Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 44 of the 44 references it cites were found in a registry or an index; 21 of the 37 sources its claims cite were checked for support.

Network integration of thermal proteome profiling with multi-omics data decodes PARP inhibition

Open access
Burtscher, Mira L, Gade, Stephan, Garrido-Rodriguez, Martin, Rutkowska, Anna, Werner, Thilo, Eberl, H Christian, Petretich, Massimo, Knopf, Natascha, Zirngibl, Katharina, Grandi, Paola, Bergamini, Giovanna, Bantscheff, Marcus, Fälth-Savitski, Maria, Saez-Rodriguez, Julio · 2024 · Molecular Systems Biology

Abstract

Complex disease phenotypes often span multiple molecular processes. Functional characterization of these processes can shed light on disease mechanisms and drug effects. Thermal Proteome Profiling (TPP) is a mass-spectrometry (MS) based technique assessing changes in thermal protein stability that can serve as proxies of functional protein changes. These unique insights of TPP can complement those obtained by other omics technologies. Here, we show how TPP can be integrated with phosphoproteomics and transcriptomics in a network-based approach using COSMOS, a multi-omics integration framework, to provide an integrated view of transcription factors, kinases and proteins with altered thermal stability. This allowed us to recover consequences of Poly (ADP-ribose) polymerase (PARP) inhibition in ovarian cancer cells on cell cycle and DNA damage response as well as interferon and hippo signaling. We found that TPP offers a complementary perspective to other omics data modalities, and that its integration allowed us to obtain a more complete molecular overview of PARP inhibition. We anticipate that this strategy can be used to integrate functional proteomics with other omics to study molecular processes.

Overview

CoChat Summary

TL;DR

The paper integrates thermal proteome profiling (TPP) with phosphoproteomics and transcriptomics using COSMOS to model olaparib-induced PARP inhibition in UWB1.289 ovarian cancer cells. It reports that TPP adds complementary functional information and improves recovery of cell cycle, DNA damage response, interferon, and hippo/YAP-TEAD signaling.

Key findings

  • After 24 h olaparib treatment, transcriptomics identified 20,493 expressed genes with 44 significantly changed; phosphoproteomics identified 11,615 phosphosites with 256 significantly changed; TPP identified 9,455 proteins with 76 thermal stability changes (FDR < 0.1).
  • Footprint analysis yielded activity estimates for 229 transcription factors and 170 kinases, of which 123/229 and 48/170 were deregulated (absNES > 1.7); top 30 of each were used for network modeling. Prominent TPP hits included CHEK2, PARP1, RNF146, MX1, cyclins (stabilized), and RRN3 (destabilized).
  • TPP hits overlapped little with kinases and transcription factors: three DDR proteins (CDK1, CDK2, CCNB1) overlapped between TPP hits and kinases, and there was no overlap between altered kinases and transcription factors. TPP hits had 149 detected phosphorylation events and appeared more phosphorylated than kinases and transcription factors; 24/30 top kinases had direct connections to at least one TPP hit, and CDK1 had direct edges to 20 TPP hits.
  • Inclusion of TPP in COSMOS networks made cell cycle and DNA damage pathways significantly more prominent (p < 0.05, Wilcoxon test), with some DDR-associated pathways significant exclusively in networks with TPP. Clustering identified interferon signaling and YAP/TEAD signaling; maximum Jaccard index between clusters with and without TPP was 0.28 for DDR and nuclear receptor signaling.

Methods

  • UWB1.289 (BRCA1-deficient ovarian cancer) cells were treated with 4 µM olaparib or DMSO for 24 h; transcriptomics, MS-based proteomics/phosphoproteomics, and 2D-TPP were generated. Three independent replicates were generated per data point per readout, though TPP used four olaparib concentrations (0.4, 1, 4, 10 µM) or DMSO with one replicate per treatment.
  • TPP was performed across 12 temperatures (42.1–63.9 °C), with TMT10 multiplexing; hits were called using an F-statistic-based method with 10% FDR.
  • COSMOS/CARNIVAL integration used Viper footprinting for TF and kinase/phosphatase activities, top 30 TFs (NES > 4), top 30 kinases/phosphatases (NES > 2), and 76 TPP hits; two prior knowledge networks were used (COSMOS, 30k interactions; MetaBase, 117k interactions), with two-round optimization and ReactomePA enrichment.

Limitations

  • Stated: the study is the first example of TPP in network-based multi-omics; more comprehensive datasets covering additional contexts and data modalities are needed to draw conclusions about generalizability. The authors also note that TPP data often contain more than binary stabilization/destabilization information, such as effect size or quantitative drug response curves, which could improve modeling.
  • Evident from the text: the analysis is limited to one cell line (UWB1.289), one PARP inhibitor (olaparib), and one treatment time (24 h). TPP was performed with one replicate per treatment, whereas transcriptomics and proteomics/phosphoproteomics had three independent replicates. The network results also depend on prior knowledge resources, and the authors note an expected influence of the prior knowledge resource.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s44320-024-00025-w

Cites

44 of 44 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Overview of the COSMOS-TPP workflow. Preprocessing and modeling steps of the involved data types are indicated by different colors. Olaparib is an inhibitor of PARP1, an essential player in DNA repair. Inhibition of PARP results in DNA damage that is not repaired in BRCA-mutated cells such as UWB1.289 ovarian cancer cells inducing apoptosis, a concept known as synthetic lethality. The different omic layers acquired after 24 h of Olaparib treatment are processed separately and combined as input for an integer linear programming (ILP) optimization. Additionally, different omics information is used to filter the underlying prior knowledge network (PKN). To obtain a network model describing the interplay of Kinases (Kin), TPP hits (TPP), and transcription factors (TF) two runs are merged into the final network.
Figure 1 Overview of the COSMOS-TPP workflow. Preprocessing and modeling steps of the involved data types are indicated by different colors. Olaparib is an inhibitor of PARP1, an essential player in DNA repair. Inhibition of PARP results in DNA damage that is not repaired in BRCA-mutated cells such as UWB1.289 ovarian cancer cells inducing apoptosis, a concept known as synthetic lethality. The different omic layers acquired after 24 h of Olaparib treatment are processed separately and combined as input for an integer linear programming (ILP) optimization. Additionally, different omics information is used to filter the underlying prior knowledge network (PKN). To obtain a network model describing the interplay of Kinases (Kin), TPP hits (TPP), and transcription factors (TF) two runs are merged into the final network.CC BY · PubMed Central
Figure EV1 Transcriptomics, phosphoproteomics, and TPP data. UWB1.289 cells were treated with 4 uM Olaparib for 24 h to generate direct and indirect treatment effects before multiple omics datasets were acquired. (A) Volcano plot of the phosphoproteomics data after limma analysis comparing Olaparib treatment to a DMSO control (n = 11610 phosphosites, adjusted p.value <0.05, t-test). (B) Volcano plot of the transcriptomics data after limma analysis comparing Olaparib treatment to a DMSO control (n = 13,710 transcripts, adjusted p.value <0.05, t-test). (C) Overview of the Olaparib TPP hits determined using the F-statistic-based hit calling approach.
Figure EV1 Transcriptomics, phosphoproteomics, and TPP data. UWB1.289 cells were treated with 4 uM Olaparib for 24 h to generate direct and indirect treatment effects before multiple omics datasets were acquired. (A) Volcano plot of the phosphoproteomics data after limma analysis comparing Olaparib treatment to a DMSO control (n = 11610 phosphosites, adjusted p.value <0.05, t-test). (B) Volcano plot of the transcriptomics data after limma analysis comparing Olaparib treatment to a DMSO control (n = 13,710 transcripts, adjusted p.value <0.05, t-test). (C) Overview of the Olaparib TPP hits determined using the F-statistic-based hit calling approach.CC BY · PubMed Central
Figure 2 Transcriptomics, phosphoproteomics, and thermal proteome profiling data exploration. (A) Normalized enrichment scores (NES) for the top transcription factors and (B) top kinases as a proxy of enzyme activity based on transcriptomics and phosphoproteomics data. In response to Olaparib treatment, cells react with a strong transcriptional activation and signaling around the DNA damage checkpoint. (C) Comparison of top transcription factors, kinases, and TPP hits which were selected for network modeling overall and subsetted for proteins involved in DNA damage response (DDR). The different protein sets show almost no overlap, the three overlapping proteins between the kinase and the TPP set are CDK1, CDK2, and CCNB. (D) We compared log2 fold-changes of phosphosite intensity upon Olaparib treatment on kinases, TFs, and TPP proteins measured in the phosphoproteomics experiment. TPP proteins show a significant shift (p.value < 0.05, Wilcoxon test) towards higher phosphosite fold-changes in comparison to Kinases and TFs. n = 244 phosphosites in total, 70 on kinases, 26 on TFSs, and 148 on TPP proteins. Boxplots indicate the median, first, and third quartiles. Whiskers extend from the hinges to the largest value no further than 1.5× the interquartile range. Data points beyond the end of the whiskers are plotted individually. (E) TPP proteins further map frequently downstream of kinases identified in the activity estimation analysis, as exemplarily shown for CDK1 with five downstream TPP proteins and related significantly affected phosphosites.
Figure 2 Transcriptomics, phosphoproteomics, and thermal proteome profiling data exploration. (A) Normalized enrichment scores (NES) for the top transcription factors and (B) top kinases as a proxy of enzyme activity based on transcriptomics and phosphoproteomics data. In response to Olaparib treatment, cells react with a strong transcriptional activation and signaling around the DNA damage checkpoint. (C) Comparison of top transcription factors, kinases, and TPP hits which were selected for network modeling overall and subsetted for proteins involved in DNA damage response (DDR). The different protein sets show almost no overlap, the three overlapping proteins between the kinase and the TPP set are CDK1, CDK2, and CCNB. (D) We compared log2 fold-changes of phosphosite intensity upon Olaparib treatment on kinases, TFs, and TPP proteins measured in the phosphoproteomics experiment. TPP proteins show a significant shift (p.value < 0.05, Wilcoxon test) towards higher phosphosite fold-changes in comparison to Kinases and TFs. n = 244 phosphosites in total, 70 on kinases, 26 on TFSs, and 148 on TPP proteins. Boxplots indicate the median, first, and third quartiles. Whiskers extend from the hinges to the largest value no further than 1.5× the interquartile range. Data points beyond the end of the whiskers are plotted individually. (E) TPP proteins further map frequently downstream of kinases identified in the activity estimation analysis, as exemplarily shown for CDK1 with five downstream TPP proteins and related significantly affected phosphosites.CC BY · PubMed Central
Figure EV2 Robustness and reproducibility. We tried to validate whether the chosen setup (combination of runs, activity estimation and filtering steps) can be used to model reasonable networks with COSMOS. (A) We integrated TPP proteins between upstream kinases and downstream transcription factors into the signaling cascade based on the correlation of phosphoproteomic and TPP data. Further, we implemented a strategy for activity estimation where we inferred activities for TPP proteins via upstream kinases and phosphorylation status. For TPP proteins without activity estimate, we inferred the activity state during the optimization supported by a weight reflecting (de)stabilization. (B) Different modeling setups (1–5) were tested to maximize the number of integrated TPP proteins which we reached using activity estimations and weights. 1: no TPP proteins as input. 2: TPP proteins without activity information. 3: TPP with activity estimation. 4: TPP with activity estimation and weights. 5: TPP with activity estimation, NA values and weights. Running the optimization without TPP data still yields few TPP proteins in the solution, due to overlaps with other input modalities or the inclusion of prior knowledge. Details of the optimization process are provided in the CARNIVAL and COSMOS publications (Liu et al, 2019; Dugourd et al, 2021). (C) In principle, the integer linear programming solver optimizes towards a local optimum and the complete search space of the problem is not known. As a consequence, multiple technical replicates of COSMOS can have similar but slightly different optimal solutions. To assess if this solver-dependent variability can be distinguished from actual differences due to the used input data, we set up a small robustness analysis. We clustered replicate networks for different TPP set sizes based on node activities (active, missed, inactive). All replicate runs of distinct TPP sets cluster together. The networks separate into networks with TPP input and networks without TPP input. Further the influence of the used prior knowledge is clearly visible (PKN).
Figure EV2 Robustness and reproducibility. We tried to validate whether the chosen setup (combination of runs, activity estimation and filtering steps) can be used to model reasonable networks with COSMOS. (A) We integrated TPP proteins between upstream kinases and downstream transcription factors into the signaling cascade based on the correlation of phosphoproteomic and TPP data. Further, we implemented a strategy for activity estimation where we inferred activities for TPP proteins via upstream kinases and phosphorylation status. For TPP proteins without activity estimate, we inferred the activity state during the optimization supported by a weight reflecting (de)stabilization. (B) Different modeling setups (1–5) were tested to maximize the number of integrated TPP proteins which we reached using activity estimations and weights. 1: no TPP proteins as input. 2: TPP proteins without activity information. 3: TPP with activity estimation. 4: TPP with activity estimation and weights. 5: TPP with activity estimation, NA values and weights. Running the optimization without TPP data still yields few TPP proteins in the solution, due to overlaps with other input modalities or the inclusion of prior knowledge. Details of the optimization process are provided in the CARNIVAL and COSMOS publications (Liu et al, 2019; Dugourd et al, 2021). (C) In principle, the integer linear programming solver optimizes towards a local optimum and the complete search space of the problem is not known. As a consequence, multiple technical replicates of COSMOS can have similar but slightly different optimal solutions. To assess if this solver-dependent variability can be distinguished from actual differences due to the used input data, we set up a small robustness analysis. We clustered replicate networks for different TPP set sizes based on node activities (active, missed, inactive). All replicate runs of distinct TPP sets cluster together. The networks separate into networks with TPP input and networks without TPP input. Further the influence of the used prior knowledge is clearly visible (PKN).CC BY · PubMed Central
Figure 3 Effect of TPP integration on network level. (A) Comparison of the network-level enrichment results of networks with TPP data and networks without TPP. The enrichment results were filtered for significant pathways (q.value < 0.05, one-tailed Fisher’s exact test) and a minimum size of four nodes. We then extracted pathways related to cell cycle or DNA damage and repair and compared their log odds ratio between networks with TPP and without TPP (p.value < 0.05, Wilcoxon test). n = 480 pathways in total, 43 cell cycle related, 28 DDR related, 409 other pathways. Boxplots indicate the median, first, and third quartiles. Whiskers extend from the hinges to the largest value no further than 1.5× the interquartile range. Data points beyond the end of the whiskers are plotted individually. (B) Graph representation of the network with TPP data. The colors indicate the seven clusters identified using a fast and greedy clustering algorithm. We performed a cluster-specific enrichment of the nodes and summarized the significant pathways to one higher-level process per cluster ranging from cell cycle to interferon signaling. (C) We compared the overlap of TPP and no TPP clusters using the Jaccard index (JI). The colors refer to the respective cluster of the network with TPP. The size of the points refers to the Jaccard index for the given comparison. The maximum overlap reached is a Jaccard index of 0.28.
Figure 3 Effect of TPP integration on network level. (A) Comparison of the network-level enrichment results of networks with TPP data and networks without TPP. The enrichment results were filtered for significant pathways (q.value < 0.05, one-tailed Fisher’s exact test) and a minimum size of four nodes. We then extracted pathways related to cell cycle or DNA damage and repair and compared their log odds ratio between networks with TPP and without TPP (p.value < 0.05, Wilcoxon test). n = 480 pathways in total, 43 cell cycle related, 28 DDR related, 409 other pathways. Boxplots indicate the median, first, and third quartiles. Whiskers extend from the hinges to the largest value no further than 1.5× the interquartile range. Data points beyond the end of the whiskers are plotted individually. (B) Graph representation of the network with TPP data. The colors indicate the seven clusters identified using a fast and greedy clustering algorithm. We performed a cluster-specific enrichment of the nodes and summarized the significant pathways to one higher-level process per cluster ranging from cell cycle to interferon signaling. (C) We compared the overlap of TPP and no TPP clusters using the Jaccard index (JI). The colors refer to the respective cluster of the network with TPP. The size of the points refers to the Jaccard index for the given comparison. The maximum overlap reached is a Jaccard index of 0.28.CC BY · PubMed Central
Figure 4 Mechanistic insight into selected molecular processes. Based on the results of the previous clustering analysis we extracted signaling related to the identified processes from networks with TPP to investigate it in detail. The red color indicates TFs, kinases, and TPP nodes which were used as input, while gray indicates predicted nodes added from the prior knowledge network during the optimization. TPP hits are indicated as stars. We chose three pathways represented by different clusters. (A) DNA damage response and cell cycle arrest, (B) Interferon signaling, and (C) YAP/TEAD signaling.
Figure 4 Mechanistic insight into selected molecular processes. Based on the results of the previous clustering analysis we extracted signaling related to the identified processes from networks with TPP to investigate it in detail. The red color indicates TFs, kinases, and TPP nodes which were used as input, while gray indicates predicted nodes added from the prior knowledge network during the optimization. TPP hits are indicated as stars. We chose three pathways represented by different clusters. (A) DNA damage response and cell cycle arrest, (B) Interferon signaling, and (C) YAP/TEAD signaling.CC BY · PubMed Central

Verification

Claims the paper makes

16 Valid · 3 Not valid · 12 Not checked · 59 Own results and interpretations

41 more claims

Create a free account to see all 90 claims and to verify the 3 references still open.

See all claims — free account
  • Probably not a substantial claim. 32
  • It may state two separate claims as one. 15
  • Borderline: one of the tests for a substantial claim is unclear. 2

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 44 references

Cited by

See all 8

Metadata

Authors
Burtscher, M. L., Gade, S., Garrido-Rodriguez, M., Rutkowska, A., Werner, T., Eberl, H. C., Petretich, M., Knopf, N., Zirngibl, K., Grandi, P., Bergamini, G., Bantscheff, M., Fälth-Savitski, M., & Saez-Rodriguez, J.
Affiliations
  • Heidelberg University (DE)
  • GlaxoSmithKline (Germany) (DE)
Venue
Molecular Systems Biology
Published
2024-03-07
Volume
20
Issue
4
Pages
458-474
ISSN
1744-4292 (ISSN-L)
OpenAlex
W4392544555
Cited by
7
References
44
Citations by year
2026: 3 · 2025: 3 · 2024: 1
FWCI
0.87
Type
journal article
Language
en
Open access
Licence
Locations
  • Molecular Systems Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
  • GlaxoSmithKline
  • Pfizer
  • Sanofi
  • Travere Therapeutics
  • Stadapharm
  • Astex Pharmaceuticals
  • Grunenthal
Topics
  • Physical Sciences › Chemistry › Spectroscopy › Advanced Proteomics Techniques and Applications
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Health Sciences › Medicine › Oncology › PARP inhibition in cancer therapy
Keywords
thermal proteome profiling, PARP inhibition, multi-omics integration, cosmos, phosphoproteomics, transcriptomics, protein thermal stability, ovarian cancer, DNA damage response, cell cycle, interferon signaling, Hippo signaling pathway, transcription factors, kinases, pharmacological activity
MeSH terms
Poly(ADP-ribose) Polymerase Inhibitors, Multiomics, Poly(ADP-ribose) Polymerases, Proteome, Proteomics

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Network integration of thermal proteome profiling with multi-omics data decodes PARP inhibition. https://app.cochat.ai/papers/10.1038/s44320-024-00025-w
The paper
Burtscher, M. L., Gade, S., Garrido-Rodriguez, M., Rutkowska, A., Werner, T., Eberl, H. C., Petretich, M., Knopf, N., Zirngibl, K., Grandi, P., Bergamini, G., Bantscheff, M., Fälth-Savitski, M., & Saez-Rodriguez, J. (2024). Network integration of thermal proteome profiling with multi-omics data decodes PARP inhibition. Molecular Systems Biology. https://doi.org/10.1038/s44320-024-00025-w

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.