Regulation of transcriptional bursting by a naturally oscillating signal

Abstract

Transcription is highly stochastic, occurring in irregular bursts. For temporal and spatial precision of gene expression, cells must somehow deal with this noisy behavior. To address how this is achieved, we investigated how transcriptional bursting is entrained by a naturally oscillating signal, by direct measurement of transcription together with signal dynamics in living cells. We identify a Dictyostelium gene showing rapid transcriptional oscillations with the same period as extracellular cAMP signaling waves. Bursting approaches antiphase to cAMP waves, with accelerating transcription cycles during differentiation. Although coupling between signal and transcription oscillations was clear at the population level, single-cell transcriptional bursts retained considerable heterogeneity, indicating that transcription is not governed solely by signaling frequency. Previous studies implied that burst heterogeneity reflects distinct chromatin states. Here we show that heterogeneity is determined by multiple intrinsic and extrinsic cues and is maintained by a transcriptional persistence. Unusually for a persistent transcriptional behavior, the lifetime was only 20 min, with rapid randomization of transcriptional state by the response to oscillatory signaling. Linking transcription to rapid signaling oscillations allows reduction of gene expression heterogeneity by temporal averaging, providing a mechanism to generate precision in cell choices during development.

Figure 1

Oscillations in Transcription and Motility (A) Two-frame displacement for individual cells tracked over time for four fields of view at 5 hr development. Each row represents one cell. Color denotes two-frame motility (μm/min; black [low] to white [high]). Dark areas indicate where a cell has not been tracked. Data from four stage positions were captured simultaneously. Cell tracks are organized by track starting time during capture. (B) Transcription spot intensity in the same cells as in (A). (C) Displacement (upper) and spot intensity (lower) averaged over each field of view reveal clear oscillations. (D) Wavelet analysis: the averaged displacement for a single field of view (top panel, solid line). Phase is indicated by background color. Bottom panel: wavelet transform of the displacement data; ridge points are indicated by black circles. (E) Displacement (top panel) and transcription (bottom panel) data grouped by motility phase. The phase lag was estimated by fitting a sinusoidal function to the cross-correlation between motility and transcription. Error bars indicate SEM.

Figure 2

Developmental Changes in Transcriptional Strength and Phasing (A) Variation of transcription intensity with developmental time. Each data point represents a field of view. Colors denote distinct experiments. (B) Phase lag between motile and transcriptional responses. The circular mean and circular SEM are displayed as a function of developmental time. (C) A simple model of transcriptional phasing with examples of average transcription state for a range of cAMP frequencies. Increasing cAMP wave frequency reduces transcription amplitude, in line with real data. (D) Cell mixing experiments address relative effects of cAMP wave timing and developmental time. (E) Representative experiments comparing motility and transcriptional phasing of 4.5 hr csaAMS2 cells (green) mixed with 6.5 hr cells (red). Four of six experimental repeats showed clear transcriptional oscillations (one of the two transcriptional nonoscillators showed no motility oscillation). (F) Comparison of motility and transcriptional phasing of 6.5 hr csaAMS2 cells (green) mixed with 6.5 hr cells (red). Five of six repeats showed clear transcriptional oscillations, with the nonoscillating repeat showing no motility periodicity. (G) Transcription phase lag (left) and relative spot intensity (right) for csaAMS2 cells in 4.5 hr and 6.5 hr mixes. Error bars indicate SEM.

Figure 3

Quantifying Extrinsic and Intrinsic Regulation of the Transcriptional Response (A) Comparison of trough and subsequent peak transcription responses for a characteristic 5 hr field, normalized by field-of-view overall mean. The dotted line shows a linear fit to the data points with gradient and intercept indicated. The solid line indicates equal peak and trough intensities. Inset: the ratio of peak to trough response as a function of trough transcription intensity. (B) Relationship between response gradient and the mean spot intensity of fields of view. The measured correlation coefficient is 0.43 (p = 10⁻⁴). (C) Relationship between response gradient and intercept. (D) Cartoon illustrating the link between the response gradient and the balance between intrinsic and extrinsic variation. The thick black line shows a typical relationship between peak and trough intensities, intermediate between a line of gradient 1 (no response to cAMP) and a horizontal line (response determined entirely by cAMP). Alternative scenarios not supported by the data are described in Figure S3C. The scenario depicted in the cartoon is a consequence of the different forms of our wave model (Figures S3D and S3E). (E) Distribution of response gradient and intercept as a function of developmental time. Whiskers extend to the most extreme data point in the distribution not flagged as an outlier. Outliers fall more than 1.5-fold outside the interquartile range beyond the upper or lower quartile.

Figure 4

Sources of Transcriptional Heterogeneity (A) Fields of view more dense than average have weaker transcription than average. Each field is denoted by a marker, with color indicating developmental time and shape denoting imaging day. Points are normalized relative to other fields captured simultaneously. (B) Average spot intensity against average local density at 5 hr. Different markers represent four different fields of view captured simultaneously. A weak negative correlation is observed both within fields of view (main figure) and between fields (inset). Error bars indicate SEM. (C) Histogram of the weak and heterogeneous negative correlation within individual fields of view for all data at all time points. (D) Transcriptional response of individual cells to the third wave of cAMP (from start of image capture) compared with their response to waves 1, 2, 3, 4, and 5. (E) Ensemble of decay curves for transcriptional persistence (light gray lines) as a function of temporal separation. The average behavior (circles) is described by exponential decay to a nonzero plateau. Error bars indicate SEM. (F) Correlation between transcriptional persistence and the response gradient.