans can significantly influence the stability, activity, and immunogenicity of biologics, making their analysis crucial for assessing the quality of antibody therapeutics, especially biosimilars. Presented here is a tool to streamline mass spectrometry-based glycan analysis automated glycan grouping with software. From released glycans to peptide-level multi-attribute method (MAM) workflows, automated glycan grouping and intuitive visualizations can transform complex datasets into actionable insights. Work presented here shows the utility of glycan grouping in bridging data from multiple analytical methods such as MAM and released glycan analysis. Additionally, glycan grouping was used to compare day of culture samples from different bioreactors using mass spectrometry (MS)

Authors
Kayleigh Voos1*, Gayatri Iyer2*, Anne Marie Weitzel3, Gina Many1, Laurie Goodyear4, Sue Bodine5, Karyn Esser6, Joshua Adkins1, Charles Burant3 and the MoTrPAC Study Group

Institutions
1 Pacific Northwest National Laboratory, Richland, WA. 2 Department of Computational Medicine and Bioinformatics, University of Michigan, Ann Arbor, MI. 3 Department of Internal Medicine , University of Michigan, Ann Arbor, MI. 4 Joslin Diabetes Center, Harvard Medical School, Boston, MA. 5 Oklahoma Medical Research Foundation, Oklahoma City, OK. 6 University of Florida, Gainesville, FL.

Abstract
Skeletal muscle plays a central role in generating force for movement and mediating adaptive responses to exercise. In this study, we examined the temporal molecular response of rat skeletal muscle to a single bout of acute endurance exercise across multiple molecular layers, from immediately post-exercise to 48 hours post-exercise. We observed that the phosphoproteome and epigenome exhibited the most robust early responses to exercise. Notably, several molecular links were identified between phosphoproteomic changes and epigenomic regulation, suggesting a coordinated interplay between these layers. Additionally, peak-to-gene correlation analyses revealed associations between epigenetic changes and corresponding alterations in transcriptional programs related to protein degradation and folding, transcription factor activity, and aerobic respiration. Multi-omics factor analysis (MOFA) across molecular layers further revealed a subset of exercise-responsive and sex-dependent modules with clear temporal dynamics. These included proteins and genes associated with mitochondrial respiration, cytoskeletal remodeling, sarcomere assembly, RNA metabolism, and inflammatory responses. Furthermore, temporal clustering of metabolomic, phosphoproteomic, and transcriptomic data revealed early, middle, and late molecular programs responsive to an acute exercise bout. Together, these data provide a time-resolved molecular characterization of the skeletal muscle response to acute exercise in untrained female and male rats. These findings contextualize early molecular responses that precede muscle adaptation to training and enhance our understanding of muscle biology on the minute-to-hour timescale following a single bout of endurance exercise.

Authors
Rose C. Pletcher, Richard Ivey, Jeffrey R. Whiteaker, Amanda G. Paulovich

Institutions
Fred Hutchinson Cancer Center, University of Washington, Seattle, WA 98109, USA

Abstract
Accurate protein quantification is critical for targeted proteomic assays that rely on normalization to protein input. Here, we present an optimized MicroBCA workflow designed to improve quantitative precision and accuracy in sample-limited, longitudinal targeted proteomics applications. Each plate incorporates a standard curve, a cell lysate QC sample to monitor precision in a complex lysate that more closely mimics the sample types, and a NIST-traceable protein standard to assess accuracy. Assay performance is tracked using Levey-Jennings analysis with one year of historical data to define acceptance criteria and establish plate pass/fail status prior to downstream processing. Sample and QC dilution schemes were systematically optimized to improve pipetting precision and maintain measurements within the assay’s linear range. All liquid transfers were performed at volumes ≥5 µL. These changes were associated with a 6.1% absolute reduction in CV, consistent with improved inter-plate reproducibility. Plate position effects were evaluated, revealing a 3.4% edge-associated deviation relative to interior wells (p < 0.0001). Exclusion of outer wells reduced positional bias and improved overall assay precision. Longitudinal monitoring of the cell lysate QC sample over five years revealed an approximately 30% decrease in measured protein concentration. This decline was gradual and non-monotonic, consistent with contributions from both drift and inter-assay variability. Because standardized lysates are commonly used as bridge samples in targeted workflows, these findings indicate that commonly used reference lysates can exhibit measurable long-term drift, introducing a potential source of bias in longitudinal normalization strategies. Ongoing work is focused on characterizing the sources of this drift and evaluating its impact on quantitative measurements in targeted proteomics assays, including comparisons between legacy and newly prepared reference materials. Overall, this work defines a QC-driven MicroBCA framework that improves quantitative performance and enables detection of both technical and longitudinal sources of variability in targeted proteomics studies.

Authors
Jesse Wilson, Jason Gilmore

Institutions
Just-Evotec Biologics

Abstract
Trypsin digestion of proteins is the dominant approach for proteomics and peptide mapping by LC-MS/MS due to the favorable length and charge characteristics of the resulting peptides. However, there are regions of therapeutic proteins where the resultant peptides are too short or too long for confident identification and quantification of post translational modifications (PTMs). To approach 100% primary sequence coverage, complimentary enzymes are used. One such enzyme is Glu-C from Staphylococcus aureus. Unlike trypsin, Glu-C digestion is less reliable due to incomplete enzymatic activity causing a divergence between expectations and empirical results. The goal of this study is to evaluate these discrepancies and provide a more accurate prediction of peptide coverage with Glu-C to properly align expectations for IgG1 antibodies.

Authors
Joshua L. Fox, Andrea I. Gutierrez, Julia Robbins, Daniele Canzani, Evan Hubbard, Sebastian J. Paez, Anastasiya Prymolenna, Lily Tatka, Gaelle Mercenne, Kyle Siebenthall, Brian McEllin, William Fondrie, Alexander Federation, Lindsay K Pino

Institutions
All contributors are employed by Talus Bioscience.

Abstract
Intro: Transcription Factors (TFs) and other chromatin-bound proteins (cofactors, chromatin remodelers, etc.) form cellular machinery for interpreting signals and controlling gene expression. These proteins play a central role in disease, drug response, and homeostasis, yet they remain difficult to measure due to their low abundance, transient binding behavior, and extensive post-translational regulation. To address these challenges, we developed a high-throughput mass spectrometry approach leveraging subcellular fractionation and quantitative chromatin proteomics to evaluate TF activity in their native cell environment. Additionally, TFs generally lack druggable cognate binding sites and small molecule approaches have had limited success. We address this by using machine learning-powered in silico compound screening to discover novel, AI-guided, therapeutics.

Methods: Human cell lines were plated in a 96-well plate using cell-line specific amounts optimized for ideal confluence (typically 20,000 - 50,000 cells) and incubated with commercially-available disease-relevant control compounds. Using regulome profiling, our semi-automated high-throughput screening platform built upon chromatin enriching salt separation coupled to DIA (ChESS-DIA), the chromatin-associated proteome was analyzed. This includes protein fractionation utilizing lectin-coated beads which recruit glycans that are highly abundant on the nuclear envelope; and digestion with our optimized cell-line specific conditions via SP3. Mass spectrometry acquisition and analysis were conducted, respectively, using timsTOF Ultra II and DIA-NN version 1.8.1 on the quantms Nextflow pipeline.

Preliminary data: Using our regulome profiling platform, we were able to map the chromatin TF landscape of a variety of cell lines spanning cancer, inflammation, and other disease indications. We used our machine learning model to identify hits, which we define as proteins knocked off of chromatin with less than a log2 fold change (log2FC) of -0.75, in a variety of these disease indicators after treatment with small molecule covalent compounds. These results allow us to quantify peptides, explore compound binding sites, and evaluate functional effects of drug treatments. Specifically, our machine learning model, Strategian, has enabled us to identify 25 preliminary hits in STAT6, a TF involved in immune signaling, tumor microenvironment modulation, and disease progression which is currently undrugged.

Novel aspect: Our use of ChESS-DIA to provide high-throughput data for our regulome atlas, a dataset on TFs in their native environment following drug treatment.

Authors
Anna B. Lin, Christopher J. Weir, Trung Nguyen, Matthew F. Bush

Institutions
University of Washington

Abstract
Introduction: The measurement and characterization of protein stability is critical to many aspects of biopharmaceutics. In the context of quality control, understanding protein stability is essential for ensuring the consistency and integrity of manufactured therapeutics. In biopharmaceutical research, this same understanding is critical to probing many life-threatening diseases associated with protein misfolding. One such example is transthyretin (TTR), a homotetrameric protein implicated in the development of familial amyloid polyneuropathy (FAP) and senile systemic amyloidosis (SSA). Transthyretin can dissociate into individual monomers that are prone to misfolding and aggregation, leading to the formation of fibrils associated with these diseases. To date, more than 150 TTR mutations have been identified each with their own stability profile. Among these, the V30M variant is one of the most recognized, exhibiting accelerated amyloid formation relative to the wild-type protein. A promising strategy to combat this process involves the use of pharmacological chaperones, which work by shifting the thermodynamic landscape of protein folding while also providing kinetic facilitation of folding the desired structure. However, pharmacological chaperones have only been successfully developed for a handful of diseases, including tafamidis for TTR amyloidosis. The small number of approved pharmacological chaperones across all diseases emphasizes how challenging it is to identify molecules with these properties, highlighting the need for novel technology capable of enabling measurements that can identify viable candidates across many proteins. To meet this need, the Bush Lab has developed programmed-temperature electrospray ionization (ptESI), a method that modulates the temperature of liquid samples across varying gradients while enabling real-time mass spectrometry analysis. In this work, ptESI is used to characterize the kinetic stability of transthyretin, and tafamidis is applied to provide preliminary validation of ptESI as a ligand screening tool, with the broader goal of extending this approach toward identifying pharmacological chaperones for other protein misfolding diseases.

Methods: Wild-type and V30M transthyretin (3 uM tetramer) solutions are prepared in a blend of 200 mM ammonium acetate and 200 mM acetic acid for a final pH of 4. Samples are buffer exchanged and loaded into the ptESI source using nano-ESI borosilicate capillaries. Temperature programs spanning 40 °C to 75 °C are used to modulate the temperatures of solutions while MS data is continuously acquired using a Waters Cyclic IM-MS system.

Preliminary Data: Under acidic conditions (pH 4) and thermal stress, wild-type TTR exhibits partial tetramer dissociation at ~75 °C, reaching a plateau over a 10 minute hold at that temperature. Under identical conditions, V30M undergoes complete and significantly faster dissociation into its monomer subunits. In preliminary analysis, an exponential decay is fitted to each replicate for both species. Individual decay rate constants are used to determine a mean apparent decay rate constant and direct comparisons of the decay rates between the two species confirm that the mutation significantly destabilizes the tetramer and complete dissociation of V30M into subsequent monomer subunits is observed prior to the onset of WT-TTR dissociation. However, all samples appear to undergo subtle behavioral changes as a function of time after sample preparation, strongly suggesting that the samples are not at equilibrium. For both WT and V30M, this is reflected in differences in the initial delay period preceding dissociation as well as a shift in decay rates during dissociation. Despite the variance that this contributes to the data there is a significant difference between the stability of WT and V30M.

To validate ptESI as a tool for pharmacological chaperone screening, tafamidis is introduced to evaluate its effect on TTR thermal dissociation. When the tafamidis bound complex is subjected to identical conditions as before, a significant stabilizing effect is observed as expected. Tafamidis bound WT-TTR remains intact until a point at which dissociation begins at a significantly reduced rate relative to apo WT-TTR. Preliminary data has shown a similar stabilizing effect of tafamidis on V30M transthyretin as well. Ongoing work aims to continue characterizing this temperature dependent behavior and begin a screening of a small library of ligands.

Authors
Julien Margaret A. Dagan, Alesi R. Escobedo, Anran Yu, Miklos Guttman

Institutions
University of Washington

Abstract
Hydrogen Deuterium Exchange-Mass Spectrometry (HDX-MS) is routinely employed for probing protein structures and conformational dynamics. However, its reproducibility is limited by back-exchange after following labeling, which complicates exchange rate measurements and data interpretation. While most back-exchange occurs in solution or during ionization, some studies have also observed deuterium loss post-ionization within the MS. We hypothesize that ambient water is entering the MS and contributing to deuterium loss. Here, we quantitatively track deuterium loss using EDTA and P1 peptide to systematically evaluate water vapor-driven back exchange in both negative and positive modes across five widely used MS platforms: Thermo Orbitrap Tribrid (Ascend and Eclipse), Waters Synapt G2-Si, Thermo Finnegan LTQ, and Bruker timsTOF fleX. This study aims to define the magnitude of water vapor effects and highlight platform-dependent differences.

Authors
Anais Gentilhomme1,  Chris Hsu1, Bo Wen1, Justin A. Sanders1, William S. Noble1, Laura Zelter2, Michael Riffle1, Susan Q. Lang3, William J. Brazelton4, and Brook L. Nunn1

Institutions
1University of Washington (Department of Genome Sciences) 2Florida State University (Department of Chemistry & Biochemistry) 3Woods Hold Oceanographic Institution (Department of Geol. and Geophys. and Marine Chem. and Geochem.) 4Blue Marble Space Institute of Science (Seattle, WA)

Abstract
We present current efforts in an ongoing investigation to develop and evaluate low-input metaproteomic workflows tailored to Lost City Hydrothermal Field (LCHF) chimneys, aiming to identify expressed microbial functions. LCHF is an ultramafic, serpentinizing submarine hydrothermal vent system characterized by high pH (9-11), temperatures of ~100℃ in venting fluids, and elevated concentrations of formate, methane, and sulfide that can support C1-metabolizing organisms (Kelley et al. 2005, Brazelton et al., 2006). These characteristics make LCHF an ideal study site and analog for serpentinizing systems predicted to be on icy ocean worlds such as Europa and Enceladus. Metaproteomics allow us to both detect and measure microbial functional responses to environmental conditions. Metagenomic datasets for LCHF chimneys and venting fluids are available (Alian et al., 2025, Brazelton et al., 2022), but no metaproteomic studies have been reported. The goal of this study is in two phases, to first identify chemosynthetic-associated peptides within these chimneys and the second is to then apply a targeted proteomics approach on 1.5 km of LCHF uncharacterized rock core to detect peptides that are relevant for life detection in the subsurface of ocean worlds with serpentinizing hydrothermal systems. Samples from five different active LCHF chimneys and three carbonate veins were sampled during the 2018 ROV Jason Lost City expedition (i.e. Brazelton et al., 2022, Moore et al., 2021). These samples are not only composed of a complex matrix but also have low cell counts ( < 2E76 cells/g of chimney rock) and therefore have inherently low protein concentration (20-100ng total protein). For phase one metaproteomic investigation, one of the chimneys with the highest amount of predicted biomass, Sombrero, was selected for methods and search development and prepared in technical triplicate. The protein fraction was isolated, and the cellular biomass solubilized by a combination of mechanical and chemical disruption using an in-house EDTA/SDS solution. Proteins were purified using High Recovery S-trap to isolate and wash the proteins embedded in the complex matrix, followed by trypsin digestion to peptides. Then digested peptides were analyzed using an Orbitrap Astral mass spectrometer coupled to a Vanquish Neo UHPLC system and the data were acquired using a data dependent acquisition (DDA). It was searched against metagenome-derived protein database that was chimney-specific and contained the top 100k most abundant contigs using Comet and Percolator (Peptide Spectral Match (PSM) False Discovery Rate < 0.01) where results were visualized and analyzed in limelight and R (4.3.2) (Chambers et al., 2012, Eng et al., 2013, Käll et al., 2007, Riffle et al., 2025a,b). Across these technical replicates, peptide-spectrum matches were detected under strict false discovery control of 0.01, though peptide recovery was low, which is consistent with the low biomass of this system. Despite these challenges, we detected about ~1400 LCFH- associated unique peptide sequences across samples. In addition, we are currently comparing our metagenomic based proteomic identification findings with an agnostic de novo approach to identifying peptides without relying on a reference database. This approach is especially relevant for life detection applications as it presents the evaluation of peptide level biomarkers in environments where genomic context may be absent.

Authors
Anna G. Duboff, Kathryn Kothlow, Tim S. Veth, Jacob H. Russell, Katrina N. Peterson, Fengchao Yu, Daniel A. Polasky, Alexey I. Nesvizhskii, Devin K. Schweppe, Nicholas M. Riley

Institutions
Department of Chemistry, University of Washington; Department of Genome Sciences, University of Washington; Department of Pathology, University of Michigan

Abstract
INTRODUCTION Targeted proteomics is the gold standard for detecting proteins of interest in complex mixtures. Major challenges of extending these methods to glycoproteomics include the limited number of glycoforms that can be targeted per acquisition and the requirement for dictating glycoforms to target a priori despite limited knowledge of what glycans would be at each glycosite. Intelligent data acquisition (IDA) is a powerful tool that can overcome these limitations, and strategies such as real-time database search (RTS) and real-time library search (RTLS) have shown promise for non-modified peptides. Here, we use heavily glycosylated integrins as a test case to explore how IDA strategies can be used to target glycoproteins of interest in real-time without prior knowledge of the glycosylation profile.

METHODS Experiments were performed on Thermo Scientific Orbitrap Ascend Tribrid MS system (Thermo Fisher Scientific) equipped with a Vanquish Neo UHPLC system. Analyses used online reversed-phase separations with a 25 cm Aurora Series reverse-phase LC column (IonOpticks) or in-house packed columns. Recombinant integrins (ACROBiosystems) were digested with trypsin and gluC. For spike-in experiments, recombinant integrins were spiked into K562 cell lysates which were then reduced, alkylated, and digested with trypsin. RTS was conducted using a modified version of MSFragger-RTS, specifically tailored for glycopeptides, within Orbiter. Searches were performed against the human proteome database using 253 glycan mass offsets. Through the Instrument Application Programming Interface (IAPI), Orbiter and Helios directed the subsequent MS/MS acquisition of glycoforms according to the output of MSFragger-RTS.

PRELIMINARY DATA Glycoproteomics can benefit greatly from on-the-fly decision making during MS data acquisition, and our preliminary work shows that the commercially available RTS platform on the Orbitrap Tribrid platform that uses Comet-based scoring can identify glycopeptides with enzymatically truncated N-glycosites. We aim to further build on this work to tackle non-truncated glycopeptides using a modified version of MSFragger-RTS specifically tailored for glycopeptides with search speeds of ~200-300 ms per scan. Previously, we showed that this tailored version of MSFragger-RTS performs comparably to an offline search on the peptide, glycan, and glycopeptide level with ~3,000-3,500 peptide spectral matches. We build on the previously developed Helios and Orbiter platforms to implement MSFragger-RTS through the IAPI, which allows us to selectively target glycopeptides from proteins of interest and capture all relevant glycoforms. With this platform, we first assess whether the identified glycopeptide is from our protein of interest and then use known monosaccharide and oligosaccharide mass shifts to generate an “ephemeral” glycoform inclusion list in real-time. Priority scans are inserted into the scan acquisition sequence to target these precursor ions for MS/MS scans corresponding to other glycoforms of the original peptide sequence, and these priority scans are not searched again in real-time. Glycoforms are added and removed to the “ephemeral” inclusion list to prevent an infinitely growing target list. Additionally, if a given spectrum is not of high enough quality to enable an identification, but contains oxonium ions, indicating the presence of a glycopeptide, then the instrument is directed to accumulate ions longer for this species, maximizing the chance of subsequent identification. This platform enables us to maximize the depth of our data without having to decide the glycan modification prior to acquisition. We demonstrate the value of this workflow using integrins, highly glycosylated cell surface proteins that are often involved in cancer metastasis.

Authors
Yuanye Chi[1], Bo Wen[1], Michael J. MacCoss[1], and William Stafford Noble[1,2]

Institutions
[1] Department of Genome Sciences, University of Washington [2] Paul G. Allen School of Computer Science and Engineering, University of Washington

Abstract
Targeted proteomics methods such as selected reaction monitoring (SRM) and parallel reaction monitor- ing (PRM) provide sensitive, reproducible quantification of specific proteins in complex biological samples. Reliable quantification depends critically on accurate peak detection, in particular, correct localization of chromatographic peak boundaries. In practice, peak boundaries are still frequently corrected by hand. Existing automated methods rely largely on heuristic rules or traditional machine learning that may not fully capture the complex patterns in mass spectrometry data. These methods also analyze each run inde- pendently, leaving information unexploited when the same peptide is measured across many samples. We present a deep learning approach to peptide peak detection based on joint, multi-run detection: by analyzing a peptide across all runs in an experiment simultaneously, a library-aware transformer borrows statistical strength from high-quality runs to localize peaks in runs with weak or noisy signal, without requiring sta- ble isotope-labeled internal standards. Evaluated on manually annotated targeted datasets, our multi-run models consistently outperformed their single-run counterparts and the default Skyline peak picker, yielding more accurate peak boundaries and lower quantitative variability across runs. To make the method readily accessible within existing workflows, we release it as a Skyline external tool for deep learning-based peak boundary refinement.

Authors
Bhavyahshree Navaneetha Krishnan 1, Michael Dammann 3, Sander Willems 2, Fabian J. Theis 3, Wout Bittremieux 4, George Rosenburger 6, William S. Noble 1

Institutions
1: University of Washington, Seattle, 2: Bruker, Kontich, Belgium, 3: Helmholtz Munich, Neuherberg, Germany, 4: University of Antwerp, Antwerpen, Belgium, 5: Bruker Switzerland AG, Fällanden, Switzerland

Abstract
Casanovo, a state-of-the-art de novo peptide sequencer powered by a transformer architecture, is currently only trained on Orbitrap data aggregated through the MassIVE-KB database. This results in unsatisfactory results when applied to timsTOF data due to the spectral variability between the two instrument types and an additional ion mobility dimension within the timsTOF data. In this work, we extended Casanovo’s functionality to timsTOF data.

A timsTOF data corpus consisting of 1.8 million unique peptides and 5.6 million PSMs from 13 organisms, including immunopeptides and non-tryptic peptides, was collected with the ion mobility dimension collapsed to align with the existing model.

To balance information retention and memory efficiency, Casanovo uses the top 150 most intense peaks per spectrum in training and evaluation. However, this is not suitable for timsTOF data as it has more peaks compared to the Orbitrap data Casanovo is currently trained on. Models with varied peak numbers (150, 300, 500) were trained to find the optimal maximum peak number.

After training Casanovo with varying max peak numbers, we found that Casanovo with 500 peaks performed the best with an AP of 0.869, hereby referred to as Casanovo-tims. Casanovo-tims performance on immunopeptidomics data was also evaluated. HLA Class I peptides were sequenced with Sage and Casanovo-tims. All predictions were separated into three groups: predictions shared between Sage and Casanovo-tims, Sage’s unique predictions and Casanovo-tims’ unique predictions. Sage-only predictions had a lower median cosine similarity compared to Casanovo-tims’ with Sage having a cosine similarity of 0.896 and Casanovo-tims having a cosine similarity of 0.931. This indicates that the predictions made by Casanovo better explain the spectra when compared to Sage. Casanovo-tims also has a higher share of peptides below the 500 nM threshold for intermediate binding calculated by NetMHCPan compared to Sage, with a percentage of 68.8% for Casanovo-tims only predictions and 37.5% for Sage only predictions. The shared peptides had the highest percentage with 85.5% of peptides.

Future work will focus on the model’s application to metaproteomics. In order to facilitate usage of Casanovo with both Orbitrap and timsTOF data, a combined model trained on both types of data is also being developed.

Authors
Chris Hsu, Lilian R. Heil, Bo Wen, Graeme McAlister, Gennifer E. Merrihew, Philip M. Remes, Deanna L. Plubell, Rafael Melani, Vlad Zabrouskov, and Michael J. MacCoss

Institutions
University of Washington, Thermo Fisher Scientific (San Jose)

Abstract
Data-independent acquisition (DIA) proteomics relies almost exclusively on beam-type collision-induced dissociation (HCD) because its short activation time supports fast acquisition rates required for DIA. However, HCD requires charge state calibration and is therefore imperfect for mixed-charge DIA isolation windows. Resonance-excitation collision-induced dissociation (reCID) offers a promising alternative to HCD for DIA because the ion activation is effectively independent of charge state. Historically, reCID’s longer activation time has been considered too slow for DIA. Here, we revisit reCID for DIA proteomics using a modified Orbitrap Tribrid Apex MultiOmics mass spectrometer (Apex) that recovers acquisition-matrix overhead as additional ion injection time, enabling reCID acquisition rates comparable to HCD. Using matched acquisition rate settings of tryptic HeLa digests, reCID achieved precursor and protein detections similar to HCD when used with Carafe fine-tuned, fragmentation-matched spectral libraries. Library fine-tuning improved reCID precursor detections more than HCD detections, 24% versus 5%, indicating that HCD-trained prediction models are suboptimal for reCID spectra. ReCID also maintained peptide-level quantitative performance, including ions measured per peptide, precision, and accuracy. Across seven NCI cancer cell lines and pooled mixtures, protein abundance rankings were highly conserved between the Apex reCID and HCD methods, and also across Orbitrap Astral Zoom platforms. These results support reCID as a practical fragmentation mode for DIA proteomics.

Authors
Vanessa Encinas, Tim S. Veth, Nicholas M. Riley

Institutions
Department of Chemistry, University of Washington, Seattle, WA, USA

Abstract
Immobilized metal affinity chromatography (IMAC) is widely used as enrichment strategy in phosphoproteomics. Coordination of negative charges in IMAC also enriches other moieties besides phosphopeptides, including sialylated glycopeptides. Recent years have seen emergence of Zr4+ IMAC in favor over other metal ions and reports of certain IMAC conditions (e.g., acid additives and specific metal ions) that can be selective for phosphopeptide over glycopeptide enrichment. Here we revisit the value of IMAC and metal oxide affinity chromatography (MOAC) using Fe3+, Zr4+, and Ti4+ metal ions for enriching glycopeptides. We explore the impact concentrations of glycolic acid in loading buffers, use of phosphatase, and other mass action-centric parameters on glycopeptide enrichment. Human cell lysates (HEK293 and K562) were used for all experiments, and glycopeptides were enriched using a DynaBead magnetic rack or using a KingFisher Flex with Fe3+, Zr4+, and Ti4+ IMAC and TiO2 and ZrO2 MOAC magnetic beads from ReSyn Biosciences or generated in-house. Glycolic acid concentrations in loading buffers ranged from 0-2 M. All LC-MS/MS acquisitions were performed using 25 cm 75 µm ID Aurora Series reversed-phase LC column packed with 1.6 µm C18 particles (IonOpticks) on a Vanquish Neo UHPLC system coupled to an Orbitrap Ascend Tribrid mass spectrometer. MS/MS data were analyzed using MSFragger-Phospho, MSFragger-Glyco, GlyCounter, pGlyco, and in-house scripts. IMAC uses transition metals like Fe3+, Zr4+, and Ti4+ to chelate phosphate groups, with Ti 4+ being particularly popular in recent years. More recently, several groups have established the benefits and complementary nature of Zr-IMAC relative to Ti-IMAC and Fe-IMAC for phosphopeptide enrichment. Zr is a stronger Lewis acid (i.e., electron-pair acceptor) with a higher coordination number, leading to unique coordination chemistry and enrichment complementarity. IMAC and MOAC strategies have previously been shown to enrich sialylated glycopeptides, but it is not well-known how the different metal ions, especially Zr4+, affect glycopeptide enrichment. Interestingly, some groups have reported that Fe-IMAC can specifically enrich phosphopeptides over glycopeptides (Caval et al., 2019), leading to open questions about how to best leverage IMAC and MOAC for glycopeptide analysis. Because Zr-IMAC has not been widely investigated for glycopeptide enrichment, we first used our freely available open-source tool GlyCounter to evaluate publicly evaluable Zr-IMAC datasets from Diez et al. (PXD018273) and Bortel et al. (PXD045601). Using diagnostic oxonium ions from HexNAc (m/z 204, 186, 168, 144, 138, and 126) and sialic acid (m/z 274 and 292) species, we saw that most enrichment conditions generated thousands of MS/MS spectra corresponding to putative glycopeptides, with many of them, surprisingly, not being sialylated. Here, we expand on this observation by exploring differences in glycopeptide populations enriched by IMAC using Fe3+, Zr4+, and Ti4+ ions and TiO2 and ZrO2 MOAC. We evaluate the presence of spectra with oxonium ions and glycopeptide identifications while varying glycolic acid concentrations in loading buffers and enzymatic treatments with phosphatase. In all we show how IMAC and MOAC can be used to enrich overlapping and complementary glycopeptides, and we establish the viability of Zr-IMAC for glycopeptide enrichment.

Authors
William I. Eisen, Haley M. Schramm, Miklos Guttman, and Nicholas M. Riley

Institutions
1. Department of Chemistry, University of Washington, Seattle, WA, USA 2. Department of Medicinal Chemistry, University of Washington, Seattle, WA, USA

Abstract
Glycan structures mediate functions of glycoproteins but defining how this information is encoded in glycan structural isomers (those that only differ in linkage between monosaccharides) remains a challenge. Recent evidence has shown that glycan structural isomers leave “linkage memory” in their fragments, where protonated monomer fragments generated from disaccharides have different proton affinities based on precursor structural configurations. Here we investigate if this encoded proton affinity can be measured using gas-phase ion-ion proton-transfer charge reduction (PTCR) reactions to distinguish glycan structural isomers. We demonstrate that subtle structural variations within glycan-specific fragments translate to differences in ion depletion kinetics with PTCR reactions, revealing a novel way to elucidate glycan structural isomers.

All experiments were performed on an Orbitrap Ascend Tribrid MS (Thermo Fisher Scientific) via nano-electrospray ionization (nanoESI). Samples analyzed included a monosaccharide,13C labeled N-acetylglucosamine (GlcNAc), and 2 disaccharides, Gal(β1-6)GlcNAc and Gal(β1-4)GlcNAc, which only differ in their linkages. First, an MS2 experiment was run on 1 µM GlcNAc in 0.1% formic acid, with fragmentation via CID, followed by PTCR on select fragments. Then, the disaccharides were similarly analyzed. The main difference between the mono- and disaccharide experiment was an extra fragmentation step. This experiment was MS3, with two rounds of CID fragmentation to produce similar fragments to the monosaccharide, followed by a PTCR experiment. Data were processed using our open-source software, GlyCounter (riley-research/GlyCounter), and in-house python and R scripts.

Recent evidence using gas-phase hydrogen deuterium exchange and tandem MS has shown that glycan structural isomers leave “linkage memory” in their fragments based on the precursor structural configuration. These experiments indicate that linkage information is encoded via proton affinities in fragment ions. Proton transfer charge reduction (PTCR) is one of the most established gas-phase ion-ion reactions that involves reacting precursor cations with reagent anions, which abstract a proton from the cation to reduce its charge state. PTCR is typically used to simplify and deconvolute complex spectra, but we were curious if different proton affinities of glycan fragments would generate measurable differences in ion-ion reaction kinetics. Fragment ions in these experiments are singly charged, meaning PTCR neutralizes them. Thus, we measure kinetics through ion depletion. Our first experiments use GlcNAc to discern feasibility of this approach, where we show decay curves of singly charged ions can generated on reasonable timescales (10s to 100s of ms) with PTCR reactions. We then use PTCR to compare ion depletion rates of fragments from Gal(β1-6)GlcNAc and Gal(β1-4)GlcNAc disaccharide ions. Our analyses of the ion depletion kinetics indicate that certain glycan-specific ions such as m/z 204 and m/z 186 do show different rates of decay. In contrast, other glycan-specific ions, including m/z 366 and 222, show no difference in decay rates under the same analysis. These ions are common to both structural isomers and thus would not be expected to have different kinetics. In all, our novel approach to use PTCR reactions for measuring glycan structural information demonstrates depletion kinetic analysis based on proton affinity, affording the ability to discern the original linkage. This work establishes a new tandem MS strategy for structural glycoproteomics that can be applied to study both glycans and glycopeptides.

Authors
Kathryn Kothlow (1), Bo Wen (2), Anna G. Duboff (1), Jacob H. Russell (1), William S. Noble (2,3), and Nicholas M. Riley (1)

Institutions
(1) Department of Chemistry, University of Washington, Seattle, WA, USA (2) Department of Genome Sciences, University of Washington, Seattle, WA, USA (3) Paul G. Allen School of Computer Science and Engineering, Seattle, WA, USA

Abstract
Introduction: Immunoglobulin G (IgG) glycosylation is a key regulator of numerous immune signaling pathways, making it a promising biomarker for diseases and aging. Glycan structure at the individual monosaccharide level contributes to IgG function, meaning detailed structural information is necessary to functionally characterize IgG glycoforms in diverse immune contexts. Although current glycoproteomic methods often fall short of delivering structural information, collision energy ramps have shown promise as a strategy to distinguish glycan structural isomers on intact glycopeptides. In this work, we use real-time library searching (RTLS) with fine-tuned machine learning-generated libraries to identify IgG glycopeptide spectra during acquisition.

This strategy prioritizes time-intensive multi-scan collision energy ramps only for precursors likely to be informative, generating glycopeptide structural information while maximizing identifications.

Methods: All experiments were performed on a Thermo Scientific Orbitrap Ascend Tribrid MS system (Thermo Fisher Scientific, San Jose, CA) equipped with a Vanquish Neo UHPLC System. All analyses used online reversed phase separations. Spectral libraries were created a custom-trained version of Carafe (https://github.com/Noble-Lab/Carafe). An IgG antibody standard (Agilent) was treated using glycosidases and glycosyltransferases (Genovis) to create known glycan structures. RTLS was performed using custom C# code which interfaces with the Thermo Instrument API. Library matches were evaluated with cosine similarity scores and used to trigger additional scans at different collision energies to gain glycan structure information.

Preliminary Data: Most glycoproteomics studies generate tandem mass spectra that can provide glycan composition in a site-specific manner but rarely generate information to resolve glycan structures. Maliepaard et al. and others have shown that collision energy ramps can generate fragments revealing connectivity of monsosaccharides and distinguishing isomers. These collision energy ramps, however, require multiple tandem MS/MS scans per precursor ion, meaning they would ideally be performed only on precursor ions where we know that the time cost will translate to a glycopeptide identification. Here, we generate fine-tuned in silico spectral libraries and use them in RTLS methods to guide this process of prioritizing identified glycopeptide precursors for structural characterization. To account for many glycoforms, we require that the library spectra be impartial to the glycan composition, instead making an identification based on the peptide fragments. To better predict the intensity of glycopeptide b- and y- ions, we generated fine-tuned peptide fragment ion intensity prediction models using Carafe on glycopeptide data. We created a custom RTLS algorithm that automatically filters glycan-specific oxonium ions and Y-ions. This decongests the spectra and allows peptide b- and y- ions to be searched more easily. Using RTLS with these modified libraries, we identify a majority of glycopeptides identified in an offline MSFraggerGlyco database search. We found that confident RTLS matches can trigger collision energy ramps to generate glycopeptide structural information. We benchmark our methods with antibody standards treated with glycan-altering enzymes, defining IgG glycopeptides with known glycan structures. We analyze intensities of glycan fragments specific to certain isomers which provided mathematical ratios to distinguish those isomers in unknown glycan structures. Finally, we create an automated pipeline to extract glycan structures from IgG glycopeptide data. In summary, a combination of real-time library search with fine-tuned libraries and post-acquisition data analysis allowed us to fully characterize IgG glycan structures.

Authors
Maddy Yeh [1], Nick Riley [1]

Institutions
[1] Department of Chemistry, University of Washington, Seattle, WA, USA

Abstract
Glycoproteomics largely relies on data-dependent acquisition for MS/MS identification of glycopeptides and MS1-based label free methods for quantitation. Dynamic range limitations of mass spectrometers, i.e., ion capacities they can process per scan, remain a challenge for identification and quantitation in glycopeptide analysis, especially considering the large degree of compositional and abundance heterogeneity across the glycoproteome. Phlairaharn, Searle, et al. recently demonstrated the benefits of Multiple Accumulation Precursor Mass Spectrometry (MAP-MS) to improve dynamic range limitation in proteomics; inspired by this, we investigated the merits of MAP-MS for glycoproteomics. We explore several multiplexing schemes and related parameters, e.g., dynamic exclusion, to evaluate effects on identification, quantification, and reproducibility of glycopeptide measurements with MAP-MS.

Authors
Isabel Mariano, Miklos Guttman, Abhinav Nath

Institutions
University of Washington

Abstract
Antibody-based therapeutics are tremendously powerful with desirable features such as high target binding affinity and specificity, but their development is challenging, complex, and time-consuming. Poor biophysical properties can drive off-target interactions observed with low specificity and poor stability due to weak, nonspecific interactions occurring in crowded biological environments. To characterize the propensity for an antibody to experience nonspecific interactions with serum, we propose the use of in-serum hydrogen deuterium exchange mass spectrometry (HDX-MS). HDX-MS will broaden our understanding of these nonspecific interactions by allowing for the analysis of the biophysical and chemical properties of the amino acids within an antibody that are driving their local conformational changes and the nonspecific interactions with serum. HDX experiments are performed in serum and buffer to compare structural dynamics of selected antibodies. Antibodies are biotinylated using a disulfide-bearing linker, attached to streptavidin-coated magnetic Dynabeads, and exposed to deuterated serum (made by reconstituting lyophilized pooled human serum in D 2O) or buffer for 1m–24h. Beads are washed with 4M urea pH 2.5 to remove serum proteins, and 4M urea, 200mM TCEP pH 3.0 to release antibodies. Samples are pHed to 2.5 with 10% formic acid before freezing to prevent back-exchange. Following in-line digestion with nepenthesin II and HPLC separation, peptides are detected using a Waters Synapt-G2 Q-TOF or Thermo Orbitrap Ascend Tribrid MS. Byonic is used to identify and map peptides, and HDExaminer to quantify deuterium uptake. In preliminary HDX studies, the biotin label appears to have minimal effect on the dynamics of sotrovimab in buffer. The detection of peptides following in-serum immobilized HDX experiments was successful. The sequence coverage of the light chain was nearly 100% in both conditions, whereas the sequence coverage of the heavy chain was less complete with about 80% coverage following in-buffer HDX and about 65% coverage following in-serum HDX.

Authors
Ruby Zhang, Nicholas M. Riley

Institutions
University of Washington

Abstract
INTRODUCTION Glycan modifications shift glycopeptides to higher m/z values than non-modified peptides which warrants re-evaluation of MS1 scan ranges used in glycoproteomic methods. Data-dependent acquisition (DDA) is a mainstay for glycoproteomics due to intricacies of glycopeptide fragmentation, and as with most DDA experiments, maximizing identifications becomes more challenging as sample complexity increases – a hallmark of the glycoproteome. Offline fractionation can alleviate this complexity to improve sensitivity, but this strategy dramatically increases acquisition time per sample. Here, we investigate strategies for online gas-phase fractionation (GPF) to improve precursor sampling and identification through the lens of glycoproteomics. We compare replicate injections using traditional DDA methods versus binning and tiling methods of GPF that differ in how they split the MS1 m/z range.

METHODS Human cell lysates (K652, HEK293) were digested with trypsin and glycoenriched using a SAX-ERLIC protocol with Oasis MAX cartridges (Waters Corporation). Peptides and glycopeptides were analyzed through 60-minute liquid chromatography-tandem mass spectrometry (LC-MS/MS) acquisitions on a Vanquish Neo UHPLC system interfaced to an Orbitrap Ascend Tribrid MS system (Thermo Fisher Scientific). We used the Autonomous Dissociation-type Selection strategy to fragment N-glycopeptides with HCD and O-glycopeptides with HCD and EThcD. Methods varied the MS1 scan range (generally between 400-1800 m/z, with specific ranges varying widely based on methods), number of injections per sample, dynamic exclusion times (ranging from one second to sixty seconds) and precursor accumulation times (ranging from ~50 to ~150 ms).

PRELIMINARY DATA Enrichment strategies such as lectin affinity and ion exchange have been successfully employed in the extraction of glycopeptides from a complex mixture. However, non-specific interactions during enrichment often lead to heterogeneous mixtures of both modified and unmodified species which are analyzed using LC-MS/MS. We show how shifting MS1 scan ranges to account for higher precursor ion m/z values that come from the added mass of glycan modifications (average 1,000-2,000 Daltons) can favor glycopeptide identifications relative to “out-of-the-box” methods built from proteomic-centric methods. In shifting the lower scan range bound to higher m/z values (e.g., 600 to 800 m/z starting values), we saw a notable increase in glycopeptide identifications and other MS/MS spectra that represent (unidentified) putative glycopeptides based on oxonium content. We also tested how splitting the m/z range of precursor ions that are sampled in a given DDA experiment can improve glycoproteome sampling without the need for offline fractionation. We investigate “binning” approaches, which are analogous to common GPF methods used in DIA library-buildng that split the MS1 m/z range into consecutive m/z ranges that are sampled in multiple LC-MS/MS acquisitions. In our case, these were determined based on ranges containing putative glycopeptides (i.e., 600-900 m/z in injection 1, 900-1200 m/z in injection 2, and 1200-1500 m/z in injection 3). We also investigate “tiling” methods, where we restrict precursor selection to several narrowed, non-consecutive m/z ranges distributed throughout the full MS1 m/z range. In changing these precursor sampling ranges, we also investigate the interplay between dynamic exclusion times and MS/MS ion accumulation times to show how judicious selection of precursor sampling parameters can improve glycopeptide identification. We compare these methods to multiple injections using traditional DDA methods, showing how GPF methods can improve glycoproteome sampling depth when offline fractionation is not feasible or practical.

Authors
Jacob H. Russell and Nicholas M. Riley

Institutions
Department of Chemistry, University of Washington, Seattle, WA, USA

Abstract
Poly-N-acetyl-lactosamine (poly-LacNAc) is a linear carbohydrate motif mainly found on branched N-glycans. Poly-LacNAc glycans contribute to numerous cell-cell interactions in the immune system (e.g., enhanced interactions with immune receptors), but detection of poly-LacNAc-modified glycoproteins is complicated by the motif’s relatively low abundance, high structural heterogeneity, and physiochemical properties. We present a strategy to improve detection and quantitation of poly-LacNAc motifs through enzymatic treatment of N-glycopeptides with endo-β-galactosidase (EBG), i.e., LacNAc-ase. EBG cleaves LacNAc from N-glycans, resulting in characteristic partially galactosylated N-glycoforms. Instead of entirely missing their LacNAc-modified glycoforms in current workflows, our method enables detection of “formerly LacNAcylated” glycopeptides unique to EBG treatment. We use our approach to discern differences in abundance of poly-LacNAc in cutaneous and uveal melanoma cells.

All experiments were performed on an Orbitrap Ascend Tribrid MS (Thermo Fisher Scientific) equipped with a Vanquish Neo UHPLC system. All samples were separated online with a 25 cm long, 75 µm ID Aurora Series Gen3 reversed-phase LC column packed with 1.6 µm C18 particles (IonOpticks). Samples analyzed included normal human serum (Invitrogen), K562, MP41, and A375 lysates. Lysates were treated with EBG in 100 mM sodium phosphate or 100 mM HEPES and 1% SDC at a protein:enzyme ratio of 1:100. Various workflows were tested, including desalting with Strata-X SPE cartridges (Phenomenex) before and after or only after EBG treatment. Glycopeptides were enriched with Oasis-MAX SPE cartridges (Waters). MS/MS data were analyzed using MSFragger-Glyco, O-Pair Search, GlyCounter, and in-house scripts.

Uveal melanoma is an ocular cancer of melanocytes and presents with metastasis in 50% or greater of cases. In contrast, cutaneous melanoma results in a far lower incidence of metastasis, at approximately 4% of cases. Although previous studies have identified poly-LacNAc N-glycoproteins in A375 cutaneous melanoma cells, no studies to date have evaluated the role of these glycoforms in uveal melanoma cells. We have previously shown poly-LacNAc N-glycoproteins to be upregulated in extravillous trophoblasts invading decidual tissue during human placentation, suggesting the involvement of these glycoforms in tissue invasion. Additionally, other studies have shown that poly-LacNAc N-glycoproteins enable cancer cells to evade immune responses in various ways. One way is that poly-LacNAc N-glycoproteins increase the stability of the PD-L1 protein, which binds to the PD-1 receptor and reduces T cell response. Another way is that the sialyl Lewis X [HexNAc(1)Hex(2)NeuAc(1)Fuc(1)] epitope on poly-LacNAc N-glycoproteins binds to selectins, promoting leukocyte-endothelial cell adhesion. These observations highlight the importance of addressing the complex analytical needs of poly-LacNAc N-glycopeptides to directly understand their role in the metastasis of aggressive malignancies. Partially galactosylated glycoforms are characterized by at least one missing hexose from a branched N-glycan. By analyzing these partially galactosylated products of EBG treatment, we enable the identification of formerly poly-LacNAcylated N-glycopeptides, which would otherwise be missed due to poor ionization efficiency, large size, and high microheterogeneity. We find that partially galactosylated glycoforms are significantly upregulated in EBG-treated compared to untreated samples. Additionally, we find that the di-LacNAc [HexNAc(2)Hex(2), m/z 731] oxonium ion signal is depleted following EBG treatment. We show how this workflow enables the study of molecules known to be involved in metastasis in cell lines from two cancers with markedly different metastasis rates, as well as how this workflow could be applied more broadly.

Authors
Andreas FR Huhmer (1), Zhengjian Zhang (1), James Joly (1), Rukshan Perera (1), MaryamJouzi (1), Vivek Budamagunta (1), Brittany Nortman (1), Sanjib Guha (1), Jessica Schwarz (2), Emily Sartori (2), Ana Forton-Juarez (2), Joel Blanchard (2), Greg Kapp (1), Kara Juneau (1), Parag Mallick (1)

Institutions
(1) Business Development, Nautilus Biotechnology, San Carlos, CA, USA, (2) The Ronald Loeb Centerfor Alzheimer’s Disease at Mt.Sinai, Icahn School of Medicine at Mt.Sinai, New York, NY, USA.

Abstract
Background: The microtubule-associated protein tau (MAPT) exhibits substantial molecular heterogeneity arising from alternative splicing and extensive post-translational modifications, particularly phosphorylation. This proteoform diversity is central to Alzheimer’s disease (AD)pathogenesis but remains incompletely characterized in human-relevant disease models. The miBrain system provides a physiologically relevant platform to study AD-associated molecular phenotypes, including those driven by the APOE4 genetic risk factor. Here, we apply a single-molecule proteome analysis approach to compare the tau proteoform landscape in APOE3/3 toAPOE4/4 miBrain models. Methods: Tau proteoforms were analyzed using the Nautilus Single Molecule Proteome AnalysisPlatform, which enables Iterative Mapping of individual protein molecules. Tau extracted fromAPOE3/3 and APOE4/4 miBrains was immobilized on a hyper-dense flow cell and sequentially probed with a panel of 12 isoform- and phosphorylation-specific affinity reagents. Each molecule’s proteoform identity was inferred from its unique binding pattern across iterative probe cycles. Computational analysis corrected for binding inefficiencies and off-target interactions to quantify proteoform abundances with single-molecule resolution. Immunostaining was used to confirm tau phosphorylation in miBrains of both genotypes prior to proteoform analysis. Results: miBrains genotypes exhibited a diverse tau proteoform landscape characterized by distinct differences in phosphorylation states and isoform compositions. Age-dependent changes were observed, including shifts in isoform distribution toward more mature tau species, suggesting progressive tau maturation with miBrain development. Single-molecule analysis revealed multiple AD-associated phosphorylated tau proteoforms, including elevated pS202/pT205 phosphorylation consistent with APOE4/4-associated pathology and AT8 immunoreactivity. Conclusion: Single-molecule Iterative Mapping enables direct, quantitative resolution of tau proteoform heterogeneity in genetically defined human miBrain models. This approach revealsAPOE3/APOE4 genotype-associated differences in tau phosphorylation and isoform maturation that are not accessible through bulk measurements. These findings demonstrate the utility of single-molecule proteoform analysis for mechanistic studies of AD and establish a framework for future investigations linking tau proteoform patterns to disease risk, progression, and therapeutic response.