Network integration and modelling of dynamic drug responses at multi-omics levels

Figures

• 

  Fig. 1. 3D cardiac microtissues and genome-wide methylation. a Essential microtissue characteristics remain stable across a time period of 28 days. Contractions per 15 s, microtissue sizes and ATP content. Bars indicate measurements per week. Red line is the average over all 4 weeks. b Inverse correlation of gene body methylation and expression. Genes were classified into six classes of expression strengths based on the median fragments per kilobase of transcript per million reads (FPKM) over all time points and treatments. Methylation values (Y-axes) were averaged for each class of genes according to the following procedure: promoters and gene bodies were binned in 75 windows, 20 bins for the promoter (−5 kb), 1 bin for the transcription start site (TSS), 4 bins for the starting exons/introns (+1 kb) and 50 bins for the rest of the gene body (TTS: transcription termination site); for each gene % methylation in each bin was averaged over all seven time points for the specific treatment and then % methylation was averaged over all genes of the respective expression strength classes. In the plots of the AC-treated experiments the respective curves for the control experiments (DMSO) are…

• 

  Fig. 2. Dynamic proteome changes upon AC treatment. a VENN diagrams (generated with Venny 2.1.0) of dynamically altered proteins after AC treatment (DOX: blue, DAU: yellow, EPI: green, IDA: red) measured with LC–MS at therapeutic and toxic doses. b Examples of polynomial regression fit (degree ≤ 2) for significantly altered sarcomere genes (ACTN2, MYL4) and transcriptional regulators (TRIM28). The curves of the regression fit are plotted as dotted lines, solid lines connect the median values of the three replicates per time point to visualize the trend over time. Dots: experimental values, red curves: DMSO control experiments, green curves: AC treatment (DAU). Y-axes: normalized protein expression, X-axes: time. c VENN diagram showing overlap of proteins dynamically altered upon AC treatment according to therapeutic (THE) and toxic (TOX) doses. Overlap is counted with respect to proteins and enriched pathways. d Hierarchical clustering of protein expression over all experiments using the h.clust() method in R with Eucledian distance as distance measure and Ward’s method (option “ward.d2”) as agglomeration method. Upper panels code for the different treatments and the different time…

• 

  Fig. 3. Dynamic transcriptome changes upon AC treatment in sarcomere and mitochondrion. Expression changes of mitochondrial response and sarcomere genes upon AC treatment at therapeutic dose measured with RNA-seq (expression changes with respect to toxic doses in Supplementary Fig. 8). Protein complexes are represented by purple, ovals with the genes encoding for subunits displayed next it in yellow rectangles. On top of each gene, the boxes display the expression change over time, where each box corresponds to a specific time point. The fill level of these boxes display the log2 fold change (completely filled boxes: log2 fold change ≥ 5), upregulations are depicted in red and downregulations in blue. Fold-changes were computed from the experimental replicates (n = 3) for each AC against the time-matched control samples. Color-codes refer to the average fold-change of the four ACs at the specific time point. Significance of temporal changes is indicated by transparency, where the lightest genes were not differentially expressed genes (DEGs) in any AC, medium transparent are DEGs with respect to only 1 AC treatment and not transparent are DEGs with respect to at least 2 AC treatments…

• 

  Fig. 4. ACT response network. ACT response network computed from integrated proteome and transcriptome data (network derived from toxic doses is described in Supplementary Fig. 16). Nodes were initialized based on the significance of the dynamic changes of respective AC treatment compared to DMSO control longitudinal data and for each AC a drug response network was computed (Supplementary Fig. 12). Nodes and their interactions that appeared in at least two of the individual AC networks were integrated. Node colors reflect the occurrence of the node in the individual AC networks. The three major enriched cellular compartments are shown in transparent brown (extracellular matrix), transparent orange (mitochondrial part) and transparent blue (sarcomere). Red circles: Proteins that are contained in the mitochondrial model (Fig. 6a). Blue circles: Proteins genetically associated with dilated cardiomyopathy according to Talay et al.40.

• 

  Fig. 5. Clinical relevance of in vitro-derived proteins. a Patient characteristics. Histogram of LVEFs for the patients under study (n = 15). Chronic cardiotoxic patients (dark blue) and control cardiomyopathy patients (light blue) are sorted after LVEF. b LVEF (X-axis) of chronic cardiotoxic patients decreases with duration of chemotherapy in days (Y-axis); color of the nodes correspond to LVEF classification of the respective patient. c Examples of in vitro-derived ACT response proteins whose expression in biopsies of cardiotoxic patients correlate with LVEF. X-axes shows LVEF, Y-axis shows normalized protein expression. Color of the nodes correspond to LVEF status similar to b. d Discordant correlation values of biopsies protein expression with the LVEFs of the two patient groups. Blue bars: correlation with respect to cardiotoxic patients, red bars: control group. X-axis: genes, Y-axis: correlation values (range [−1,1]). e Pairwise comparison of protein expression fold-changes of measured time points (8, 24, 72, 168, 240, and 336 h) against the earliest time point (2 h) in 3D cardiac microtissues using Student’s t-test on n = 3 experimental replicates. Y-axis: bars indicate log2-fold…

• 

  Fig. 6. Computational modeling of in vitro-derived mitochondrial protein expression. a Mitochondrial model schematic, with subsystems indicated with the dashed boxes: electron transport chain (ETC), reactive oxygen species scavenging (ROS), and the tricarboxylic acid cycle (TCA). b, c Model predictions for the steady-state mitochondrial ATP concentration and membrane potential, as functions of exposure time, normalized to their initial value, for each drug under therapeutic (solid curves) and toxic doses (dashed curves). d, e Functional measurements of ATP concentration performed on spheroid microtissues at discrete time points, following either therapeutic (solid lines) or toxic (dashed lines) drug exposure as implemented using a PBPK model. MEAS measured, MODEL predicted by model.