Roles of G1 cyclins in the temporal organization of yeast cell cycle - a transcriptome-wide analysis

Oscillating gene expression is crucial for correct timing and progression through cell cycle. In Saccharomyces cerevisiae, G1 cyclins Cln1-3 are essential drivers of the cell cycle and have an important role for temporal fine-tuning. We measured time-resolved transcriptome-wide gene expression for wild type and cyclin single and double knockouts over cell cycle with and without osmotic stress. Clustering of expression profiles, peak-time detection of oscillating genes, integration with transcription factor network dynamics, and assignment to cell cycle phases allowed us to quantify the effect of genetic or stress perturbations on the duration of cell cycle phases. Cln1 and Cln2 showed functional differences, especially affecting later phases. Deletion of Cln3 led to a delay of START followed by normal progression through later phases. Our data and network analysis suggest mutual effects of cyclins with the transcriptional regulators SBF and MBF.


Introduction
major question is thereby whether Cln1 and Cln2 are really fully redundant or whether we can 91 identify specific influences on the cell cycle timing for each of them. To do so, we characterized 92 the roles of the G1 cylins in organizing global oscillatory gene expression. We analyzed the 93 transcriptome of wild type as well as single and double knockouts of Cln1, Cln2 and Cln3 to 94 identify timing effects, such as overall delays or temporal shifts, in the expression patterns. For 95 enhancing functional differences between Cln1 and Cln2, we additionally perturbed the cells 96 with osmotic stress. Based on the results, we propose mechanistic links that could cause the 97 observed timing effects depended on altered transcription factor activities.

98
We found that the deletion of the earliest cyclin Cln3 leads to a systematic shift of expression 99 times, with an elongated G1 phase followed by a wild type like timing of the subsequent phases As was shown before (Soifer & Barkai, 2014;Talia et al, 2007), Cln genes act as regulators of cell 119 size and morphology (Figure 2A). The loss of two out of three G1 cyclins causes the cells to 120 increase in size compared to the wild type before division ( Figure 2B). The loss of the initial 121 cyclin Cln3 as single deletion already increases cell size comparable to double deletions.

122
Changes in cell size can hint towards a deregulation of cell cycle timing, with altered cell cycle 123 phase durations resulting in longer or shorter growth periods. We, therefore, set out to   The estimated peak times of the individual oscillating genes were then used to systematically 171 analyze quantitative differences in the expression timing. For each mutant, peak times were 172 compared to the wild type estimates in a two-dimensional scatterplot (as an example general behavior of the entire group of oscillating genes. As genes occurring close to the bisecting line have a conserved peak time in mutant and wild type, we can use the slope of the than one). Shifted curves with a conserved slope around one then indicate a overall delay with a 178 conserved peak timing, for example caused by a delayed onset of the cell cycle.

179
With all peak times gathering around the bisecting line, no effect of the loss of Cln1 alone on 180 the overall cell cycle timing was evident ( Figure 3C, middle, blue line). In contrast, cells lacking 181 Cln2 (as single deletion (yellow) or in combination with Cln1 deletion (green)) showed a 182 conserved timing of the early peaks, but a decreased slope for later times -arguing for a slower 183 progression through later cell cycle phases such as S and G2 phase. In the cln1Δcln3Δ double 184 mutant (purple) and, slightly less prominent, in the cln3Δ single deletion (red), we observed that 185 the overall sequence of the peak times was retained (slope close to one) with a delay of around 186 5 and 20 minutes, respectively. However, a subset of genes shifts their peak time backwards to 187 the first third of the cell cycle, visible as a negative slope of the lowess curve. A thus elongated 188 G1 phase, in which many genes show their peak expression, is a behavior often observed in cells 189 that react to a stress stimulus.

190
Following this thought, we repeated the peak time analysis with cells that had been exposed to 191 osmotic stress following their release from synchronization. Osmotic stress is among others later phases is neglected here, since cells were exposed to osmostress in early G1. Since also 195 our target cyclins act primarily in the G1 phase, we hypothesized that an additional 196 perturbation in this phase could reveal further aspects of their action. In the stress experiments,

197
we observed a delay of oscillating gene expression for wild type and all mutants ( Figure 3C, 198 right). Also in the cln1Δ strain, which did not show changes in expression timing in the 199 unstressed experiment, we could now observe a stronger delay than in the wild type (11.5 mins 200 delay caused by osmotic stress in the wild type, 16.9 mins in cln1Δ), highlighting the role of Cln1 201 in reentering the cell cycle after osmostress.

202
In summary, the peak time analysis revealed specific characteristics of the mutant gene    Table 2) consists of four groups, whose peak expression 234 occurs at the transitions between G1/S, S/G2, G2/M and M/G1 phase (wild type example in 235 Figure 4B, mutants in Supplementary Figure 4). From the classification set, we then estimated 236 the cell cycle phase durations in all stressed and unstressed mutants ( Figure 4C). It is important 237 to notice that we detected the peak expression times here with a simple peak detection based 238 on the smoothed target trajectories as opposed to using the MoPS peak time estimates. The 239 reason for this is that especially for the stress experiments, but also due to the influence of the

287
MBF and SBF are regulated by slightly different mechanisms that share part of their actors. The 288 explicit roles of the cyclins in this regulation, especially for MBF, are, however, still unknown.

289
Based on the high temporal resolution of our data, we could analyze the expression pattern of

294
In α-factor synchronized cells, MBF targets are expressed some minutes before SBF targets 295 (Eser et al, 2011). We observe the same behavior in the wild type, in which the SBF peak occurs 296 around 6 minutes after the MBF peak. Also our mutant strains showed the same temporal 297 order, but the delays between the two target cluster expressions were affected by the 298 mutations ( Figure 5C). SBF expression was delayed by 8 min and 12 min, relative to the 299 expression of MBF target genes, in strains lacking Cln2 as single and double deletion together 300 with Cln1, respectively. Contrarily, in cells lacking Cln3, MBF expression was delayed strongly 301 compared to the wild type, such that it almost coincided with the expression of SBF target 302 genes and the difference between the two clusters was shortened to around 3 minutes.

303
Applying osmotic stress the observed behavior becomes even more pronounced. Because the 304 definition of MBF and SBF target genes is not necessarily unique and also based on

324
We, therefore, hypothesized that possibly already the release from α-factor synchronization 325 could be affected in the cln2Δcln3Δ strain, resulting in an asynchronous re-entry to the cell 326 cycle. In the RNAseq data we would then fail to detect oscillations because we are actually 327 measuring an unsynchronized cell population.

328
To test this hypothesis, we followed α-factor release of the cells in a microfluidic device,

332
To understand the effect also on the transcriptional level, we could harness the results of the 333 cluster analysis again. Besides the previously described oscillating clusters, we also found a set 334 of genes whose expression is highest at the beginning of the experiment and decreases steadily 335 afterwards. The cluster behavior was robust in all mutants, but showed the weakest decay in decaying response to α-factor arrest. Zooming in even further, we identified a group of 16 340 genes related to α-factor signaling and its cellular response (Supplementary Table 3), which 341 remained at levels 2.5 -7.7-fold higher in the cln2Δcln3Δ strain than in the wild type ( Figure 6C).

342
We, therefore, concluded that the double mutant lacking both Cln2 and Cln3 fails to silence α-343 factor signaling in a coordinated fashion once the pheromone is removed from the medium.

344
Individual cells, therefore, needed different times to exit the cell cycle arrest, which resulted in 345 a smeared mRNA expression in the population that did not reveal oscillations. Other 346 coordinated processes, which are largely independent of the cell cycle, can hence still be seen in 347 the transcriptome such as here the response to osmotic stress. Another crucial point is that the defective release occurred in the cln2Δcln3Δ but not the cln1Δcln3Δ mutant, additionally cycle. As before, loss of Cln2 had much more dramatic consequences for the cells than loss of          normalized to total expression reads, which was defined as 10 6 and genes with expression below 10 reads were excluded from the analysis. Time points were removed if total reads were 495 less than 7⋅10 4 (Supplement Figure S10). All experiments were done in replicates. Since the 496 correlation between replicates was very high (Supplement Figure S11)  the plates incubated at 30 ∘ C and exposed to 5 μg/ml α-factor in SD for 3 h with a pressure of 2 550 psi. Afterwards α-factor was removed by washing plates with SD for 10 min with a pressure of 5 551 psi. Imaging was performed every 10 min for 7h using a Olympus IX83 inverted microscope with Detection of bud emergence was performed manually utilizing BudJ: an ImageJ plugin to   between the phases were assigned to the previous phase. To calculate differential expression,

844
WT expression was linearly interpolated at the now normalized mutant time points.

845
Genes were then deemed differentially expressed if at least n  = 3 time points were changed 846 more than 3.36-fold (log 2 (1.75)) or a low p-value in a Kolmogorov-Smirnow-test compared to 847 the WT expression (<10 -8 for n  = 0, <8x10 -5 for n  =1 and <0.002 for n  = 2).  Table 3: Group of -factor induced genes with higher expression over the experiment in cln2Δcln3Δ (also see Figure 6)