Nonparametric Analysis of Thermal Proteome Profiles Reveals Novel Drug-binding Proteins*

Dorothee Childs,Karsten Bach,Holger Franken,Simon Anders,Nils Kurzawa,Marcus Bantscheff,Mikhail M Savitski,Wolfgang Huber

doi:10.1074/mcp.tir119.001481

Abstract

Detecting the targets of drugs and other molecules in intact cellular contexts is a major objective in drug discovery and in biology more broadly. Thermal proteome profiling (TPP) pursues this aim at proteome-wide scale by inferring target engagement from its effects on temperature-dependent protein denaturation. However, a key challenge of TPP is the statistical analysis of the measured melting curves with controlled false discovery rates at high proteome coverage and detection power. We present nonparametric analysis of response curves (NPARC), a statistical method for TPP based on functional data analysis and nonlinear regression. We evaluate NPARC on five independent TPP data sets and observe that it is able to detect subtle changes in any region of the melting curves, reliably detects the known targets, and outperforms a melting point-centric, single-parameter fitting approach in terms of specificity and sensitivity. NPARC can be combined with established analysis of variance (ANOVA) statistics and enables flexible, factorial experimental designs and replication levels. An open source software implementation of NPARC is provided.

Highlights

Determining the cellular interaction partners of drugs and other small molecules remains a key challenge [1, 2, 3, 4]
We present non-parametric analysis of response curves (NPARC), a statistical method for Thermal proteome profiling (TPP) based on functional data analysis and nonlinear regression
Panobinostat is a broad-spectrum histone deacetylase (HDAC) inhibitor known to interact with HDAC1, HDAC2, HDAC6, HDAC8, HDAC10, and tetratricopeptide repeat protein 38 (TTC38) [6]

Summary

Introduction

Determining the cellular interaction partners of drugs and other small molecules remains a key challenge [1, 2, 3, 4]. The method, nonparametric analysis of response curves (NPARC), is based on a branch of statistical data analysis that works on continuous functions rather than individual numbers, termed functional data analysis [22] It considers the measured melting curves as samples from an underlying stochastic process with a smooth mean function—which can be modelled parametrically or non-parametrically [23]—and constructs its hypothesis tests directly on these samples. Reliable estimates of the null distribution of this statistic can be obtained, it shows higher sensitivity for small but reproducible effects, and failures due to model misspecification or outliers are reduced This increases proteome coverage, which can make the difference between missing or detecting an important drug target. While the cancer drugs interact with limited sets of proteins, the two other compounds are promiscuous binders and affect the thermostability of a large fraction of the cellular proteome

Experimental Procedures

Results

Discussion