Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2024 May 9;15(1):3909.
doi: 10.1038/s41467-024-47957-3.

Network-based elucidation of colon cancer drug resistance mechanisms by phosphoproteomic time-series analysis

George Rosenberger #  1 , Wenxue Li #  2 , Mikko Turunen #  1 , Jing He #  1   3 , Prem S Subramaniam  1 , Sergey Pampou  1   4 , Aaron T Griffin  1   5 , Charles Karan  1   4 , Patrick Kerwin  1 , Diana Murray  1 , Barry Honig  1   6   7   8   9 , Yansheng Liu  10   11 , Andrea Califano  12   13   14   15   16   17
Affiliations

Network-based elucidation of colon cancer drug resistance mechanisms by phosphoproteomic time-series analysis

George Rosenberger et al. Nat Commun. .

Abstract

Aberrant signaling pathway activity is a hallmark of tumorigenesis and progression, which has guided targeted inhibitor design for over 30 years. Yet, adaptive resistance mechanisms, induced by rapid, context-specific signaling network rewiring, continue to challenge therapeutic efficacy. Leveraging progress in proteomic technologies and network-based methodologies, we introduce Virtual Enrichment-based Signaling Protein-activity Analysis (VESPA)-an algorithm designed to elucidate mechanisms of cell response and adaptation to drug perturbations-and use it to analyze 7-point phosphoproteomic time series from colorectal cancer cells treated with clinically-relevant inhibitors and control media. Interrogating tumor-specific enzyme/substrate interactions accurately infers kinase and phosphatase activity, based on their substrate phosphorylation state, effectively accounting for signal crosstalk and sparse phosphoproteome coverage. The analysis elucidates time-dependent signaling pathway response to each drug perturbation and, more importantly, cell adaptive response and rewiring, experimentally confirmed by CRISPR knock-out assays, suggesting broad applicability to cancer and other diseases.

PubMed Disclaimer

Conflict of interest statement

A.C. is the founder, equity holder, and consultant of DarwinHealth Inc., a company that has licensed some of the algorithms used in this manuscript from Columbia University. Columbia University is also an equity holder in DarwinHealth Inc. and assignee of patent US10,790,040 (“Virtual inference of protein activity by regulon enrichment analysis”), which covers some components of the algorithms used in this manuscript. The other authors declare no competing interests.

Figures

Fig. 1
Fig. 1. Methodological overview.
a VESPA assesses the activity of protein kinases and phosphatases based on the phosphostate of their substrates. As an input, dVESPA requires a matrix representing the phosphopeptide or phosphosite abundance of a collection of samples representing different conditions of a specific cellular context, including missing values (black). The signaling network reconstruction module (blue box) analyzes this matrix to first identify candidate signal transduction interactions by assessing the significance of the mutual information between enzymatic regulator and candidate target phosphopeptides, second, remove indirect interactions by applying a signal transduction-specific form of the Data Processing Inequality (stDPI), and third, generate signalons for each KP-enzyme representing the probability of each interaction with a substrate and the mode of regulation (kinase activation: red, phosphatase deactivation: blue). b mVESPA first uses these signalons to assess KP-enzyme activity at the “phosphostate-level” (green box). The resulting KP-enzyme activity matrix then becomes the input to an additional protein activity assessment step of dVESPA (pink box) which uses the (standard, non-signal transduction) Data Processing Inequality (DPI) to generate more abstract signalons (i.e. representing activation/deactivation instead of phosphorylation/dephosphorylation), which are then in turn used by mVESPA for activity-level inference (purple box). Methodological differences between phosphostate- and activity-level signaling networks. At the phosphostate-level, ST-Ks (e.g., GSK3A, green) are primarily associated with direct phosphorylation targets, whereas TKs (e.g., ERBB2, orange) can frequently not be directly associated with (unenriched) tyrosine-phosphorylated sites. On activity-level, more abstract “activation/deactivation” events can better associate targets for both ST-Ks and TKs.
Fig. 2
Fig. 2. Benchmark and validation of dVESPA and mVESPA.
a Comparison of different mutual information (MI) estimation strategies based on imputation (iMI), depletion (dMI) and hybrid partitioning (hpMI) and the MI – Spearman correlation relationship using the CPTAC-S45 dataset. MI was measured and Spearman correlation was computed for each KP-enzyme/target pair using data from the CPTAC-S45 dataset (Methods). b Precision-recall curves were used to evaluate the difference between regular Data Processing Inequality (DPI) and its adaptation to signal transduction networks (stDPI), as well as compared to not using the DPI step at all (noDPI). c Baseline profiles of six diverse CRC cell lines were acquired and used with the GDSC reference database to identify sensitive and resistant or insensitive cell lines for each drug. The most differentially active KP-enzymes, as induced by treatment with each drug, were then assessed using mVESPA. d The predictive performance of the analysis results of (c), comparing dVESPA and other reference networks (Pathway Commons (PC), Hijazi et al. and Johnson et al.), was evaluated using receiver-operating-characteristics (ROC). For each differential comparison, ROC metrics were computed, where the sensitivity represents the mVESPA scores, weighted by GDSC drug sensitivity, and the selectivity represents a normalized rank of the top VESPA hits (see Methods). The individual ROC curves were then averaged. Statistical comparison of the differential comparison AUC metrics was conducted using an unpaired, right-tailed Wilcox’ tests. VESPA (red), comprising the mVESPA and dVESPA steps, significantly outperformed mVESPA when run using non-context-specific SigNets, including Johnson et al. (blue), Hijazi et al. (green) and Pathway Commons (purple). e Benchmark against established algorithms and applicability to datasets with N = 1 samples. The KSTAR benchmark was extended according to the original publication. Algorithm performance on S/TKs or TKs is depicted, computed as the fraction of conditions (specific cell line perturbed by a specific drug) for which a perturbed kinase was assessed as differentially active (Phit,), i.e., either ranked in the top 10 most differentially active (translucent bars) or based on statistical significance (FDR < 0.05) (opaque bars). Source data are provided as a Source Data file.
Fig. 3
Fig. 3. Representation of CRC subtypes by cell line models.
a Uniform Manifold Approximation and Projection (UMAP) embedding of KP-enzyme activity color-coded according to different classification systems (phosphoproteome-based VESPA; VC and the CRC Consensus Molecular Signature; CMS). b The most informative proteins and their VESPA inferred normalized enrichment scores (NES) were selected for visualization (full datasets: Supplementary Fig. 7–9). CPTAC clinical profiles and cell lines were grouped according to the Consensus Molecular Classifier (CMS), VESPA clusters (VC), and microsatellite instability (MSI). Samples are grouped according to VC. c Gene Set Enrichment Analysis (GSEA) using a signal transduction-specific subset of the Reactome database. Only terms significant in at least one sample (FGSEA ES-test two-tailed BH-adj. p < 0.05) are shown. The colors represent GSEA NES and are linked to the legend in b). Source data are provided as a Source Data file.
Fig. 4
Fig. 4. Targeted drug perturbations in CRC cell lines.
a Visualization of VESPA’s inferred normalized enrichment scores (NES) across the full drug perturbation dataset (336 samples), comprising six CRC cell lines (CL), 7 drug perturbations (Drug compound; DC) and vehicle control (DMSO), across 7 time points (TP). b Gene Set Enrichment Analysis (GSEA) using a signal transduction-specific subset of proteins in the Reactome database. Only terms significant in at least one sample (FGSEA ES-test two-tailed BH-adj. p < 0.05) are shown. The colors represent GSEA NES and are linked to the legend in a). Source data are provided as a Source Data file.
Fig. 5
Fig. 5. Time-dependent response of known primary targets to drug perturbation.
VESPA normalized enrichment scores (NES) of the top 5 most downregulated proteins among known primary targets are visualized, and grouped according to drug perturbations and cell lines. Source data are provided as a Source Data file.
Fig. 6
Fig. 6. Context-specific nature of signal transduction networks.
a Analysis of VESPA-inferred protein activity by DeMAND identifies KP-enzyme whose interactions with other proteins are most dysregulated across cell lines (CL) and drug compound perturbations (DC). The heatmap color scale represents the statistical significance of the DeMAND-assessed dysregulation (-log10(BH-adj. p); one-tailed empirical p-values computed by DeMAND). Only known drug targets (bold) and proteins with significant score (black: BH-adj. p < 0.05; one-tailed empirical p-values computed by DeMAND) in the cell-line-unspecific DeMAND analysis are visualized. b Dysregulated proteins described above were selected and grouped according to drug perturbations. The heatmap depicts VESPA-inferred activities of the aggregated early time points. c Dysregulated proteins described above were selected and grouped according to drug perturbations. The heatmap depicts VESPA-inferred activities of the aggregated late time points. d Network dysregulation and drug mechanism of action (MoA) for the EGFR inhibitor osimertinib. Nodes indicate the most affected regulators with inner circle colors indicating cell line type and outer circle color and node size indicating VESPA activity. VESPA activity color legend is shared with subfigures b and c. Edges identify dysregulated, undirected interactions between KP-enzymes (Methods). Line thickness represents the statistical significance of each dysregulated interaction. Proteins highlighted in green indicate known primary/secondary targets. Source data are provided as a Source Data file.
Fig. 7
Fig. 7. Context-specific adaptive stress resistance mechanisms.
a Effect of drug perturbation vs. vehicle control at late time points is shown for each cell line (CL) and drug compound (DC). The effect is assessed based on the differential VESPA paired two-tailed t-test t-statistic between drug perturbations at three late-time-point (24 h, 48 h, 96 h) vs. vehicle control (DMSO) treated samples. Only statistically significant KP-enzymes are shown based on multiple-testing corrected q < 0.05 and avg(t-statistic) > 0, after integrating all p-values across all comparisons by Stouffer’s method (Methods). b Enrichment of predicted KP-enzymes in proteins validated by CRISPRko assays is shown using receiver operating characteristics (ROC), area under the curve (AUC), and statistical significance (one-tailed Mann-Whitney-U test). Results are shown for HCT-15 cells treated with linsitinib (LI; C2: 4.0 μM) and trametinib (TR; C2: 0.7 μM) vs. DMSO, as well as NCI-H508 cells treated with trametinib (TR; C2: 0.01 μM) vs. DMSO perturbation. Source data are provided as a Source Data file.
See this image and copyright information in PMC

Update of

Similar articles

See all similar articles

References

    1. Kholodenko BN. Cell-signalling dynamics in time and space. Nat. Rev. Mol. Cell Biol. 2006;7:165–176. doi: 10.1038/nrm1838. - DOI - PMC - PubMed
    1. Efeyan A, Comb WC, Sabatini DM. Nutrient-sensing mechanisms and pathways. Nature. 2015;517:302–310. doi: 10.1038/nature14190. - DOI - PMC - PubMed
    1. Broyde J, et al. Oncoprotein-specific molecular interaction maps (SigMaps) for cancer network analyses. Nat. Biotechnol. 2021;39:215–224. doi: 10.1038/s41587-020-0652-7. - DOI - PMC - PubMed
    1. Hanahan D, Weinberg RA. Hallmarks of cancer: The next generation. Cell. 2011;144:646–674. doi: 10.1016/j.cell.2011.02.013. - DOI - PubMed
    1. Kuipers EJ, et al. Colorectal cancer. Nat. Rev. Dis. Prim. 2015;1:15065. doi: 10.1038/nrdp.2015.65. - DOI - PMC - PubMed

Publication types

MeSH terms