INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Ribosome biosynthesis is a complex and demanding process that depends directly upon multiple metabolic pathways, including the activities of three different RNA polymerases (reviewed in references 21 and 27). In Saccharomyces cerevisiae there are 137 ribosomal protein genes (RP genes), and they are transcribed by RNA polymerase II to yield 78 ribosomal proteins. Because the RP genes are transcribed at such a high level, together they account for nearly 50% of the total RNA polymerase II-mediated transcription initiation events (18). The 25S, 18S, and 5.8S rRNAs are synthesized by RNA polymerase I, first as a large 35S transcript that subsequently gets processed into the three smaller, mature species. Synthesis of the 5S rRNA is distinct from the other rRNAs and is carried out by RNA polymerase III. In order to achieve the high levels of rRNA production that are required during rapid cell division, yeast cells contain roughly 150 repeats of the rRNA genes in a tandem array on chromosome XII. Together, these repeats represent 10% of the genome, and rRNA production alone accounts for some 60% of the total cellular transcription.

Ribosome biogenesis also depends upon the activities of a large number of protein and RNA molecules that are not themselves components of the final ribosome. The complex processing pathway that converts the 35S precursor rRNA into the mature 25S, 18S, and 5.8S rRNA species requires a multitude of factors, including RNA endonucleases, exonucleases, RNA helicases, base modification enzymes, and small nucleolar RNAs (24). Many of these processing factors are nucleolar proteins that were identified through the characterization of mutants that exhibit defects in ribosome biosynthesis. For example, Ebp2p is an essential, nucleolar protein that is required for processing of the 27S pre-rRNA (13). Temperature-sensitive ebp2-1 mutants become depleted of the mature 25S and 5.8S rRNAs at the restrictive temperature, and this diminution leads to a decline in ribosome production and the cessation of cell division. Similarly, there are dozens of other genes whose essential functions relate to the roles they play in rRNA biosynthesis.

Given the importance of ribosome biogenesis to the total economy of cellular metabolism, it is perhaps not surprising that cells have evolved mechanisms to regulate this process. Yeast cells can modulate ribosome production in response to nutrient availability, heat shock, and defects in the secretory pathway (27). The major mechanism whereby cells effect this regulation is through transcriptional control, and both heat shock and secretory defects cause a rapid repression of rRNA and RP gene transcription (18, 19). Most RP gene promoters contain two Rap1p binding sites (17) and Rap1p can act both as an activator and as a silencer of transcription (20). Although promoter swap experiments have demonstrated that the Rap1p binding sequences from the RPL30 promoter are sufficient to confer the repression response when placed upstream of the ACT1 gene, they are not the only cis-acting elements involved in this response (18).

The recent development of microarray hybridization technologies has provided the opportunity to investigate the regulation of gene expression on a genome-wide scale (10). By this approach researchers have been able to compile very large data sets that include transcription profiles for nearly all of the known yeast genes. These data sets include expression profiles derived from cells proceeding through the cell cycle, as well as expression responses to external stresses and drug treatments (5, 7, 11, 23). Analysis of these data sets has identified clusters of genes that exhibit similar expression profiles. The members of these clusters could potentially represent common targets of a particular transcriptional control pathway, such as genes containing the MCB element in their promoters. Alternatively, the clusters could represent sets of genes that are regulated by different mechanisms to achieve a coordinated response to a particular cellular requirement or stimulus. In either case, the identification and characterization of these clusters provide insights into how cells coordinate and regulate gene expression networks in response to diverse stimuli.

In this study we have used microarray expression data (7, 9, 23) to identify a novel cluster of coregulated genes. This regulon is enriched for genes that encode for proteins that have been implicated to play a role in RNA metabolism and rRNA processing. Promoter analysis reveals that this set of genes is highly enriched for two sequence motifs that are preferentially located between 50 and 200 bp upstream of their respective translation initiation sites. Site-directed mutagenesis was used to demonstrate that both promoter elements were important in vivo for regulating the expression of the rRNA processing-related EBP2 gene after release from alpha factor arrest.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Strains and media. The yeast strains and plasmids used in this study are described in Table 1. Standard yeast genetic and molecular biology techniques were used throughout (1, 12), and a list of the oligonucleotides used here can be found in Table 2.

Cluster analysis. The microarray time-series data was analyzed by using the Partitioning Around Medoids (PAM) algorithm provided by the SPLUS statistical analysis software package (22). PAM is a variant of the well known "k-means" cluster algorithm for grouping multidimensional data. The relative expression measurements at T time points for each gene on the microarray correspond to a single point in a T-dimensional space. Nearby genes in this T-dimensional space will have similar expression profiles over time. The goal of the cluster algorithm is to identify the groups of points (genes) that are close together but far from other groups. The user specifies the number, k, of groups or clusters to be found and then the program automatically optimizes the membership of the clusters to minimize the distance between members of the cluster and to maximize the distance from other clusters (see reference 15 for details.).

Statistical analysis of clusters across experiments. To estimate the numbers of overlapping genes from a purely random sampling of N  3,000 genes, we first computed the probability that any particular set of m genes found in a cluster in the first experiment would be found in a cluster of size n₂ in the second experiment:

(1)

However, since there are
different ways of choosing a particular set of m genes from the n₁ genes in the original cluster in the first experiment, the probability that a cluster with n₁ genes in the first experiment shares m genes with a cluster of size n₂ in the second experiment can be estimated by the equation:

(2)

Note that equation 2 is an overestimate, since it double counts the occurrence of the same m + 1, m + 2, ... genes in both clusters. As a consequence, this estimate erroneously gives a probability of >1 when m is small. Finally, since there are n_c = 24 clusters in the first experiment that can be compared with any of the n_c = 24 clusters in the second experiment, P₂(N, n₁, n₂, m) should be multiplied by a factor of n_c² to estimate the odds of finding m genes in common between any two clusters in the two experiments by chance.

62

3

Alpha factor arrest, heat shock, and tunicamycin treatment. Yeast strains were grown to a density of 3 × 10⁷ cells/ml in either yeast extract-peptone-dextrose (YPD; pH 5.5) (yMM13 and yMM177) or synthetic complete (SC)-Leu liquid medium (yMM354, yMM355, and yMM373) at 30°C. The cultures were arrested by the addition of alpha factor for 2 h at a final concentration of 5 µg/ml. The cells were washed free of the alpha factor and released into fresh medium at 30°C. At the indicated time intervals, 10-ml aliquots of the cultures were collected on ice, harvested with a table-top centrifuge, and flash-frozen in a dry ice-ethanol bath.

RNA preparation and relative quantitative RT-PCR. Total RNA was prepared from the samples by using a hot acidic phenol extraction protocol (1). The concentrations of the RNA preparations were determined by spectrophotometry, and aliquots were tested by electrophoresis on 1.2% agarose-2.2 M formaldehyde gels. The samples were treated with DNase I (Ambion Co.) and tested by PCR to confirm that there was no contaminating DNA. Then, 2 µg of RNA from each sample was incubated with reverse transcriptase (RT) to yield cDNA as per the instructions in the Ambion RETROscript kit. PCRs were then performed as described by Ambion's QuantumRNA Universal 18S rRNA Internal Standards Kit. Briefly, for each gene, PCRs were performed to determine the linear range of amplification that would permit a quantitative assessment of expression levels (typically between 14 and 24 amplification cycles). We then empirically determined the 18S rRNA primer-to-competimer ratio that would allow for an appropriate level of amplification for the 18S rRNA internal control. PCRs containing two pairs of oligonucleotides (one for the 18S rRNA control and one for the gene of interest) were performed in the presence of [-³²P]dATP, and the fragments were separated by gel electrophoresis. The PCR products were quantified by PhosphorImager analysis and ImageQuant software. For each time point of each gene series the number of counts in the PCR product of interest were divided by the number of counts in the internal 18S control. Expression levels across a time course were then normalized to a mean of expression level of 1.

Construction of the motif 1 and motif 2 promoter element substitution strains. The two promoter motif elements located upstream of the EBP2N62 allele were altered by a PCR-based substitution strategy. Briefly, pairs of oligonucleotides were used to amplify separate DNA fragments that flanked the motif sequences within plasmid pMM147. Each of these fragments contained altered sequences including an AatII site at one end in place of the normal motif sequences. The two DNA fragments were digested with either AatII and NcoI or AatII and HindIII to produce compatible sticky ends. These two inserts were then ligated with an NcoI and HindIII fragment from plasmid pMM147 and transformed into Escherichia coli. The resulting motif 1 and motif 2 substitution plasmids (pMM322 and pMM341, respectively) were recovered, sequenced, and transformed into yeast strain yMM13.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Clustering of microarray expression profiles. Genome-wide changes in gene expression have been monitored by microarray analysis as yeast cells progress through a variety of conditions, including traversal through the cell cycle (23), through sporulation (7), and through diauxic shift (9). Each of these studies produced large data sets containing time course expression profiles for more than 6,000 yeast genes. In the effort to identify groups of genes that share similar transcription patterns, these data sets were independently analyzed by using a partitioning cluster algorithm (described in Materials and Methods). This algorithm is designed to divide the data set into groups of genes that are similar to one another and yet distinct from other groups. As initial parameters for this partitioning, we limited the analysis to the expression profiles that varied the most within each data set (ca. 3,000 genes in each case), and we set the number of clusters to be 24. Although somewhat arbitrary, these parameters were chosen to exclude those genes that varied the least over the respective time courses and to allow for the potential discrimination of many different expression profiles. The resulting clusters typically contained 75 to 175 genes each, and the relative "strengths" of membership for each gene within a cluster could be assigned based on the similarity of their profiles to the most representative or medoid gene.

View larger version (44K): [in this window] [in a new window]   FIG. 1.   Yeast gene expression profiles derived from microarray experiments were analyzed by a partitioning algorithm and grouped into 24 clusters. One gene expression data set came from cells after release from alpha factor arrest (23) (A), and another was derived from cells progressing through sporulation (7) (B). Expression profiles of two clusters from each set of 24 groups are depicted, including the number of genes within the cluster and the expression profiles of representative members. The gene expression profile corresponding to the medoid of each cluster is indicated by "M."

38

25

A set of rRNA biosynthesis genes are coregulated. Since the theoretical calculations indicated that the set of 65 genes were highly unlikely to have been cogrouped into two distinct clusters by chance, we investigated this subset further to ascertain whether there was a physiologically relevant basis for their apparent coregulation. To do this, we first scanned the gene set to determine whether the members have been implicated to function in common areas of metabolism. There is little known about nearly half of the 65 genes in this set; however, for those genes that have reported activities, many encode proteins that function in aspects of rRNA metabolism (Table 3). There are six known or putative RNA helicases, three subunits of either RNA polymerase I or III, and twelve other proteins that have been indicated to function in rRNA processing or ribosome biogenesis. This observation is significant, since in a random selection, less than 10% of the characterized genes would be expected to have functions related to rRNA biosynthesis. Of particular interest to us was the coinclusion of the EBP2 and RRS1 genes in this gene set. Ebp2p and Rrs1p interact, and both are essential proteins that function in the rRNA processing pathway (13, 25). Because this set of 65 coregulated genes is enriched for genes that are required for ribosome and rRNA production, we suggest that they may represent members of an RRB regulon. It is worth noting that this set does not include any of the RP genes themselves.

View larger version (21K): [in this window] [in a new window]   FIG. 2.   Gene expression levels can be assessed by a relative-quantitative RT-PCR technique. (A) RT-PCRs with primers specific to the HHT2 gene were performed on a total RNA sample derived from a logarithmically growing culture of strain yMM177. The products were labeled with [-32P]dATP, separated by gel electrophoresis, and quantified by phosphorimaging analysis. In order to determine the linear range of the assay, levels of [-32P]dATP incorporation were measured for up to 26 amplification cycles. (B) RT-PCR assays were performed simultaneously with primers sets specific to both the HHT2 and 18S rRNA sequences. Increasing amounts of nonamplifying competimer 18S primers were added to dampen the 18S amplification reactions down to a level comparable to the HHT2 reaction.

View larger version (23K): [in this window] [in a new window]   FIG. 3.   Gene expression levels change as cells progress through the cell cycle. RT-PCR assays were performed on RNA samples derived from yMM13 cells after release from alpha factor arrest. Relative gene expression levels were determined as a ratio of the expression values for the gene of interest divided by the 18S values. Expression profiles for the HHT2 (A), SWI4 (B), and the EBP2, RRS1, ROK1, NOP2, and YML093W genes (C) are indicated. Expression profiles as determined by microarray hybridization studies were converted to a similar scale and are depicted as indicated for the HHT2 and SWI4 genes (array).

View larger version (30K): [in this window] [in a new window]   FIG. 4.   RRB cluster members are coregulated following heat shock and tunicamycin treatment. RT-PCR assays were performed on RNA samples derived from yMM13 cells after a shift from 25 to 37°C (A) or after treatment with 1 µg of the drug tunicamycin/ml (B). Heat shock expression profiles of the HHT2 and EBP2 genes as determined by microarray hybridizations (11) (array) are also indicated.

RRB gene promoters share common sequence motifs. One mechanism by which the RRB cluster of genes could be regulated is through their sharing of common regulatory sequences within their promoters. To identify any such sequences, we used the MEME program (2) to scan through the 500 bp of sequence that are immediately upstream of the coding sequences for each gene member. This program identified two enriched sequence motifs, each of which was present in at least 41 of the 65 RRB cluster promoters (Fig. 5 and Table 3). Motif 1 is a 13-base G(C/A)GATGAG(A/C)TGA(G/A) consensus sequence that is found 50 times (upstream of 44 genes) within the RRB cluster promoters. The corresponding P values for each of these occurrences range from 2.3 × 10⁶ to 3 × 10⁹. Within motif 1 one can recognize the 11-base GATGAGATGAG tandem repeat sequence. There are 67 perfect matches to this 11-base sequence within the entire yeast genome, and remarkably, 21 of these sequences can be found within the 65 RRB cluster promoters (i.e., upstream of only 1% of the total number of yeast genes). Typically, motif 1 sequences are located between 50 and 150 bases upstream and on the same strand (60%) of their respective initiator codons. Motif 2 is the 11-base TGAAAA(A/T)TTTT sequence. It is present 48 times (upstream of 41 genes) within the RRB promoters, and it is usually found on the same strand as the ATG codon (67% of the time). Of the 65 RRB promoters, 29 contain both motif 1 and motif 2 sequences, with a typical arrangement of motif 2 occurring some 15 to 50 bp upstream of motif 1. 

View larger version (18K): [in this window] [in a new window]   FIG. 5.   RRB cluster promoters are enriched for two sequence motifs that cluster to positions between 50 to 200 bp upstream of their respective genes. (A) The MEME program was used to identify sequence motifs that are highly represented within the 500 bp of sequence immediately upstream of the RRB cluster genes. The consensus sequences of the two identified motifs are indicated, along with the two examples of the promoter occurrence set that correspond to the highest and lowest P values, as defined by the MEME program. (B) The relative positions of the two motif sequences are represented with respect to the initiator ATG codons of their respective genes.

Motifs 1 and 2 are important for the regulation of EBP2 gene expression in vivo Given the high representation of sequence motifs 1 and 2 within the RRB cluster gene promoters, we sought to determine whether they played a role in transcriptional regulation in vivo. Since the promoter of the EBP2 gene contains both of these consensus sequences, we chose it as a reporter gene. To test the elements, we constructed a yeast strain that would allow us to simultaneously monitor expression levels in EBP2 alleles that both contained and that lacked the consensus sequences (Fig. 6). In this way, our experiments would contain an internal control consisting of the integrated wild-type EBP2 gene and promoter. To do this, we constructed a CEN ARS LEU2 plasmid that contained a 62-codon N-terminal EBP2 deletion construct that was driven by the natural EBP2 promoter. This allele produced a truncated but functional Ebp2 protein. By choosing an appropriate single pair of oligonucleotides, we could amplify and distinguish RT-PCR fragments derived from transcripts of the normal, integrated EBP2 allele and from the truncated, plasmid-borne EBP2N62 construct (443 and 257 bp, respectively).

View larger version (25K): [in this window] [in a new window]   FIG. 6.   A plasmid-borne EBP2 construct can be used to simultaneously monitor gene expression levels in two EBP2 alleles. A plasmid-borne EBP2N62 allele was created that lacked the first 62 codons of the EBP2 gene but retained the natural EBP2 promoter (pEBP2, represented by a boldface line). The EBP2 primer set used for the RT-PCRs could simultaneously amplify fragments of different sizes (dashed lines) that were derived from both the integrated EBP2 and the plasmid-derived EBP2N62 transcripts. This plasmid was subsequently modified, replacing the EBP2 promoter sequences within motifs 1 and 2 (boxed) with heterologous sequences containing an AatII restriction site (underlined).

EBP2N62

EBP2N62

View larger version (39K): [in this window] [in a new window]   FIG. 7.   Motifs 1 and 2 are important for the regulated expression of the EBP2 gene after release from alpha factor arrest. Yeast strains containing either the EBP2N62, the p-M1 EBP2N62, or the p-M2 EBP2N62 plasmid-borne alleles were arrested with alpha factor and released into the cell cycle (A) or else subjected to heat shock (B). RNA samples were prepared for RT-PCR analysis and the amplification products derived from the 18S rRNA, the integrated EBP2, and the EBP2N62 alleles were separated and quantitated. Each expression time point value represents the mean expression level derived from at least two experiments.

-M1 EBP2N62

-M2 EBP2N62

-M2 EBP2N62

EBP2N62

View larger version (16K): [in this window] [in a new window]   FIG. 8.   The RRB and RP gene set expression profiles are similar and yet distinguishable. The average expression profiles for a set of 59 RRB and 128 RP genes are depicted as determined by microarray hybridization analysis (11) after heat shock (A) or diamide treatment (B).

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

In this study we have used computational analysis of gene expression profiles to identify a set of coregulated genes that function in various aspects of rRNA metabolism. However, because the strategy used to define this gene set relied upon both experimental data and computational assumptions, it is likely that additional RRB genes exist and that some of the listed 65 genes may not have rRNA-related functions. For example, the original expression profile data sets were compiled from single microarray hybridizations on over 6,000 genes for each time point. The inevitable experimental variations that arise within or between hybridizations could significantly affect the apparent expression profile for a given gene. Because the RRB cluster profiles were typified by early and transient gene expression changes, measurement errors in one or two early expression time points would be particularly significant. Additionally, by changing the parameters of the clustering algorithm, one would tend to either merge (for low k values) or split (for larger k values) sets of dissimilar expression profiles. The choice to divide the expression data sets into k = 24 clusters was somewhat arbitrary, and it did produce clusters that contained similar expression profiles. Therefore, rather than being a definitive membership list for the RRB regulon, the identified set of 65 coregulated genes from the overlap of a single cell cycle cluster and a single sporulation cluster is perhaps best considered as representatives of the RRB regulon.

Even though the exact membership of the RRB regulon as defined in this study is likely to be incomplete, the identification of this class of genes is useful. One prediction is that the large fraction of uncharacterized genes within the RRB cluster may also have rRNA-related functions. Likewise, it is possible that even the genes reported to have functions in other aspects of cellular metabolism could also play a role in rRNA metabolism. For example, the Cbf5 protein, originally identified as a centromere-binding factor, is now believed to be involved in rRNA pseudouridylation (16). Similarly, since translation initiation factor 6 (eIF6) has recently been shown to be involved in pre-rRNA processing (3), eIF3 could have a similar role.

One can also investigate other known ribosome and rRNA metabolism-related genes and determine whether they, too, exhibit expression profiles characteristic of the RRB cluster members. An investigation of 24 ribosome and rRNA-related genes that were not among the original 65 RRB genes revealed that several (i.e., DBP8, DIM1, NOP8, SPB1, NIP7, MTR4, SQT1, and NMD3) exhibited alpha factor release and sporulation microarray expression profiles that were similar to the RRB cluster members. Two genes (XRN1 and RSA1) exhibited profiles that were unlike the RRB members, and the remaining 14 genes (RNT1, FAL1, RRP3, DBP6, NOP4, MAK5, DRS1, DBP3, RRP5, RAT1, DBP10, SPB4, RRP7, and NOP3) had expression profiles that were somewhat similar to the RRB genes. One must be careful with these categorizations, however, because the microarray profile for any given gene may contain either widely varying or missing data points that could significantly alter the shape of the expression profile. Although these individual variations may not be as problematic when one considers multiple, large data sets covering thousands of genes, they are relevant to single gene investigations. In this regard, the RT-PCR approach can be a useful, independent method for monitoring changes in mRNA levels.

Although for the most part, we observed that the gene expression profiles as determined by RT-PCR assays were very similar to those reported in microarray studies, there was one notable exception. We consistently found that the expression profiles of the RRB cluster members peaked 10 min after release from alpha factor arrest. In contrast, the expression profiles for the same gene set, as determined by a microarray hybridization method (23), exhibited a transient decline in expression levels. This discrepancy could reflect the underlying fundamental differences in the way that the relative mRNA levels are measured in the two different methodologies, or it could be due to variations in strains or experimental conditions. Support for the latter explanation comes from the fact that 59 of the 65 RRB cluster members were identified as being part of the set of 600 genes known collectively as the ESR repression subgroup (11). This group of genes covers a range of biological pathways and exhibits expression profiles that declined transiently upon stressful environmental conditions (see below). Therefore, depending on the methods used to release cells from alpha factor arrest (i.e., sample harvesting, medium changes, etc.), one could potentially trigger a stress response. We suggest that it would be reasonable to expect cells to increase expression of rRNA metabolism related genes as they entered the cell cycle. In either case, it is significant that both the microarray and RT-PCR assays indicated that the expression profiles of the RRB cluster gene members were similar to one another.

Our results indicate that motifs 1 and 2 are strong candidates for promoting expression of members of the RRB regulon after release from alpha factor arrest. We also noted that these two motif sequences are frequently located adjacent to one another and within 200 bp upstream of their respective start codons. Sequences related to these two motifs have been described previously. The G(C/A)GATGAGAT sequence was recognized as being one of 17 positionally biased and over-represented sequences in a study designed to identify DNA-binding motifs for novel yeast transcription factors (14). That same study identified a subsequence of motif 2, TGAAAA(A/T)TTT (named RRPE), that was preferentially associated with genes involved in rRNA processing. Likewise, similar sequences were recognized as potential regulatory sequences within a group of genes that share a common response to environmental changes (5), as well as within the ESR gene set (11). The motif 1 G(C/A)GATGAG(A/C)TGA(G/A) consensus contains a tandem overlapping repeat of the GATGAG sequence. The related TG(C/A)GATGAG sequence was previously named the PAC element because it was frequently found upstream of subunits of RNA polymerase A (I) and C (III) (8). However, the specific functions of these motifs were not previously defined, nor have their presumptive binding factors been identified.

In the effort to determine whether motif 1 could stimulate gene expression within the context of a heterologous promoter, the corresponding sequences were inserted into a reporter construct bearing the lacZ gene under control of the minimal CYC1 promoter (6). lacZ expression levels were monitored by liquid -galactosidase assays (1), and it was observed that motif 1 sequences did not stimulate expression either in logarithmically growing cells or in cells released from alpha factor arrest (data not shown). While this result may not provide much insight into the function of motif 1 (since there are many reasons why motif 1 might not be active in this artificial context), it does suggest that the one-hybrid approach could potentially be used to identify the putative factor that binds to motif 1 sequences. If this putative factor could bind to motif 1 and act as a transcriptional activator on its own, one would not be able to select for such an activity from a one-hybrid fusion library.

The dual-allele relative quantitative RT-PCR strategy that we have developed should be generally useful for testing putative promoter elements for their activities under a variety of conditions. The advantage of this system is that it allows for the simultaneous assessment of expression levels for both a wild-type control and a mutated allele of a given gene at the same time. The mutated allele could be integrated instead of plasmid-borne and could contain deletions, insertions, or even novel restriction sites that would allow its RT-PCR product to be distinguished from the wild-type product. Once the appropriate alleles and strains were generated, one could test expression levels under a variety of conditions. Given the increasing potential for mapping out genetic networks based on genome-wide expression and promoter sequence analysis data (5, 10, 28, 29), it will be useful to have a reliable, internally controlled, and flexible experimental approach to test putative promoter elements for their functions in individual test genes in vivo.