A total of 3000 clones from three cDNA libraries were single-pass sequenced from their 5' ends. Sequence cleaning processes as outlined in the Materials and Methods yielded 2735 ESTs. These sequences had an average trimmed length of 519 base pairs and were assembled into 348 clusters and 1332 singlets, resulting in the final annotation of 1686 unigenes. Clusters ranged in membership from 67 ESTs (one) to two ESTs (253). The redundancy levels were 15.0%, 20.3% and 62.8% within the normalized, standard and subtracted libraries, respectively. A high redundancy level of cDNA libraries constructed using a similar subtraction method from Thellungiella plants under abiotic stress conditions has also been observed previously 19.

The translated products of the 1686 unigenes were searched against the non-redundant protein database provided by GenBank 20 to putatively assign their functions. At the time of writing, 918 (54.4%) unigenes showed significant similarity (E value cut-off of 10^-5) to genes of known or putative function, whereas 549 (32.6%) ESTs were assigned to transcripts with unknown function, which includes hypothetical genes predicted in genomes of model organisms (Table 1). The remaining 219 ESTs (13.0%) had no homologues in the public protein sequence database. A further BLASTN search against GenBank EST collections revealed that 62 of the 219 ESTs were likely to be novel sequences.

More than 50% of the SAM unigenes could be assigned as genes with a known or putative function based on sequence similarity. However, the lack of information on the encoded products meant that many of these transcripts could not be functionally categorized according to the Gene Ontology Consortium (GO). Using BLAST2GO 21, we successfully classified 593 and 557 unigenes in terms of GO molecular functions (Figure 1A) and biological processes (Figure 1B), respectively. A single gene product might be assigned to more than one GO term, and hence the total number of GO mappings in each of the ontologies exceeded the number of ESTs.

The successfully classified unigenes cover a broad range of GO functional categories (Figure 1). Under a molecular-function classification, most of the genes (53%) were assigned to the "binding" class (Figure 1A). This class includes sequences with putative involvement in mainly nucleic acid binding, a substantial number of which are predicted to encode transcription factors that are known to be essential to the regulation of plant development. Unigenes predicted to encode histone subunits and histone-modification proteins, chromatin remodelling factors and DNA methyltransferases represent another group of sequences linked to nucleic acid binding.

To investigate the different types of transcription-factor families represented by our EST libraries, a search using the best-matching Arabidopsis locus for SAM unigenes (based on a BLASTX search against the TAIR Arabidopsis protein database) was performed at the Arabidopsis Gene Regulatory Information Server 22. There are 50 families of transcription factors currently listed in the database, 19 of these are represented by the SAM unigenes (Table 2). Six members of the family of homeobox transcription factors that play key roles in the regulation of development may represent interesting candidate genes for further studies.

In the biological-processes category, 40% and 35% of the unigenes were involved with cellular physiological processes and metabolism, respectively (Figure 1B). The former contains gene products that play an important role in cell organization and biogenesis, while the latter has sequences related mainly to protein metabolism. The sequences relevant to protein metabolism ranged from those associated with protein biosynthesis, such as different ribosomal subunits, to those that modify and degrade proteins, including various sequences involved in the ubiquitin-proteasome pathway.

An oligonucleotide microarray has been developed using our EST resource and about 500 pea sequences randomly retrieved from the GenBank pea protein database, which contains approximately 2000 entries (see Materials and Methods). We utilized this array to compare the transcriptional profiles of the SAM and NM. Four independent replications of balanced-block-design dual-label experiments were performed [Materials and methods, 23] and the resulting data were depicted in Figure 2.

Differentially expressed genes that were detected using LimmaGUI 24 (at an adjusted probability value of < 0.05) were subjected to further selection based on the relative change, with up- and down-regulated transcripts defined using cut-offs of greater than 1.3-fold or less than 0.7-fold, respectively. Based on these criteria, we identified 184 and 175 transcripts that were significantly up- or down-regulated in the pea SAM relative to the NM. These transcripts were annotated based on the best BLASTX match, and exhibited changes relative to the NM of 0.1- to 10.7-fold (see Additional File 1 &2). The change of 0.1-fold was for a gene encoding a type 1 metallothionein (AB176564), while that of 10.7-fold was for a sequence annotated as vegetative lectin (AAA33691). A study of similar genes encoding type 1 metallothionein in cotton revealed their abundant (although not exclusive) expression in the root 25. Meanwhile, the high expression of a sequence encoding vegetative lectin in the pea apex has been observed 26. These and other studies listed in Table 3 provide independent verification of our microarray data. We also performed RT-PCR analysis on five selected transcripts and as shown in Figure 3, the outcome is generally in good agreement with the microarray data.

Further comparison of the functional categories identified to be differentially regulated in maize data 25 and this study (Figure 4) revealed that transcripts associated with the categories of transcription, chromatin and cell division are highly represented in the list of up-regulated genes, while sequences categorized as being related to metabolism, stress and photosynthesis are down-regulated in the SAM. However, our data also highlight differences between the transcriptional repertoires of the monocot SAM [maize, 27] and the dicot SAM (pea, present study). In particular, the high retrotransposon-related transcriptional activity reported in maize 27 does not appear to be a conserved feature of the SAM, since it was not present in the pea SAM ESTs.

Since the SAM we investigated consists of distinct functional zones, the averaging of signals across the whole SAM probably attenuated signals associated with any given stem-cell region. However, our experiment showed that a sequence annotated as encoding PROTODERMAL FACTOR 1 was up-regulated 1.8-fold in the SAM relative to the NM (Table 3). An Arabidopsis counterpart gene has been found to be exclusively expressed in the outermost (L1) layer of the SAM 28. This indicates the ability to identify transcripts that are specifically expressed in certain domains of the SAM, though the magnitude of the relative change might have been reduced and low abundance transcript could have been missed detection.

A closer inspection of the differentially expressed transcripts revealed that about 60% of the identified down-regulated sequences were derived from GenBank (Table 4). In contrast, only 12% of the genes with higher expression in SAMs were retrieved from GenBank, while the other up-regulated ESTs were from clones derived from our library collection. This is not surprising since the sequences from GenBank were generated primarily from tissues other than the SAM, whereas our EST collection was derived from dissected SAMs. This indicates the significance of constructing a library from specialized tissues and further suggests the utility of our EST collection as a valuable resource in studying the molecular processes underlying the functions of the plant meristem. Meanwhile, 30% of the differentially expressed genes detected were not identified, consisting of genes annotated as hypothetical or expressed protein as well as sequences that have no BLASTX matches in the public database. This list of unknowns could represent intriguing candidate genes for functional analysis.

Putative functions could be assigned to 116 of the up-regulated transcripts based on protein sequence similarity. Manual inspection of the corresponding transcripts revealed a high representation of ESTs predicted to encode proteins associated with cell division and proliferation (Table 4; Figure 4).

Table 4 indicates that ESTs encoding all five subtypes of histones, minichromosome-maintenance proteins and cell-proliferating nuclear antigens were among the transcripts whose expression was higher in the SAM than in the NM. The expression of these genes is known to be associated with DNA synthesis and cell proliferation, and they are thus abundant in the meristematic tissue 2930 since this region consists of actively dividing cells.

In the same category, there is a transcript predicted to encode HIGH MOBILITY GROUP (HMG) protein. HMG are proteins that are known to play an architecture role in modifying DNA conformation to facilitate the assembly of multiprotein-DNA complexes. They may serve only to maintain physical orders but the involvement of these proteins in the network regulating SAM activity seems plausible. This is in view of recent studies that demonstrate the binding of HMG proteins to functionally important regions of plant gene promoter and stimulate transcriptions 31.

Other up-regulated sequences included transcripts predicted to encode cyclin D (AB008188) and cyclin B (EX570164), which are involved in the progression of the cell cycle. D-cyclins are one of the main rate-limiting factors for cell proliferation, and several of them may play key roles in the association between the cell cycle and meristem function, in particular primordia formation [reviewed in 32].

Increases in transcript abundance were also found for genes predicted to encode histone-modification protein (EX569897), chromatin remodelling factors (EX570366, EX570306 and DQ026703) and DNA methyltransferases (EX568764 and AF034419; Table 4). These proteins are known to play a role in the epigenetic regulation of gene expression by participating in mechanisms that alter chromatin structure so as to activate or repress particular sets of genes [reviewed in 33].

There is an emerging recognition of the significance of the chromatin remodelling process in regulating the activity of plant stem cells [reviewed in 34]. For instance, mutation of the Arabidopsis FASCIATA1 (FAS1) and FASCIATA2 (FAS2) genes that encode subunits of the chromatin assembly factor leads to dysfunction of the SAM 35. This was associated with the down-regulation of WUS gene expression in both mutants, suggesting that regulation of meristem function and organogenesis by chromatin remodelling factors is primarily achieved through regulation of the expression of the homeobox transcription factor 35. However, direct links between chromatin remodelling factors and the regulation of the expression of key meristem genes remain to be established. Nonetheless, the up-regulation of the expression of these sequences in the SAM suggests that the mechanism of epigenetic regulation is important to maintaining the identity of stem cells in plants, as has been reported in animals 36.

We also identified a few putative transcription factors among the genes whose expression was higher in the SAM than in the NM. These included an EST (EX570025) annotated as a putative plant-specific transcription factor from a YABBY-family protein – members of this family are reportedly involved in the abaxial cell-fate specification in lateral organs of Arabidopsis 37. Also on the list was a sequence predicted to encode AUXIN RESPONSE FACTOR 9 (ARF9; EX570238). Similar members of the auxin response factor group of proteins are known to regulate auxin-mediated transcript activation or repression. For example, the expression of several genes, such as those encoding members of LATERAL ORGAN BOUNDARIES domain proteins and AUXIN-REGULATED GENE INVOLVED IN ORGAN SIZE, are disrupted in the double mutant of ARF7 and ARF19, implicating their roles in auxin-mediated plant development 38.

Other gene products that might be related to auxin include a putative auxin-efflux carrier (PIN1), auxin-conjugating protein (DWARF IN LIGHT1) and a ribosomal protein L24B (Additional File 1, Table 4). Similar genes in Arabidopsis have been implicated in auxin-mediated developmental regulation 394041. For example, DWARF IN LIGHT1 is involved in auxin signal transduction, and inhibits shoot and hypocotyl-cell elongation 39. The presence of several auxin-related transcripts on our list reflects the well-established roles of auxin in organ initiation and positioning at the meristem 4243.

Intriguingly, miRNAs that are endogenous 21-nucleotide riboregulators have been shown to target several mRNAs implicated in auxin responses, including DWARF IN LIGHT1 44. There is increasing amount of evidence indicating that miRNA-mediated repression plays an important role in the spatial expression of plant cell-fate regulatory genes [e.g. 45]. A protein called ARGONAUTE is known to function as a catalytic component of the RNA-induced silencing complex, which targets mRNA for degradation using miRNA as a guide 46. Nevertheless, the precise role for the up-regulation of a transcript annotated as encoding a similar ARGONAUTE in our dataset (Table 4) awaits further study.

Many of the down-regulated transcripts were potentially associated with biotic and abiotic stress responses (Table 5), including transcripts predicted to encode dehydrin-related protein (AY065655), pathogenesis-related protein (AJ586324), disease-resistance-response protein (AF139018), antimicrobial defensin (AF525685) and chitinase (AB037832). This might be attributable to the stems, leaves and roots generally being exposed to greater biotic or abiotic stress than the well-shielded SAM, with the former therefore requiring the constitutive presence of these gene products at a higher level than in the SAM in order to maintain successful defence responses. However, we found at least one other sequence (EX571142) potentially related to stress responses whose expression was higher in the SAM (Table 4). Although the molecular basis for this is unknown, it is possible that the corresponding encoded product plays dual roles in both stress responses and development. This is supported by a recent study finding a network of rice genes associated with stress responses and seed development 47.

Surprisingly, the down-regulated transcripts included various sequences encoding proteins that scavenge or generate ROS, such as thioredoxin (AJ319808), catalase (EX568770), Cu-Zn superoxide dismutase (AB189165), copper amine oxidase (AB026253) and peroxidase (AB087838). This implies that the level of ROS is lower in the SAM than in the NM, and hence the expression of genes encoding ROS scavengers is lower in the former. The absence of photosynthesis (a source of ROS) in the SAM might explain the lower ROS level therein. However, there is increasing evidence that plants use ROS as signalling molecules for regulating development and various physiological responses, and for mediating abscisic-acid-induced stomatal closure, as well as in auxin signalling and gravitropism in roots 48. Whether this indicates that the regulatory role of ROS is less prominent in the SAM than in the NM awaits further investigation.

Several of the genes more abundant in the NM than in the SAM were predicted to be related to photosynthesis, including subunits of photosystem I and II and chlorophyll-a/b-binding protein. It is well known that meristematic cells do not contain differentiated plastids, which may explain the lower expression level of genes associated with photosynthesis in the SAM.

Some of the sequences predicted to encode products that play roles in cellulose synthesis, such as cellulose synthase (EX569643), pectin methylesterase (AF056493), xyloglucan endotransglycosylase (AB042531), were found to be down-regulated in the SAM. This probably reflects the cell-wall structure in the stem and leaf being more complex that the thin primary cell wall in meristematic cells 4.

Garden pea (Pisum sativum) cultivar Torsdag was grown in a greenhouse located at the University of Melbourne, Australia. SAMs were micro-dissected from 10-day-old peas under the dissecting microscope at 40× magnification. Any leaf primordia were excluded in order to create a meristem-enriched tissue collection and the location of tissue sample is indicated in Figure 5. Dissected samples were quickly frozen in liquid nitrogen and stored at -80°C until used for RNA extraction and cDNA library synthesis 49. For the subtracted library, the driver sequences were derived from an equal mix of RNA extracted from non-meristematic (NM) tissue consisted of primary stem (without axillary meristems), mature leaf lamina, primary roots (without root apical meristem and root hairs), whereas the tester sequences consisted of RNA harvested from dissected SAM. All cDNAs were cloned into pBlueScriptIISK+ plasmid vector.

A total of about 500, 1000 and 1500 randomly picked clones from each of the standard (C), subtracted (S) and normalized (N) libraries were sequenced at Australian Genome Research Facility (AGRF), Australia and subsequently at Macrogen Korea using T7 primer. These sequences have been deposited in GenBank under the accession numbers EX568682 to EX571416.

Sequence data were trimmed off vector, adaptor and low quality sequences using SEQTools 50. Trimmed sequences that were shorter than 100 basepairs were excluded from further analysis. Blast score-based clustering method with a score cut-off of 0.6 from SEQTools was then used to assemble the sequences. All clusters and singletons resulting from this automated clustering were considered to be the best estimation of a minimal gene set for our EST library and we have called this set as "unigenes".

All sequences were then imported into Blast2GO, a web-based Gene Ontology (GO) annotation and analysis tool 21 for subsequent analysis. This involved automated retrieval of GO terms associated to the hits obtained after a BLASTX search of the corresponding unigene sequence against NCBI (nr) protein database. The e-value cut off was set at 1e-5.

Total RNA was extracted from dissected SAM (approximately 80 SAMs per extraction) or other plant parts (primary stem, primary roots and mature leaves) using Qiagen RNeasy Mini Kit. Four independent tissue collections and RNA extractions (designated A, B, C and D) were performed for each of the microarray hybridization experiment.

The Combimatrix arrays are semiconductor-based oligonucleotide microarrays and are generated based on CombiMatrix technology (hyperlink) of in situ synthesis 51. The CustomArray™ 4x2k is a microarray that is divided into 4 sectors, each of which can contain up to 2,240 different oligonucleotide probes (spots) and can be hybridized individually with different targets using a provided sectored hybridization cap. A total of 1686 pea sequences (290 sequences from C library, 300 sequences from S library and 1086 ESTs from the N library) together with 500 pea sequences randomly selected from GenBank pea protein database were submitted for probe design using the open source CombiMatrix probe design system. The length of probes ranged from 35–40 bases in length. A variety of control elements were also arrayed on the slide and these include blank spot, housekeeping genes (actin) as well as non-plant transgenes.

Target preparation and hybridization were performed in Australian Genome Research Facility Ltd (AGRF) according to the standard CombiMatrix protocol described in detail at http://www.combimatrix.com/docs/PTL005_00_4x2K_%20Hyb_%20Imaging.pdf. One microgram of total RNA from SAM or NM was labelled using the Kreatech's ULS™ RNA ampULSe kit to generate Cy5 or Cy3 labelled targets. The Cy5- or Cy3-labelled cDNA was then hybridized to different sector of the chip according to a balanced block design dual label experiment scheme 23:

Sector 1: Cy3-SAM A vs Cy5-NM A

Sector 2: Cy5-SAM B vs Cy3-NM B

Sector 3: Cy3-SAM C vs Cy5-NM C

Sector 4: Cy5-SAM D vs Cy3-NM D

The Cy5- and Cy3-hybridized chip was then scanned using Genepix 4000B microarray scanner (Axon Instruments, CA, USA) according to manufacturer's instructions. The generated tiff image files were then imported into Combimatrix Microarray Imager to produce intensity data. The LimmaGUI software, which is an implementation of the Empirical Bayes linear modelling approach, was used for subsequent statistical analysis of the resulting data 24. A robust spline method was chosen for within array normalization and a least-square linear model fit was computed with the p-value adjusted using the Benjamini-Hochberg procedure. The entire differentially-expressed transcripts (p value < 0.05) in the SAM are listed in Additional File 1 and Additional File 2. Microarray data have been deposited in the Gene Expression Omnibus database 20 under accession number GSE9278.

The one-tube, two enzyme Access RT-PCR system (Promega, Annandale, New South Wales, Australia) was used according to manufacturer's instructions in all RT-PCR analysis. Ten ng of RNA isolated from the SAM, mature stem, mature leaf and primary root of 10-day-old pea seedlings were used as a template in a 10 μl reaction volume. The pea actin gene was used as an internal control. The number of cycles used for the transcripts investigated was routinely between 25–28 and 80 % of the PCR reaction was separated on 1% agarose gel containing 0.1 μg/μl ethidium bromide and visualized under UV light.