The ino and ant mutants were chosen for array analysis due to their ovule phenotypes. Ovules of strong ino mutants have only an inner integument, and do not curve as in wildtype (Figure 1D, E) 735. The general role of the ANT gene in all above ground organs is to promote the formation and growth of primordia, and to regulate that growth to control the size of plant organs 363738394041. For most organs these functions are partially redundant with other genes, but in ant mutants the integument primordia fail to initiate, and mature ovules have only a slightly enlarged chalaza between the funiculus and nucellus (Figure 1G, H).

In addition to the integument phenotypes, both mutants are affected in formation of mature embryo sacs, leading to partial and complete sterility for ino and ant respectively. The viability of mature embryo sacs was estimated using decolorized aniline blue staining for callose, which accumulates in defective embryo sacs 4243 (Figure 1C, F, I). 13% of ino mutant embryo sacs (n = 100) did not show an accumulation of callose staining and approximately 5 seeds per silique (compared with 45–50 for wild type) were formed under experimental growth conditions indicating that only approximately one in ten embryos sacs were functional. This is in agreement with microscopic analysis indicating that structurally normal embryo sacs could be formed in ino mutants 7. For the ant-4 allele, there are fewer ovules per carpel than in wild type, and sporogenesis was not observed to proceed beyond the megaspore mother cell stage 7 resulting in complete female sterility 3637. In addition, pistil size and stigma cell number were reduced, and carpels could be partially unfused 4445.

The pistil expression profiles of ino and ant were compared with each other and to wild type (Landsberg erecta, Ler), with the following predicted gene expression profiles. As the outer integument is missing in both mutants, genes that are preferentially expressed there should exhibit absent or significantly decreased expression in both ino and ant samples, relative to wild type. In contrast, an inner integument-expressed gene should exhibit absent or decreased expression in ant but would be unchanged in ino samples as the inner integument is still present. A gene that is expressed in both integuments is likely to show reduced expression in both mutants, with a greater reduction in ant samples as these lack both integuments. However, gene expression changes may be also caused by defective embryo sac formation in ino and ant and by the reduction in ovule number and effects on carpels in ant. By noting the expression level changes of genes that are known to be expressed in these areas it may be possible to define a pattern that identifies such genes for exclusion.

Three different developmental classes of pooled pistils were used. The FULL (F) pool, collected from wild type (WT) and ino, contained pistils from the stage at which ovule primordia are emerging (floral stage 9, ovule stage 1-II), up to mature ovules, just prior to anthesis (Figure 2; floral stage 12, ovule stage 3-IV). Samples containing fewer stages were collected to decrease the complexity of the samples to provide better resolution of expression differences and to evaluate the temporal expression patterns of selected genes. The EARLY (E) pool, collected from all three genotypes, included the youngest stages described above up to the point when the integuments first enclose the nucellus (floral stage 11, ovule stage 3-I), while the LATE (L) pool, collected from WT, captured the remaining stages after the integuments enclose the nucellus up to anthesis. Three biological replicates of each sample were used, with the exception of the WT L arrays, which had two replicates.

Affymetrix ATH1 Genome Arrays representing approximately 23,000 genes 4647 were hybridized with the WT, ino and ant samples. The data from these arrays were processed using Robust Multiarray Averaging (RMA) 48, as well as with dchip and Microarray Suite 5.0 (MAS) in order to determine which method would be most appropriate for the data analysis. Scatterplots between replicates (Additional file 1) and correlation coefficients (Additional file 2) showed that the replicates were very similar to each other (r = 0.9930 - 0.9979 for RMA) and that RMA produced the smallest variance between replicates, particularly at low levels of expression. Comparisons between genotypes were made with the RMA processed data using a moderated t-test in the limma program (affylmGUI) 49505152, which stabilizes variances when few replicates are used, and has been used to analyze data in other plant microarray experiments with similar numbers of replicates 34535455. A multiple testing adjustment of p-values was obtained by conversion to q-values, where a confidence level of 0.01 was used, giving a false discovery rate of 1% 56. The dchip processed data were also used for genotype comparisons using the modified fold change method, where the fold change threshold was 1.2, using the lower bound of the 90% confidence interval for fold change. Using the two statistical tests, there were more genes identified as significantly changed between ant E and wildtype with the RMA-limma test than with the dchip test (3537 and 1672 respectively) and greater than 50% of these genes were uniquely identified by a single method (Additional file 3). The limma test with RMA processed data was successful at identifying seventeen genes known to be expressed in ovules while the dchip fold change method was not as successful (five of seventeen genes failed to be identified) (Additional file 4). This indicated that the RMA-limma method was more appropriate for the task of identifying ovule-expressed genes than the dchip method. Based on these results, RMA processed data were used for further expression analysis described below.

Pairwise comparisons between the mutant and the baseline wild type arrays (WT F and ino F, WT E and ino E, and WT E and ant E) showed that many more genes were changed in the ant E samples (up to 14 times the number identified with ino E) (Table 1), which was predictable from the more perturbed phenotype of ant mutants and the wider spectrum of action of the ANT gene. In addition, there were more genes identified in the ino F comparison than in the ino E comparison, which indicates that there are several identified genes expressed in later stages of ovule development. Finally, there were at least as many genes with increased expression in the EARLY mutant arrays as there were with decreased expression. This is in contrast to the ino F comparison, where more genes exhibited decreased expression in the mutant.

Based on the EARLY array hybridizations, expression of 1717 genes was significantly decreased in the ant mutant relative to WT, and these genes formed a set from which putative inner and outer integument genes were identified based on their levels of expression in ino. From this set, eight hundred putative inner integument genes were identified based on their showing a significant decrease in ant E samples relative to ino E, but no difference between WT E and ino E (WT = ino > ant). Of the 1717 genes reduced in ant, eighty-nine genes showed a decrease in ino E relative to WT and these were further divided into a putative outer integument set of fifty-eight genes that showed no difference between ino E and ant E (WT > ino = ant) and a set of twenty-five genes that showed a further significant decrease in ant E and are therefore likely to have expression in both integuments (WT > ino > ant). These groups of genes were clustered with Kohonen self-organizing maps as implemented in GeneCluster 2.1.7 (SOM) 57, and Broad Institute http://www.broad.mit.edu/cancer/software/genecluster2/gc2.html using all the arrays (Additional file 5). Observing gene levels in the ino F arrays allowed for extrapolation of the ino E inferences and use of the LATE arrays showed whether expression of a particular gene is maintained later in development. Genes with known expression were used to understand the nature of the observed expression patterns and to set thresholds that reflect specific expression patterns, as described below.

The set of genes our analysis indicated were expressed in the inner integument included several genes previously shown to be expressed in ovule primordia or integuments. These included PHB, an inner integument-expressed gene, indicating that the comparisons could identify desired genes. There was minor overlap with putative gametophyte expressed genes (83 of 1278) identified using mature nozzle/sporocyteless (nzz/spl) mutant ovules 335859 and coatlique mutant gynoecia 34. However, some genes expressed in specific cells of the embryo sac, such as WUSHCEL-RELATED HOMEOBOX 2 (WOX2) and WOX8 60 or during meiosis (seven genes including homologs of SPO11 and RAD51) 6162, show no differential expression in our ant-hybridized arrays. A few genes were previously identified as expressed in stigmatic papillae and transmitting tract using arrays (seven of 140 identified) 31. Therefore, most of the differentiation that occurs in the stigma, transmitting tract and gametophyte was not captured by this experiment, leaving the loss of the integuments, reduction in ovule number and reduced growth of the medial regions as likely causes of the identification of the large number of genes with small changes in ant.

The interpretation of the cluster profiles for the 800 putative integument genes depended on whether the mutant expression was considered absent (which indicated the specificity of expression), the inclusion of known indicator genes for each cluster (Additional file 6) and evaluation of expression levels in ino. At least six genes fall to very low levels in ant, and therefore could be integument specific, while two thirds of the genes appear to be expressed more highly during early stages of pistil development since the WT E level is higher than the WT L level.

SOM clusters 4, 5 and 8 – 10 (Figure 3A) contain a total of 310 genes that show little or no change between WT and ino arrays even at later stages and that have higher WT E than WT L levels (except cluster 10), indicating little or no outer integument or embryo sac expression later in development, and more expression early in development. Known genes in this group are expressed in medial regions, placenta and ovule primordia, for example CUC2, PERIANTHIA (PAN) and NUBBIN 196364, while others also show some expression in integument primordia, such as BEL1, SPATULA (SPT) and PIN-FORMED 1 (PIN1) 106566. For these genes, outer integument expression is too low to be discerned in the ino arrays relative to the overall expression levels in the pistil. The patterns can be roughly separated on the basis of fold change: expression in ovule primordia regions results in smaller changes (approximately -1.4) and ovule and integument expression results in slightly higher fold changes.

The remaining clusters all show changes in ino E arrays which were not considered statistically significant but do show a consistent pattern, and are split into two groups by their WT L and ino F expression levels, which are significantly lower than WT F levels in some cases. For clusters 0, 1, 2 and 6 the WT L expression is less than the WT E level (Figure 3A) indicating that gene expression does not rise or expand in later stages and the decrease in ino levels may indicate some expression in the outer integument. Accordingly, this group contains genes such as AINTEGUMENTA-LIKE 5 (AIL5) and ERECTA-LIKE 2 (ERL2), expressed in placenta, ovule and integument primordia 6768 and L1 specific genes such as PROTODERMAL FACTOR 2 (PDF2) and MERISTEM LAYER 1 (ATML1) expressed in ovule primordia and the L1-derived integuments 6970. In addition, twenty seven embryo sac genes identified either by Yu et al 33 or by Johnston et al 34 were found in this set of genes, which could also be a cause of decreased ino E levels. Cluster 2 also contains PHB, previously shown to be downregulated in ant gynoecia 40, whose slight decrease in expression in the ino arrays could be reflecting post-transcriptional regulation of the mRNA in the outer integument 1471.

The third group, clusters 3, 7 and 11, in which expression seems to increase towards later stages of pistil development, also shows decreases in the ino F arrays, with larger decreases in clusters 3 and 7 (Figure 3A). Such genes are likely to have early carpel or ovule expression as well as later outer integument expression, and remain expressed in later stages, as is seen with the cluster 3 gene, FIDDLEHEAD (FDH) 72, and cluster 11 genes ERL1, which is expressed throughout early carpels and later resolves to expression in the ovules 68, and PRETTY FEW SEEDS (PFS2), expressed in carpel and ovule primordia, and in the chalaza, integument primordia and nucellus 73. These genes have more ovule specific expression and also show higher fold changes, which is likely to be correlated.

In summary, the known genes in this set of 800 genes encompass a wide variety of expression patterns, whose common thread seems to be expression in ovule primordia. On the basis of the clustering results, genes expressed primarily in the inner integument would be expected to have patterns similar those genes in clusters 4, 5, 8, 9, and 10. These groups are also likely to be populated with genes that have expression in placenta and ovule primordia. Any decrease in ino arrays appears to signify a wider expression pattern that includes ovule and integument primordia, and expression that is maintained later in development likely shows that the expression is not limited to primordial cells. With a few exceptions, more specific or prolonged ovule expression leads to higher fold changes between wild type and ant.

The fifty-eight genes that are decreased in both ino and ant to a similar level (Figure 3C) are considered good candidates for expression in the outer integument, and accordingly, the APETALA 3 (AP3) gene, known to be expressed in the outer integument, was identified 74. Similar to the genes described above, lower fold changes could imply general carpel expression combined with elevated or more specific outer integument expression, as seen with the SHP2 gene, also found in this group. SHP2 acts with related genes to specify ovule development, and is also expressed early in carpel development 117576. Mutations in RABBIT EARS (RBE), produce a phenotype in which the growth of the outer integument is aberrant, with the inner integument being affected at later stages 7778. Expression of this gene has been described as being in both integuments 78, but this is not reflected by the measurements on the arrays, which show similar decreases in both ino and ant relative to wild type. There are at least three other uncharacterized transcription factor genes that would be good candidates for activity in the outer integument. These encode an AP2 domain protein, a ZF-HD protein and a myb domain protein.

Putative outer integument genes are also contributed by a comparison of the WT F and ino F arrays, with one hundred forty three genes identified only by these arrays. These are good candidates for being expressed in the later stages of outer integument development, and may be involved in differentiation of the cell layers in preparation for pollen reception or seed coat development. However, genes that are predominantly and strongly expressed in the embryo sac at late stages will share this pattern, as evidenced by the overlap (50 genes, 35%) with putative gametophyte expressed genes 3334. Therefore all the identified genes will require further validation of expression pattern. A total of seventeen putative outer integument genes dropped to very low levels in ino indicating specific expression in the outer integument.

Twenty-five genes exhibited expression patterns expected for expression in both integuments, with ino expression lower than wild type and ant lower than ino (Figure 3B). Most such genes were greater than two-fold changed from wild type to ant, and all showed a reduction in ino F as well as in ino E, with variation in WT L levels. As expected, this group contains the INO gene, which is expressed briefly in the ino mutant 79. Also detected here is the SUPERMAN (SUP) gene, involved in regulation of INO 7980 and integument growth, although expression in integuments has not been observed 818283. For At4g12960, only inner integument expression has been described, at late stages 30, but the array measurements predict expression in both the inner and outer integuments at earlier stages. For this gene and SUP it is possible that the loss of the outer integument affects gene expression in the inner integument.

While all the above genes are candidates for expression in the integuments, only those genes with significant expression changes from wild type in both ino and ant, or those with a 2-fold change level from wild type were examined closely (207 genes). This selection was made based on the observation that known expression patterns that were most specific to ovules had higher fold changes. This leaves 132 putative inner integument genes (Additional file 7), retaining known ovule expressed genes such as BEL1, PFS2, ETTIN (ETT), PAN and AIL5, but excluding genes with wider carpel expression (such as PHB and CUC2), L1 expressed genes, and other placenta and primordia genes such as FDH, MONOPTEROS, ATCEL2 and SPT. The outer integument group is reduced to 50 genes, removing genes such as SHP2 (FC = -1.3) whose expression is not specific to the outer integument, but retaining the AP3 and RBE genes (Additional file 8). The 25 genes that show decreases in both ino and ant are all retained and listed in Additional file 9. The expression profiles of these genes from the arrays are shown in Figure 3, grouped by general expression changes into 'outer integument', 'both integuments' and 'inner integument and primordia expression' groups, and then into subgroups based on analysis information above.

The sets of genes described above were analyzed for their putative functions, as listed at The Arabidopsis Information Resource http://arabidopsis.org, using gene ontology searches http://www.geneontology.org and published literature. Divisions into broad functional classes are shown in Figure 4. The proportions of the different categories vary little between the putative expression groups. The most prominent categories are proteins with unknown function and proteins involved in metabolism. There are also many putative transcription factors and DNA binding proteins (Table 2), which are good candidates for regulators of ovule development. The proportion of transcription factors is approximately 20%, which is higher than estimates for the proportion of transcription factors found in the genome (6–7%) 848586.

Within the set of transcriptional regulators, several families are represented, including different types of Zn finger (9), B3 (5), homeodomain leucine zipper (3), myb (2), bZIP (1), HMG/ARID (2), MADS (1), bHLH (1), YABBY (2), homeodomain (3), ANT-like (2), WRKY (2), ARF (2), ERF (1), TCP (1), and GARP/KANADI (1) proteins 84. These encompass both characterized and uncharacterized proteins, and make good candidates for ovule development regulators. Groups of gene family members form good targets for analysis, as these genes, if they act redundantly in ovule development, would not be found through mutant screens.

Of the five identified B3 domain family proteins, four are part of the reproductive meristem (REM) family 87. Two of the REM genes are very similar to each other (63% amino acid identity) and occur close to each other on Chromosome 5: At5g18000 (REM18) 88 and At5g18090. REM18 is regulated by the ovule identity complex formed by STK, SHP1/2 and SEP 768889.

Most of the seventeen class I homeodomain leucine zipper proteins (HD-ZIP I) are uncharacterized. All 3 members of the δ subclass, ATHB40 (At4g36740), ATHB21 (At2g18850) and ATHB53 (At5g66700), show decreased expression in the mutants. This group of genes is expressed in inflorescences, and is induced by ABA and NaCl treatment in seedlings and ovules 9091. The related subclass ε contains two proteins ATHB51 (At5g03790) and ATHB22 (At2g36610). ATHB51/LATE MERISTEM IDENTITY 1 (LMI1), which was identified as a putative outer integument gene, is activated by LFY in meristems and regulates CAULIFLOWER expression and leaf/bract formation 929394. ATHB22 shows no evidence of expression in this experiment, but the MPSS database 24 shows low expression in inflorescences that drops to near zero in agamous inflorescences, implying carpel or stamen expression. No ovule mutant phenotypes were observed from putative insertional knockouts of ATHB51/LMI1 or ATHB40 (data not shown), and mutant combinations might be required to expose a role in ovule development, possibly as developmental regulators or environmental response factors.

Proteins with TAZ zinc fingers and BTB/POZ protein binding domains were shown to bind calmodulin and the BET class of chromatin binding and modification proteins that contain a bromodomain 95969798. Two of these genes are predicted to be expressed in regions affected in the ant mutant and show almost identical expression profiles, being greater than 4-fold decreased in ant and 2.5-fold decreased in ino. BTB AND TAZ DOMAIN PROTEIN 2 (BT2) (At3g48360) induces telomerase activity in response to auxin 99, while BT5 (At4g37610) has no described function. Another pair of related genes, the HMG ARID transcription factors At1g76110 and At1g04880, are putative chromatin binding proteins 84 and putative inner integument or primordia genes. The specific functions of these genes are not known, and they represent interesting candidate genes for ovule development.

Several genes that encode proteins involved in protein modification and proteolysis were identified in this analysis. These include proteins involved in ubiquitin-mediated proteolysis, as well as RING proteins, protein kinases and proteases. A group of enzymes involved in trehalose metabolism, (4 of 11 trehalose-6-P synthase genes and a trehalase) were also identified. Trehalose synthesis has been shown to affect trichome morphology and plant architecture in Arabidopsis through regulation of cell shape 100, and could be acting similarly in the gynoecium.

The effectiveness of the array methodology and experimental design was evaluated using three approaches, quantitative RT-PCR (qRT-PCR), in situ expression analysis of select genes, and detection of known, expected genes within the integument groups as detailed above.

qRT-PCR was used to obtain an independent assessment of a sample of the microarray results 101, to test for spurious results due to cross hybridization, alternative splicing, or technical problems leading to inaccurate measurement of expression. Twelve genes that represented a range of fold changes and absolute expression levels were tested. The relative expression levels determined by qRT-PCR are compared with fold changes from the arrays in Figure 5, and for all the genes tested the direction of the fold change was confirmed, although there was variability in the magnitude of fold changes. When fold changes were small (< 2) the microarray and qRT-PCR results were more similar than when fold changes were larger. The rankings of genes by fold change did not vary widely between the two methods, so that relative differences in expression levels were also confirmed by the qRT-PCR method.

Analysis of mRNA expression patterns with in situ hybridization tests the predictive value of the expression profiling groups and provides important information for understanding gene function. In situ hybridizations were performed for At3g55560 (AT-HOOK PROTEIN OF GA FEEDBACK 2, AGF2), that encodes an At-hook DNA binding protein 102. This gene showed a 2.7 fold decrease in ant relative to wild type, and a slight decrease (1.5 fold) in ino in the FULL arrays, and was in cluster 2, predicting expression in primordia, medial regions or inner integument with later embryo sac or outer integument expression. In confirmation of this prediction, early expression was seen in the placenta and ovule primordia, as well as the inflorescence meristem and flower primordia, and in the outer integument and distal funiculus of the ovule later. Serial sections indicated that expression was highest in the anlagen and primordia in the outermost 2 to 3 cell layers of flower primordia (Figure 6A). Expression was observed in the floral organ primordia, and persisted in growing carpels, stamens and petals (Figure 6B). The petal expression was highest in the edges of the petals, and expression in the anthers was highest in the center of each locule, prior to pollen formation. After microsporogenesis, expression in the tapetum and pollen decreased and was undetectable at maturity (not shown). In carpels, expression was limited very early to the parietal placental regions, before fusion of the septum (Figure 6C). Expression remained high in the ovule primordia as they formed as protrusions from the placenta (Figure 6D, E), and localized to the distal funiculus and outer integument after integument initiation (Figure 6F). By maturity, expression could not be detected in any part of the ovule (data not shown). A sense probe made from the same construct showed a distinct pattern confined to sporogenous cells, with a high level of expression seen in the tapetum and pollen and in the developing embryo sac (Figure 6G). In addition, MPSS signatures exist in this genomic region for this strand which have a different distribution pattern from the signatures for the coding strand 24.

In situ hybridization was also performed for At5g42630 that had shown a 7-fold decrease in ant and a 3.5-fold decrease in ino relative to wildtype. These array results, that predicted expression in both integuments, were used in combination with a separate map-based cloning effort (that had narrowed the search to fourteen candidates) to identify this gene as ABERRANT TESTA SHAPE (ATS) 103. Full characterization of ATS is published elsewhere 104. The ats mutant is affected in both integuments, and in situ hybridization showed initial expression in both integuments that subsequently resolves to expression in the inner integument, confirming the prediction from the array results (Figure 7G – 7I) 104.

The gene with highest fold change observed between wild type and ino, At2g34700, was categorized as an outer integument gene, (Figure 3C). In situ hybridization confirmed this result, showing that this gene is expressed solely in the outer integument in inflorescences, and likely in the outermost layer of the outer integument (Figure 7A – 7D). MPSS data for this gene indicates that this protein is limited to expression in inflorescences and therefore is likely to be an outer integument-specific gene. The protein product of At2g34700 is similar to the pollen Ole e 1 allergen from olive 105 and is a member of an uncharacterized group of plant-specific proteins that are likely secreted and may act as extensins. Expression was first visible in the basal/proximal region of the outer integument (Figure 7A) and initially localized to only one or two cells (Figure 7B). Expression spread to most cells of the outer cell layer as the outer integument grew (Figure 7C) and close to maturity, expression was retained at high levels in the outer cell layer (Figure 7D).

As the At2g34700 protein is predicted to be secreted, (TargetP 1.1) 106, subcellular localization was investigated using a C-terminal fusion of the GFP coding sequence to a cDNA encoding the At2g34700 protein, under control of the CaMV 35S promoter 107. GFP was observed most strongly in trichomes and guard cells of transformants and in these cells the fluorescence was localized to the cell wall (Figure 7E). Two Ds transposon insertion lines 108 were obtained and lines containing homozygous insertions were identified using PCR. Neither line showed any differences in phenotype from wild type when examined with SEM (data not shown). There are two similar genes (63% amino acid similarity) in the Arabidopsis genome that could be redundant with At2g34700, alternatively, loss of function of this gene may not produce noticeable effects under normal conditions.

Wild type plants were Landsberg erecta (Ler) ecotype, and the ant-4 and ino-1 mutants were in the Ler background. Plants were grown in 24 hour light at 19°C in growth chambers, using a 1:1 mixture of Premier Pro-Mix 'BX' potting soil (Premier Horticulture, Oceanside, CA) and vermiculite. Once germinated, plants were fertilized using a complete nutrient solution once per week 125. All plants used in the array analysis were grown in a single growth chamber and flats of pots were rotated three times per week to different shelves and orientation to help ensure even growth conditions. Pots containing a particular genotype were placed randomly in flats. For each genotype and stage, pistils from more than 20 plants were collected over several days, during the same time period each day (10 AM – 1 PM), and pooled. Pistils were collected into tubes on dry ice and stored at -80°C.

Total RNA was extracted using the Qiagen RNeasy Plant kit (Qiagen, Valencia, CA). aRNA synthesis was performed using the Ambion MessageAmp kit (Ambion, Austin, TX) with biotin-11-CTP (Perkin Elmer, Boston, MA) and biotin-16-UTP (Roche, Indianapolis, IN). aRNA was fragmented following Affymetrix protocol and total RNA, aRNA and fragmented aRNA were checked for fragmentation and purity by gel electrophoresis and absorbance. Samples were hybridized to Affymetrix ATH 1 Genome Arrays (Cat # 900385; Affymetrix, Santa Clara, CA). All arrays were processed by the Core Facility in the UC Davis Department of Medical Microbiology & Immunology using an Affymetrix Hybridization Oven 640, the Affymetrix 450 Fluidics Station, and an Affymetrix GeneChip 3000 Scanner. The array data sets were named for the genotype (WT, ino or ant), replicate number and ovule stage pool (E, F, L). All array hybridization data were deposited in the ArrayExpress database with accession number E-MEXP-1920.

Raw CEL data was generated using MAS 5.0 (Affymetrix). An RMA measure of gene expression was calculated using the affylmGUI (affy and limma) package implemented for the Bioconductor project running in the R environment 5051126. Perfect match values only were used with quantile normalization 127. Raw CEL files were also loaded into dchip (v1.3 release date: 07/20/2005) and were normalized and modeled using the PM-MM and PM-only algorithms. Scatter plots of the log₂ expression measures between a set of replicates processed with different methods and Pearson correlation coefficients were prepared using Excel 2003 (Microsoft, Redmond, WA).

For the statistical tests, affylmGUI was used to compute the moderated t-statistic 4952, and the log fold changes. P-values were adjusted for multiple testing using the Storey q-value method 56. For dchip, the PM-only expression values of sets of replicates were used to compare with other genotypes using the 'compare samples' function, with the following criteria: means separated by at least 20 and a fold change of 1.2 (using the lower bound of 90% confidence interval). SOM cluster analysis was carried out in GeneCluster 2.1.7 (Broad Institute), chosen as the number of likely patterns was low and it was not important to identify sub-clusters. The 12 clusters that result from SOM clustering of 800 inner integument genes are shown in additional file 5.

Genes were considered putatively absent in a sample if the average value was below 12, which was chosen by assessing the values of a set of putatively root specific genes 128 in the pistil samples (additional file 10).

The RNA samples for RT-PCR were subjected to DNase treatment (Promega, Madison, WI) and digestion by two four-base cutter restriction enzymes to ensure complete digestion of any contaminating DNA. Two of the three biological replicates used for the microarrays were used in the reverse transcription and quantitative PCR. 1 μg of each RNA was used in reverse transcription reactions with either 3.75 units of Thermoscript (Invitrogen, Carlsbad, CA) or no reverse transcriptase as a -RT control, which was tested for contaminating genomic DNA in PCR reactions. 2 μl of a 1:40 dilution of the RT reactions were used in each quantitative PCR (qPCR) reaction. Primers were designed using the SYBR Green option of the Beacon 2.0 primer design software (Premier Biosoft, Palo Alto, CA) (Additional file 11). qPCR reactions were carried out using an iCycler (Biorad, Hercules, CA) and the following PCR reaction mix: 20 mM Tris pH 8.4, 50 mM KCL, 3 mM MgCl₂, 4% glycerol, 20 nM fluorescein diacetate, 0.5× BSA (New England Biolabs, Beverly, MA), 1:50 000 diluted SYBR GREEN I (Cambrex Bio Science, Rockland, ME), 0.2 mM each dNTP, 0.24 μM each primer and 0.6 U iTaq (Biorad, Hercules, CA).

The fluorescence threshold at which the cycle number (C_t) was calculated was set at 25 CF RFU (curve-fit relative fluorescence units) for all experiments, close to the automatically determined threshold for each plate. The 60S ribosomal protein RPL14B gene (At4g27090) was used as a reference and showed very similar C_t values (range: 19.17 – 19.70) in all sample types tested. The relative starting quantity of cDNA for a particular gene was determined in GENEX (Biorad) using the following equation: relative quantity = efficiency ^{(control Ct-experimental Ct)} based on Livak 129 and Vandesompele 130. The mean PCR efficiencies of the primer sets were determined using LinRegPCR 131, using a linear regression model.

Plasmids for production of probes for in situ hybridization were constructed as follows. The 1 kb At3g55560 coding region was amplified from Columbia genomic DNA using c55560F1: AGAATGGCGAATCCTTGG and c55560R1: CTAATCAATACGAAGGAGG and cloned into the pCR4-TOPO vector (Invitrogen, Carlsbad, CA) to form pDS148. The At2g34700 cDNA was amplified from the cDNA clone U20928 132 using the primers c34700F3: ATACTAGTAATGGGTCTGGTAACAAAAGCTC adding the restriction site Spe1 and c34700R4ns: ATAGGATCCGTCTTCCAAGAGCACAGGCAGGCTC which removes the STOP codon, and adds the restriction site BamH1. This fragment cut with BamH1 and Spe1 was subcloned into pLitmus28 (New England Biolabs, Beverly, MA) cut with the same enzymes to form pDS137. This fragment was also cut and used in a three-way ligation with pLitmus28 cut with Spe1 and Sac1, and the GFP sequence subcloned from the pGFP1.1.5 plasmid 133 using BamH1 and Sac1. The resulting plasmid, pDS138.3, was digested with Spe1 and Sac1 and the GFP-fusion fragment inserted into pMON999 79 cut with Xba1 and Sac1 to give pDS142. This formed a sequence verified expression cassette using the 35S promoter driving expression of a chimeric gene that encodes a C-terminal fusion of the GFP protein to the At2g34700 protein. This plasmid was tested for transient expression by blasting into onion cells as described previously 134. The resulting fusion protein formed aggregates of protein that were likely not localized correctly. Therefore, this plasmid was subcloned using Not1 sites into a transformation plasmid, pMLBART, forming pDS146. This clone was transferred using three-way mating into the Agrobacterium strain ASE and transformed into wild type Arabidopsis (Ler) plants.

Wild type Ler inflorescences were fixed in FAA: formaldehyde (10%), ethanol (50%) and acetic acid (5%) overnight at 4°C and embedded in Paraplast Xtra (Electron Microscopy Sciences, Philadelphia, USA). Probe preparation and in situ hybridization were performed using a modification 135 of the protocol of Ferrándiz et al. 136. Digoxigenin-labeled RNA probes were prepared from the clones above, linearized with appropriate enzymes and transcribed with T3 or T7 RNA polymerases. As a control for all in situ hybridization experiments, an antisense probe for the INO gene was used simultaneously. The INO hybridizations confirmed the expression pattern reported previously 3579.

Mutant and wildtype pistils were fixed for scanning electron microscopy (SEM) as described 137, using 5% glutaraldehyde with postfixation in 2% osmium tetroxide. Pistils were dissected following critical point drying to allow observation of ovules. SEM images were collected using a Hitachi S3500N microscope and processed using Photoshop v. 7.0.

For callose staining of embryo sacs, wild type and mutant pistils at anthesis were preprocessed by cutting the pistil just below the style, and at the base to allow entry of the stain, and immersed in 65°C 5 M NaOH for 5 minutes. Pistils were rinsed with water, stained with decolorized aniline blue for 2 hours and examined with fluorescence microscopy using a UV laser on a Zeiss (Oberkoche, Germany) Axioplan microscope and images were acquired with a MDS290 digital camera (Kodak, New Haven, CT) and edited in Photoshop v. 7.0 (Adobe, San Jose, CA).

Transgenic plants expressing the At2g34700-GFP fusion protein and wild type non-transgenic plants were examined on an Olympus (Orangeburg, NY) Confocal FV1000 microscope and digital images obtained with the integral Olympus camera and edited in Photoshop v. 7.0 (Adobe, San Jose, CA).