Previous Article |
Table of Contents
| Next Article 
PHYSICAL SCIENCES / BIOLOGICAL SCIENCES / APPLIED MATHEMATICS / CELL BIOLOGY
High-resolution timing of cell cycle-regulated gene expression
Department of Biochemistry, University of Texas Southwestern Medical Center, 5323 Harry Hines Boulevard, Dallas, TX 75390
Communicated by Steven L. McKnight, University of Texas Southwestern Medical Center, Dallas, TX, June 27, 2007 (received for review February 23, 2007)
 |
Abstract |
|---|
 Top Abstract ResultsDiscussionMethodsAcknowledgementsReferences |
|---|
1% of the cell cycle time). The set of 1,129 cell cycle-regulated genes was identified by a comprehensive analysis encompassing all available cell cycle yeast data sets. Our results reveal distinct subphases of the cell cycle undetectable by morphological observation, as well as the precise timeline of macromolecular complex assembly during key cell cycle events.
mitosis | microarrays | yeast | maximum entropy
Previous whole-genome studies of the yeast cell cycle relied on microarray data from artificially synchronized cultures grown under nutrient-rich laboratory conditions (1, 3, 5). Under these typical conditions, cell divisions happen immediately, one after another, with almost overlapping cell division phases, thereby hindering accurate timing [see supporting information (SI) Appendix]. Moreover, Fourier methods have typically been used for timing of genes (3, 5, 8), but these methods are limited to genes with one expression peak per cycle and are most accurate for temporal profiles resembling a sinusoid. A model-based approach to timing was also developed (12), but it remained unable to deal with multiple peaks or to achieve a high temporal resolution. Because of these experimental and methodological limitations, the timing of the cell cycle transcriptional program by using logarithmically growing yeast and Fourier methods (3, 5, 13) lacks within-phase resolution (i.e., the variance of expression peaks of individual genes is comparable to the lengths of cell cycle phases; see SI Appendix). The timing of gene expression by using such data and methods, therefore, is not interpretable beyond assigning transcripts to main cell cycle phases in which their expression peaks: G1 (gap 1), S (DNA synthesis), G2 (gap 2), M (mitosis), and M/G1 (1, 5).
Here we present a unique approach to the timing of cell cycle-regulated gene expression. We rely on a recent data set, in which a naturally synchronized continuous yeast culture proceeds through 5-h metabolic cycles (yeast metabolic cycle; YMC) (14) while exhibiting strong and stable cell cycle synchronization. Using the characteristic profile shape of regulated genes (Fig. 1) to align cell cycles of the whole culture by maximum-entropy deconvolution, we reveal a high-resolution timeline of cell cycle transcriptional events in fine detail.
|
|
Results |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
100 cycles, whereas synchronization achieved by other methods deteriorates noticeably during the first three cycles (5). The stable repetition of temporal expression patterns over consecutive cycles is an essential requirement for applying our deconvolution-based, accurate timing method. Therefore, we chose to base high-resolution timing solely on the YMC data (14), whereas other data sets (1, 3, 5) were used for identification of the set of cell cycle transcriptionally regulated (CCTR) genes.
|
To recover the mRNA concentration in the typical individual cell, we deconvolved the measured profile by using the common shape. The intrinsic noise in budding yeast gene expression is low (18), so we expect our estimated average individual cell expression timing to be reflective of the majority of actual single cells. We implemented a deconvolution algorithm adapted to microarray data analysis (see SI Appendix), with regularization based on the maximum-entropy principle (19). We thus accurately determined the moment of the gene expression peak by aligning cell cycles of the whole culture by deconvolution of the observed expression profiles (see SI Appendix). This method allows the recovery of single cell expression profiles, which in the microarray measurement are distorted because of averaging mRNA levels of imperfectly synchronized cells (Fig. 3).
|
|
Error Estimation. To estimate the robustness of the timing procedure, we have investigated the extent to which the obtained peak times are affected by expected errors in the measured mRNA concentrations (see SI Appendix). We confirmed that the timing method is very robust; the median value of estimated total error for predicted peak times is 2 min (see Table 6 in SI Appendix). Such high-accuracy timing allows annotating of each transcript to a small fraction of a cell cycle phase and reveals otherwise undetectable differences in gene expression times. The complete list of CCTR genes, together with their expression peaks and error estimates, is available online at http://cellcycle.info.
Phase and Subphase Assignment. We defined time intervals corresponding to the main cell cycle phases (Fig. 4E) by using expression peaks of known cell cycle genes (see Table 5 in SI Appendix). The histogram of expression peaks of CCTR genes (Fig. 4F) reveals two main waves of transcription, separated by intervals of almost no CCTR transcriptional activity, between late S and late G2 phases and in most of the G1 phase. This dramatic variation in transcriptional activity between stages of the cell cycle has not been described previously (5) (SI Fig. 6). Beyond prominent expression waves in G1/S–S and G2/M–M, the histogram in Fig. 4F reveals a previously unidentified (1, 3, 5), distinct expression wave preceding the start of DNA replication. In YMC, this wave spans 45 min and encompasses 19% of CCTR genes. Because the majority of subunits of the prereplicative complex are expressed in this phase (Fig. 4A), we propose to designate it the "prereplicative" or "G1 (P)" phase. Other genes expressed in G1 (P) are involved in preparation for budding (e.g., RSR1, BUD13, GIC2, RAX1, PEA2, and BNI4) and in synthesis of cell wall components (e.g., FKS1, GAS1, and GAS5). In previous studies, G1 (P) genes were perhaps incorrectly assigned to different cell cycle phases (1, 5). More than one-third had been annotated as being expressed in mitosis or M/G1 (1, 5), including all subunits of the MCM complex and the G1 cyclin CLN3 (1, 5). Our timing places expression of these genes at the beginning of the new cycle, suggesting involvement in preparation for a round of division rather than for entry into extended G1 phase, which is more consistent with their known biological function (6, 24).
Another gene expressed in G1 (P) phase is CDC28, the catalytic subunit of the main yeast cyclin-dependent kinase, which drives progress through the cell cycle (25). Our study, which classifies CDC28 as periodic, challenges the established view (3, 5, 8, 12, 25) that CDC28 is constitutively expressed. We determined that CDC28 expression peaks twice per cycle, first in G1 (P) phase and again in early mitosis (Fig. 4C), precisely coinciding with expression waves of its predicted targets (26) (Fig. 4H). The periodicity of CDC28 expression in YMC is strikingly clear (SI Fig. 10; P < 0.00003); it also is not an artifact of metabolic regulation, because a similar profile of CDC28 had been earlier observed under different cell cycle synchronization (1) (SI Fig. 5). The lack of earlier acceptance of CDC28 as transcriptionally regulated seems to be rather an artifact of the Fourier methods used (3, 5, 8), which, although convenient, are unable to deal appropriately with genes expressed twice per cycle (see SI Appendix).
Our timing results have also clarified when some key cell cycle genes are expressed. For instance, on the basis of experiments with rapidly growing yeast (1, 5), CLB1, SWI5, and CLN3 have all been thought to be expressed during mitosis (1, 5), whereas our data suggest that G2 cyclin CLB1 is expressed in G2, SWI5 in G2/M, and CLN3 upon reentry to the cell cycle, in G1 (P) (see Fig. 2). Similarly, we find that the Swi5-activated cyclin, PCL9, is expressed in mitosis and MCM3 is expressed in G1 (P), and consequently their expression can be delayed by an arbitrarily long G1 phase, although, on the basis of experiments with rapidly growing yeast (1, 5), they were both believed to be expressed in M/G1 (Fig. 2). Our timing results consistently place the expression of these key genes just before the time their products are needed within the cell cycle (6, 24).
Initiation of DNA Replication.
The initiation of DNA replication occurs at the beginning of S phase and requires the prior assembly and subsequent modifications of the prereplicative complex (24), which starts in G1 (P) (Fig. 4B). Strikingly, our timing of the expression peaks of CCTR subunits of MCM, replicative complex and elongation complex, corresponds with the exact order in which their gene products are needed (Fig. 4 A and B). The subunits of the origin of replication complex, ORC2–6, have not previously been classified as transcriptionally regulated, nor did they pass the stringent criteria of being accepted as CCTR in this study. Still, applying the deconvolution timing to their expression profiles reveals that these genes have expression peaks 10 min before the MCM subunits, exactly when their products are needed. This observation raises the possibility that the transcription of ORC2–6 is regulated as a function of the cell cycle, contrary to established beliefs (1, 3, 5, 8) (Fig. 4A).
A more detailed view reveals that subunits of the MCM complex are expressed in two groups of three, separated by an 8-min interval, with each predicted expression group containing one MCM subunit with a nuclear localization signal (27) (Fig. 4A). These results may provide insight into the dynamics of MCM complex assembly and transport from cytoplasm into nucleus (28).
The precision of our timing data reveals that the SBF- and MBF-activated expression programs, thought to be identically timed during the mitotic cell cycle (29), actually differ (Fig. 4G). Unlike MBF, whose targets peak predominantly in G1/S phase, targets of SBF are also activated in G1 (P) phase and are generally characterized by a broader time distribution (Fig. 4G). This conclusion holds, independent of whether SBF and MBF targets are defined based on evolutionary analysis of conserved binding sites in 17 related fungus species or on various experimental studies (29, 30).
Cell Cycle-Regulated Complexes. We also timed expression of several other complexes, such as the spindle pole body (SPB) (Fig. 4C) (see Table 5 in SI Appendix). We observed especially tight transcriptional coregulation for complexes active in late G1 and S phase; the elapsed time between expression peaks of the first and last CCTR subunits of a complex is between 5 min (RFA) and 22 min (histones) (Fig. 4C). Transcription of many non-CCTR subunits of the cell cycle complexes also exhibits variability, allowing for timing, albeit weak. For example, only three subunits of RFC are CCTR, although expression of all five subunits occurs in the same 12-min interval (Fig. 4C). ORC2–6 exhibit even weaker modulation, but interestingly their expression timing is nevertheless very consistent with the time in which they function (Fig. 4 A and B). The anaphase-promoting complex (APC) contains only two CCTR subunits, although the expression peaks of most of its 16 subunits appear in a time interval broadly corresponding to mitosis (Fig. 4C). However, some APC subunits, e.g., Cdh1, seem to be only posttranscriptionally regulated (31).
We generally find more subunits of the cell cycle-involved complexes to be CCTR than was the case in previous studies (2, 5, 8) (Fig. 4 and see SI Appendix and http://cellcycle.info). This difference may be explained by the increased quantity and improved accuracy of cell cycle expression data, together with our comprehensive approach to identifying CCTR genes. In addition to the examples discussed above and complexes involved in DNA replication initiation, we find more components of the septin ring of the mother-bud neck to be CCTR. Previous studies (2, 3, 5, 8) each classified only one septin (either CDC11 or CDC10) as cell cycle-regulated, whereas we classify three components of the septin ring (CDC11, CDC12, and CDC3) as CCTR. Our classification of CDC11, CDC12, and CDC3 as coregulated is independently supported by timing results (not used for classification), which places their expression peaks within an 6-min interval in late S phase.
|
Discussion |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
It is possible, however, that some of the CCTR genes periodic in the YMC are regulated as a part of the metabolic oscillation, independent of the cell cycle. Therefore, we have used expression patterns from all budding yeast cell cycle data sets for classification of genes as CCTR. Still, the YMC condition increases the number of genes in our CCTR set, but we view this as a reflection of the cell cycle under a different set of natural conditions, rather than as an artifact of metabolically induced synchrony.
In previous cell cycle studies, two types of synchrony have been used (32–34): (i) induced synchrony, which forces cells to synchronize by some intervention such as pheromone arrest or the use of temperature-sensitive cdc mutants, and (ii) selection synchrony, which selects a cohort of cells at the same cell cycle stage, as in the elutriation method. Although it has been argued that selection methods are theoretically superior (35), so far for budding yeast they have resulted in whole-genome cell cycle data of inferior quality as compared with the intervention method (5). The cell cycle synchronization method we use here is of a third type. It is more a natural state of metabolic synchrony than an intervention method because, although an initial starvation is necessary to induce YMC synchrony, unlike in all other intervention methods (1, 5, 32), the synchrony remains stable (14).
How is the timing of gene expression that we describe here mirrored at the level of proteins? The key factor here is protein half-life: the more short-lived the protein, the closer the protein concentration correlates with transcript concentration. A recent whole-genome study of protein half-lives in budding yeast (36) reports that known cell cycle proteins are especially short-lived. These data show as well that our CCTR set is significantly enriched in rapidly degraded proteins, confirming that studying the timing of CCTR gene expression is a valuable proxy for understanding the temporal orchestration of the cell cycle proteome.
The deconvolution-based timing method we developed can also be applied to cell cycle data from other species (1, 4, 13, 37–39) and to analysis of other temporal phenomena with models of cell population synchrony, such as those during circadian rhythms (40) or development (41). One can judge the potential advantages of applying our methods by examining the power spectrum of the temporal profiles in question. The advantages of applying deconvolution will be greatest when squared magnitudes of discrete Fourier modes corresponding to multiplicities of the primary frequency are a measurable contribution to the power spectrum.
In summary, the precise timing of gene expression, using data from a highly synchronized continuous culture, has revealed the sequence of cell division events in fine detail, thereby providing a reference for defining cell cycle stages and studying individual gene functions. We have observed a prereplicative expression wave, occurring at the crucial time before the start of DNA replication. Our work has also enriched the description of eukaryotic cell cycle transcription, supplying detailed information on cell cycle progress under nutrient-poor conditions, which are perhaps more reflective of natural yeast growth conditions in the wild. From our analysis, we infer that just-in-time transcription is more prevalent in cell cycle regulation than previously recognized. The confinement of expression of genes sharing the same function to specific time intervals within the cycle (Fig. 4H) can be helpful in predicting gene function (SI Fig. 9). Moreover, the striking correlation between genes having a related function and the timing of their expression could result from the need for economical use of transcripts and proteins. Although phenotypes corresponding to the loss of such temporal optimizations could be difficult to observe, they might still be subject to natural selection in species such as yeast, which have a large effective population size.
|
Methods |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
Deconvolution Algorithm.
The measured profile, M, is the convolution of the individual cell profile, f, with the time-shift distribution h:
 |
To find f, we optimize the following functional consisting of goodness-of-fit and regularization terms:
 |
where the sum is over all measurements, A corresponds to the amount of regularization applied, and 
is the expected error of each measurement (see SI Appendix and SI Fig. 7).
Timing. For each of the CCTR genes, we calculated the deconvolved profiles and identified the peaks of expression. Significant peaks were selected based on a heuristic score that included peak height, width, and shape (see SI Appendix). Using deconvolved gene expression profiles, we were able to observe secondary (lower-scoring) expression peaks, even when they were not evident in the raw data.
Error Estimation.
To estimate the robustness of the timing procedure, we have investigated the extent to which the obtained peak times are affected by expected errors in the measured mRNA concentrations. Using Monte Carlo simulations (3,456 mock microarrays) we estimated the accuracy of peak time determination for all predicted CCTR genes (see SI Appendix). We confirmed that the timing method is very robust; the median value of estimated total error for predicted peak times is 2 min.
Physiological Time. To align precisely the three measured cycles in the YMC (which vary in length up to 11.5 min), we converted all measurement times into physiological time, using the rapid changes in dissolved oxygen concentrations as a reference (see SI Fig. 7 and SI Appendix). Thus, every measurement is assigned a physiological time between 0 and 300 min, with minute 0 corresponding to the point of maximal dissolved oxygen consumption. The resulting correction in timing is relatively small. To correct for residual long-term changes, logarithmic detrending has been applied to transcript concentrations. All timing results presented in SI Appendix and Fig. 4 are expressed in physiological time.
|
Acknowledgements |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
|
Footnotes |
|---|
Abbreviations: CCTR, cell cycle transcriptionally regulated; SPB, spindle pole body; YMC, yeast metabolic cycle.
*To whom correspondence may be addressed. E-mail: zo{at}work.swmed.edu, andrzej{at}work.swmed.edu, or maga{at}work.swmed.edu
Author contributions: M.R. and A.K. contributed equally to this work; M.R., A.K., B.P.T., and Z.O. designed research; M.R., A.K., and Z.O. performed research; M.R., A.K., B.P.T., and Z.O. contributed new reagents/analytic tools; M.R. and A.K. analyzed data; and M.R., A.K., B.P.T., and Z.O. wrote the paper.
Conflict of interest statement: S.L.M. declares a conflict of interest (such as defined by PNAS policy). "The manuscript titled ‘High-resolution timing of cell cycle-regulated gene expression’ is coauthored by a group of scientists working in the Department of Biochemistry that I chair here at University of Texas Southwestern Medical Center. I am highly familiar with the research because it makes extensive use of data coming from my own laboratory."
This article contains supporting information online at www.pnas.org/cgi/content/full/0706022104/DC1.
© 2007 by The National Academy of Sciences of the USA
|
References |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
