Large-scale characterization of drug mechanism of action using proteome-wide thermal shift assays

  1. Jonathan G. Van Vranken
  2. Jiaming Li
  3. Julian Mintseris
  4. Meagan Gadzuk-Shea
  5. Steven P. Gygi
  6. Devin K. Schweppe

  • Curated by eLife

    eLife logo

    eLife assessment

    This study is a useful showcase of a workflow to perform large-scale characterization of drug mechanisms of action using proteomics. The work is backed by solid evidence, however, more statistical analyses and a user-friendly interface to enhance data mining by the readers are recommended. The strengths of this study include the large number of compounds tested within a common workflow and well-described experimental protocols. This will be of broad interest to medicinal chemists, toxicologists, and biochemists.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

In response to an ever-increasing demand of new small molecules therapeutics, numerous chemical and genetic tools have been developed to interrogate compound mechanism of action. Owing to its ability to characterize compound-dependent changes in thermal stability, the proteome-wide thermal shift assay has emerged as a powerful tool in this arsenal. The most recent iterations have drastically improved the overall efficiency of these assays, providing an opportunity to screen compounds at a previously unprecedented rate. Taking advantage of this advance, we quantified 1.498 million thermal stability measurements in response to multiple classes of therapeutic and tool compounds (96 compounds in living cells and 70 compounds in lysates). When interrogating the dataset as a whole, approximately 80% of compounds (with quantifiable targets) caused a significant change in the thermal stability of an annotated target. There was also a wealth of evidence portending off-target engagement despite the extensive use of the compounds in the laboratory and/or clinic. Finally, the combined application of cell- and lysate-based assays, aided in the classification of primary (direct ligand binding) and secondary (indirect) changes in thermal stability. Overall, this study highlights the value of these assays in the drug development process by affording an unbiased and reliable assessment of compound mechanism of action.

Article activity feed

  1. Version published to 10.7554/elife.95595.1 on eLife
  2. Version published to 10.7554/elife.95595 on eLife
  3. eLife

    eLife assessment

    This study is a useful showcase of a workflow to perform large-scale characterization of drug mechanisms of action using proteomics. The work is backed by solid evidence, however, more statistical analyses and a user-friendly interface to enhance data mining by the readers are recommended. The strengths of this study include the large number of compounds tested within a common workflow and well-described experimental protocols. This will be of broad interest to medicinal chemists, toxicologists, and biochemists.

  4. eLife

    Reviewer #1 (Public Review):

    Summary:

    This is an interesting and potentially important paper, which however has some deficiencies.

    Strengths:

    A significant amount of potentially useful data.

    Weaknesses:

    One issue is a confusion of thermal stability with solubility. While thermal stability of a protein is a thermodynamic parameter that can be described by the Gibbs-Helmholtz equation, which relates the free energy difference between the folded and unfolded states as a function of temperature, as well as the entropy of unfolding. What is actually measured in PISA is a change in protein solubility, which is an empirical parameter affected by a great many variables, including the presence and concentration of other ambient proteins and other molecules. One might possibly argue that in TPP, where one measures the melting temperature change ∆Tm, thermal stability plays a decisive or at least an important role, but no such assertion can be made in PISA analysis that measures the solubility shift.

    Another important issue is that the authors claim to have discovered for the first time a number of effects well described in prior literature, sometimes a decade ago. For instance, they marvel at the differences between the solubility changes observed in lysate versus intact cells, while this difference has been investigated in a number of prior studies. No reference to these studies is given during the relevant discussion.

    The validity of statistical analysis raises concern. In fact, no calculation of statistical power is provided. As only two replicates were used in most cases, the statistical power must have been pretty limited. Also, there seems to be an absence of the multiple-hypothesis correction.

    Also, the authors forgot that whatever results PISA produces, even at high statistical significance, represent just a prediction that needs to be validated by orthogonal means. In the absolute majority of cases such validation is missing.

    Finally, to be a community-useful resource the paper needs to provide the dataset with a user interface so that the users can data-mine on their own.

  5. eLife

    Reviewer #2 (Public Review):

    Summary:

    Using K562 (Leukemia) cells as an experimental model, Van Vracken et. al. use Thermal Proteome Profiling (TPP) to investigate changes in protein stability after exposing either live cells or crude cell lysates to a library of anti-cancer drugs. This was a large-scale and highly ambitious study, involving thousands of hours of mass spectrometry instrument time. The authors used an innovative combination of TPP together with Proteome Integral Solubility Alternation (PISA) assays to reduce the amount of instrument time needed, without compromising on the amount of data obtained.

    The paper is very well written, the relevance of this work is immediately apparent, and the results are well-explained and easy to follow even for a non-expert. The figures are well-presented. The methods appear to be explained in sufficient detail to allow others to reproduce the work.

    Strengths:

    Using CDK4/6 inhibitors, the authors observe strong changes in protein stability upon exposure to the drug. This is expected and shows their methodology is robust. Further, it adds confidence when the authors report changes in protein stability for drugs whose targets are not well-known. Many of the drugs used in this study - even those whose protein targets are already known - display numerous off-target effects. Although many of these are not rigorously followed up in this current study, the authors rightly highlight this point as a focus for future work.

    Weaknesses:

    While the off-target effects of several drugs could've been more rigorously investigated, it is clear the authors have already put a tremendous amount of time and effort into this study. The authors have made their entire dataset available to the scientific community - this will be a valuable resource to others working in the fields of cancer biology/drug discovery.

  6. eLife

    Reviewer #3 (Public Review):

    Summary:

    This work aims to demonstrate how recent advances in thermal stability assays can be utilised to screen chemical libraries and determine the compound mechanism of action. Focusing on 96 compounds with known mechanisms of action, they use the PISA assay to measure changes in protein stability upon treatment with a high dose (10uM) in live K562 cells and whole cell lysates from K562 or HCT116. They intend this work to showcase a robust workflow that can serve as a roadmap for future studies.

    Strengths:

    The major strength of this study is the combination of live and whole cell lysates experiments. This allows the authors to compare the results from these two approaches to identify novel ligand-induced changes in thermal stability with greater confidence. More usefully, this also enables the authors to separate the primary and secondary effects of the compounds within the live cell assay.

    The study also benefits from the number of compounds tested within the same framework, which allows the authors to make direct comparisons between compounds.

    These two strengths are combined when they compare CHEK1 inhibitors and suggest that AZD-7762 likely induces secondary destabilisation of CRKL through off-target engagement with tyrosine kinases.

    Weaknesses:

    One of the stated benefits of PISA compared to the TPP in the original publication (Gaetani et al 2019) was that the reduced number of samples required allows more replicate experiments to be performed. Despite this, the authors of this study performed only duplicate experiments. They acknowledge this precludes the use of frequentist statistical tests to identify significant changes in protein stability. Instead, they apply an 'empirically derived framework' in which they apply two thresholds to the fold change vs DMSO: absolute z-score (calculated from all compounds for a protein) > 3.5 and absolute log2 fold-change > 0.2. They state that the fold-change threshold was necessary to exclude non-specific interactors. While the thresholds appear relatively stringent, this approach will likely reduce the robustness of their findings in comparison to an experimental design incorporating more replicates. Firstly, the magnitude of the effect size should not be taken as a proxy for the importance of the effect. They acknowledge this and demonstrate it using their data for PIK3CB and p38α inhibitors (Figures 2B-C). They have thus likely missed many small, but biologically relevant changes in thermal stability due to the fold-change threshold. Secondly, this approach relies upon the fold-changes between DMSO and compound for each protein being comparable, despite them being drawn from samples spread across 16 TMT multiplexes. Each multiplex necessitates a separate MS run and the quantification of a distinct set of peptides, from which the protein-level abundances are estimated. Thus, it is unlikely the fold changes for unaffected proteins are drawn from the same distribution, which is an unstated assumption of their thresholding approach. The authors could alleviate the second concern by demonstrating that there is very little or no batch effect across the TMT multiplexes. However, the first concern would remain. The limitations of their approach could have been avoided with more replicates and the use of an appropriate statistical test. It would be helpful if the authors could clarify if any of the missed targets passed the z-score threshold but fell below the fold-change threshold.

    The authors use a single, high, concentration of 10uM for all compounds. Given that many of the compounds likely have low nM IC50s, this concentration will often be multiple orders of magnitude above the one at which they inhibit their target. This makes it difficult to assess the relevance of the off-target effects identified to clinical applications of the compounds or biological experiments. The authors acknowledge this and use ranges of concentrations for follow-up studies (e.g. Figure 2E-F). Nonetheless, this weakness is present for the vast bulk of the data presented.

    The authors claim that combining cell-based and lysate-based assays increases coverage (Figure 3F) is not supported by their data. The '% targets' presented in Figure 3F have a different denominator for each bar. As it stands, all 49 targets quantified in both assays which have a significant change in thermal stability may be significant in the cell-based assay. If so, the apparent increase in % targets when combining reflects only the subsetting of the data. To alleviate this lack of clarity, the authors could update Figure 3F so that all three bars present the % targets figure for just the 60 compounds present in both assays.

    Aims achieved, impact and utility:

    The authors have achieved their main aim of presenting a workflow that serves to demonstrate the potential value of this approach. However, by using a single high dose of each compound and failing to adequately replicate their experiments and instead applying heuristic thresholds, they have limited the impact of their findings. Their results will be a useful resource for researchers wishing to explore potential off-target interactions and/or mechanisms of action for these 96 compounds, but are expected to be superseded by more robust datasets in the near future. The most valuable aspect of the study is the demonstration that combining live cell and whole cell lysate PISA assays across multiple related compounds can help to elucidate the mechanisms of action.

  7. Version published to 10.1101/2024.01.26.577428v3 on bioRxiv
  8. Version published to 10.1101/2024.01.26.577428v2 on bioRxiv
  9. Version published to 10.1101/2024.01.26.577428v1 on bioRxiv