Dynamic epistasis analysis reveals how chromatin remodeling regulates transcriptional bursting

Yeast strains and plasmids

All strains were derived from BY4741 and BY4742 anchor-away background strains54. FRB-tags for anchor-away were introduced either by transformation with a polymerase chain reaction (PCR) product containing the FRB-yEGFP1-hphMX4 cassette (pTL100) or the FRB-mScarlet-hphMX4 cassette (pTL329). Alternatively, a CRISPR–Cas9-based approach was used55, and strains were transformed using a plasmid expressing Cas9, a guide RNA and double-stranded PCR repair template from the same plasmids, followed by removal of the Cas9 plasmid by 5-fluoroorotic acid selection. PP7 loops were introduced by transformation with a PCR product containing the PP7 loop cassette and loxP-kanMX-loxP (pTL031), and subsequent removal of the kanMX marker by CRE recombinase expression (pTL014 or pTL191). PP7-coat protein was integrated at URA3 by transformation of a PacI-digested single-integration plasmid (pTL174)56. TATA-mut-int238 was introduced in the GAL10 TATA by the CRISPR–Cas9 approach described above using a single-stranded oligo as the repair template. The GAL4/gal4Δ strain was created by mating a BY4741 wt GAL4 anchor-away strain with a BY4742 anchor-away gal4Δ haploid strain that was constructed by the CRISPR–Cas9 approach described above, using a single-stranded oligo as repair template. The GAL4-3V5/GAL4-3V5 strain was created by mating a BY4741 and a BY4742 anchor-away GAL4-3V5 strain that were constructed by the CRISPR–Cas9 approach described above using a repair template created by a PCR product from genomic DNA from a GAL4-3V5 strain (YTL1446) and verified using Sanger sequencing. The GAL4-3V5/gal4Δ strain was created by mating the same BY4741 GAL4-3V5 strain with a BY4742 gal4Δ haploid strain. The hht2Δ-hhf2Δ strains were constructed by the CRISPR–Cas9 approach described above using a single-stranded oligo as the repair template. For all strains, at least two replicates were constructed independently, which were verified by PCR, growth plates (Extended Data Fig. 2a), microscopy (Extended Data Fig. 2b) and, if applicable, sequencing and smFISH (Extended Data Fig. 2c). All strains, plasmids and oligos used in this study are listed in Supplementary Tables 1, 2 and 3, respectively. The yeast strains and plasmids are available on request.

Live-cell imaging of transcription dynamics

Live-cell imaging of transcription dynamics was performed as previously described in detail4,57 with minor modifications. In brief, cells were treated with 7.5 μM rapamycin or dimethyl sulfoxide (DMSO) for 60 min and subsequently imaged at mid-log (optical density (OD600nm) 0.2–0.4) on a coverslip with an agarose pad consisting of 2% agarose and synthetic complete medium containing 2% galactose and 7.5 mM DMSO or rapamycin. Imaging was performed on a setup consisting of an AxioObserver inverted microscope (Zeiss), an alpha Plan-Apochromat ×100 numerical aperture (NA) 1.46 oil objective, an sCMOS ORCA Flash 4v.3 (Hamamatsu) with a 475–570 nm dichroic (Chroma), 570 nm longpass beamsplitter (Chroma) and 515/30 nm emission filter (Semrock), a UNO Top stage incubator (OKOlab) at 30 °C and light emitting diode (LED) excitation at 470/24 nm (Spectra X, Lumencor) at 20% power and an neutral density (ND) 2.0 filter, resulting in a 62 mW cm−2 excitation intensity. Widefield images were recorded for 1 h at 15 s intervals, with z-stacks (nine slices, Δz 0.5 μm) and 150 ms exposure using Micro-Manager software58. For each condition, at least two and often three replicate datasets were acquired with at least 80 cells in total.

Microscopy of anchor-away nuclear depletion

Before measuring the transcription dynamics, proper nuclear depletion was ensured in each sample by imaging of FRB-mScarlet. Imaging was performed on the setup described above but with 475–570 nm dichroic (Chroma), 570 nm longpass beamsplitter (Chroma) and 600/52 nm emission filter (Semrock) and LED excitation at 550/15 nm (Spectra X, Lumencor) at 100% power and ND 2.0 filter, resulting in 413.0 mW cm−2 excitation intensity. A single widefield image was recorded as a z-stack (nine slices, Δz 0.5 μm) using either 150 ms exposure (for YTL1178, YTL1179, YTL1281, YTL1309) or 500 ms exposure (for YTL1505, YTL1506, YTL1508, YTL1510, YTL1448, YTL1450, YTL1470, YTL1591, YTL1626, YTL1750, YTL1752, YTL1751).

Analysis of transcription dynamics

For the analysis of the transcription dynamics imaging data, a similar approach was used to that previously described4. All analysis was implemented as custom-written Python software (https://github.com/Lenstralab/livecell). First, the images were corrected for xy-drift in the stage using an affine transformation on the maximum intensity projection. Next, the cells were segmented using Otsu thresholding and watershedding. The intensity of the TS was calculated by fitting a two-dimensional Gaussian mask after the local background subtractions as described previously59. Initially, a threshold of six times the standard deviation (s.d.) of the background was used. For frames where no TS was detected, a second fit was made in the vicinity of the high-intensity spots detected in that cell, using a threshold of four times the s.d. of the background. For frames where still no TS was detected, the intensity was measured at the location of the previous frame where a TS was detected. The tracking within each cell was inspected visually, and the endpoint of each trace was manually set at the last frame with a visible TS. Cells without a TS, cells that were segmented incorrectly, and cells in which the track contained tracking errors were excluded from analysis.

Binarization was performed using a threshold set at five times the s.d. of the background, determined for each cell by fitting a Lorentzian distribution to intensities measured at four points at a fixed distance from the TS in each frame. This threshold reliably distinguished on and off periods at the single-transcript level. Subsequently, the binarization was improved by removing the bursts that lasted a single frame and merging bursts that were separated by a single frame. From these binarized traces, the burst durations, time between bursts and induction time were directly calculated. The burst intensity was measured as the average intensity of all frames in which the cell was on. The fraction of active cells was determined by the manual scoring of whether the cells show a TS during the 1 h acquisition. Reported values for burst duration, time between bursts, induction time and burst intensity were determined by bootstrapping with 1,000 repetitions. Reported error bars are the s.d. from the same bootstrap. Error bars in the number of active and inactive cells are given by the square root of the number of cells. To determine whether the active fractions are significantly different between conditions, a two-sided Fisher’s exact test was used. For the other parameters, we have used bootstrap hypothesis testing using equation (4) from ref. 53 to determine the achieved significance level.

Dynamic epistasis analysis

For the dynamic epistasis analysis, the fractional change in each parameter of the transcriptional bursting was determined as the ratio between the bootstrap mean of this parameter in the perturbed population and the unperturbed population. For the effect of the hht2Δ-hhf2Δ perturbations (Fig. 5 and Extended Data Fig. 7), the fractional change in the presence and absence of RSC was calculated relative to +RSC and −RSC, respectively. The error bars were calculated from the same bootstraps and propagated under the assumption that the measurements are independent between conditions. To calculate the expected effect of a double perturbation on each parameter, fractional changes of the individual perturbations are multiplied, analogous to the way phenotypic growth effects caused by pairwise genetic interactions are assessed36. The error bars are calculated by error propagation of the errors of individual perturbations.

Fitting of induction time distributions

To determine whether gene induction depends on a single or multiple rate-limiting steps, a least-squares fit was performed on the histogram of the distribution of induction times, with a binsize of 1 min. The following parameterization of the probability density function of the Gamma distribution was used:

$$}\left( \right) = A\frac}\left( k \right)}}x^\mathrm^}$$

Here, Γ(k) is the Gamma function, defined as: \(}\left( k \right) = _0^\infty \mathrm^}t}\)

Here t is the variable of integration. The amplitude parameter A was added because there is a dead-time between addition of galactose and actual start of the image acquisition. Free parameters in the fit are A (with lower bound 1 and initial guess 10), k (with lower bound 0.0001 and initial guess 10 for non-memory induction or 1.0001 for re-induction conditions) and θ (with lower bound 0 and initial guess 1). The scale parameter k is a measure for the number of rate-limiting steps.

Testing for subpopulations

Dynamic epistasis analysis relies on the use of the bootstrap mean to describe each parameter in a given condition. This is valid if the cells behave as a single population, but masks the potential effects of the specific subpopulations. To determine whether there are subpopulations with different behavior, we tested whether the shapes of these distributions were well described by a theoretical distribution (Extended Data Fig. 6). For induction time a Gamma function was fit as above. For the time between bursts and burst duration a Gamma function was fit (with A = 1, k with lower bound 1 and initial guess 1.0001 and θ with lower bound 0 and initial guess 1). Theoretically, the burst duration should be described by the sum of a deterministic time (the elongation time for GAL10) and an exponential distribution describing the initiation kinetics, but this was approximated with a Gamma distribution for simplicity. For the burst intensity a log-normal distribution was fit:

$$}\left( \right) = \frac }}\mathrm^}}}$$

Free parameters in the fit are σ (with lower bound 0 and initial guess 0.6) and μ (unbounded with initial guess 5.5). The goodness of all fits was determined by calculating the R2 between the data and the fit according to the following formula, where y1…yn are the observed values, with an average value of ȳ, and f1…fn are the fitted values:

$$R^2 = 1 - \frac\nolimits_i (y_i - f_i)^2}}\nolimits_i (y_i - \bar y)^2}}$$

For the time between bursts, Gamma fits consistently showed k between 1 and 1.6 with R2 > 0.8 (Extended Data Fig. 6d), indicating these distributions are approximately described by single populations with an exponential distribution. The fits to the burst duration and burst intensity distributions showed high R2 values (>0.97 and >0.94, respectively, Extended Data Fig. 6c,e). In contrast, for the induction time distributions, we found four experiments with R2 < 0.8 (Extended Data Fig. 6b), which could be caused by the presence of multiple populations. However, careful analysis of these distributions did not reveal signs of subpopulations. We noted that three of the low R2 experiments describe cells where TBP was tagged for depletion. Together with the finding that TBP tagging resulted in unexpected faster induction (Fig. 4) and showed deviating shapes of the induction time distribution, this suggested that the anchor-away tag may partially interfere with TBP function. For the fourth experiment, GAL4/gal4Δ upon RSC depletion, the low R2 value appeared to be caused by data sparsity, even though we included 181 cells. This data sparsity also appeared to cause inconsistencies in the k values of the Gamma fits to the induction times (Extended Data Fig. 6b), preventing proper interpretation. In addition, the inconsistent k values could arise because the Gamma distribution assumes different rate-limiting steps with equal rates, which may not be valid in all conditions. Overall, this analysis shows no signs of subpopulations, justifying our dynamic epistasis analysis based on the bootstrap mean.

smFISH

The smFISH was performed as previously described4,60 with minor modifications. In brief, yeast cultures were grown to an early mid-log (OD600nm 0.5), treated with either 7.5 μM rapamycin or DMSO for 60 min for anchor-away before fixation with 5% paraformaldehyde (Electron Microscopy Sciences, 15714-S) for 20 min. Then cells are washed three times with buffer B (1.2 M sorbitol and 100 mM potassium phosphate buffer pH 7.5), permeabilized with 300 U of lyticase (Sigma-Aldrich, L2524-25KU) and washed with buffer B. Cells were immobilized on poly-ʟ-lysine-coated coverslips (Neuvitro) and permeabilized with 70% ethanol overnight or for up to 3 days. Coverslips were hybridized for 4 h at 37 °C with hybridization buffer containing 10% dextran sulfate, 10% formamide, 2 × SSC, and 5 pmole probe. Four PP7 probes labeled with Cy3 (for YTL1178, YTL1179, YTL1281 and YTL1309) or Cy5 (for YTL1448, YTL1450, YTL1505, YTL1506, YTL1508, YTL1510, YTL1470, YTL1591, YTL1626, YTL1750, YTL1752, YTL1751) were targeted to the loops, or 48 probes labeled with Quasar670 (for YTL524, YTL525, YTL526, YTL527, YTL528, YTL529) were targeted to coding region of GAL10 (Supplementary Table 4). Coverslips were washed 2× for 30 min with 10% formamide, 2× SSC at 37 °C, 1× with 2 × salium sodium citrate (SSC) and 1× for 5 min with 1× PBS at room temperature. Coverslips were mounted on microscope slides using ProLong Gold mounting media with DAPI (Thermo Fisher Scientific, P36934).

The imaging was performed on two similar microscope setups consisting of an AxioObserver inverted microscope (Zeiss), a Plan-Apochromat ×40 NA 1.4 oil differential interference contrast (DIC) ultraviolet (UV) objective, a ×1.25 optovar, an sCMOS ORCA Flash 4v.3 (Hamamatsu). For Cy3, we used a 562 nm longpass dichroic (Chroma), 595/50 nm emission filter (Chroma) and 550/15 nm LED excitation at full power (Spectra X, Lumencor), with an excitation intensity at the two microscopes of 6.8 W cm−2 or 8.8 W cm−2. For Cy5, we used a 660 nm longpass dichroic (Semrock or Chroma), 697/60 nm emission filter (Chroma) and 640/30 nm LED excitation at full power (Spectra X, Lumencor), with an excitation intensity at the two microscopes of 4.9 W cm−2 or 6.7 W cm−2. For DAPI, we used either a 410 nm/490 nm/570 nm/660 nm dichroic (Chroma), a 430/35 nm, 512/45 nm, 593/40 nm, 665 nm longpass emission filter (Chroma) or a 425 nm longpass dichroic (Chroma) and a 460/50 nm emission filter (Chroma) and LED excitation at 395/25 nm at 25% power (Spectra X, Lumencor), with an excitation intensity at the two microscopes of either 1.2 W cm−2 or 1.9 W cm−2. For each sample, for at least 50 fields-of-view, a z-stack (21 slices, Δz 0.3 μm) was recorded for DAPI and Cy3 or Cy5, using 25 ms exposure for DAPI and 250 ms exposure for Cy3 and Cy5 using Micro-Manager software.

Analysis of smFISH

Images were analyzed using custom-written Python software (https://github.com/Lenstralab/smFISH). Here, the cells and nuclei were segmented using Otsu thresholding and watershedding. The spots were localized by fitting a three-dimensional Gaussian mask after local background subtraction59. Cells in which no spots were detected were excluded from further analysis, because visual inspection indicated these cells were not properly segmented or not properly permeabilized, such that smFISH probes did not enter the cells. For each cell, the TS was defined as the brightest nuclear spot and the number of RNAs at each TS was determined by normalizing the intensity of each TS to the median fluorescent intensity of the cytoplasmic RNAs detected in all cells. Cells with fewer than five RNAs at the TS were classified as inactive, and cells with five or more RNAs at the TS were classified as active cells. Subsequently, the fraction of active cells and the mean number of RNAs at the TSs of active cells were determined. For each condition, at least three replicate experiments were performed with in total at least 5,000 cells, and the average value and standard error of the mean were determined for both the active fraction and the number of RNAs at the TSs of active cells. The fractional changes of these parameters upon nuclear depletion of indicated factors were determined from these mean values.

For the classification of cells into G1, S and G2 cell-cycle stages, the sum of the nuclear DAPI intensity in each cell is calculated from a maximum intensity projection. Subsequently, a histogram of all nuclear DAPI intensities (with 50 equally spaced bins) is fit with a Gaussian mixture model consisting of two peaks. Cells are classified as G1 stage if they are in a window of (s.d.1, 0.75 × s.d.1) around the center of the first peak, as G2 stage if they are in a window of (0.5 × s.d.2, 1.5 × s.d.2) around the center of the second peak and as S stage if they are in between the two peaks, where s.d.1 and s.d.2 are the s.d. of the first and second peak, respectively. Fractional changes in active fraction and number of RNAs at the TSs of active cells are determined as described above for each cell-cycle stage separately.

MNase-seq

The preparation and analysis of mono-nucleosomal DNA was performed as described previously4,54 with minor modifications. Briefly, cells were grown in SC + 2% raffinose or SC + 2% galactose from OD 0.3 to OD 0.75 and then treated with 7.5 μM rapamycin or DMSO for 60 min. Then, cells were fixed in 1% paraformaldehyde, washed with 1 M sorbitol, treated with spheroplasting buffer (1 M sorbitol, 1 mM β-mercaptoethanol, 10 mg ml−1 zymolyase 100T (US Biological, Z1004.250)) and washed twice with 1 M sorbitol. Spheroplasted cells were treated with 0.01171875 U (low MNase) or 0.1875 U (high MNase) micrococcal nuclease (Sigma-Aldrich, N5386-200UN) in digestion buffer (1 M sorbitol, 50 mM NaCl, 10 mM Tris pH 7.4, 5 mM MgCl2, 0.075% NP-40, 1 mM β-mercaptoethanol, 0.5 mM spermidine) at 37 °C. After 45 min, reactions were terminated on ice with 25 mM ethylenediaminetetraacetic acid (EDTA) and 0.5% sodium dodecyl sulfate (SDS). The samples were treated with proteinase K for 1 h at 37 °C and decrosslinked overnight at 65 °C. Digested DNA was extracted with phenol/chloroform (PCI 15:14:1), precipitated with NH4-Ac, and treated with 0.1 mg ml−1 RNaseA/T1. The extent of digestion was checked on a 3% agarose gel. For all conditions, two independent experiments were performed, with similar outcomes, except for the SWI/SNF depletion strain treated with high MNase concentration, where one replicate of the DMSO condition was underdigested. For this condition, only one replicate was used for analysis.

Sequencing libraries were prepared using the KAPA HTP Library Preparation Kit (07961901001, KAPA Biosystems) using 1 μg of input DNA, 5 μL of 10 μM adapter, double-sided size selection before and after amplification using 10 cycles. Adapters were created by ligation of Universal adapter to individual sequencing adapters (Supplementary Table 5). Libraries were checked on a Bioanalyzer High Sensitivity DNA kit (Agilent). Sequencing was performed on a NextSeq550. Paired-end 2 × 75 bp reads were aligned to the reference genome SacCer3 (January 2015) using bowtie2 (ref. 61) with the settings ‘–sensitive–end-to-end −3 15 −5 5 -X 1980–no-contain–no-discordant -p 40 -x’. The data have been deposited in National Center for Biotechnology Information’s Gene Expression Omnibus (GEO)62 and are accessible through GEO Series accession number GSE190737.

Analysis of MNase-seq

Analysis of MNase-seq data was carried out using custom-written Python software (https://github.com/Lenstralab/MNase_analysis). First, the aligned reads were filtered for length and only reads between 95 and 225 bp were retained for analysis. Subsequently, the read coverage was determined on a chromosome-by-chromosome basis by counting the number of reads covering each base, and normalized to the total coverage on the chromosome. Next, the coverage along each gene was determined using all verified open reading frames in the Saccharomyces Genome Database63. TATA and TATA-mismatch genes were identified as ‘TATA-containing’ and ‘TATA-less’ as previously described64. The coverage in TATA or TATA-mismatch regions was determined as the sum of the coverage in the 8 bp region spanning the TATA or TATA-mismatch sequence.

For metagene plots, genes were aligned at the +1 nucleosome in unperturbed conditions. To determine the position of the +1 nucleosome for each gene in these unperturbed conditions, the (prenormalization) coverage in a 4,000 bp window around the TSS of each gene was extracted from all experiments performed in DMSO using the high MNase concentration (combining the data for all yeast strains, that is YTL524, YTL525, YTL1306 and YTL1584). If the gene was on the Crick strand, the coverage was flipped to facilitate alignment of all genes. For each gene, these coverages were subsequently summed and smoothed using a Gaussian filter with a 40 bp window. The minimum of this smoothed coverage was determined, and a peak-calling function was used to detect nucleosome peaks. The −1 nucleosome was defined as the first peak before the coverage minimum, and the +1 nucleosome as the first peak after the coverage minimum. Genes for which fewer than two peaks were detected or for which the −1 or +1 nucleosome was detected more than 1,000 bp away from the TSS were excluded from the analysis. To generate metagene plots, the normalized coverage of all genes in a window of 2,000 bp centered at the location of the +1 nucleosome in unperturbed conditions of that gene was averaged. To generate heatmaps of the log2-fold-change of the coverage upon depletions, genes were sorted by the NDR width as determined by the distance between the −1 and +1 nucleosomes in unperturbed conditions. Subsequently, for each gene, the log2-fold-change between the coverage in each depletion (rapamycin) condition and the average coverage between two replicate experiments in the nondepleted (DMSO) condition was calculated. These data were represented as a heatmap. To determine the shift in the +1 position, the location of the +1 nucleosome was determined in each depletion dataset independently using the same steps as performed on the summed coverage to detect the position of the +1 nucleosome in unperturbed conditions. The shift in +1 nucleosome was then defined as the difference between the +1 nucleosome in depleted conditions and the +1 nucleosome as determined from all unperturbed high MNase datasets.

Protein detection by immunoblot and antibodies

Yeast cultures were grown to early mid-log (OD600nm 0.5), washed in MilliQ, pelleted and snap-frozen on dry ice. For protein extraction, cells were resuspended in 300 μl MilliQ, incubated with 300 μl 0.2 M NaOH for 7 min at room temperature, centrifuged and resuspended in 500 μl 2× SDS-PAGE sample buffer (4% SDS, 20% glycerol, 0.1 M dithiothreitol (DTT), 0.125 M Tris-HCl pH 7.5 and EDTA-free protease inhibitors). Samples were incubated at 95 °C for 5 min while shaking and centrifuged at 800g for 10 min at 4 °C. Then 20 μl of lysate with loading buffer was run on a NuPAGE 3–8% gradient TAC gel (V5) or 16% polyacrylamide gel (histone H3 and H3K79me3), and transferred to a 0.45 μm nitrocellulose membrane for 4 h (V5) and 2 h (histone H3 and H3K79me3). For blocking, the membrane was washed with tris-buffered saline-tween (TBS-T), incubated with PBS containing 5% milk for 1 h at room temperature and washed briefly with TBS-T. The membrane was incubated with PBS containing 2% milk and primary antibody (1:2,000 for αV5, 1:5,000 for αPgk1 and 1:1,000 for αH3 and αH3K79me3) overnight at 4 °C, washed 3× for 10 min with TBS-T, incubated with 2% mild and secondary antibody (1:5000) for 1 h at room temperature, washed 3× for 10 min with TBS-T and 1× for 10 min with PBS, and imaged using an LI-COR Odyssey infrared imager (Biosciences). Western blot analysis was performed using antibodies against V5 (Thermo Fisher Scientific R960-25, RRID: AB_2556564), Pgk1 (Invitrogen 22C5D8, RRID: AB_2532235), histone H3 (RRID:AB_2631108, a kind gift of the F.v.L. laboratory)65 and histone H3K79me3 (RRID: AB_2631107, a kind gift of the F.v.L. laboratory)65. Secondary antibodies used were IRDye 800CW Goat anti-Mouse IgG 925-32210 Li-COR (RRID AB_2687825), IRDye 800CW Goat anti-Rabbit IgG 926-32211 Li-COR (RRID:AB_621843) and IRDye 680RD Donkey anti-Mouse IgG 925-68072 Li-COR (RRID AB_2814912).

Growth assay

The growth assay used to assess growth rapamycin- or DMSO-containing plates was performed as described previously4 with minor modifications. Serial fivefold dilutions of YTL559, YTL658, YTL047, YTL525, YTL524, YTL1306, YTL1281, YTL1391, YTL1413, YTL014, YTL1397, YTL1506, YTL1584, YTL1510, YTL1613, YTL1615, YTL1394, YTL1588, YTL1744, YTL1749, YTL1508, YTL1617, YTL1745, YTL1751 and YTL1747 strains were spotted on Yeast Extract Peptone (YEP) + 2% glucose + 7.5 μM rapamycin, YEP + 2% glucose + DMSO, YEP + 2% galactose + 20 μg μl−1 ethidium bromide + 7.5 μM rapamycin, YEP + 2% galactose + 20 μg μl−1 ethidium bromide + DMSO, and YEP + 2% raffinose + 2% galactose + 40 mM lithium chloride + 0.003% methionine + 7.5 μM rapamycin and YEP + 2% raffinose + 2% galactose + 40 mM lithium chloride + 0.003% methionine + DMSO. The growth was assessed after 3 days at 30 °C.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

留言 (0)

沒有登入
gif