IRI Life Sciences & Institute for Theoretical Biology, Humboldt Universität Berlin, Berlin, GermanyInstitute of Pathology, Charité – Universitätsmedizin Berlin, Berlin, GermanyBerlin Institute of Health, Berlin, Germany
Figure 1.Expression kinetics from a synthetic model system for ERK signal duration
HEK293 with a stably transfected ∆RAF1:ER fusion protein were treated with ER antagonist 4‐hydroxytamoxifen (4OHT, ON scenario in B). To generate pulses, ERK signalling was turned off using the MEK inhibitor U0126 (ON/OFF scenario in B). To distinguish between primary and secondary response genes, translation was blocked with cycloheximide (CYHX) in parallel to 4OHT stimulation (ON/CYHX scenario in B). In addition, we used actinomycin D (ActD) to determine mRNA half‐lives via transcriptional shutdown and 4‐thiouridine (4SU) to determine mRNA half‐lives via metabolic labelling.
Log2 gene expression fold changes of significantly induced genes (FDR = 1%) across different treatment scenarios. Gene induction of immediate, delayed and late responding genes is sustained upon constant activation (ON scenario) and transient upon two‐hour pulse activation (ON/OFF scenario). Genes still significantly induced upon parallel CYHX treatment were considered primary response genes. Genes were ranked by their model‐derived response time.
Figure 2.Simulation of primary response gene dynamics upon different signalling durations
Different activation patterns of signalling molecules (input functions, left) can elicit multiple different response profiles (right) with different response times (r) depending on mRNA half‐lives (t1/2) and transcriptional delays (∆t). Rapid induction requires short half‐lives (red lines). Late induction can be caused by transcriptional delays (blue lines), long half‐lives (yellow lines) or combinations thereof. Decoding of signal duration depends on mRNA half‐life. Short‐lived mRNAs relay signal duration to response duration, whereas long‐lived mRNAs decode signal duration to response amplitude (yellow lines).
Response amplitude for all simulated combinations of mRNA half‐life and transcriptional delay. Response amplitude is shown over time (columns) and in respect to input function (rows). For sustained signalling, all primary response genes exceed their half maximum response amplitude. Pulse and transient signalling inputs are only sufficient for immediate–early genes and short‐lived delayed–early genes. Long‐lived mRNAs with half‐lives greater 120 min require sustained signalling inputs to exceed their half maximum response amplitude. Example parameter sets displayed in (A) are marked with asterisks in (B). Dashed lines indicate cluster borders. IEG: immediate–early genes, t1/2 ≤ 120 min and ∆t ≤ 30 min. ILG: immediate–late genes, t1/2 > 120 min and ∆t ≤ 30 min. DEG: delayed–early genes, ∆t > 30 min.
Figure 3.Model fitting and classification of primary response genes
HEK293∆RAF1:ER cells were treated with 4OHT for constitutive induction of ERK signalling. Phosphorylation levels of ERK2 were measured with bead‐based ELISA (Bio‐Plex). RNA time course expression data were measured using microarrays. These data were the basis to train gene‐wise model parameters.
pERK2 log2 fold change upon sustained activation (left) and deduced input function (right) used for model fitting. Average pERK2 log2 fold change upon 4OHT treatment equals 100% signalling amplitude.
Measured gene expression kinetics and the resulting maximum‐likelihood fit of the gene expression model of all significantly induced primary response genes (FDR = 1%). Gene expression is shown as percentage of response amplitude. Mean error is calculated as the mean of absolute residuals serving as a goodness of fit measure.
Classification of primary response genes. For each gene, response time r is calculated as the sum of deduced transcriptional delay ∆t and mRNA half‐life t1/2 and used for ranking. In general, immediate–early genes (IEGs) have short response times, whereas both immediate–late genes (ILGs) and delayed–early genes (DEGs) have longer response times. For ILGs, long response times are due to long mRNA half‐lives. For DEGs, long response times are mainly due to long transcriptional delays (cf. boxplots).
Response times for IEGs, DEGs and ILGs. Genes on the same trajectory have the same response time, but response times are composed differently. For IEGs and ILGs, response times are solely determined by mRNA half‐life, whereas DEGs have response times that are mixtures of half‐life and transcriptional delay.
Sum of weighted squared residuals (wRSS) for simple (immediate) and complete (delayed) model. The complete model was rejected for genes with χ2 < 3.84 and Δt < 30 min to only accept significantly better fitted genes for the complete model and to reflect time intervals in sampling.
Figure 4.Semi‐quantitative prediction of mRNA log2 fold changes upon different signalling scenarios
Gene expression upon different stimulations was predicted based on fitted model parameters and measured pERK2 levels. Predictions were verified with gene expression time course data.
Signalling input conditions (left side shows deduced input function, and right side shows pERK2 measurements): Sustained ERK signalling (4OHT), 2‐h pulse ERK signalling (4OHT + U0126), growth factor signalling (EGF: epidermal growth factor, FGF: fibroblast growth factor). Deduced input functions: 100% signalling amplitude corresponds to mean induction in training condition (4OHT). Growth factor‐induced input functions are linear interpolations of pERK2 log2 fold changes relative to mean induction in test condition.
Predictions are verified with actual gene expression data. Heat maps show log2 fold changes of induced mRNAs. P: model prediction. D: gene expression data. E: mean error = mean of absolute residuals.
Figure 5.Immediate–late genes (ILGs) have long mRNA half‐lives, are transcribed immediately and have GC‐rich promoters
Boxplot comparison of mRNA half‐life estimates based on modelling of gene induction (model‐derived), transcriptional shutdown (ActD‐derived) and metabolic labelling (4SU‐derived). Estimates from 4OHT‐pretreated HEK293∆RAF1:ER cells (ON panel) are more appropriate to characterise induced genes than estimates from unstimulated cells (OFF panel). Genes not assigned to any cluster are shown in grey.
Promoter GC content in IEGs, DEGs and ILGs. Calculated for TSS ± 1,000 bp in hg19. Wilcoxon rank sum was used to check for significant differences (n.s.: not significant, ***: P‐value < 0.001).
Log2 fold changes of transcription rate in 4OHT‐treated HEK293∆RAF1:ER cells derived from metabolic labelling (4SU) RNA‐sequencing data document immediate transcription of IEGs and ILGs but delayed transcription of DEGs. Dashed horizontal line indicates doubling of transcription rate.
Data information: Boxplots show median and inter‐quartile range. IQR is extended with whiskers to the largest and smallest value respectively, but no further than 1.5× IQR from hinges.
Figure EV5.Anti‐correlation of transcription rate and mRNA half‐life and comparison of nascent and total mRNA levels in 4OHT‐stimulated HEK293∆RAF1:ER
Comparison of absolute log2 transcription rate [TPM/h = transcripts per million per hour] after different periods of 4OHT treatment and model‐derived mRNA half‐life in HEK293∆RAF1:ER. Anti‐correlation indicates that short mRNA half‐lives in IEGs are compensated with high transcription rates.
Comparison of nascent and total mRNA levels after different periods of 4OHT treatment in HEK293∆RAF1:ER. IEGs have higher nascent mRNA levels after stimulation than ILGs and DEGs but end up at similar total mRNA levels after prolonged activation.
Data information: Boxplots in (B) show median and inter‐quartile range. IQR is extended with whiskers to the largest and smallest value respectively, but no further than 1.5× IQR from hinges.
Figure 6.Immediate–late genes (ILGs) translate signal duration into response amplitude
Upper panel: pERK2 log2 fold changes upon different input scenarios (sustained: 4OHT; 2‐h pulse: 4OHT + U0126). Lower panel: response amplitude across temporal clusters and signal durations. Bold lines show median cluster amplitude at each time point.
Capacity to decode signal duration. ILGs translate ERK signal duration into response amplitude. IEGs are only partially able to do so, as many of these genes are still strongly induced upon shortened signal durations.
qPCR validation to test different ERK signal durations. HEK293∆RAF1:ER cells were treated with 4OHT and U0126 for different periods of time to generate signal duration scenarios of 0.5–8 h (cf. Fig EV1C). mRNAs of IEGs EGR1 and FOS relay signal duration to response duration, whereas ILGs CLU and FOSL1 decode signal duration to response amplitude (qPCR data for all 17 validated mRNAs is shown in Fig EV6A).
Relation between signal duration and response amplitude for IEGs and ILGs derived from qPCR validation data. Median amplitude is based on six qPCR‐validated ILGs and six qPCR‐validated IEGs.
Quantification of Western blots to present protein log2 fold changes of sample genes upon sustained ERK signalling (4OHT‐induced) and transient ERK signalling (EGF‐induced) in HEK293∆RAF1:ER cells.
Conservation of signal duration decoding to response amplitude in two prominent model systems for ERK signal duration: NGF and HRG cause more sustained ERK signalling compared to EGF treatment in PC12 and MCF7 cells, respectively. Decoding of signal duration to response amplitude is clearly governed by ILGs.
Figure EV6.Signal duration effects on mRNA and protein level in 4OHT‐treated HEK293∆RAF1:ER cells, and conservation of mRNA response dynamics in rat PC12 and human MCF7 cells
mRNA log2 fold changes in qPCR time course data for five different ERK signal durations in HEK293∆RAF1:ER cells (cf. Fig EV1C for treatment scheme).
Representative Western blot for protein fold changes shown in Fig 6E. CLU and FOSL1 were measured on the same membrane; hence, lower GAPDH control corresponds to both blots.
Comparison of HEK293∆RAF1:ER and PC12 cells. Left panel: Spearman correlation of maximum mRNA log2 fold changes in 4OHT‐treated HEK293∆RAF1:ER cells and corresponding homologues in NGF‐treated PC12 cells. Middle panel: Spearman correlation of mRNA response times in 4OHT‐treated HEK293∆RAF1:ER cells and peak expression time points of corresponding homologues in NGF‐treated rat PC12 cells. Right panel: Spearman correlation of median mRNA half‐lives in HEK293∆RAF1:ER cells and peak expression time points of corresponding homologues in NGF‐treated rat PC12 cells.
Figure EV7.Gene Ontology enrichment and apoptosis in sustained versus transiently induced HEK293∆RAF1ER cells
Gene Ontology Biological Process term enrichment for IEGs, DEGs and ILGs. Score corresponds to significant enrichment in respective cluster, where Score = −log10 (P‐value). Significant enrichments are highlighted with coloured border. Background colour intensities correspond to denoted fractions and are normalised column‐wise.
FACS data to detect cleaved Casp3‐positive cells among untreated or treated HEK293∆RAF1:ER cells as a marker for apoptosis. EtOH: no ERK signalling. 4OHT: sustained ERK signalling. 4OHT+U0126: 2‐h pulse ERK signalling.