|Home | About | Journals | Submit | Contact Us | Français|
In budding yeast, the Start checkpoint integrates multiple internal and external signals into an all-or-none decision to enter the cell cycle. Here, we show that Start behaves like a switch due to systems-level feedback in the regulatory network. In contrast to current models proposing a linear cascade of Start activation, transcriptional positive feedback of the G1 cyclins Cln1,2 induces the near-simultaneous expression of the ~200-gene G1/S regulon. Nuclear Cln2 drives coherent regulon expression, while cytoplasmic Cln2 drives efficient budding. cln1,2-deleted cells frequently arrest as unbudded cells, incurring a large fluctuation-induced fitness penalty due to both the lack of cytoplasmic Cln2 and insufficient G1/S regulon expression. Thus, positive-feedback-amplified expression of Cln1,2 simultaneously drives robust budding and rapid, coherent regulon expression. A similar G1/S regulatory network in mammalian cells, comprised of non-orthologous genes, suggests either the conservation of regulatory architecture or convergent evolution.
Positive feedback in genetic control networks can ensure that cells do not slip back and forth between either cell cycle phases or developmental fates. For example, commitment to sporulation in budding yeast is driven by transcriptional positive feedback of the meiotic inducer IME11–3. In Xenopus laevis, positive feedback underlies the all-or-none characteristics of oocyte maturation4, 5 and mitotic entry6, 7, suggesting the frequent use of positive feedback to regulate cellular transitions.
Absent from this list of examples is the well-studied Start checkpoint controlling cell cycle commitment in budding yeast. Nutrient limitation and pheromone exposure arrests cells prior to DNA replication, while size control extends G1 in small daughter cells8–11. Beyond Start, cells proceed through division almost independently of size and environment9. Previous experiments suggested that Start represents a feedback-free cascade of events12 (see schematic in Fig. 1a; omitting red arrows). The transition is initiated by the G1-cyclin Cln313–15, which in complex with Cdc28 activates the transcription of about 200 genes16 by phosphorylating promoter-bound protein complexes that include the transcription factors SBF and MBF17 and the transcriptional inhibitor Whi518, 19. Phosphorylation and inactivation of Whi5 is rate-limiting, and phosphorylated Whi5 rapidly exits the nucleus. The G1/S regulon, which includes two additional G1-cyclins, CLN1,2, contributes to the activation of B-type cyclins, DNA replication, spindle pole body duplication, and bud emergence. Mitotic B-type cyclins then inactivate SBF20 and, with NRM1, inactivate MBF21, thus turning off the G1/S regulon.
Any one of the three G1-cyclins suffices to activate the regulon, suggesting the potential for transcriptional positive feedback of CLN1,2 on their own expression22, 23. However, analysis of synchronized populations led to the conclusion that positive feedback, defined as Cln1,2 advancing transcription from the CLN2 promoter, did not occur in WT; rather, Cln3 was the sole activator of firing14, 15.
In sharp contrast to the prevailing linear model, we demonstrate that Cln1,2-dependent positive feedback is central to Start control. We use single-cell time-lapse fluorescent microscopy to show that Cln1,2 advance timing and reduce variability in the activation of CLN2, and of the entire G1/S regulon. We further explore the mechanisms and functional significance of this control.
Positive feedback of Cln1,2 on their own transcription should yield faster accumulation of CLN2 mRNA in WT cells than in cln1Δ cln2Δ cells. Although Cln1,2-dependent positive feedback was clearly demonstrated in the absence of Cln322–24, this does not imply that WT cells function similarly. In synchronized populations, near-identical timing of onset of CLN2 promoter activity was observed in the presence or absence of CLN1,2, leading to the linear model14, 15. Here, we revisit this issue using single cell assays. As a reporter for CLN2 transcription, we use unstable GFP driven by the CLN2 promoter24, 25 (see Methods and Fig S1–2). Birth time was determined using a marker for cytokinesis (disappearance of the Myo1-GFP myosin ring11, which did not influence the CLN2pr-GFP signal). The timing of CLN2 promoter induction in individual cells is sharp and easily quantified computationally (see Methods, Fig 1d,e and Fig. S1–2). Since cln1Δ cln2Δ cells are larger than WT, we integrated MET3pr-CLN2 in both strains, to conditionally express Cln2 prior to time-lapse imaging so that initially sizes were comparable14 (see methods; Fig. S3&S12 for controls). Thus, we can assay for positive feedback by comparing the time interval from birth to transcriptional activation of CLN2pr-GFP transcription in the first cell cycle after MET3pr-CLN2 turnoff in WT and cln1Δ cln2Δ cells.
Positive feedback should advance CLN2 promoter activation in WT compared to cln1Δ cln2Δ cells14, 15. Strikingly, in daughter cells, the average time between birth and CLN2 promoter activation (τon; Fig. 1d–e,f) was much shorter for WT (41 min) than for cln1Δ cln2Δ (83 min). Furthermore, CLN2pr-GFP activation was much less variable for WT than for cln1Δ cln2Δ cells (standard deviation of 21 vs. 47 min). CLN2pr-GFP transcription was Cln3-dependent in cln1Δ cln2Δ cells. Qualitatively similar results were obtained in mother cells and also in cells growing in glycerol/ethanol instead of glucose. In all cases, the interval from birth to CLN2pr-GFP activation was smaller and less variable in WT than in cln1Δ cln2Δ, indicating strong positive feedback of Cln1,2 on their own transcription independent of nutrient conditions or cell type (Table S3; P < 10−4).
We explored the potential redundancy of CLN1 and CLN2 in activating the feedback loop. Although budding is slightly delayed in cln1Δ CLN2, and CLN1 cln2Δ cells compared to WT, the timing of CLN2 promoter activation is similar (Table S3), indicating that CLN1 and CLN2 form redundant conduits for positive feedback.
Our data can be reconciled with previous work14, 15 arguing against positive feedback because measurements averaged over a population of cells necessarily lose information. In addition to delayed onset of transcription, cln1Δ cln2Δ cells express a more intense and prolonged CLN2pr-GFP signal. The larger peaks are likely due to a delay in the Clb2-mediated repression of SBF/MBF14, 15, 20, 21 (Fig. 1d–e), since the average time between induction of CLN2 and CLB2 was much larger in cln1Δcln2Δ strains, (measured using a CLB2pr-GFP cassette) (Fig. 1f; Fig. S13), and Clb2p accumulation is known to be delayed in cln1Δcln2Δ strains14.
Therefore, imperfect synchrony11 allows the high and lengthened transcriptional response from the first cln1Δcln2Δ cells firing the CLN2 promoter to mask the delayed response of the majority. This effect is reconstituted in Fig. 1g by averaging our measured single-cell data, and explains why positive feedback was not detected in measurements of mRNA levels in populations of synchronized daughter cells14, 15.
Once a cell senses the signal to initiate the cell cycle, it must actuate all the machinery effecting the cell cycle transition. At Start, this requires activating many SBF and MBF regulated genes16 encoding proteins involved in DNA replication and bud site formation. However, noise in protein expression in individual cells26 could interfere with expression of this large regulon. In particular, the delayed and variable induction of the CLN2 promoter in cln1Δ cln2Δ cells suggested that G1/S regulon expression might be severely disrupted in these feedback-free cells.
To investigate regulon expression in individual cells, we compared induction of CLN2pr-GFP and RAD27-mCherry, another member of the regulon16 (Fig. 2a–d; Fig. S7–8). RAD27 expression is Cln-dependent (Fig. S11). CLN2 and RAD27 are synchronously induced in WT, while there is a long and variable period of time between the inductions of the two genes in the cln1Δ cln2Δ mutant (Fig. 2e–f). Indeed, out of the 86 cln1Δ cln2Δ cells studied, 11 failed to produce a detectable increase in Rad27-mCherry and 4 failed to produce a detectable increase of either marker. We performed identical experiments on strains containing CLN2pr-GFP and RFA1-mCherry, another regulon member16, and obtained similar results (Fig. 2g–h). Our conclusions are valid even after excluding outlying points (P<0.01). Thus, Cln1/2 dependent positive feedback likely promotes coherent and efficient transcription across the SBF/MBF regulon.
Further comparison of these three promoters in cln1Δ cln2Δ cells reveals that CLN2 is almost always the first of the three to be activated, while the times to subsequent RFA1pr and RAD27pr inductions are significantly different from each other (P=0.004; Table S3). This suggests that the CLN2 promoter is the easiest for Cln3 to induce, followed by the RFA1 promoter, followed by the RAD27 promoter. We note that two MBF targets27–29, RAD27 and RFA1, exhibit different induction timing.
To ask whether lack of coherence in cln1Δ cln2Δ cells might simply come from low G1 cyclin levels, we analyzed cln1Δ cln2Δ 6xCLN3 cells. Although expression of both the CLN2 and RAD27 promoters was significantly accelerated by extra CLN3, these cells still exhibited strongly incoherent expression compared to WT (Fig 2i).
To directly short-circuit the proposed positive feedback loop, we examined gene expression in cln1Δ cln2Δ cln3Δ MET3pr-CLN2 cells on methionine-free medium (MET3pr-CLN2 on). Although induction of CLN2pr-GFP and RAD27-mCherry were strongly accelerated by constitutive CLN2 expression, incoherent expression compared to WT was still observed (Fig. 2j). Intriguingly, this incoherence was due to RAD27-mCherry induction prior to CLN2pr-GFP, compared to nearly simultaneous expression in WT (−8±2 min compared to 2±1 min; P<10−3), perhaps due to differential loading of SBF (CLN2) and MBF (RAD27) regulated genes21, 27–30.
Overall, these experiments suggest that the positive feedback architecture is a particularly effective way to promote coherent regulon expression.
In addition to exhibiting incoherent gene expression, 26% of cln1Δ cln2Δ cells fail to bud (Fig. 3a). We hypothesized that incoherent gene expression plays a role in this sporadic unbudded arrest. 20 out of 143 assayed cln1Δ cln2Δ cells were ‘strongly incoherent’: they failed to transcribe one or both of their two transcriptional markers (Fig. 2f,h). 90% of the ‘strongly incoherent’ cells arrested unbudded compared to 26% of all cln1Δ cln2Δ cells (P<0.003; Fig. 3a). Thus, a lack of coherence in the SBF/MBF regulon is a strong predictor of unbudded arrest within the cln1Δ cln2Δ population. 6X CLN3 reduced unbudded arrest in cln1Δ cln2Δ cells, perhaps because of accelerated regulon expression (Fig. 2i). Thus, unbudded arrest in cln1Δ cln2Δ cells may result from highly delayed expression of some regulon members.
We hypothesized that in strongly incoherent cells, activation of only some regulon members might lead to activation of mitotic Clbs, which would then inactivate further SBF/MBF regulated expression20 (Fig. 1a; Fig. S9). If genes required for budding in the absence of CLN1,2, such as PCL1,231, had not yet been expressed, unbudded arrest might ensue. Indeed, 20/20 arrested cln1Δ cln2Δ cells contained large amounts of nuclear Clb2-YFP (Fig. 3b–c).
To further test the role of transcription in unbudded arrest, we deleted the rate-limiting SBF inhibitor CLB2 in a MET3pr-CLN2 cln1Δ cln2Δ strain and observed a decrease in unbudded arrest from 26% to 13% (Fig. 3d). Additionally, we integrated unphosphorylatable Cdh1 under GAL control (GALL-HA3-CDH1-m11) into a cln1Δ cln2Δ MET3pr-CLN2 strain to induce rapid degradation of all mitotic cyclins upon galactose induction32. This reduced the unbudded arrested fraction to 4% in the first cell cycle following GAL induction (Fig. 3d). Since the timing of CLB2pr-GFP induction in cln1Δcln2Δ cells was similar whether they arrested or not (P = 0.91), the unbudded arrest was not due to unusually early CLB2 induction.
Thus, mitotic cyclins promote unbudded arrest specifically in highly incoherent cln1Δ cln2Δ cells, probably due to insufficient regulon expression before Clb-dependent SBF/MBF inactivation.
We wanted to determine if Cln1,2-dependent positive feedback operated through Whi5, a transcriptional inhibitor of the G1/S regulon18, 19. Whi5 inactivation is rate-limiting for CLN2 transcription and occurs via Cln-dependent phosphorylation, which leads to nuclear exclusion19.
First, we developed a quantitative assay for nuclear levels of Whi5-GFP by marking the nucleus with HTB2-mCherry (histone H2B) and measuring the difference between nuclear and cytoplasmic GFP fluorescence intensity(Fig. 4a–c). Whi5 entered the nucleus rapidly in both WT and cln1Δ cln2Δ cells. In WT cells, Whi5 also exited very rapidly. In cln1Δ cln2Δ cells, Whi5 exited much more slowly (Fig. 4d–g,i) consistent with biochemical data showing that Whi5 remains on the CLN2 promoter longer in cln1Δ cln2Δ than in WT cells18. Since in cln1Δ cln2Δ cln3Δ cells, Whi5-GFP remained nuclear (Fig. 4h), the slow Whi5 exit in cln1Δ cln2Δ cells is Cln3-dependent (this also excludes photobleaching artefacts). Thus, Cln3 initiates the slow exit of Whi5 from the nucleus, while Cln1/2 rapidly removes the remainder.
Since Whi5 exit and CLN2 induction are tightly correlated in WT (Fig. 4j), translocation occurs shortly after Whi5 inactivation and coincides with activation of transcriptional positive feedback. CLN2 promoter activation and Whi5 exit were less tightely correlated in cln1Δ cln2Δ cells consistent with the gradual exit of Whi5 (Fig. 4k; Fig. S5–6).
To examine the role of Whi5 phosphorylation in positive feedback and regulon coherence, we used a WHI5(6A) allele19 lacking 6 of 12 Cln-dependent phosphorylation sites. Although Whi5(6A) was reported to be constitutively nuclear19, we observed significant, but slower and incomplete, shuttling of Whi5(6A)-GFP out of the nucleus at Start and again at nuclear division (10/10 cells; Fig. 4l). CLN2 and RAD27 induction are less coherent in WHI5(6A) than in WT (Fig. 4m; but more coherent than cln1Δ cln2Δ), correlating with the poor nuclear transport of Whi5(6A). Thus, interfering with the positive feedback loop by reducing the ability of Cln proteins to phosphorylate Whi5 reduces regulon coherence, even with all three G1 cyclins present.
The addition of WHI5(6A) to cln1Δ cln2Δ cells increased the frequency of unbudded arrest from 26% to 51%, consistent with the idea that unbudded arrest is a consequence of incoherent regulon expression in cln1Δ cln2Δ cells.
Overall, these results strongly suggest that Whi5 is a Cln1,2 substrate in WT cells, and that this phosphorylation contributes to positive feedback. To see if Whi5 was the only such substrate, we compared timing of CLN2 promoter activation for whi5Δ and cln1Δ cln2Δ whi5Δ cells (Fig. S14; Table S3). Deletion of WHI5 advances CLN2 promoter induction in both WT and cln1Δ cln2Δ cells. Since cln1Δ cln2Δ whi5Δ cells delayed CLN2pr induction relative to whi5Δ cells, Cln1,2 likely act through a Whi5-dependent and a Whi5-independent mechanism to promote positive feedback. Previous results suggested a Whi5-independent Cln3 requirement for SBF activation19, possibly acting through Swi619, 33; a similar mechanism may be employed by Cln1,2.
Cln1,2 are pleiotropic effectors of Start with important nuclear and cytoplasmic functions34, 35, complicating interpretation of cln1Δ cln2Δ phenotypes. Therefore, we tested forced-localization CLN2 alleles, expressed from the wild-type CLN2 promoter, that restrict Cln2 to either the nucleus (CLN2-NLS) or the cytoplasm (CLN2-NES)34. cln1Δ cln2Δ CLN2-NLS cells exhibit coherent regulon expression (P=0.45 compared to WT), but cln1Δ cln2Δ CLN2-NES cells are highly incoherent compared to WT (P<10−7), implying that coherent gene expression is primarily a nuclear function of CLN2 (Fig. 6a–b; compare to Fig. 2; Table S3).
Consistent with a role of cytoplasmic Cln2 in budding34, 35, integration of CLN2-NES into cln1Δ cln2Δ cells strongly reduces arrest (to 3%) in spite of less coherent gene expression. Furthermore, exogenous expression of CLN2 drives cell cycle progression in previously blocked cln1Δ cln2Δ cells (Fig. S10) and restores viability of mbp1Δ swi4Δ cells, which lack SBF and MBF and have very low regulon expression36, 37. The localization mutants also have different efficacy for relieving unbudded arrest. Integration of CLN2-NLS into cln1Δ cln2Δ cells, providing coherent gene expression, leads to a partial but significant reduction of unbudded arrest (from 26% to 19%; P=0.04).
Thus, cell morphogenesis and budding can be driven by two partially redundant pathways: via cytoplasmic Cln1,234, 38 or other genes in the G1/S regulon such as Pcl1,231 (Fig. 6c). Having Cln1,2 coherently activate the G1/S regulon and directly drive bud emergence provides a compact solution to ensure efficient and timely morphogenesis and G1/S regulon expression, before subsequent Clb activation.
The regulatory architecture of the G1/S regulon provides an effective design to promote coordinated activation. The promoters are pre-loaded during G1 with a complex of factors that are subsequently rapidly activated by phosphorylation17, 24, 30 removing a potentially rate-limiting step. Furthermore, the upstream cyclin Cln3 is intrinsically more capable of triggering the CLN2 promoter compared to two other randomly selected promoters from the regulon (RFA1 or RAD27; Fig 2e–h). High sensitivity of CLN1/2 to Cln3 means that positive feedback from the initial burst of Cln1,2 will ensure that all other genes fire together. Indeed, in our experiments in WT cells, the genes are expressed too synchronously to evaluate which comes first. We find it likely that positive feedback will be a recurring motif within genetic control networks responsible for the coherent temporal coordination of multiple downstream events.
The sharpness of the Start switch, defined by the rapid exclusion of the transcriptional inhibitor Whi5 and the coherent expression of the G1/S regulon, is principally due to CLN1,2-dependent positive feedback (Fig. 6c, red lines) rather than a linear Cln3-Whi5-SBF pathway14, 15, 18, 19. Our data are inconsistent with the sharpness of Start being primarily due to non-linear increases in CLN3 translation39 or nuclear translocation40, or cooperative phosphorylation of Whi5 by Cln319, since these mechanisms all predict a sharp switch in feedback-free cln1Δ cln2Δ cells.
In budding yeast, Start is a fundamental point of commitment where physiological inputs such as nutrients, mating factor, size and cell type are integrated to produce an all-or-none decision. We show here that positive feedback provides robust switch-like cell cycle entry. Our single-cell data suggest that the point of commitment to the cell cycle, Start, is a very brief interval coinciding with the initiation of positive feedback and Whi5 exclusion. Subsequent Cln-dependent events, such as Sic1 phosphorylation and degradation41 leading to DNA replication, could then be viewed as dependent on, rather than part of, Start.
This work also provides a molecular basis for understanding the modular structure of G111. Two temporally uncorrelated processes in G1 are separated by the molecular event of Whi5 inactivation and nuclear exit. The upstream module is responsible for cell size control, while the downstream size-independent module actuates cell cycle progression11. Here, we showed that rapid Whi5-exit coincides with initiation of Cln1,2-dependent positive feedback. Once feedback is initiated, the rapidly accumulating Cln1,2 likely dominates cellular Cln-kinase activity and Cln3, the rate-limiting upstream activator, is rendered unimportant. In general, we expect modularity, best revealed by single-cell analysis, to be a signature of feedback-driven cellular control networks.
Our systems-level analysis of Start provides a template for further studies of other checkpoints in yeasts or the G1/S transition in mammals. The utility of feedback at Start leads us to expect similar regulatory architecture across eukaryotes, even if the enabling genes are not homologous.
Standard methods were used throughout. All strains are W303-congenic.
Preparation of cells for time-lapse microscopy was performed as previously described24. Since mutant cells are larger than WT, we integrated MET3pr-CLN2 to conditionally express Cln214. On media lacking methionine (MET3pr-CLN2 on), cells bud and divide at comparable sizes (Fig. S3). By pre-growing cells without methionine before plating on media containing methionine (MET3pr-CLN2 off), we are able to begin our time-lapse imaging experiments with similarly sized WT and cln1Δ cln2Δ cells. We imaged the first Start in cells that were budded at the time of transfer,and that divided least 30 minutes after methionine addition, to allow degradation of Cln213, 42 made before MET3 promoter turnoff.
Automated image segmentation and fluorescence quantification of yeast grown under time-lapse conditions were performed as previously described11, 24. We added a function to previously described custom software24 to identify nuclei labeled with Htb2-mCherry (histone). The red signal was smoothed, disconnected fragments were eliminated and the cells with nuclei too small, or dim, or oddly shaped (area vs. minimally enclosed rectangle) were eliminated. After background subtraction, the nucleus was defined to be where the fluorescence was greater than 70% of maximum, which controls for cell variability and vertical movement of the nucleus. The nuclear Whi5-GFP signal was the difference between the average nuclear and cytosolic intensities.
Fluorescence time series were extracted from movies as previously described24. Time-series were fit using smoothing splines (MATLAB) with a smoothing parameter of 0.001. We defined the onset of transcription for a G1/S fluorescent reporter by the maximum in the second derivative that fell between birth and budding (scored separately), which accurately locates rate-changes in spite of noisy data and slow changes in the background fluorescence (Fig. S3–4).
Standard methods were used throughout. All strains are W303-congenic. In synchronized WT cells, GFP mRNA from the CLN2 promoter and CLN2 mRNA follow similar kinetics, and accumulation of cellular fluorescence follows with a slight delay24. WHI5(6A) and WHI5(6A)-GFP strains with modified WHI5 at the endogenous locus were a gift from M. Tyers. Plasmids for introduction of CLN2-NES and CLN2-NLS under control of the CLN2 promoter were obtained from B. Futcher, and integrated at the ura3 locus in a cln1Δ cln2Δ background. Histone H2B (HTB2) was C-terminally tagged with mCherry using PCR-mediated tagging, with the template plasmid pKT35543 by J. Bean and B. Timney. RAD27 and RFA1 were tagged similarly. All other alleles were from laboratory stocks described previously.
Preparation of cells for time-lapse microscopy was performed as previously described24. Since mutant cells are larger than WT, we integrated MET3pr-CLN2 to conditionally express Cln214. On media lacking methionine (MET3pr-CLN2 on), cells bud and divide at comparable sizes (Fig. S3). By pre-growing cells without methionine before plating on media containing methionine (MET3pr-CLN2 off), we are able to begin our time-lapse imaging experiments with similarly sized WT and cln1Δ cln2Δ cells. We imaged the first Start in cells that were budded at the time of transfer, and that divided least 30 minutes after methionine addition, to allow degradation of Cln213, 42 made before MET3 promoter turnoff. Briefly, growth of microcolonies was observed with fluorescence time-lapse microscopy at 30ºC using a Leica DMIRE2 inverted microscope with a Ludl motorized XY stage. Images were acquired every 3 minutes for cells grown in glucose and every 6 minutes for cells grown in glycerol/ethanol with a Hamamatsu Orca-ER camera. Custom Visual Basic software integrated with ImagePro Plus was used to automate image acquisition and microscope control.
Automated image segmentation and fluorescence quantification of yeast grown under time-lapse conditions were performed as previously described24. Budding was scored visually, and cell birth was scored by the disappearance of Myo1-GFP at the bud neck, generally with single frame accuracy. Background was measured as the average fluorescence of unlabelled cells and subtracted from the measured pixel intensities. We added a function to previously described custom software24 to identify nuclei labeled with Htb2-mCherry (histone). The red signal was smoothed, disconnected fragments were eliminated and the cells with nuclei too small, or dim, or oddly shaped (area vs. minimally enclosed rectangle) were eliminated. After background subtraction, the nucleus was defined to be where the fluorescence was greater than 70% of maximum, which controls for cell variability and vertical movement of the nucleus. The nuclear Whi5-GFP signal was the difference between the average nuclear and cytosolic intensities.
P-values using appropriate tests yielded P<0.001 for all comparisons in the text, except where noted. Fluorescence time series were extracted from movies as previously described24. Time-series were fit using smoothing splines (MATLAB) with a smoothing parameter of 0.001. We defined the onset of transcription for a G1/S fluorescent reporter by the maximum in the second derivative that fell between birth and budding (scored separately). This method was chosen because it accurately locates rate-changes in spite of noisy data and slow changes in the background fluorescence. The onset time was nearly unchanged over a range of 103 in smoothing parameter (Fig. S3–4).
This work was supported by the National Institute of Health (J.M.S., E.D.S., F.R.C.), the Burroughs Wellcome Fund (J.S) and the National Science Foundation (E.D.S.). We thank N. Buchler, G. Charvin, B. Drapkin and J.E. Ferrell for insightful conversations, and J. Widom and C. Wittenberg for thoughtful comments on the manuscript. We thank J.M. Bean, B. Timney and J. Robbins for help with strain/plasmid construction, M. Schwab for the plasmid pWS358, B. Futcher for the CLN2-NES and CLN2-NLS plasmids, E. Bi for the pKT355 mCherry tagging plasmid, and M. Tyers for WHI5 phosphorylation site mutant strains and plasmids. The authors declare no competing financial interests.