Decoding mechanism of action and sensitivity to drug candidates from integrated transcriptome and chromatin state

  1. Caterina Carraro
  2. Lorenzo Bonaguro
  3. Jonas Schulte-Schrepping
  4. Arik Horne
  5. Marie Oestreich
  6. Stefanie Warnat-Herresthal
  7. Tim Helbing
  8. Michele De Franco
  9. Kristian Haendler
  10. Sach Mukherjee
  11. Thomas Ulas
  12. Valentina Gandin
  13. Richard Goettlich
  14. Anna C Aschenbrenner
  15. Joachim L Schultze
  16. Barbara Gatto

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This article applies a systems biology approach to understand the mechanism of action of 3-chloropiperidines (a class of anticancer drugs) in cancer cells and evaluate their sensitivity to drugs. It integrates transcriptomic and open-chromatin data and utilizes sound statistical frameworks for building a sensitivity model. The author's methodology can be applied to early-stage drug discovery. This paper will be of interest to the large class of people who tried to understand how omics data will help drug discovery. It sets a new framework to integrate transcriptome and chromatin accessibility data to identify the key mechanisms of action and provide potential disease targets, which will help speed up the early phases of drug discovery.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. All three reviewers agreed to share their names with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Omics-based technologies are driving major advances in precision medicine, but efforts are still required to consolidate their use in drug discovery. In this work, we exemplify the use of multi-omics to support the development of 3-chloropiperidines, a new class of candidate anticancer agents. Combined analyses of transcriptome and chromatin accessibility elucidated the mechanisms underlying sensitivity to test agents. Furthermore, we implemented a new versatile strategy for the integration of RNA- and ATAC-seq (Assay for Transposase-Accessible Chromatin) data, able to accelerate and extend the standalone analyses of distinct omic layers. This platform guided the construction of a perturbation-informed basal signature predicting cancer cell lines’ sensitivity and to further direct compound development against specific tumor types. Overall, this approach offers a scalable pipeline to support the early phases of drug discovery, understanding of mechanisms, and potentially inform the positioning of therapeutics in the clinic.

Article activity feed

  1. Version published to 10.7554/elife.78012 on eLife
  2. eLife

    Evaluation Summary:

    This article applies a systems biology approach to understand the mechanism of action of 3-chloropiperidines (a class of anticancer drugs) in cancer cells and evaluate their sensitivity to drugs. It integrates transcriptomic and open-chromatin data and utilizes sound statistical frameworks for building a sensitivity model. The author's methodology can be applied to early-stage drug discovery. This paper will be of interest to the large class of people who tried to understand how omics data will help drug discovery. It sets a new framework to integrate transcriptome and chromatin accessibility data to identify the key mechanisms of action and provide potential disease targets, which will help speed up the early phases of drug discovery.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. All three reviewers agreed to share their names with the authors.)

  3. eLife

    Reviewer #1 (Public Review):

    They adopted a comprehensive experimental and analytic approach to understand molecular and cellular mechanisms underlying tissue-specific responses against 3-CePs. They used two cell lines - BxPC-3 and HCT-15 - as example models for responsive and non-responsive cell lines, respectively. Although mutation rates didn't differ by the drug treatment, they observed changes in cell cycle and expression of genes involved in DNA damage, repair and so on. Furthermore, they combined RNA-seq and ATAC-seq data and applied two approaches, pairwise and crosswise, to identify a number of gene groups that are altered in each cell line upon the drug treatment. Finally, they calculated enrichment of up/down genes in different cell lines, tumor types and samples to estimate potential responsitivity against the drug. This study is unique in in-depth analysis of RNA-seq and ATAC-seq data in identifying genetic signature underlying drug treatment. This study has the potential to be applied to different drugs and cell lines.

    However, several major concerns need to be resolved. First of all, the biological and clinical performance of 3-CePs is not clearly described. They referenced several papers but they seem to have focused on the chemical properties of the drug. Without proven activity of 3-CePs against cancers in vitro and in vivo, the rationale of the study would be compromised.

    Their RNA-seq analysis was focused on discovering differentially expressed genes between cell lines, time points, etc. Interestingly, they found that DNA damage and repair signal was specifically increased in HCT-15. But is this approach capable of finding signals that are constitutively expressed in different cell lines? In other words, what if the differential responsiveness to 3-CePs was already there even before the drug was introduced?

    Are there any overlapping signals between pairwise vs crosswise approaches?

    Probably a similar question with the above: is this methodology applicable to other drugs in addition to 3-CePs?

  4. eLife

    Reviewer #2 (Public Review):

    Carraro et al. describe a framework to understand MoA and susceptibility of drug candidates by integrating RNA-seq and ATAC-seq information. More specifically, by collecting drug responses from high-sensitive and low-sensitive cell lines, the authors identified a key set of pathways with co-expression analysis, and further predicted sensitivity of different cancer cell lines.

    The authors provided a new bioinformatics pipeline to integrate multi-omics data (RNA-seq and ATAC-seq) in a drug response study. This approach increased detection power and identified additional key pathways that are associated with drug 3-CePs. This framework has the potential to be applied to the general drug discovery process.

    However, the current manuscript failed to describe the integration methodology in a clear and concise way. Without a full understanding of the methodology, it's tough to evaluate the downstream results in an unbiased manner. In addition, the authors didn't mention how much additional value this multi-omics approach provided compared to the single-omic data set, as multi-omics approaches are more expensive and labor-intensive.

  5. eLife

    Reviewer #3 (Public Review):

    Carraro et al utilize systems biology approaches to decode the mechanism of action of 3-chloropiperidines (a novel class of cancer therapeutics) in cancer cell lines and build a drug-sensitivity model from the data that they evaluate using samples from The Cancer Genome Atlas and cancer cell lines. The approach provides a framework for integrating transcriptomic and open-chromatin data to better understand the mechanism of action of drugs on cancer cell types. The author's approach is of sound design, is clearly explained, and is bolstered by validation via holdout sets and analysis in new cell lines which lends the findings and approach credibility.

    The major strength of this approach is the depth of information provided by performing RNA-seq and ATAC-seq on cells treated with 3-CePs at various time points, and the author's utilization of this data to perform pairwise and crosswise analyses. Their approach identified gene modules that were indicative of why one cell type was more sensitive to a particular drug compared to another. The data was then used to build a sensitivity model which could be applied to samples from The Cancer Genome Atlas, and the authors evaluated their sensitivity predictions on a set of cancer cell lines which validated the predictions.

    The major drawback to this type of approach is that it relies on next-generation sequencing (somewhat costly) and requires intricate bioinformatics analyses. While I agree with the author's perspective that this approach can be applied to additional classes of drugs and cancer samples, I disagree with their view that it is efficient and versatile. However, for research teams with the means to perform both transcriptomic and open-chromatin studies, I think this integrated approach has promise for evaluating novel classes of drugs, particularly in cancer cell lines that are easy to manipulate in vitro.

    While there are examples of similar frameworks being applied to drug development, this work will add to the body of literature utilizing an integrated systems biology approach for pairing drugs with specific tumor or cancer types and understanding their mechanism of action on an epigenetic level.

  6. eLife

    Author Response

    Reviewer 1

    They adopted a comprehensive experimental and analytic approach to understand molecular and cellular mechanisms underlying tissue-specific responses against 3-CePs. They used two cell lines - BxPC-3 and HCT-15 - as example models for responsive and non-responsive cell lines, respectively. Although mutation rates didn’t differ by the drug treatment, they observed changes in cell cycle and expression of genes involved in DNA damage, repair and so on. Furthermore, they combined RNA-seq and ATAC-seq data and applied two approaches, pairwise and crosswise, to identify a number of gene groups that are altered in each cell line upon the drug treatment. Finally, they calculated enrichment of up/down genes in different cell lines, tumor types and samples to estimate potential responsitivity against the drug. This study is unique in in-depth analysis of RNA-seq and ATAC-seq data in identifying genetic signature underlying drug treatment. This study has the potential to be applied to different drugs and cell lines.

    We thank the reviewer for the precise and kind summary of our work.

    However, several major concerns need to be resolved. First of all, the biological and clinical performance of 3-CePs is not clearly described. They referenced several papers but they seem to have focused on the chemical properties of the drug. Without proven activity of 3-CePs against cancers in vitro and in vivo, the rationale of the study would be compromised.

    We apologize for not being clear enough when introducing previous findings on the differential sensitivity of HCT-15 and BxPC-3 cancer cell lines to 3-CePs. In the revised manuscript, we now cite references on the preferential activity of these agents against the pancreatic cancer cell line in 2D and 3D in vitro cancer models (see lines 71-74, 128-129). These compounds have been selected to exemplify the use of the pipeline in drug discovery and early-stage of drug development: indeed, only cellular data are available for these molecules, which have not yet been characterized in vivo. The pipeline itself offered a final perspective on directions to take for their further development, i.e. most sensitive tumor types to target (PAAD, KIRC).

    Their RNA-seq analysis was focused on discovering differentially expressed genes between cell lines, time points, etc. Interestingly, they found that DNA damage and repair signal was specifically increased in HCT-15. But is this approach capable of finding signals that are constitutively expressed in different cell lines? In other words, what if the differential responsiveness to 3-CePs was already there even before the drug was introduced?

    We thank the reviewer for pointing out such key concept. The premise for the developed approach is that factors determining the overall cellular sensitivity to a treatment must be determined by intrinsic characteristics of the cell line. For this reason, we built the sensitivity signature on basal transcriptome profiles, where we prioritized a subset of genes based on perturbational evidence (perturbation-informed basal signature).

    Beyond signature genes, we show in figure R1 (see above) the results of a GSEA analysis on the whole overlap (300 genes) between DE genes from the baseline comparison (BxPC-3 ctrl vs HCT-15 ctrl) and those from the 6 h M treatment comparison, in the sensitive cell line (BxPC-3 M 6 h vs BxPC-3 ctrl). Pathways like ribosome biogenesis, ROS metabolism, UPR also arise, attesting that genes activated in response to the treatment also have a constitutively different expression in unperturbed cells.

    Are there any overlapping signals between pairwise vs crosswise approaches?

    We thank the reviewer for this question. To make it easier for the reader to compare the output from the two types of integration and to intuitively grasp their functional overlap, we changed the visualization of the results from the pairwise approach (Figure 4 D).
    Indeed, some functional pathways both new or already emerging from previous analysis, arise from both integrations. This overlap has now been directly discussed from the functional point of view in the main text (from line 348 and in the following crosswise integration paragraph).

    Genes used as input in both types of integration are DE or DAR-associated, so this means that many of the hits that we find having the same double regulation (pairwise) also appear in CoCena modules. Among them, only few hits show both 1) the same double regulation in a specified comparison (as suggested by crosswise) and also 2) end up having the similar pattern of regulation across all conditions (contributing to the same CoCena module, one of the strengths of the crosswise integration). Indeed, while the pairwise integration checks one single comparison per time, CoCena checks the pattern throughout conditions providing a more holistic view of the gene regulation (e.g one gene can have a different pattern across conditions at the transcriptional and chromatin level). This is due to the biological fact that RNA and chromatin regulation is not 1:1 (also, for instance, from a timing perspective).

    The major added value of the two approaches consists in their intrinsically different output information. Within a specific comparison, the pairwise integration detects genes consistently activated at the transcriptome and chromatin level. At this information level gene set enrichment can simplify the coherent functional role of this set of genes; we now report this extra information in figure 4 to provide a more granular description of the pairwise integration. Instead, CoCena analyzes the pattern throughout conditions, and clusters together genes and peaks that behave similarly. Functional annotation of genes behaving similarly can put together promoters and/or transcripts that together may orchestrate a specific process (as highlighted by GSEA on each module).

    Probably a similar question with the above: is this methodology applicable to other drugs in addition to 3-CePs?

    To address this extremely important point, that we agree with the reviewer would be key to prove the versatility of our approach, we further applied the pipeline to the prediction of cancer cell lines’ sensitivity to cisplatin, a thoroughly reported broad-acting chemotherapeutic also acting as a DNA damaging agent. Results strongly supported the broad applicability of our approach, which was able to predict sensitivity to this reference drug with extremely high accuracy.

    Reviewer 2

    Carraro et al. describe a framework to understand MoA and susceptibility of drug candidates by integrating RNA-seq and ATAC-seq information. More specifically, by collecting drug responses from high-sensitive and low-sensitive cell lines, the authors identified a key set of pathways with co-expression analysis, and further predicted sensitivity of different cancer cell lines.

    The authors provided a new bioinformatics pipeline to integrate multi-omics data (RNA-seq and ATAC-seq) in a drug response study. This approach increased detection power and identified additional key pathways that are associated with drug 3-CePs. This framework has the potential to be applied to the general drug discovery process.

    We thank the reviewer for the precise summary of our study.

    However, the current manuscript failed to describe the integration methodology in a clear and concise way. Without a full understanding of the methodology, it’s tough to evaluate the downstream results in an unbiased manner.

    We apologize for not having included sufficient details in describing the difference between CoCena and the other two horizontal and vertical approaches. As already discussed in the response to Reviewer 1, we now included a more detailed description not only in the Methods section (from line 894) but also in the main text (lines 393-400).

    In addition, the authors didn’t mention how much additional value this multi-omics approach provided compared to the single-omic data set, as multi-omics approaches are more expensive and labor-intensive.

    We thank the reviewer for this valuable point. To better support the claim for multi-omics approaches, we have extended the Introduction (lines 96-98), as successful integration of information derived from multiple omic layers usually strengthens the determination of the major observed cellular responses. Here, this information helps dissecting and predicting how perturbations (here by drugs) can affect the overall cellular dynamics and mechanisms underlying a certain niveau of sensitivity. We agree with the reviewer that current costs are still prohibitive for large scale use of multi-layer omics in many settings, mainly when it comes to clinical use or drug development. However, significantly less expensive technologies (90% cost-reductions, lines 53-55) have recently been announced, which assures us that approaches as outlined here, will be applicable to many more clinical questions in the near future. Further, we show evidence that some cellular responses to the drug-induced perturbation was only revealed by applying multilayer analysis, but not by a single omics layer, e.g. TGF beta and EMT signaling (see lines 456-459).

    Reviewer 3

    Carraro et al utilize systems biology approaches to decode the mechanism of action of 3chloropiperidines (a novel class of cancer therapeutics) in cancer cell lines and build a drugsensitivity model from the data that they evaluate using samples from The Cancer Genome Atlas and cancer cell lines. The approach provides a framework for integrating transcriptomic and open-chromatin data to better understand the mechanism of action of drugs on cancer cell types. The author’s approach is of sound design, is clearly explained, and is bolstered by validation via holdout sets and analysis in new cell lines which lends the findings and approach credibility.

    The major strength of this approach is the depth of information provided by performing RNA-seq and ATAC-seq on cells treated with 3-CePs at various time points, and the author’s utilization of this data to perform pairwise and crosswise analyses. Their approach identified gene modules that were indicative of why one cell type was more sensitive to a particular drug compared to another. The data was then used to build a sensitivity model which could be applied to samples from The Cancer Genome Atlas, and the authors evaluated their sensitivity predictions on a set of cancer cell lines which validated the predictions.

    We thank the reviewer for the accurate recapitulation of our work.

    The major drawback to this type of approach is that it relies on next-generation sequencing (somewhat costly) and requires intricate bioinformatics analyses. While I agree with the author’s perspective that this approach can be applied to additional classes of drugs and cancer samples, I disagree with their view that it is efficient and versatile. However, for research teams with the means to perform both transcriptomic and open-chromatin studies, I think this integrated approach has promise for evaluating novel classes of drugs, particularly in cancer cell lines that are easy to manipulate in vitro.

    We thank the reviewer for this insightful comment. As with almost every technology, the early years are more difficult and at times adventurous. However, we have seen enormous improvements in robustness of the technology and significant cost reduction with more to come. Only recently sequencing technologies have been introduced into the market with a further 90% cost reduction (as stated in line 53-55). We are convinced that due to their increasing affordability and robustness, RNA-seq and ATAC-seq will be implemented routinely into clinical contexts. As a group working at the cross-section between drug discovery and bioinformatics, we hope that our current work, accompanied by a fair and detailed sharing of our scripts, will become a head start to run this type of analysis also by others in the field who are not (yet) so close to bioinformatics and computational biology.

    While there are examples of similar frameworks being applied to drug development, this work will add to the body of literature utilizing an integrated systems biology approach for pairing drugs with specific tumor or cancer types and understanding their mechanism of action on an epigenetic level.

    We thank the reviewer for this very positive statement and the support for our approach and her/his interest in the described pipeline.

  11. Version published to 10.1101/2022.02.21.481294v1 on bioRxiv