MSstatsTMT improves accuracy of thermal proteome profiling by trading off temperature treatments and biological replicates.

Thermal proteome profiling investigates protein-protein, protein-nucleic acid, or protein-drug interactions, and the impact of metabolite binding and post-translational modifications on these interactions. The experiments quantitatively characterize biological samples treated with small molecules versus controls, and subjected to timed exposures to multiple temperatures. Typically, each enzymatically digested sample is labeled with a tandem mass tag (TMT), where each TMT channel corresponds to a specific temperature treatment, and profiled using liquid chromatography coupled with mass spectrometry in data-dependent data acquisition mode. The resulting mass spectra are processed with computational tools to identify and quantify proteins, and filter out noise. Protein-drug interactions are detected by fitting curves to the protein-level reporter ion abundances across the temperatures. Interacting proteins are identified by shifts in the fitted curves between treated samples and controls. In this manuscript, we focus on data processing and curve fitting in thermal proteome profiling. We review the statistical methods currently used for thermal proteome profiling, and demonstrate that such methods can yield substantially different results. We advocate for the statistical analysis strategy implemented in the open-source R package MSstatsTMT, as it does not require subjective pre-filtering of the data or curve fitting, and appropriately represents all the sources of variation. It supports experimental designs that trade-off temperatures for a larger number of biological replicates, and handles multiple drug concentrations or pools of samples treated with multiple temperatures, thus increasing the sensitivity of the results. We demonstrate these advantages of MSstatsTMT as compared to the currently used alternatives in a series of simulated and experimental datasets, which include conventional thermal proteome profiling and its OnePot counterpart that pools the samples treated at multiple temperatures into one sample, and incorporates multiple doses of a drug. The suggested MSstatsTMT-based workflow is documented in publicly available and fully reproducible R vignettes.