Construction and screening of subtracted cDNA libraries from limited populations of plant cells: a comparative analysis of gene expression between maize egg cells and central cells


  • Quyên Lê,

    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
    Search for more papers by this author
    • Present address: Zentrum für Molekulare Neurobiologie, Institut für Neurale Signalverarbeitung, Universität Hamburg, Falkenried 94, 20251 Hamburg, Germany.

  • José F. Gutièrrez-Marcos,

    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
    Search for more papers by this author
  • Liliana M. Costa,

    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
    Search for more papers by this author
  • Stephanie Meyer,

    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
    Search for more papers by this author
  • Hugh G. Dickinson,

    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
    Search for more papers by this author
  • Horst Lörz,

    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
    Search for more papers by this author
  • Erhard Kranz,

    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
    Search for more papers by this author
  • Stefan Scholten

    Corresponding author
    1. 1 Biozentrum Klein Flottbek und Botanischer Garten, Entwicklungsbiologie und Biotechnologie, Universität Hamburg, Ohnhorststraße 18, 22609 Hamburg, Germany, and
      2 Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, UK
      (fax +49 40 42816 229; e-mail
    Search for more papers by this author

(fax +49 40 42816 229; e-mail


The analysis of cell type-specific gene expression is an essential step in understanding certain biological processes during plant development, such as differentiation. Although methods for isolating specific cell types have been established, the application of cDNA subtraction to small populations of isolated cell types for direct identification of specific or differentially expressed transcripts has not yet been reported. As a first step in the identification of genes expressed differentially between maize egg cells and central cells, we have manually isolated these types of cell, and applied a suppression-subtractive hybridization (SSH) strategy. After microarray screening of 1030 cDNAs obtained from the subtracted libraries, we identified 340 differentially expressed clones. Of these, 142 were sequenced, which resulted in the identification of 62 individual cDNAs. The expression patterns of 20 cDNAs were validated by quantitative RT–PCR, through which we identified five transcripts with cell type-specific expression. The specific localization of some of these transcripts was also confirmed by in situ hybridization on embryo sac sections. Taken together, our data demonstrate the effectiveness of our approach in identifying differentially expressed and cell type-specific transcripts of relatively low abundance. This was also confirmed by the identification of previously reported egg cell- and central cell-specific genes in our screen. Importantly, from our analysis we identified a significant number of novel sequences not present in other embryo sac or, indeed, in other plant expressed sequence tag (EST) databases. Thus, in combination with standard EST sequencing and microarray hybridization strategies, our approach of differentially screening subtracted cDNAs will add substantially to the expression information in spatially highly resolved transcriptome analyses.


Global expression analyses have revealed that distinct patterns of gene expression characterize different phases of development and differentiation states (Smith and Greenfield, 2003; Sreenivasulu et al., 2004). To help understand such biological processes, methods are required for studying changes in gene expression within distinct cell types or groups of cells. These studies often require the isolation of specific cell types, and several such approaches have been reported to date. In plants, these include protoplast preparations in which uniform populations of cells are accumulated, an example being the isolation of AtSUC2 promoter–green fluorescent protein (GFP)-marked companion cells for patch-clamp studies and expression profiling (Ivashikina et al., 2003). Another approach, adopted by Karrer et al. (1995), involved a microcapillary method to isolate individual cells. The microcapillary method has since been used in conjunction with GFP constructs under the control of specific promoters to visualize and isolate various cell types (Brandt et al., 1999). In addition, manual microdissection has enabled the isolation of specific cell types from the female gametophyte in cereals (Kranz et al., 1991; Kumlehn et al., 1999). Recently, laser capture microdissection (LCM) has been employed to isolate individual plant cell types (Asano et al., 2002; Kerk et al., 2003; Nakazono et al., 2003) as well as specific domains of the developing embryo (Casson et al., 2005).

Substantial insights into the transcriptional profiles of isolated cell types have been generated by cDNA library construction, expressed sequence tag (EST) sequencing and microarray hybridization (Birnbaum et al., 2003; Casson et al., 2005; Schnable et al., 2004; Sprunk et al., 2005). However, because many genes important in phenotypic determination are expressed at very low levels in eukaryotic cells (Kuznetsov et al., 2002), the identification of genes expressed differentially between diverse cell types remains challenging. New strategies are therefore required to help identify weakly expressed genes that may not be represented in large-scale EST sequencing projects (Journet et al., 2002), or those that evade characterization by microarray hybridizations (Casson et al., 2005). The usefulness of suppression-subtractive hybridization (SSH) (Diatchenko et al., 1996) as a complementary transcript-profiling tool, especially in identifying novel genes and transcripts of low abundance, was recently demonstrated in a large-scale comparative study of SSH and Affimetrix GeneChip microarrays with human cells (Cao et al., 2004). Suppression-subtractive hybridization has also proven effective in identifying differentially expressed genes in plants. Examples include monitoring of transcriptional changes in response to stress (Zheng et al., 2004); along primary root tips (Bassani et al., 2004); and in plants overexpressing a Dof (DNA-binding with one finger) transcription factor (Kang et al., 2003).

In this technical report, we demonstrate how SSH coupled to microarray analysis can be applied to very limited numbers of isolated plant cells. Because our interest is in plant reproductive biology, we chose the female gametophyte as an experimental system. The female gametophyte plays a critical role in the reproductive process of angiosperms, and contains a number of different cell types that are technically difficult to access. Our current knowledge of genes required for female gametophyte development and the early stages of embryo and endosperm development has come mostly from mutant screens (Christensen et al., 2002; Pagnussat et al., 2005; Yadegari and Drews, 2004), and it is likely that genes expressed differentially between egg cells and central cells have specific functions in conferring identity on cells of the embryo or endosperm. Because a large proportion of female gametophytically expressed genes are expected to have partially or completely redundant functions (Drews and Yadegari, 2002), reverse-genetic approaches are necessary to characterize these genes.

With this in mind, we demonstrate how subtracted cDNAs can be obtained from small amounts of isolated egg and central cells, and be used for differential screening to identify transcripts that were either specifically or differentially expressed at low levels. The usefulness and applications of this approach for the plant community, and the possible substitution of alternative methods for isolating other cell types, are discussed.


Cell isolation and optimization of cDNA synthesis from a limited number of cells

Specific cell types of the female gametophyte were isolated by manual microdissection without any contaminating tissue traces. Representative egg and central cells used for cDNA synthesis and subtraction are shown in Figure 1(a). PolyA+ RNA was directly isolated from these cells with oligo dT(25) magnetic beads, and first-strand synthesis was performed with mRNA bound to the beads to minimize loss of mRNA. Despite this, we anticipated that the yield of RNA from very few cells would be insufficient to perform cDNA subtraction solely after first- and second-strand cDNA synthesis. We therefore applied a template-switch mechanism to the 5′ end of the RNA template (SMART) during reverse transcription and amplified the first-strand cDNA by long-distance PCR (LDPCR). In preliminary experiments, we found that 25 egg cells and 20 central cells were sufficient to generate cDNA of broad size range (data not shown). To ensure the original relative transcript abundance was preserved, the LDPCR reaction was not allowed to reach saturation. Consequently, we tested the cycle number required to reach the plateau for each individual reaction and used one cycle less for bulk cDNA amplification. The final conditions produced 6–8 μg cDNA with a size distribution of between 0.2 and 4 kb, after 24 cycles. To test the quality of the cDNA generated under these conditions, egg cell cDNA was analysed by Southern hybridization and probed with Zmcdc2A/B. Two distinct bands of approximately 1.4 and 1.3 kb were identified (Figure 1b), corresponding to the full-length transcripts for these genes. Generally, the appearance of egg and central cell cDNAs on gels was reproducible. The central cell generated more cDNA fragments between 0.5 and 0.75 kb, and fewer fragments of size range below 0.5 kb, while egg cell cDNA lengths were distributed more equally over the whole size range. In addition, both sets of cDNA populations showed declining fragment abundance between 2.0 and 4.0 kb (Figure 1c).

Figure 1.

cDNA synthesized from mRNA isolated from microdissected egg and central cells.
(a) Bright-field image of isolated maize central cell (CC) and egg cell (EC). Scale bar = 50 μm. Approximately 100 ng cDNA synthesized from 25 egg cells and 20 central cells was run on a 1.2% agarose gel alongside 100 ng 1 kb ladder (M). (b) Egg cell cDNA was blotted and hybridized to a Zmcdc2A/B-specific probe (Hyb). (c) Comparison of cDNAs synthesized from egg cells and central cells.

Identification of differentially expressed genes between two distinct cell types

We employed SSH to identify genes differentially expressed between egg and central cells. The cDNA populations were used for forward and reverse subtractions to enrich those genes specifically or predominantly expressed in one or other of these two cell types. To gain an indication of the subtraction efficiency, we compared ubiquitin transcript abundance before and after subtraction by PCR. The abundance of this non-differentially expressed gene was estimated to be 40-fold less in both subtracted cDNA populations (data not shown). A total of 519 egg cell clones, 511 central cell clones and two controls were arrayed and differentially screened by microarray hybridizations, using subtracted cDNAs and unsubtracted control cDNAs of both cell types as targets. When using subtracted cDNAs as targets, 175 and 165 clones from the central cell and egg cell libraries, respectively, showed a twofold or higher difference in expression between the two cell types (Figure 2). In contrast, when unsubtracted cDNA populations were used as targets, 82 and 131 of the central cell and egg cell clones, respectively, showed a twofold or higher difference in expression (data not shown).

Figure 2.

Differential screening by microarray hybridization.
The histogram represents the distribution of normalized log2 ratio values for the hybridizations of 511 central cell and 519 egg cell cDNA clones with subtracted cDNA targets. Log2 ratio values were calculated by dividing the fluorescence intensity of central cell clones with fluorescence intensity of egg cell clones. Positive log2 values indicate preferential expression in the central cell; negative log2 values indicate preferential expression in egg cells.

Based on our microarray hybridization data, we selected clones within the range of twofold or higher differential expression between egg cell and central cell, and with at least one target combination. Out of these, we further selected clones that displayed moderate to low relative fluorescence intensities with unsubtracted targets. The dynamics of these hybridizations ranged up to 51 063 relative fluorescence units (RFU) within a maximum system range of up to 65 000 RFU. Half the clones selected with predominant expression in egg cells had a mean of 10 369 RFU and the other half a mean of 5673 RFU. Similarly, half the central cell clones selected had a mean of 13 873 RFU, and the other half a mean of 2556 RFU. After plasmid isolation and sequencing of the selected clones, a total of 142 sequences were generated and edited to remove vector sequences. Sequence sizes ranged from 215 to 752 bp, with a mean of 502 bp. Cluster analysis revealed a redundancy of 56.3%, resulting in 62 unique sequences, half specific to central cells and the other half to egg cells. We submitted each unique sequence to the blast program (Altschul et al., 1997) to search the GenBank non-redundant public sequence databases. A total of 30 (48.4%) differentially expressed genes showed close matches [blastx or blastn expectation (E) values <10−13] to database entries with assigned identities. We classified these sequences into eight groups based on functional categories established by the Munich Information Center for Protein Sequences (MIPS) for Arabidopsis (Rudd et al., 2003). The proportion of genes in each discrete category for egg and central cells is shown in Figure 3, while blastx results and log2 ratios of the differential screening are shown in Table 1 for sequences with homologies to either known genes or functionally annotated genes. Among the fragments identified, three clones represented previously characterized genes that are expressed exclusively in egg or central cells (Table 1). E039 corresponds to Zea mays transparent leaf area peptide gene (blastnE 10−149), which was originally isolated from an egg cell-specific cDNA library (Dresselhaus et al., 2005). C068 and C075 are ZmEBE-1 and ZmEBE-2 genes, respectively (blastn E 0.0), which are reportedly expressed in central cells and endosperm, but not in egg cells (Magnard et al., 2003).

Figure 3.

Classification of differentially expressed transcripts between egg and central cells.
Sequences originating from egg and central cells are indicated by grey and black bars, respectively. Digits indicate number of transcripts.

Table 1.  Differentially expressed cDNAs between egg and central cells
CloneLengthRatioaACC numberblastx sequence similarity (accession number)bE
  1. aLog2-transformed ratio of fluorescence intensities from differential screening with subtracted targets.

  2. bblastn sequence similarity, if indicated.

  3. cRelative fluorescence intensities (RFU) of differential screening with unsubtracted targets, see text for details.

  4. C062, C089 and C190 are different cDNA fragments corresponding to the same gene.

Egg cell cDNAs with moderate RFUc
E001700−1.26DN591121Putative cig3, Oryza sativa (BAC92461.1)8.00E-62
E007734−1.54DN591123Aspartate transaminase, Proso millet (S53303)e-111
E013712−1.27DN591125Heat shock protein 82, O. sativa (P33126)2.00E -95
E021746−1.53DN591129Putative meiosis protein mei2, O. sativa (BAD12869.1)2.00E-93
E023711−1.00DN591130Ankyrin-like protein-like protein, Sorghum bicolor (AAO16699.1)6.00E-71
E026723−1.44DN591132Anthocyanin biosynthetic gene regulator PAC1, Zea mays (blastn, AY115485.1)6.00E-40
E034444−1.71DN591136Heat-shock protein 80, Triticum aestivum (AAD11549.1)1.00E-75
E039281−2.98DN591137Transparent leaf area peptide mRNA, Z. mays (blastn, AY211982.1)e-149
E047351−1.03DN591140Auxin response factor 10, O. sativa (BAB85919.1)1.00E-13
E048631−1.46DN591141Putative NTGP4, O. sativa (BAD16419.1)4.00E-73
Egg cell cDNAs with low RFUc
E106538−1.17DN591143SMC6 protein, O. sativa (CAD59413.1)6.00E-33
E107542−3.39DN591144Phosphoglyceride-transfer family protein, Arabidopsis thaliana (NP_173637.3)1.00E-28
E112669−1.58DN591147Putative auxin-response factor, O. sativa (AAO34491.1)2.00E-29
E113390−1.7DN591148Alpha glucosidase-like protein, A. thaliana (BAB02784.1)4.00E-42
E146475−1.2DN591149Protein kinase-like protein, A. thaliana (CAA22964.2)5.00E-77
E148452−2.25DN591150Putative methyl-binding domain protein MBD108, Z. mays (AAK40309.1)8.00E-44
E149389−1.23DN591151Cyclin A-like protein, Z. mays (T02746)2.00E-28
Central cell cDNAs with moderate RFUc
C0532351.12DN591090Chloroplast rRNA-operon, Z. mays (blastn, Z00028.1)e-124
C0563472.37DN591091P-type ATPase, Hordeum vulgare (CAC40035.1)4.00E-21
C0586371.57DN591092Ae1 protein, Z. mays (CAB56550.1)2.00E-67
C0626761.64DN591094Catechol O-methyltransferase, H. vulgare (S52015)3.00E-54
C0685252.21DN591095ZmEBE-1 protein, Z. mays (CAD24795.1)3.00E-80
C0723772.96DN591096Putative sterol glucosyltransferase, O. sativa (NP_915652.1)6.00E-57
C0757391.29DN591097ZmEBE-2 protein, Z. mays (CAD24798.1)e-133
C0893281.28DN591098Catechol O-methyltransferase, H. vulgare (S52015)5.00E-16
Central cell cDNAs with low RFUc
C1542901.20DN591100Non-photosynthetic NADP-malic enzyme, Z. mays (AAQ88396.1)1.00E-32
C1735411.11DN591111Origin recognition complex subunit 3, Z. mays (blastn, AF417483.1)0.0
C1753921.10DN59111318S small subunit ribosomal RNA gene, Z. mays (blastn, AF168884.1)0.0
C1762931.02DN591114Putative coatomer protein complex, subunit beta, A. thaliana (NP_175645.1)3.00E-25
C1774271.71DN591115Acetoacetyl CoA thiolase, Z. mays (AAD44539.1)1.00E-36
C1782881.22DN591116Putative protein kinase, O. sativa (NP_908679.1)7.00E-18
C1793090.98DN591117Plectin-related protein, A. thaliana (NP_180245.1)2.00E-21
C1906521.88DN591118Catechol O-methyltransferase, H. vulgare (S52015)3.00E-54
C1933761.61DN591119Ae1 protein, Z. mays (CAB56550.1)1.00E-16

Validation of differential expression patterns by RT–PCR identifies cell type-specific transcripts

To validate the microarray expression profiles, RT–PCR was used to study 20 differentially expressed clones. Ten of these clones (E013, E023, E047, E106, E112, E146, E148, E149, C089 and C176) were selected for their similarities to genes involved in regulatory processes (Table 1). The other 10 sequences, which either shared no homology with any other sequence, or showed homology with hypothetical proteins, were chosen at random. To estimate differential expression levels of these genes, gene-specific oligonucleotides were used in PCR reactions containing equal amounts of both cDNAs as template. Products of incremental cycle numbers were subsequently analysed. The difference in the number of cycles required for equal amplification of the corresponding PCR product in egg cell and central cell samples was used to estimate levels of differential expression. In addition, three constitutively expressed genes (ubiquitin, TATA box binding factor and actin) were used as controls as their expression levels were found to be near identical in both egg and central cells (Figure 4a). In contrast, the 20 subtracted library transcripts tested exhibited either differential or specific expression in egg and central cells (Figure 4b,c). For instance, three transcripts (C176, C170 and C195) were expressed specifically in central cells (Figure 4b), whereas two (E023 and E017) were expressed specifically in egg cells (Figure 4c). These cell type-specific expression patterns were confirmed using three independent cDNA samples. We refer to these transcripts hereafter as egg cell-specific or central cell-specific. For the other 15 transcripts tested, amplification products exhibited subtle or more pronounced differences of three to 18 cycles, indicating an estimated seven- to 46-fold difference in expression levels between egg cell and central cell (Figure 4b,c). In most cases, the cycle number needed to amplify a visible PCR product from original cDNA populations could be correlated to the RFU value in the microarray hybridizations with unsubtracted targets. For the 20 transcripts tested, we found a significant correlation of r = −0.595.

Figure 4.

RT–PCR analysis of transcripts differentially expressed between egg and central cells after incremental PCR cycles.
Equal amounts of egg and central cell cDNA was used as template for each PCR reaction using specific oligonucleotides. PCR products were analysed at each cycle number indicated. Difference in numbers of cycles required for equal amplification in egg cell and central cell samples was considered as an indication of the level of differential expression. (a) Ubiquitin (Ubi), TATA box binding protein (Tbbp) and Actin (Act) were used as positive controls. (b,c) Transcripts identified by differential screening that are preferentially expressed in central cells or in egg cells.

In situ hybridization of female gametophyte serial sections confirm specific expression patterns

We performed in situ hybridization to further confirm the expression patterns of two transcripts from the subtracted egg cell library, chosen for their predominant (E012) and specific (E017) expression patterns, and of two central cell-specific transcripts (C170 and C195) from the subtracted central cell library. E012 transcript was localized predominantly in egg cells, while a weak signal was detected in the sporophytic nucellar tissue (Figure 5b). E017 transcript was detected exclusively in egg cells (Figure 5d), whereas C170 transcript was localized in the central cell (Figure 5f) but not in the egg cell (Figure 5g). Transcript C195 was detected in nucellar tissue as well as in the polar nuclei and, to a lesser extent, the cytoplasm of central cells only (Figure 5i).

Figure 5.

In situ hybridization analysis of transcripts differentially expressed in egg and central cells.
Embryo sac sections were hybridized with sense (a,c,e,h) or antisense (b,d,f,g,i) digoxigenin-labelled RNA probes for E012 (a,b); E017 (c,d); C170 (e–g); C195 (h,i) transcripts; (f,g) show serial sections of the same embryo sac. Scale bar (a–i) = 100 μm.

Differentially expressed egg and central cell transcripts show diverse expression patterns within female and male reproductive cells

The expression patterns of transcripts specific to, or predominantly expressed in, egg or central cells were analysed by RT–PCR using cDNA obtained from egg cells, central cells, synergids, antipodals and sperm cells. Interestingly, none of the transcripts identified as being specific to central cells was expressed in sperm cells, and instead exhibited differential expression within cells of the female gametophyte (Figure 6). For instance, C195 was expressed exclusively in central cells, whereas C170 and C176 were co-expressed in antipodal cells and in synergids, respectively. Two transcripts (C059 and C089) predominantly expressed in central cells, were also expressed in egg cells, sperm cells and synergids, but not in antipodal cells. One transcript (C162), also with predominant expression in central cells, was expressed in all cells of the female gametophyte, but not in sperm cells (Figure 6). Interestingly, none of the genes expressed exclusively or predominantly in egg cells was expressed in sperm cells. The egg cell-specific transcript E017 was detected only in egg cells, whereas egg cell transcript E023 was also detected in synergids, but not in the central cell. Two transcripts with predominant expression in egg cells (E041 and E012) were also expressed in central cells only. In contrast transcript E148, which exhibited predominant expression in egg cells, was found to be also expressed in all other cells of the female gametophyte.

Figure 6.

RT–PCR analysis of transcripts expressed differentially within cells of the female gametophyte and sperm cells.
cDNA synthesized from central cells (CC), egg cells (EC), synergid cells (SY), antipodal cells (AP) and sperm cells (SC) were used as template for PCR amplification using gene-specific oligonucleotides for a selected number of transcripts. The ubiquitin (Ubi) gene was used as a positive control. An H2O sample served as a negative control (−) in PCR reactions.


Amplification of cDNA from small numbers of distinct cell types

To compare the mRNA populations of specific cell types, synthesis of high-quality cDNA is required. For this reason, and to minimize mRNA loss, we isolated polyA+ RNA with oligo dT(25) magnetic beads and carried out first-strand synthesis with the mRNA bound to the beads (Adjaye et al., 1999). In addition, we chose SMART cDNA amplification to produce sufficient amounts of cDNA for our purpose, because it has been shown to generate representative cDNA samples from microscopic amounts of tissues in animals (Matz, 2003). By taking into consideration the amplification guidelines of Matz (2003), as well as the cycle number and concentrations of cDNA obtained, we calculated that 0.65–1.31 × 106 individual transcripts with a mean length of 1 kb would be amplified in our reactions. Although we cannot rule out the possibility of having lost the least abundant transcripts, these estimations indicated that the transcripts of egg and central cells were well represented. Further, by limiting the cycle number for bulk cDNA amplification (Diatchenko et al., 1998; Endege et al., 1999), we were able to preserve the original relative transcript abundance. This was indicated by the effective identification of a significant number of differentially expressed transcripts from these cDNAs. Similarly, the preservation of relative transcript abundance has been demonstrated previously when using SMART-amplified cDNA as targets for gene expression profiling by array technology (Livesey et al., 2000; Wang et al., 2000).

Subtracted cDNA libraries from limited amounts of different cell populations allow effective identification of low-abundance, differentially expressed and cell type-specific genes

To identify differentially expressed genes and, at the same time, avoid sequencing of genes with identical expression levels, subtracted cDNA libraries have been screened differentially by microarray hybridization (Bangur et al., 2002; Welford et al., 1998; Yang et al., 1999). Here, we demonstrate that subtracted and unsubtracted cDNAs generated during the SSH procedure can be used as targets for microarray hybridizations. For this approach, we applied two criteria for selecting differentially expressed clones with moderate and low expression levels after screening and data normalization. For the first criterion, a minimum threshold of twofold expression difference in the microarray experiments was taken to indicate cDNAs as being differentially expressed. These results were confirmed by RT–PCR, mRNA in situ hybridization, and comparison with known expression data published in the literature for some of the genes identified in our study (Dresselhaus et al., 2005; Magnard et al., 2003).

The second criterion was based on the fluorescence intensity with unsubtracted cDNAs. We noticed a correlation between the relative signal intensities with unsubtracted targets and transcript abundance. We found that sequences putatively involved in signal transduction or cell growth, division and DNA synthesis generally possessed low RFU values. This is in contrast to transcripts with homology to heat-shock proteins (E013, E034), chloroplast rRNA-operon (C053) and ATPase (C056), which are probably expressed at higher levels, and displayed moderate RFU values (Table 1; Figure 3). This correlation was confirmed for 20 clones by quantitative RT–PCR. Further, we found that a higher proportion of genes belonging to functional classes, which are generally expressed at low levels, were identified through our method compared with conventional EST sequencing of cDNA libraries. For instance, it is known that signal transduction and cell growth, division and DNA synthesis represent the two least abundantly expressed functional classes in yeast (Jansen and Gerstein, 2000). These two classes constituted 4 and 2%, respectively, of functionally annotated EST clones that were randomly sequenced from a wheat egg cell cDNA library (Sprunk et al., 2005). In contrast, these classes constituted 23.5 and 17.6%, respectively, in our subtracted and screened egg cell library. Together these data clearly demonstrate that genes with relatively low expression levels, which are either differentially expressed or cell type-specific, can be effectively identified through our method. That this is achieved solely with cDNAs generated during the SSH procedure provides a great advantage when starting material is limited.

Identification of novel transcripts expressed in the female gametophyte

Given that around 5700 EST sequences from cDNA libraries prepared from whole maize and barley embryo sacs have been deposited in GenBank (H. Yang and S. McCormick, USDA Plant Gene Expression Center, Albany, CA, USA, unpublished data; R. Waugh and D.T. Marshall, Scottish Crop Research Institute, Dundee, UK, unpublished data), it was surprising to find that only 17 of our 62 sequences (27.4%) are represented in these collections. This indicates that our combined approach is capable of identifying novel transcripts with high efficiency. Moreover, while nearly half the sequences obtained from central cells are represented in current embryo sac libraries, only three of the 31 egg cell sequences are represented in such collections. This discrepancy might be explained by the different sizes of the two cell types: the diploid central cell fills the majority of the volume of the embryo sac, whereas the haploid egg cell is much smaller and hence might account for the minority of embryo sac mRNAs.

From our analysis, we found cDNAs with homologies to metabolic genes that are predominantly expressed in central cells (Table 1; Figure 3). This could be explained by the high energy demand during initial free-nuclear development of the endosperm, which commences 12 h before the first zygotic division (Mòl et al., 1994). We also identified the predominant expression of three cDNAs with homology to plant hormone-responsive genes in egg cells, two of which shared homology with auxin-responsive genes (Table 1). The latter probably reflects the immediate, important role of auxin in regulating embryonic patterning (Kepinski and Leyser, 2003). In contrast, none of the transcripts predominantly expressed in central cells had homology to known plant hormone-responsive genes.

Importantly, some of our sequences (11%) are not featured in any of the current plant EST databases, indicating that although more than 4 000 000 plant ESTs have been sequenced (, this figure is by no means exhaustive. On the other hand, 25.8% of our sequences showed significant similarity to unknown or hypothetical genes with no assigned function (Figure 3). This suggests that our approach is highly effective for identifying transcripts that putatively encode novel proteins. Indeed, these candidates might represent proteins involved in gamete identity determination or post-fertilization differentiation.

Wider applications of this technique for cell type-specific expression analysis

In this report we have demonstrated that the combination of SSH and microarray analysis is an effective method to analyse differential expression of moderate- to low-abundance genes when using small or limited quantities of isolated cells as starting material. Although our approach can be applied to most areas of plant transcriptomic research, the only foreseeable limitation is in the cell isolation technique employed. Here we describe the isolation of maize egg cells, central cells, antipodals, synergids and sperm cells by manual microdissection. Although highly time-consuming and technically demanding, this method has also enabled the isolation of other reproductive cells in maize (Scholten et al., 2002); barley (Mogensen and Holm, 1995); rice (Khalequzzaman and Haq, 2005); and wheat (Kumlehn et al., 1999). However, manual microdissection of other cell types might be impractical, in which case the microcapillary method and the use of protoplasts both serve as good alternatives. For instance, microcapillary aspiration was used to isolate various cell types from leaf to generate specific cDNA libraries (Karrer et al., 1995) and targets for cDNA array hybridizations (Brandt et al., 2002). In addition, Engel et al. (2003) demonstrated how high numbers of protoplasts could be purified for construction of a sperm-specific cDNA library by fluorescence-activated cell sorting. Although the identification of certain cell types is, in general, not easily accomplished, cell-specific promoters are becoming increasingly available and can be fused to visible markers to aid cell isolation (for examples see Brandt et al., 1999; Ivashikina et al., 2003). A combination of protoplast preparation and automated cell sorting has been used for detailed gene expression profiling in Arabidopsis root (Birnbaum et al., 2003). Another alternative method is laser capture microdissection. This method is particularly attractive when specific markers are not available, because it can be used to dissect out cells precisely from heterogeneous tissues (Asano et al., 2002; Casson et al., 2005; Inada and Wildermuth, 2005; Kerk et al., 2003; Nakazono et al., 2003).


Here we present the valuable use of SSH coupled to microarray hybridization to enable the efficient identification of genes with either specific or differential expression between maize egg and central cells. Although the functions of the genes identified in this study remain to be established, we expect that future characterization of these genes will add significantly to our understanding of how egg and central cells differentiate during female gametophyte development, and what factors contribute to the divergent developmental pathways of embryo and endosperm.

Importantly, we have demonstrated that the construction and differential screening of subtracted cDNA libraries allows the biased identification of moderate- to low-expression transcripts. This property makes our approach appealing as a powerful technique that is complementary to the conventional EST sequencing and microarray hybridization procedures currently used for spatially highly resolved transcriptome analyses. Alternative methods for the isolation of other cell types, such as LCM, can be readily substituted for manual microdissection to render our combined approach applicable to many other lines of plant transcriptomic research.

Experimental procedures

Cell isolation

All plant material was isolated from line A188 (courtesy of A. Pryor, Commonwealth Scientific and Industrial Research Organization, Canberra, Australia). Plants were grown in the glasshouse with a 16-h illumination period (25 klux) at 25/20°C (day/night). Individual central cells, eggs, synergids and sperm cells were isolated and selected as described by Kranz et al. (1991, 1998). Antipodal cells were isolated in groups of approximately 15 cells. After washing in mannitol solution (0.650 osmolar), cells were transferred into tubes, snapped frozen in liquid nitrogen and stored at −80°C until use.

Synthesis of cDNAs

Isolation of mRNA from 25 synergids and egg cells, 20 central cells, four clusters of approximately 15 antipodal cells, and 400 sperm cells was carried out with oligo dT(25) magnetic beads (Dynal, Hamburg, Germany). For mRNA isolation, 2× concentrated lysis/binding buffer (200 mm Tris-HCL pH 7.5; 1 m LiCL; 20 mm EDTA pH 8.0; 10 mm DTT; 2% LiDS) was added to the frozen cells/tissue in mannitol solution or 10 μg total RNA. The volume was adjusted with ddH2O to give 1× lysis/binding buffer final concentration in 20–40 μl total volume. For each cell type, 15 μl oligo dT(25) magnetic beads were washed once with lysis/binding buffer, and the lysed cells were added and incubated at 21°C for 15 min under continuous rotation. Washes were carried out twice each with 50 μl washing buffer (10 mm Tris-HCl pH 7.5; 0.15 m LiCl; 1 mm EDTA) and 50 μl first-strand buffer (50 mm Tris-HCl pH 8.3; 75 mm KCl; 3 mm MgCl2). Beads with mRNA were resuspended in 3 μl ddH2O and used for first-strand cDNA synthesis, applying the template switch mechanism at the 5′ end (SMART) and long-distance PCR (LDPCR) (BD Bioscience Clontech, Erembodegem, Belgium) following the manufacturer's instructions. For each cell type the optimal LDPCR cycle number was determined empirically with an aliquot of the reaction to ensure the cDNA remained in the exponential phase of amplification. At all times, one cycle below that required to reach the plateau was used to amplify the remaining cDNA. These LDPCR conditions led to equal cDNA concentrations of approximately 20 ng μl−1 for each cell type. Egg and central cell cDNAs were used for SSH, whereas all cDNAs were used as template for gene-specific expression analysis. The latter was repeated three times using three independent sets of cDNA samples.

Southern blot analysis

Transfer of 100 ng egg cell cDNA to nylon membranes and hybridization with Zmcdc2A/B-specific (Colasanti et al., 1991) digoxigenin-labelled probes was performed as described by Scholten et al. (2002).

Suppression-subtractive hybridization

The generation of forward- and reverse-subtracted cDNAs and unsubtracted control cDNAs from egg and central cells was performed using the PCR-select cDNA subtraction kit (BD Bioscience Clontech) following the manufacturer's recommendations. The subtracted egg and central cell cDNAs were cloned with the pGEM-T Easy Vector System (Promega, Mannheim, Germany) and transformed into Escherichia coli DH5α cells (Invitrogen, Groningen, the Netherlands). The transformed bacteria were plated onto Luria Bertani (LB) agar plates containing ampicillin, X-gal and isopropyl-beta-d-thiogalactopyranoside (IPTG). For each subtracted cDNA, 672 recombinant white colonies were selected and cultured in 100 μl LB broth containing ampicillin in a 96-well format. After overnight culture, 100 μl LB, 50% glycerol was added and the plates were stored at −70°C.

Microarray production

From each bacterial culture of the subtracted libraries, 2 μl served as template to amplify individual cDNA inserts using a primer pair corresponding to the flanking adaptor sequences (NP1, NP2R; BD Bioscience Clontech). The PCR products were visualized on 1% agarose gels to test the quality and quantity of amplification, followed by purification with Multiscreen PCR plates (Millipore, Eschborn, Germany). In total, 76.6% of the inserts of subtracted cDNA clones amplified successfully with a single fragment (519 and 511 for egg and central cells, respectively). Fragments of the E. coli uidA gene and the human estrogen receptor (er) gene were amplified and processed as the PCR products from the subtracted libraries to serve as control elements for the microarray hybridizations. The control spots were amplified fourfold and distributed over the whole array. The clean PCR products were reformatted in 384-well plates, dried and resuspended in 15-μl spotting buffer (3 × SSC, 1.5 m betain). Four replicates for each PCR product were printed onto poly-l-lysine-coated glass slides (Sigma, Taufkirchen, Germany) with a Microgrid II robot (BioRobotics, Boston, MA, USA).

Differential screening by microarray hybridization

For differential screening, the cDNA microarrays were hybridized with fluorescent dye-labelled forward- and reverse-subtracted cDNAs, or with egg and central cell unsubtracted control cDNA as targets in the same hybridization reaction. Hybridizations were repeated twice with reverse dye labelling. Indirect random prime labelling was used for target preparation. After purification with GFX PCR DNA and gel-band purification columns (Amersham, Freiburg, Germany), 2 μg cDNA spiked with 10 ng uidA and er control fragments was labelled after denaturation in 50 μl reactions. Each reaction contained 6 μm 5′-amino-modified random hexamer oligonucleotide; 50 mm Tris-HCl pH 8.0; 5 mm MgCl2; 1 mm DTT; 500 μm each dATP, dCTP, dGTP; 200 μm dTTP; 300 μm aminoallyl-dUTP (Sigma); 10 U Klenow fragment (MBI Fermentas, St Leon-Rot, Germany), and incubated at 37°C for 12–16 h. Labelled cDNAs were washed three times with 450 μl ddH2O in Microcon-30 concentrators (Millipore) dried under vacuum, resuspended in 0.05 m sodium bicarbonate pH 9.0, added to Cy3 or Cy5 mono-reactive dye (Amersham) and incubated for 1 h at room temperature. After addition of 4.5 μl of 4 m hydroxylamine and a further 15 min incubation, the reaction was purified with GFX PCR DNA and Gel-band purification columns (Amersham), and eluted with 100 μl 10 mm Tris-HCl pH 8.5.

The absorbance at 260 and 550 or 650 nm was read for Cy3- or Cy5-labelled samples, respectively. The dye incorporated was calculated using the formula: pmol Cy3 = A550 × 100/0.15; pmol Cy5 = A650 × 100/0.25. The frequency of incorporation (f.o.i., number of labelled nucleotides per 1000 nucleotides) was calculated for Cy3: f.o.i. = pmol Cy3 × 324.5/(A260 × 37 × 100) and for Cy5: f.o.i. = pmol Cy5 × 324.5/(A260 × 37 × 100). For each hybridization, 20 pmol of two samples with f.o.i. ≥8 were combined, dried under vacuum and resuspended in 45 μl hybridization solution (25% formamide, 5 × SSC, 0.1% SDS, 100 μg ml−1 sonicated salmon sperm DNA, 400 μg ml−1 polyA DNA; 300 ng ml−1 equal amount of NP1, NP2R and the complementary oligonucleotides). After denaturation, the targets were hybridized at 42°C with spotted arrays, prehybridized for 1 h in 5 × SSC, 0.1% SDS, 1% BSA for 16 h. Stringent washing conditions were: 5 min, 42°C (2 × SSC, 0.1% SDS); 10 min, room temperature (0.1 × SSC, 0.1% SDS); 3 × 1 min, room temperature, 0.1 × SSC. Slides were dried in a stream of nitrogen.

For scanning the hybridized slides, an ArrayWorx microarray reader (AppliedPrecision, Issaquah, WA, USA) was used. Analysis of the images was conducted with genepix 4.0 software (Axon, Union City, CA, USA). For data preparation, all spots with high background values were removed from further analysis, and the local background of each spot was subtracted and normalized according to the spiking controls. In a second step, genesight software (Biodiscovery, El Segundo, CA, USA) was used to combine replicated experiments, remove spots with relative fluorescence values below 500, and perform global normalization using LOWESS regression (Cleveland and Devlin, 1988) with a smoothing parameter of 0.2 and one degree of fitness. The resulting ratios of both channels were log2-transformed.

Differential gene expression analysis by RT–PCR with incremental cycle numbers

To quantify the difference of expression levels between egg and central cells, equal amounts (20 ng) of cDNA, prepared as described above, from each cell type were used for PCR with gene-specific oligonucleotides in 100-μl reactions. These oligonucleotides were designed with primerselect software (Lasergene, GATC Biotech AG, Konstanz, Germany) and the sequences are available on request. Cycling parameters were as follows: 95°C for 30 sec, 56°C for 30 sec, and 72°C for 1 min. Samples of 10 μl were taken every three cycles from 12 to 36 cycles and analysed by agarose gel electrophoresis.

Control amplifications were performed with primers specific to the maize ubiquitin gene (accession number U29159, 5′-CAGGGGTGGCATGCAGATTTTTG-3′ and 5′-CACCAGACGACGCAGGCACATC-3′); the TATA box binding protein gene (accession number X90652, 5′-AATTTTGGTCTGCTTCGTCTTTGAG-3′ and 5′-GTTACATACAGCGCCAGCCTTTTC-3′); and the actin gene (accession number J01238, 5′-TCCTGACACTGAAGTACCCGATTGA-3′ and 5′-CGTTGTAGAAGGTGTGATGCCAGTT-3′). PCR conditions were as described above, except that the annealing temperature was 68°C for ubiquitin and 58°C for TATA box binding protein and actin genes.

To estimate the difference in expression levels, plasmid dilution series with template concentrations of 10 pg, 100 pg, 1 ng and 100 ng were used with the experimental procedure described above and repeated nine times. An average of 3.9 additional cycles were required to amplify equal amounts of PCR product in 10-fold diluted templates. Thus, under our experimental conditions, three cycles corresponded to a 7.7-fold concentration difference.

In situ hybridization

Riboprobes for the transcripts E017, E023, C170 and C195 were labelled with DIG-UTP (Roche GmbH, Mannheim, Germany) according to the manufacturer's instructions, and in situ hybridization was performed on 8-μm-thick wax serial sections of mature ovules fixed, as described by Costa et al. (2003). Serial sections containing the embryo sac were incubated in a 1:3000 Anti-DIG-Antibody (Roche) and transcripts were detected colorimetrically after an overnight incubation in Sigma Fast BCIP/NBT substrate buffer pH 9.5. Slides were viewed with a Zeiss AxioPhot microscope under DIC3-5/5-3 optics and images were digitally recorded.

Gene expression analysis by RT–PCR

PCR analysis was conducted in 25-μl reactions with approximately 10 ng cell type-specific cDNA as template. For expression analysis, a 35-cycle PCR was performed with the cDNA. Thermal cycling parameters were as follows: 95°C for 1 min, 56°C for 30 sec, and 72°C for 1 min and 45 sec. Primer pairs were as for quantitative RT–PCR. Half of each reaction was loaded for analysis by agarose gel electrophoresis. To show the absence of genomic DNA in cDNA preparations, primers gZf 5′-CCCTGTCGTGGCGTCCTCCTG and gZr 5′-CTACGCGCCGGCCTCTTTCTCT, specific for a non-transcribed region of the maize 27-kDa zein gene (accession number X58197), were used (data not shown). All reactions were repeated twice with template cDNA from independent plant material.


We thank Marlis Nissen and Petra von Wiegen for excellent technical assistance in cell isolation. We thank Dr Manfred Gahrtz for helpful discussions. Research was funded by the EU framework V (MAZE) initiative.

Accession numbers for the sequences mentioned in this article are DN591090DN591120 (central cell) and DN591121DN591151 (egg cell).