Identifying the precise protein targets of small-molecule drugs is crucial for drug development. Existing methods, such as those using chemical probes, thermal proteome profiling (TPP), stability of proteins from rates of oxidation (SPROX), and drug affinity responsive target stability (DARTS), each have limitations. Chemical probes can perturb interactions, while TPP, SPROX, and DARTS lack enrichment steps and may miss low-abundance targets. A significant gap remains in approaches that map interactions proteome-wide without drug labeling and identify binding sites with peptide-level resolution. The authors previously used limited proteolysis (LiP) coupled with mass spectrometry (LiP-SMap) to map metabolite binding proteins in microbial organisms. This study adapts this LiP-based approach for complex eukaryotic proteomes, introducing a machine learning framework called LiP-Quant to enhance target identification and reduce noise. The study focuses on identifying drug targets, a high-impact application in drug discovery, using various known and novel drug compounds to evaluate LiP-Quant's performance and capabilities.

The introduction thoroughly reviews existing chemoproteomics techniques for drug target identification. It highlights the strengths and weaknesses of methods like chemical probes, TPP, SPROX, and DARTS, emphasizing the need for a method that overcomes their limitations, specifically the need for a label-free approach capable of proteome-wide analysis with peptide-level resolution for binding site identification. The authors’ previous work with LiP-SMap in microbial systems is cited as the foundation for the development of LiP-Quant. This review sets the stage for introducing LiP-Quant as a novel solution to the existing challenges.

LiP-Quant integrates limited proteolysis (LiP) with mass spectrometry (MS) and machine learning. The methodology involves a drug dose titration series on protein lysates, followed by limited protease cleavage with proteinase K. Changes in proteolytic patterns upon compound binding are measured using label-free Data Independent Acquisition Mass Spectrometry (DIA-MS). A machine learning algorithm analyzes the resulting data to identify four key features contributing to a 'LiP-Quant score': (I) correlation to a sigmoidal dose-response curve, (II) frequency of protein identification in experiments where it is not a confirmed target (protein frequency library, PFL), (III) number of peptides from a protein with high dose-response correlations, and (IV) statistical significance of differential peptides. A composite LiP-Quant score is generated, ranking peptides and proteins by their likelihood of being a genuine drug target. The threshold for a putative target is set at 1.5, which corresponds to the median non-target peptide score plus three standard deviations. The methodology includes detailed descriptions of sample preparation, MS acquisition, data analysis using Spectronaut and SpectroMine software, and machine learning training using linear discriminant analysis (LDA). The study also details methods for estimating EC50 values, approximating drug binding site positions, and performing cellular thermal shift assays (CETSA). Different LC-MS gradients (2h and 4h) were used to compare the performance of the standard and a "Deep" LiP-Quant method. The paper clearly outlines the methods for data analysis, including filtering criteria, statistical tests, and the algorithms used for target deconvolution. Detailed procedures are given for all experiments, including those involving yeast and human cell lines, various drug compounds, and additional validation techniques (e.g., CETSA and kinase assays).

LiP-Quant successfully identified known drug targets across different compound classes (kinase inhibitors, phosphatase inhibitors, membrane protein drugs), demonstrating its broad applicability and effectiveness. Using rapamycin, a drug with a single known target (FKBP1A), LiP-Quant correctly identified FKBP1A as the top-ranking target in both HeLa cell lysates and live cells. The method also performed well with FK506, another FKBP1A inhibitor. Consistent results were obtained in *S. cerevisiae*, identifying the known rapamycin target FRP1 as the top hit. The study compared LiP-Quant with TPP and kinobeads using staurosporine, a promiscuous kinase inhibitor. While TPP showed higher sensitivity, identifying more kinases, LiP-Quant exhibited comparable performance for identifying true positive targets amongst the top-ranking candidates. The use of longer LC-MS gradients (Deep LiP-Quant) improved LiP-Quant's performance, increasing the number of kinases identified. LiP-Quant accurately profiled selumetinib's specificity as a MAPK inhibitor, unlike the broad-spectrum kinase inhibitor staurosporine, highlighting its ability to differentiate between promiscuous and specific binders. The method successfully discriminated between highly homologous protein targets, like calyculin A and fostriecin which demonstrated distinct specificity profiles towards serine/threonine phosphatases. LiP-Quant provided peptide-level resolution, allowing for the approximation of drug binding sites. The distance between the center of mass of the highest-scoring LiP-Quant peptides and the known drug binding sites was within Van der Waals distance in most cases. Finally, LiP-Quant successfully identified the previously unknown target of a fungicide research compound (BAYE-004) as a *B. cinerea* homolog of casein kinase I, confirmed by CETSA and kinase assays. The predicted binding site of BAYE-004 was mapped to the ATP-binding site of this kinase.

LiP-Quant offers a significant advancement in chemoproteomics by combining limited proteolysis, mass spectrometry, and machine learning to identify drug targets and binding sites without requiring drug modifications. The method's ability to discriminate between direct and indirect targets and to distinguish between highly homologous proteins with subtle differences in binding affinity is a key advantage. The comparison with TPP and kinobeads demonstrates LiP-Quant's comparable performance and its complementarity to existing methods. The successful identification of the unknown fungicide target highlights the potential of LiP-Quant for drug discovery. The ability to estimate EC50 values directly from cell lysates provides a more physiologically relevant measure of drug-target affinity than in vitro assays. Improvements in proteome coverage via longer LC-MS gradients enhance LiP-Quant's sensitivity, particularly for membrane proteins and kinases.

LiP-Quant represents a powerful new tool for drug target identification. Its label-free approach, peptide-level resolution, and ability to estimate binding affinities make it a valuable addition to the drug discovery pipeline. Future work could focus on further enhancing proteome coverage, improving the accuracy of EC50 estimations, and expanding the application to other biological systems.

The study notes that LiP-Quant's sensitivity might be limited by protein sequence coverage, particularly for membrane proteins. While the use of longer LC-MS gradients (Deep LiP-Quant) addresses this issue to some extent, it increases experimental time. EC50 values estimated from cell lysates might be higher than those measured in vitro, potentially due to competition from other targets or other factors in the complex cellular environment. The study's reliance on cell lysates could miss targets requiring intact cellular structures or processes for interaction.