Figure 7. Comparison of prolactin transcription dynamics in different pituitary states.

(A) The stochastic switch model enables comparison of transcription dynamics in different pituitary states. Example plots of transcription rates estimated using the stochastic switch model. Analyses of populations of cells are shown from different states of pituitary tissue development. Each cell is associated with a single transcriptional profile obtained by random sampling of all the possible profiles for that cell (shown as in Figure 2A iii, with transcription rate and switch timings depicted). Data shown were from a single experiment representative of independent experiments (i: E18.5 n = 181 cells, ii: P1.5 n = 217 cells, iii: Adult n = 305 cells). (B–E) Characterisation of transcription dynamics estimated using the stochastic switch model. (B) The cumulative distribution function of the frequency of different transcription rates was estimated using a weighted kernel. No clear difference between pituitary developmental states was evident. (C) Weighted histogram of the number of switches for each developmental state. Data, pooled from independent experiments, show the mean + SD, calculated by weighting the number of switches in each profile by the probability of occurrence. P1.5 has significantly fewer switches to Adult and E18.5 (p<0.01 Mann-Whitney U test of the pooled distributions). (D) The duration of transcription rates for each tissue state. Transcription rates were binned into quartiles, with the first three lower quartiles grouped together (<75) as one bin and the highest quartile as the other bin (>75). The duration of transcriptional states was calculated for each bin. The duration represents the minimum duration spent in each transcriptional state, as not all transcriptional states were completely observed within the time-frame of the imaging experiment. Both E18.5 datasets showed a significant difference in the duration spent in low transcriptional states (lower 3 quartiles) to high transcriptional states (upper quartile) as did one P1.5 dataset and two Adult datasets. Moreover, the duration spent in the upper quartile of the E18.5 (pooled data) was significantly different to the duration spent in the upper quartile of either the pooled data of P1.5 or Adult. Significance was assessed through the distribution of the p-values calculated from a Mann-Whitney bootstrap (Figure 7—figure supplement 2). Boxplots were calculated by a random sampling of the weighted transcription rate distributions and the associated durations. (E) The graph shows a kernel density estimate of the time between any two switches using data from panel D, for cells that displayed more than one switch so that the total duration spent in a single transcriptional state was estimated. The inter-switch time in the adult tissues is longer than the inter switch times of either the E18.5 or P1.5 tissues. E18.5 = orange, P1.5 = red, Adult = black. All boxplots represent the median and interquartile range (IQR), with whiskers drawn 1.5xIQR away from the lower and upper quartile. SD, standard deviation.

DOI: http://dx.doi.org/10.7554/eLife.08494.016

Figure 7—figure supplement 1. Characterisation of prolactin transcription dynamics.

(A) Characterisation of transcription rates, determined using the stochastic switch model, in E18.5, P1.5, and adult pituitaries. Rates were classified as ‘Low’ if they were preceded and followed by higher rates; all other rates were classified as ‘Active’ even if they were preceded by a higher level of activity, given that they were always followed by a further down-switch. Cells with no switches were removed from the analysis. Graphs are of the kernel density estimate of the cumulative distribution of transcription rates and show that low transcription rates were reduced compared to the active transcription rates. There is some evidence of reduced transcriptional rates in the ‘active’ states (lower rates relative to fold change of overall tissue) in the P1.5 datasets, but no clear difference in the distribution of rates employed by cells can be seen between Adult and E18.5. (B) Durations spent at 'low' (L) and 'active' (A) transcription rates. Boxplots of the duration spent in either an 'active' or 'low' transcriptional state. These are obtained by sampling from the weighted density of transcriptional states and calculating the associated duration. These show that 'active' transcription in E18.5 and P1.5 pituitaries occurs over shorter time periods than in adult pituitaries, as calculated through bootstrap Mann-Whitney U-tests. All boxplots represent the median and interquartile range (IQR), with whiskers drawn 1.5xIQR away from the lower and upper quartile.

DOI: http://dx.doi.org/10.7554/eLife.08494.017

Figure 7—figure supplement 2. Significance testing of transcriptional state durations.

The boxplots in Figure 7D and Figure 7—figure supplement 1B are a single sample obtained from the weighted distribution of transcriptional rates. In order to test for significance difference between groups, bootstrap samples were obtained and a Mann-Whitney U-test was performed on each sample. The histograms of the p-values of each of these tests are shown in Figure 7—figure supplement 2A (corresponding to the tests on Figure 7D) and in Figure 7—figure supplement 2B (corresponding to the tests on Figure 7—figure supplement 1B). The vertical line indicates a p-value of 0.5% (an overall significance level of 5% with Bonferroni correction to account for multiple testing). Significance is indicated when the empirical histogram of the sampled p-values lies to the left of the red line. (A) Significant differences were found when comparing the duration spent in the lower transcriptional rates (lower three quartiles) or higher transcriptional rates (upper quartile) for datasets; Adult-2, Adult-3, E18.5-1 and E18.5-2. Significant differences were also found between the duration spent in the high transcriptional states of pooled datasets for E18.5 vs Adult and E18.5 vs P1.5. (B) Significant differences were found when comparing the duration spent in low (L) to active (A) transcriptional rates for datasets; Adult-1, Adult-3, P1.5-1, P1.5-2, E18.5-1 and E18.5-2. In addition, there is a significant difference in the duration spent in the high transcriptional states of the pooled datasets for E18.5 vs Adult and P1.5 vs Adult.

DOI: http://dx.doi.org/10.7554/eLife.08494.018

Figure 7—figure supplement 3. Spatial organisation of transcription switch profiles in developing pituitaries.

The spatial organisation of transcriptional switch profiles determined by the stochastic switch model in developing pituitaries was performed as outlined in Figure 4. (A) Graph of boxplots of switch timing intervals in cells that switch in the same direction and cells that switch in different directions, binned by the distance between cells. All pairwise switches are considered. An increase in the time interval between switches in transcription, which occur in the same direction, is seen with increasing distance between the cells in adult pituitary tissues but not in E18.5 or P1.5 pituitary tissues. (B) The cumulative distribution of the time interval between switch events. In adult pituitary tissue, cells within 30 µm that switch activity in the same direction do so within a smaller time frame than cells located greater than 30 µm apart, the unsorted population and cells that switch activity in opposite directions (confirmed by significant p-value <0.01 of Kolmogorov-Smirnov tests). In E18.5 and P1.5 pituitary tissues, this trend is not seen, as cells within 30 µm that switch activity in the same direction do so within a similar time frame to the whole cell population. In pooled data from P1.5 tissues, a significant difference (p-value <0.01 of Kolmogorov-Smirnov tests) is seen in the distribution of time intervals between cells that switch transcription activity in the same direction that are less than 30 µm apart and those that are greater than 30 µm apart, but this was not consistent across all data sets. All boxplots represent the median and interquartile range (IQR), with whiskers drawn 1.5xIQR away from the lower and upper quartile.

DOI: http://dx.doi.org/10.7554/eLife.08494.019