Stochastic tuning of gene expression enables cellular adaptation in the absence of pre-existing regulatory circuitry

We calculated the distribution of mRuby fluorescence values for each sample at each time point by smoothing the observed values using a kernel density estimator. We then measured the pairwise mRuby fluorescence distribution overlap of the top 20%, bottom 20% and the complete distribution at each time point for each growth condition. The distribution overlap was calculated by numerically integrating the area under the (normalized) kernel density distribution estimates of both populations being compared.

An increasing overlap relative to t = 0 signifies the amount that the two populations have moved towards each other, and therefore the higher the overlap, the more mixed the two populations have become. Therefore, we calculated ft=(max⁡x-xt)xt=0, where x is the overlap between the two distributions and max(x) is the maximum observed overlap. f(t) can be modeled as an exponential decay process according to:f(t)=ae-tτwhere τ provides a timescale for the mixing time (in particular, τ ln(2) is the half-life of the decay process). We used nonlinear curve fitting in Matlab to estimate the values of the parameters in the above equation for cells grown under each of the physiological conditions described above and report the estimated half-lives to give insight into the mixing times active in the populations studied here.

The images shown and analyzed in Figure 6, Figure 7, and panel A of Figure 7—figure supplement 2 were obtained on a Zeiss Axio Observer Z1, using a 40x objective lens. PHSP12-URA3-mRuby/PADH1-DHFR-GFP cells were grown overnight in SC +glu liquid media, and then back-diluted 100x into SC/loflo + glu media and grown four additional hours with shaking at 30°C. Cells were spun down, and then incubated in ura-/loflo/6AU5 liquid for 12–13 hr. The cells were then pipetted onto the prepared slides. In order to prepare slides, we added 200 µL of ura-/loflo/6AU5 media containing 1% agar to each well of a two-well slide. Using a 22 µm coverslip, the surface of the media in the wells containing the solid media was flattened. After adding the cells on to the wells, we allowed extra media to be absorbed and then added a cover slip on top. The cells were imaged on DIC, GFP, brightfield, and mRuby channels; snapshots were taken once every 30 min for approximately 24 hr.

The additional imaging time series analyzed in panels B-D of Figure 7—figure supplement 2 were obtained for PHSP12-URA3-mRuby/PADH1-DHFR-GFP cells immobilized to thin-bottomed growth chambers and grown in ura-/6AU5 media. To prepare the slides, cells were grown overnight in SC +glu liquid media, and then back-diluted 100x into SC/loflo + glu media and grown four additional hours with shaking at 30°C. During that incubation, a coverslip/incubation chamber (Nunc) was treated for five minutes with poly-D-lysine solution (MPI Biomedical), washed three times with sterile deionized water, and then allowed to dry.

After the pregrowth period, cells were diluted 10x into additional prewarmed SC/loflo + glu, and then pipetted onto the poly-D-lysine treated cover slip and allowed to settle for 30 min at room temperature. The media was removed, and non-adherent cells were washed away with two 1 mL rinses of sterile deionized water. The cells were then covered with 2 mL of ura-/loflo/6AU5 media, and then placed in a preheated microscopy incubation chamber (OKO) at 30°C and 90% relative humidity. Cells were imaged on DIC, GFP, and mRuby channels; snapshots taken once every 30 min for 24 hr on a Nikon Eclipse Ti microscope using a 20x objective.

For comparative visualization purposes (Figure 7A–B), the DIC or brightfield channel of each image was rescaled using the ImageMagick ‘normalize’ operator, and the fluorescence channels were normalized by subtracting the minimum pixel value within a given field of view, and then subjecting the remaining data to a median filter over a 5 × 5 pixel window. The fluorescence channels were then stacked on the DIC or brightfield to generate the images shown. Un-normalized data were used for all quantitative analysis.

For the quantitative analysis in Figure 7C–E and Figure 7—figure supplement 2, segmentation and lineage tracking were performed manually to identify cell division events and define cell interiors at the plotted timepoints. The fluorescence of each cell for each channel was then taken to be the average value of all pixels within the defined cell interior, with the mode value of all pixels in a defined window around the cell subtracted as background. For the purpose of classifying cells based on their division state, a cell was classified as ‘dividing’ if it gave rise to a daughter cell before the next analyzed snapshot. Timepoints prior to three hours were excluded from quantitative analysis of dividing vs. nondividing cells for the populations pregrown in SC/loflo + glu, as a large fraction of cells in all of our microscopy experiments did undergo a single division before arresting, likely using residual nutrients from their previous growth in complete media.