Gene activity fully predicts transcriptional bursting dynamics

Abstract

Transcription commonly occurs in bursts, with alternating productive (ON) and quiescent (OFF) periods, governing mRNA production rates. Yet, how transcription is regulated through bursting dynamics remains unresolved. Here, we conduct real-time measurements of endogenous transcriptional bursting with single-mRNA sensitivity. Leveraging the diverse transcriptional activities in early fly embryos, we uncover stringent relationships between bursting parameters. Specifically, we find that the durations of ON and OFF periods are linked. Regardless of the developmental stage or body-axis position, gene activity levels predict individual alleles' average ON and OFF periods. Lowly transcribing alleles predominantly modulate OFF periods (burst frequency), while highly transcribing alleles primarily tune ON periods (burst size). These relationships persist even under perturbations of cis-regulatory elements or trans-factors and account for bursting dynamics measured in other species. Our results suggest a novel mechanistic constraint governing bursting dynamics rather than a modular control of distinct parameters by distinct regulatory processes.

FIG. 1.. Live single-cell transcription rate measurements of endogenous gap genes.

(A) Live fluorescence imaging of nascent transcripts using MS2 stem-loops measures single allele transcriptional activity (green hotspots) along the anterior-posterior (AP) axis of the fly embryo (also see Fig. S1A and Methods). (B) Transcription time series for 10 alleles (blue) of the gap gene hunchback (hb) at position $x/L=0.435\pm 0.010$ sampled every 10 s. Low embryo-to-embryo variability (Fig. S1F) enables pooling alleles from multiple spatially and temporally aligned embryos ($n=10-20)$ to average over 200–350 alleles at a given position (black). (C) Calibration of transcriptional activity in absolute units performed by matching mean spatial activity profiles from previously calibrated fixed smFISH measurements [23] (black) with 5-min-interval averages (gray shade in B) of live time series at each AP position (color) for all examined gap genes. A single global conversion factor matches live and fixed profiles to within 5% error (Fig. S1B), defining a unit for transcriptional activity (i.e., the cytoplasmic unit, C.U. [24]) equivalent to a fully tagged transcript. (D) Reconstruction of transcription initiation events (gray bars) and underlying bursts from single allele transcription time series (black). (Top) Bayesian deconvolution enables sampling possible configuration of initiation events (see Methods and Fig. S2A–B). (Bottom) Clustering of sampled initiation events (using a moving average of width ~1 min (gray curve) and a threshold at two mRNA/min (dashed line)) identifies individual bursts (blue). (E) Autocorrelation (AC) functions of single allele hb transcription rates averaged over time for different positions along the AP axis (color). AC functions are normalized by the variance; uncorrelated (${\mathrm{\Sigma }}_{\mathrm{A}\mathrm{C}}$) and time-correlated $\left({\tau }_{\mathrm{A}\mathrm{C}}\right)$ components of the rate fluctuations are highlighted (see Fig. S5). (F) Correlation time ${\tau }_{\mathrm{A}\mathrm{C}}$ (from fitted exponentials, see Methods) of the single allele transcription rate as a function of mean transcription rate $R$ (color code as in C). Dashed line corresponds to overall mean correlation time ${\bar{\tau }}_{\mathrm{A}\mathrm{C}}=1.37\pm 0.31\mathrm{m}\mathrm{i}\mathrm{n}$. Error bars are 68% confidence intervals.

FIG. 2.. Direct estimation of instantaneous mean transcription parameters.

(A) Hb mean transcription rate $R$ as a function of time in NC14 (color encodes AP position). (B) Control parameters $P_{\mathrm{O}\mathrm{N}}$ and $K$ as a function of mean transcription rate $R$ in $\mathrm{l}\mathrm{o}\mathrm{g}$ space. Since $\mathrm{l}\mathrm{o}\mathrm{g}(R)=\mathrm{l}\mathrm{o}\mathrm{g}(K)+\mathrm{l}\mathrm{o}\mathrm{g}\left(P_{\mathrm{O}\mathrm{N}}\right)$ by construction, changes in $P_{\mathrm{O}\mathrm{N}}$ determine changes in $R$ below the dashed line $\left(R~8.5\mathrm{m}\mathrm{R}\mathrm{N}\mathrm{A}/\mathrm{m}\mathrm{i}\mathrm{n}\right.$, corresponding to $\left.P_{\mathrm{O}\mathrm{N}}=0.75\right)$. (C) hb mean OFF-time $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ and mean ON-time $T_{\mathrm{O}\mathrm{N}}$ as a function of time in NC14 for all AP positions (color code as in (A)). (D) Mean OFF-time $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ and mean ON-time $T_{\mathrm{O}\mathrm{N}}$ as a function of $P_{\mathrm{O}\mathrm{N}}$, for all positions and time points beyond the 7.5 min mark in (C). Dotted line corresponds to mean $T_{\mathrm{C}}=1.1\pm 0.2$ (see Fig. S8E), which sets a lower bound on possible $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ and $T_{\mathrm{O}\mathrm{N}}$ values.

FIG. 3.. Transcription parameters collapse for all gap genes.

(A) Kymographs of ON-probability $P_{\mathrm{O}\mathrm{N}}$ for all gap genes as a function of position and time for NC13 and NC14. The spatiotemporal transcriptional pattern of the gap genes arises from a complex regulation of $P_{\mathrm{O}\mathrm{N}}$ (color map). (B-E) Transcription parameters collapse for all gap genes across time and position. Transcription rate R (B), Mean OFF-time $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ (C), ON-time $T_{\mathrm{O}\mathrm{N}}$ (D), and switching correlation time TC (E) as a function of $P_{\mathrm{O}\mathrm{N}}$. Colored data points represent individual gap genes (same color code as in (A), see Fig. S11A–B for gt male data). Each panel shows all the remaining genes in gray. (F) Density (color) of all data points across space and time of the transcription parameters for all gap genes, normalized by the maximum density. Potentially accessible space (gray shade) for plausible ranges of K (0.1–30 mRNA/min) and TC (0.1–10 min). $P_{\mathrm{O}\mathrm{N}}$ almost fully determines R and sets the combinations of $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ and $T_{\mathrm{O}\mathrm{N}}$. For $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ and $T_{\mathrm{O}\mathrm{N}}$, the dashed lines are the 2-state model predictions based on TC, and the solid lines take the finite recording length into account (see Fig. S11D).

FIG. 4.. Effect of cis- and trans-perturbations on ON and OFF times.

(A) Distal hb enhancer removal from the fly line carrying MS2 stem-loops in the endogenous hb locus. (B) Quantification of hb wild-type and mutant phenotypes. Both transcription rate R (left) and $P_{\mathrm{O}\mathrm{N}}$ (right) levels display significantly different expression patterns for the enhancer deletion mutant. Black arrows indicate the time points (15 and 40 min into NC14) in the kymograph at which rate profiles are depicted. “o” and “⋆” mark two bins with predominant $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ modulation and predominant $T_{\mathrm{O}\mathrm{N}}$ modulation, respectively. (C) Transcription parameters for hb enhancer deletion (cyan) collapse on corresponding wild-type parameters (gray). (D) Transcription parameters for kni (green) in a hb null background (i.e. a trans-mutation, Fig. S14D–E) collapse on corresponding wild-type parameters (gray). Solid black lines in (C) and (D) correspond to the endogenous bursting relationships from Fig. 3F. (E-F) Verification of predicted changes in $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ and $T_{\mathrm{O}\mathrm{N}}$ (E) (or changes in burst size B and frequency F (F)) for all wild-type and mutant $P_{\mathrm{O}\mathrm{N}}$ pairs (two example pairs shown in (A) for hb). For most pairs (> 85%) the prediction is correct (see also Fig. S15A–C. (G) Transcription parameters computed from two other Drosophila studies (yellow circles [16] and pink circles [21], respectively) are consistent with the gap genes relationships (black lines, data density color coded as in Fig. 3F).

FIG. 5.. Generality of bursting relationships.

(A) Scatter plots of the transcription parameters for a single gene copy (see Method) versus $P_{\mathrm{O}\mathrm{N}}$ (color code as in Fig. 3F). As in Fig. 4G, transcription parameters computed from two other Drosophila studies (yellow circles [16] and pink circles [21]. Transcription parameters resulting from multiple perturbations performed on the yeast GAL10 gene (orange circles [19]) also closely follow our relationships (black lines), suggesting that these may apply beyond Drosophila. Transcription parameters estimated using our deconvolution approach on 11 human genes (red squares [9, 45]) further highlight the constant nature of $T_{\mathrm{C}}$ and $K$ (see Methods). (B) Transcription rate $R$, burst frequency $F$ and burst size $B$ for a single gene copy versus $P_{\text{ON}}$. Transcription parameters estimated from single-cell RNA-seq in mouse cells (gray circles [22]) are mostly consistent with our relationships (black lines), though parameters are likely underestimated due to low mRNA recovery rate (~10–30%) as demonstrated in original study. Black dotted lines correspond to expected relationships in scRNA-seq data assuming constant $K$ and $T_{\mathrm{C}}$.

FIG. 6.. Gene activity as a predictor of bursting dynamics

(A) Heatmaps of simulated single allele activity (top) and of corresponding ON-OFF gene state (bottom) for a low and high $P_{\mathrm{O}\mathrm{N}}$ regime, as predicted from our bursting relationships (see B). (B) The bursting relationships extracted from our data characterize single-allele bursting dynamics, as shown in A. (C) Fractional changes in activity explained by $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$ and $T_{\mathrm{O}\mathrm{N}}$ (left), and explained by burst frequency $F$ and size $B$ (right). For low $P_{\mathrm{O}\mathrm{N}}<0.5$, changes in activity are predominantly dictated by changes in $T_{\mathrm{O}\mathrm{F}\mathrm{F}}$, while for high $P_{\mathrm{O}\mathrm{N}}>0.5$ by changes in $T_{\mathrm{O}\mathrm{N}}$. Similarly, for low $P_{\mathrm{O}\mathrm{N}}<1/3$, changes in activity are mainly dictated by changes in $F$, while for high $P_{\mathrm{O}\mathrm{N}}>1/3$ by changes in $B$.