INTRODUCTION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

Gene expression covers a complex series of distinctive processes. So far, studies on the regulation of gene expression in plants mainly have been focused on mechanisms that underlie the process of transcription. As a consequence, the architecture and mode of action of promoter sequences of various genes from different plant systems have been investigated extensively (for review, see Novina and Roy, 1996).

Despite the importance of transcription, it becomes more evident that posttranscriptional processes also perform a key function in the regulation of plant gene expression (for review, see Gallie, 1993; Fütterer and Hohn, 1996; Bailey-Serres, 1999). In precise terms, posttranscriptional processes comprehend all steps downstream of transcription, i.e. from pre-mRNA modification to protein turnover. In many cases, the main determinant for posttranscriptional regulation is the control of translation efficiency. In eukaryotes, control of translation efficiency often occurs at the translation initiation level by either posttranslational modification of translation initiation factors or by posttranscriptional modification of individual or sets of transcripts (for review, see Pain, 1996; Bailey-Serres, 1999; Kozak, 1999). In the latter case, structural properties of the 5'-untranslated region (UTR) of mRNA molecules often play an important role. Examples of these properties are length (Gallie et al., 2000), the presence of secondary structures (Klaff et al., 1996; Gallie et al., 2000) or upstream open reading frames (Lukaszewicz et al., 1998; Wang and Wessler, 1998), and the composition of the sequence that surrounds the translation initiation codon (Geballe and Morris, 1994; Joshi et al., 1997). In addition, the presence of specific sequence elements that serve as interaction sites for antisense RNAs (Shayig, 1997; Hu et al., 1999) or RNA-binding proteins (for review, see Burd and Dreyfuss, 1994; Albà and Pagès, 1998) can also contribute to the regulatory capacity of 5'-UTRs.

To identify putative regulatory sequence elements in the 5'-UTR of coregulated genes, we focused on genes that are highly expressed during the development and germination of the male gametophyte (pollen). During pollen development, a large number of genes are transcribed (Willing and Mascarenhas, 1984; Willing et al., 1988; Guyon et al., 2000; F. Cnudde, unpublished data), which leads to the appearance of distinctive sets of transcripts (Stinson et al., 1987; Schrauwen et al., 1990; Hulzink, 2002). These mRNA sets can be found in pollen from both mono- and dicotyledonous plant species, which argues for a conservation of genetic programs that underlie pollen gene expression. The 5'-UTRs of several pollen transcripts have been shown to alter gene expression at the transcriptional (Curie and McCormick, 1997) or posttranscriptional level (Bate et al., 1996; Hulzink, 2002; Hulzink et al., 2002). With regard to the evolutionary conservation of genetic programs in pollen and the important role of the 5'-UTR in pollen gene expression, we hypothesize that the 5'-UTRs of pollen-expressed genes share regulatory sequence elements.

To identify these shared (overrepresented) regulatory sequence elements in the 5'-UTRs of pollen-expressed genes, a statistical analysis has been carried out. Two different sequence sets were collected: a test set containing 5'-UTR sequences of pollen-expressed genes (pollen sequences) and a reference set containing 5'-UTR sequences of genes that have been isolated from sporophytic tissues (reference sequences). Both sequence sets were used to identify overrepresented sequence elements (oligonucleotides) in the pollen sequences (oligo-analysis; Van Helden et al., 1998, 2000b). Although genetic programs in pollen are conserved in different plant species, it may well be that the presence of several sequence elements are associated to genes that originate from specific plant species or from subsets of plants that share similar taxonomic classifications or morphological features. Hyper-geometric statistics were applied to investigate whether the presence of the pollen elements was associated to genes from specific plant species or from sets of plants that are distinctive in the number of cotyledons (monocots or dicots), stigma type (wet or dry), or pollen type (bicellular or tricellular).

	RESULTS

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

The 5'-UTR of Pollen-Expressed Genes Shares Several Sequence Elements

To investigate whether the 5'-UTRs of pollen-expressed genes share sequence elements (overrepresented oligonucleotides), two datasets were collected containing 5'-UTR sequences of either pollen-expressed genes (Table I, pollen sequences) or genes that have been isolated from sporophytic tissues (reference sequences). The background oligonucleotide frequencies were estimated by calculation of the relative frequencies of all oligonucleotides within the reference set. Oligonucleotide occurrences were counted in the pollen sequences, and their statistical significance was estimated on the basis of the background frequencies (for a description of the followed methodology, see "Materials and Methods").

Table II shows the hexanucleotides that are significantly overrepresented in the pollen sequences compared with the reference sequences. From the 4,096 possible hexanucleotides, 31 sequence elements are preferentially present in the 5'-UTRs of pollen-expressed genes (Table II). Similar results were obtained for penta-, hepta-, and octanucleotides (for these data, see http://rsat.ulb.ac.be/rsat/). The majority of the overrepresented oligonucleotides (pollen elements) are A rich, i.e. more than 80% of the oligonucleotides contain four or more A residues. The most significant overrepresented oligonucleotide is AAAAAA, which is 74 times present (O_occ) when 20.4 occurrences are expectable on the basis of the background model (E_occ). When the total number of possible oligonucleotides in each dataset is taken into account (4,096), the occurrence probability value converts into the occurrence significance index (sig_occ). For the oligonucleotide AAAAAA, the sig_occ is 15.9, which means that a hexanucleotide with a similar high-significance value is expected to occur in 10^15.9 datasets of random sequences. Highly significant values are also found for several AAAAAA single-substitution variants. To examine whether the overrepresented oligonucleotides are positional biased in the pollen sequences, the distribution of each hexanucleotide was determined, and a test of homogeneity was applied (Table III). None of the hexanucleotides passed the significance threshold of 64.64, which indicates that the overrepresented sequence elements are not positional biased in the pollen 5'-UTRs.

View this table: [in this window] [in a new window]   Table III.   Position analysis of the pollen 5'-UTR sequence elements Analysis of the distribution of each overrepresented sequence element was performed with the program "position-analysis" (Van Helden et al., 2000b). Chi-square values ( 2) were calculated using the parameters df = 30 and  = 0.000244. Sequence elements with  2  64.64 do not exhibit a significant biased position in the pollen sequences. For details, see "Materials and Methods."

In summary, these data clearly show that several sequence elements are preferentially present in the 5'-UTR of pollen-expressed genes. Within the pollen 5'-UTRs, the overrepresented sequence elements are not positional biased.

Assembly of Sequence-Related Pollen Elements Gives Rise to Several Consensus 5'-UTR Elements

Several of the identified pollen elements share sequence similarity, and their mutual overlap might reveal consensus sequence elements. To identify these consensus elements, we assembled the sequence-related pollen elements using the program "pattern-assembly" (Van Helden et al., 2000a). As shown in Table IV, assembly of sequence-related pollen elements gave rise to several consensus 5'-UTR sequence elements. Highly significant values are found for the consensus elements CAAATAAAAAT and AAAAAA.

Several Pollen Elements Are Associated with Genes from Arabidopsis, Brassica napus, Dicotyledonous Plants, or Plants Containing a Wet Stigma or Bicellular Pollen

The pollen-expressed genes that were used for the 5'-UTR analysis are derived from 25 different plant species (Table I). With regard to their taxonomic classification, the plants were separated in subsets containing mono- or dicotyledonous species. On the basis of their stigma type, the plant species were grouped further into the subsets wet and dry stigma. A wet stigma is covered with a liquid secretion layer, whereas a dry stigma is covered with less or no secretion material (Heslop-Harrison and Shivanna, 1977). Furthermore, the plant species were grouped into the subsets bi- and tricellular pollen. Sperm cells in tricellular pollen are formed during pollen maturation, whereas sperm cells in bicellular pollen are arranged during pollen tube growth (Brewbaker, 1967). On the basis of the hyper-geometric probability, we tested to what extent the pollen elements were preferentially associated to genes from each subset of plant species. Furthermore, we analyzed to what extent the pollen elements were associated to genes from specific plant species. For each sequence element, the number of matching pollen sequences in a given subset (B) was compared with the number of matching sequences in a given set (M), and the corresponding E value was calculated (for a detailed description of the methodology, see "Materials and Methods"). The analysis revealed the presence of 12 associations with an E value < 0.1 (Table V). From the 31 overrepresented sequence elements, seven are significantly associated to dicotyledonous plant species. The most significant example is the sequence element AAAAAT, which is present in 32 dicotyledonous pollen sequences. The pollen element ATCAAA is significantly associated to genes from both wet stigma and bicellular pollen plants. Interestingly, the sequence elements AAAAAA and AAAAAT are significantly associated to B. napus, whereas AAGAAG is associated to Arabidopsis.

	DISCUSSION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

A systematic approach based on the statistical analysis of oligonucleotide occurrences has led to the identification of several oligonucleotides (sequence elements) that are significantly overrepresented in the 5'-UTR of pollen-expressed genes (Table II). It is obvious that the choice of appropriate reference and test datasets is an important determinant for a reliable outcome of the analysis. Genes from both datasets were selected on the basis of the origin of their respective cDNAs (male gametophytic or sporophytic tissues) without a priori consideration of the composition of the 5'-UTRs. The extent of expression of pollen genes during pollen development and tube growth was ascertained by data from the available literature. From the analysis, we conclude that the 5'-UTRs of genes that are expressed during male gametogenesis share pollen-specific sequence elements (pollen elements). Statistical analyses of the presence of overrepresented sequence elements have led to a successful identification of regulatory elements in promoter (Van Helden et al., 1998, 2000c; Sinha and Tompa, 2002) and 3'-UTR (Jacobs-Anderson and Parker, 2000; Van Helden et al., 2000b) sequences of coregulated yeast (Saccharomyces cerevisiae) genes. To our knowledge, the present study is the first report that describes the in silico identification of sequence elements that are shared in the 5'-UTRs of coregulated plant genes.

Although many sequence elements are significantly overrepresented in the 5'-UTR of pollen-expressed genes, their statistical significance does not necessarily imply that they are functional in the regulation of pollen gene expression. However, several observations indicate that some of the pollen elements exhibit a pollen-related regulatory function. Figure 1 shows a schematic representation of the 5'-UTR of the pollen-expressed gene ntp303. A sequence region at the 5' end of the ntp303 5'-UTR has been shown previously to affect translation efficiency, whereas a sequence region at the 3' end was found to modulate mRNA stability (Hulzink, 2002; Hulzink et al., 2002). As shown in Figure 1, assembly of several of the pollen elements in the ntp303 5'-UTR leads to several extended sequence regions. We assume that some of these extended pollen sequence regions comprise regulatory elements because several of these regions are also present in the functional ntp303 5'-UTR regions. In addition, pattern assembly analysis (Table IV) highlighted the presence of consensus sequence elements that are conserved in various pollen-expressed genes. Because some of these consensus pollen elements are also localized in the functional ntp303 5'-UTR regions, it is assumable that at least the consensus sequence elements in the functional ntp303 5'-UTR regions resemble regulatory sequences. Regulatory sequence elements are often concentrated as short and highly conserved core elements (for review, see Novina and Roy, 1996). A representative example of such a regulatory sequence element is the consensus sequence AAGAAG, which is repetitive present in the 5'-functional region of the ntp303 5'-UTR. Deletion analysis strongly suggests that the AAGAAG repeat is involved in directing pollen gene expression (Hulzink et al., 2002). In this respect, the identification of the AAGAAG sequence as a pollen-specific 5'-UTR element provides additional clues for its regulatory function.

View larger version (14K): [in this window] [in a new window]   Figure 1.   Schematic representation of the distribution of overrepresented sequence elements (pollen elements) in the 5'-UTR of the pollen-expressed gene ntp303. Distribution of the pollen elements is presented above the graphic representation of the ntp303 5'-UTR. The small arrows indicate the start of a pollen element. The numbers correspond to sequence elements as presented in Table II (counted from up to down). Sequences below the ntp303 5'-UTR illustration indicate the position of consensus pollen sequence elements as presented in Table IV. The gray regions represent the 5' (left) and 3' (right) regulatory regions of the ntp303 5'-UTR (see text for description). The arrow at the 3'-end of the ntp303 5'-UTR indicates the position of the translation initiation site.

The pollen 5'-UTR sequences that were used in the present study originate from different plant species. Several pollen elements are preferentially associated to genes from dicotyledonous plant species, whereas none of the elements exhibit a significant association to genes from monocots (Table V). These results indicate co-evolution of sequence elements in the 5'-UTRs of pollen-expressed genes. It is plausible that the significant association of pollen 5'-UTR elements is reflected by specific properties of dicots, which are absent in monocots. In addition, the pollen element ATCAAA is found to be preferentially associated to genes from plants containing a wet stigma and bicellular pollen. With regard to self-incompatibility systems in angiosperms, a clear relationship exists between pollen and stigma characteristics, i.e. plant species with a wet-type stigma often have bicellular pollen (Brewbaker, 1967; Heslop-Harrison and Shivanna, 1977). Such a relationship might explain the preferential association of the ATCAAA element to genes from wet-type stigma and bicellular pollen plants. Besides the pollen elements that are preferentially associated to subsets of plant species, the presence of several other 5'-UTR sequence elements are not related to the number of cotyledons, stigma type, or pollen type. It is assumable that these elements play a more general role in the regulation of pollen gene expression, independent of the phylogenetic background. On the contrary, the preferential association of three other sequence elements to pollen-expressed genes from Arabidopsis and B. napus indicates a strong phylogenetic dependency of these elements for these Brassicaceae spp.

It is obvious that in silico identification of conserved sequence elements in the 5'-UTRs of coregulated genes has to be validated experimentally. Nevertheless, computational identification of shared 5'-UTR sequence elements provides useful indications for new functional studies. Although gene expression studies in pollen have been prosperous in several ways, a systematic analysis of the 5'-UTRs of the growing number of isolated pollen-expressed genes was still lacking. With the increasing number of publicly available genomic and EST sequences, we will be able to gain a better understanding of the new intriguing functional and evolutionary clues provided by our analysis.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

5'-UTR Sequence Datasets

The 5'-UTR sequences of pollen-expressed genes (pollen sequences) were collected from the GenBank database. The pollen 5'-UTR dataset consists of 5'-UTR sequences from 132 different genes that are highly expressed in mature pollen or pollen tubes. Expression of the genes in pollen or pollen tubes was ascertained by data from the available literature. Because of their expression in anther tissues, genes related to pollen coat proteins have been excluded from the analysis. The total number of nucleotides in the pollen 5'-UTR dataset is 16,645. The average length of the full-length or partial pollen 5'-UTRs is 126 nucleotides; the smallest UTR sequence consists of six nucleotides, whereas the longest UTR is 620 nucleotides in length.

The detection of overrepresented oligonucleotides (sequence elements) in the pollen sequences relies on the prior definition of a background model, which is used to estimate the random expectation for each oligonucleotide. As a background model, we used a set of non-pollen 5'-UTR sequences (named reference sequences). The reference sequences were obtained from the GenBank database (March 2001) by selecting the first 1,076 cDNA entries that did not contain any of the key words "pollen," "gametophyte," "flower," and "bud." With regard to the extraction of the reference sequences, the complete sequence upstream of the translation initiation site was selected. This procedure resulted in the collection of 113,481 nucleotides. The average length of the full-length or partial reference 5'-UTRs is 105 nucleotides; the smallest UTR sequence is 14 nucleotides in length, whereas the longest UTR consists of 1,396 nucleotides.

Sequence Purging

The presence of homologous 5'-UTR sequences in the datasets might lead to the inclusion of large conserved sequence regions. These conserved sequence regions can bias the analysis by duplicating all the oligonucleotides that are found in the redundant sequences. To avoid this phenomenon, the pollen and reference sequences were purged to remove sequence repeats larger than 50 nucleotides (containing a maximum of three mismatches) using the programs "mkvtree" and "vmatch" (Kurtz and Schleiermacher, 1999).

Oligonucleotide Analysis

A pattern discovery approach (oligo-analysis; Van Helden et al., 1998, 2000b) was used to detect overrepresented oligonucleotides in the pollen sequences. All statistics were performed on nondegenerated oligonucleotides, without the acceptation of mismatches. To calculate the expected frequency of each oligonucleotide, oligo-analysis was first applied to the reference sequence set. The obtained expected frequency values were used to estimate the expected number of occurrences for each oligonucleotide in the set of pollen sequences. The detection of overrepresented oligonucleotides is based on an estimation of the significance of the observed occurrences (O_occ). For each oligonucleotide, the P value P_occ = P(X  x) was calculated on the basis of the binomial distribution. Because the analysis comprised multiple tests (4,096 in the case of hexanucleotides), the possibility existed that even low P values appeared by chance. To correct for such a multitesting effect, the P values were multiplied by the number of oligonucleotides. This correction resulted in an expected value, the E value (E_occ). The significance index sig_occ = log(E_occ) reflects the degree of overrepresentation of each oligonucleotide on a logarithmic scale. All oligonucleotides with a positive significance index (E_occ  1) were selected. To prevent a bias due to self-overlapping occurrences (Kleffe and Borodovsky, 1992), a nonoverlapping counting mode was adopted. This means that when a self-overlapping hexanucleotide was found at a given position, its occurrence at the next five positions was ignored. The number of possible positions was corrected according to the calculation of the binomial.

Overrepresentation of an oligonucleotide can be due to either its frequent presence in most of the pollen sequences or in a subset of these sequences. The probability of the first occurrence was taken into account by calculating an additional index of overrepresentation. For each oligonucleotide, the number of sequences that contained at least one occurrence (matching sequence) was determined, and the statistical significance was calculated with the binomial probability. For a single sequence, the first occurrence probability was calculated as:
where p_W represents the probability to observe the oligonucleotide W at any position of the sequence, X_S represents the number of matches on sequence S, and T_S represents the number of positions that was calculated from the sequence length L_S and the oligonucleotide size w(T_S = L_S  w + 1). The matching sequence probability P_ms was calculated by taking the right tail of the binomial probability:
where m represents the number of sequences that contain at least one copy of the oligonucleotide, whereas S represents the total number of sequences. The E value (E_ms) and the significance index (sig_ms) were calculated from the P value in the same way as for the occurrences. A positive sig_ms value indicates that the oligonucleotide is present in more sequences than what would be expected by chance alone.

Using the 1,076 non-pollen gene sequences as reference, a problem of statistical sampling was observed. Some oligonucleotides were present in the pollen sequences and not in the reference set. As a consequence, their probability was estimated to be zero and their respective significance was infinite, even if they were found in a low copy number in the pollen sequences. The problem occurred because the reference sequence set was too small to reflect all possibilities. It is likely that these oligonucleotides will appear in much larger reference datasets. To circumvent this problem for the current reference sequence set, pseudo-weights were used. This means that the oligonucleotide frequencies that were calculated from the reference sequence set contributed for 90% to the estimation of prior oligonucleotide probabilities, whereas the remaining 10% were left for the potential presence of additional oligonucleotides in the pollen sequences.

Analysis of oligonucleotide distributions in the pollen sequences was performed with the program "position-analysis" (Van Helden et al., 2000b). Occurrences were regrouped by intervals of 20 nucleotides. Given the sequence sizes, the positional distributions contained 31 classes with a class interval of 20 nucleotides. The expected distribution was calculated according to a homogeneous model, i.e. a position-independent distribution for each oligonucleotide. Observed and expected positional distributions were compared using the chi-square statistics. To take the number of oligonucleotides (n = 4096) into account, the threshold was adapted according to the Bonferoni rule, which recommends a first error risk  < 1/n. The degrees of freedom (df) of the chi-square test depended on the number of position classes (c), which in turn depended on the class interval. For the class interval of 20 nucleotides, the resulting probability value ( = 0.000244) corresponded to a theoretical value of X_theor = 64.64.

Association Analysis

To examine to what extent the oligonucleotides were associated to specific plant species or to subsets of plants containing one or two cotyledons, a wet- or dry-type stigma, or bi- or tricellular pollen, the hyper-geometric probability test was applied. The hyper-geometric probability estimates the significance of association between the overrepresented oligonucleotides and different subsets of pollen sequences. Considering a set of n sequences (e.g. the 132 pollen sequences) that contains a subset of size S (e.g. the 99 sequences from dicotyledonous plants) and the observation that a given oligonucleotide is present in M pollen sequences, the probability that exactly B of these pollen sequences belongs to the given subset is:
The P value is the probability to observe at least B matching sequences in the given subset:
In our analysis, each oligonucleotide was compared with T = 31 different subsets of pollen sequences (dicots/monocot, wet stigma/dry stigma, bicellular pollen/tricellular pollen, and 25 different plant species). To correct for this multitesting, the P value was converted to an E value by:
Finally, the E value was converted to a logarithmic significance index:

Availability

The "Regulatory Sequence Analysis Tools" are available at http://rsat. ulb.ac.be/rsat/. The complete sets of data and calculation procedures are available on the same site.

Distribution of Materials

Upon request, all novel materials described in this publication will be made available in a timely manner for noncommercial research purposes, subject to the requisite permission from any third party owners of all or parts of the material. Obtaining any permission will be the responsibility of the requestor.