Seed after‐ripening is a discrete developmental pathway associated with specific gene networks in Arabidopsis

1Seed survival following shedding from the mother plant is a key component in the life history of flowering plants. Under natural conditions, seed survival in the soil and cycling through states of dormancy are major ecological characters determining entry and persistence in ecosystems, and seed dormancy is a major trait altered during domestication of wild species. Embryo dormancy is a genetically and environmentally determined developmental state imposed following imbibition of mature seeds, in which cell metabolism is active, but growth processes are repressed. The potential for embryo dormancy is overcome through the time- and environment-sensitive process of after-ripening (AR) that occurs in the dry seed. Germination proceeds through three phases of water uptake: non-dormant seeds successfully complete the transition from phase two to three (radicle elongation), whereas dormant seeds remain within phase two. Following imbibition, the AR status of the seed determines whether individual seeds complete germination or remain dormant. AR therefore represents a key characteristic of seeds in the determination of germination potential, which can be defined as the capacity of seeds to complete germination once imbibed. A recent definition of AR suggested that it is a widening or increased sensitivity of perception of seeds to environmental conditions promoting germination, at the same time as a narrowing or decrease in sensitivity of perception of conditions repressing germination. Thus, environmental conditions that favour germination will not trigger germination at early stages of AR, but may do so following the appropriate AR time. Despite a huge literature describing the physiological aspects of AR, nothing is known about the molecular mechanisms that control AR in the dry seed or about the downstream regulation of gene expression by AR in the imbibed seed.

2Measurement of water contents of unimbibed seeds suggests that cellular environments are compatible with the operation of complex biochemical processes, and that AR rate is affected by seed water content. Biochemical processes can occur in the unimbibed seed during storage, including changes in metabolites, proteome and gene expression, providing possible mechanisms through which the AR process may work.

3A large number of loci that regulate germination potential have been identified through genetic studies in Arabidopsis thaliana. Mutation of these loci range from subtle changes in germination potential and dormancy capacity, to extreme conversion of embryo characteristics into those of seedlings, as in severe abi3, lec and fus3 mutants. Many loci are involved in synthesis or signal transduction of phytohormones, principally gibberellins, ethylene and abscisic acid (ABA; ). Genetic evidence shows that synthesis of ABA during late stages of embryogenesis is required for dormancy. Arabidopsis mutant aba1 seeds, which are deficient in zeaxanthin epoxidase, the first enzyme in ABA-specific biosynthesis, are non-dormant. Recently, it was shown that mutant seeds deficient in ABA 8′-hydroxylase, which catabolizes ABA, show increased dormancy levels, suggesting that ABA breakdown is required to remove dormancy. Signal transduction following ABA synthesis is also important for the repression of germination, and mutants have been identified in Arabidopsis that are either insensitive (e.g. abscisic acid insensitive1, abi1) or hypersensitive (e.g. lipid phosphate phosphatase2, lpp2) in their germination response to applied ABA. A general property of AR seeds is lowered levels of, and decreased sensitivity to, ABA compared with freshly harvested seeds. Application of exogenous ABA has previously been used as an experimental manipulation to mimic dormancy in AR seeds, and it has been shown that ABA added to imbibed AR seeds delays endosperm rupture in Arabidopsis. However, ABA-treated seeds may not be developmentally equivalent to dormant seeds, and studies have shown that these seeds still carry out metabolic processes associated with germination, including fatty acid breakdown.

4In this study we have used a transcriptome profiling approach to analyse AR as a specific developmental process. We have successfully used mutant Arabidopsis seeds to uncouple AR and germination potential by identifying AR-associated and AR-independent sets of genes. The behaviour of these gene sets defines AR as a specific developmental pathway that is separate from germination potential. These observations have implications for the study of genetic mechanisms regulating dormancy in cultivated crops, and for processes controlling seed behaviour during dormancy cycling in natural ecosystems.

5Previous studies have defined AR as a widening or increased sensitivity of perception of seeds to environmental conditions promoting germination, at the same time as a narrowing or decrease in sensitivity of perception of conditions repressing germination. This definition ties the function of dry AR (occurring in the dry seed) to the physiological result of AR (which occurs in the imbibed seed). This makes it impossible to detect changes in imbibed wild-type (WT) seeds resulting from AR-specific functions (rather than those changes that are purely a consequence of being in a physiological state, such as ‘dormant’ or ‘germinating’). We searched for experimental tools to analyse AR per se. Within this study we define ‘germination potential’ as the capacity of seeds to complete germination once imbibed. We reasoned that, in artificially-induced mutants with increased germination potential (that lack dormancy), these processes may be uncoupled, i.e. changes in the dry seed that would lead to the promotion of germination (AR) in WT seeds may still occur, even though mutant seeds have no dormancy. The advantage of this as an experimental tool would be that dry-stored mutant seeds, although exhibiting the same germination potential as freshly harvested mutant seeds, would have gone through the dry AR process. In this way, AR-specific gene sets could be identified as being shared between mutant and WT imbibed seeds either before or following dry storage. Because ABA has been shown to have a profound effect on germination potential, we chose to address this possibility using two mutants with impaired synthesis (aba1-1) or perception (abi1-1) of ABA, which have been shown to have constitutive effects and not to be restricted to a specific stage of development.

6Seeds of WT Arabidopsis accessions, and mutants aba1-1 and abi1-1, were compared for AR capacity and germination potential (Figure 1). Dormancy is removed after appropriate times of storage of WT dry seeds, after which imbibed seed lots complete germination (Figure 1a, accessions Ler and Cvi). Both aba1-1 and abi1-1 freshly harvested (F) seeds were able to complete germination (Figure 1b). Analysis of imbibed seeds over 7 days showed that the velocity of germination was not appreciably different between F and stored (S) mutant seeds, or between AR WT seeds and F or S mutant seeds (Figure 1b). In all cases, no germination was evident at 24 h of imbibition. Therefore at the physiological level, the germination potential of mutant F seeds is indistinguishable from AR WT seeds. However, as discussed above, it is possible that AR could still occur if it were a discrete developmental process, uncoupled from germination potential. Mutant aba1 and abi1 seed therefore provide material to address this question.

7We used transcript profiling to define AR-related gene sets and to determine whether transcriptional events in imbibed seeds associated with AR still occur in mutants with altered germination potential. RNA samples for transcriptome analysis were derived from highly synchronous WT or mutant seed populations with controlled environment histories, and whole genome expression was analysed using the ATH1 Affymetrix Genechip. Transcriptomes of seeds were assayed at 24 h of imbibition, because this captures gene expression in phase II of germination (prior to radicle elongation, Figure 1), which results from the AR status of the dry seed. Sets of genes that showed enhanced expression in specific developmental and mutant backgrounds were identified (Table S1). Whereas for most comparisons 500–1700 genes were differentially regulated, for both aba1 and abi1 F seeds very few genes were differentially regulated in comparison with Ler dormant (D) seeds, suggesting that these seeds are similar at the transcriptome level. Conversely, comparisons of aba1 and abi1 F and S samples showed these to be very different, suggesting that changes occur in these non-dormant mutant seeds under the same conditions that promote AR in D seeds, and therefore that AR may occur in these non-dormant mutant seeds. This observation was confirmed through principal component analysis (PCA) of the datasets (Figure 1c), which showed that mutant ‘S’ and WT AR transcriptomes cluster closely, as do mutant ‘F’ and WT D transcriptomes. Using this approach, we were also able to show that the transcriptome of the F imbibed non-dormant rdo2 mutant also clustered with Ler D. Non-dormant abi3-4 and fus3-8 seeds are physiologically unlike aba1-1 and abi1-1 in that they demonstrate post-germination seedling characteristics at the end of embryo maturation (including vascularization and apical meristem activation). Accordingly, PCA of transcriptomes from F abi3-4 and fus3-8 seeds clustered together, and far from other F, D or S AR samples.

8To distinguish cohorts of genes encoding proteins related to biological processes associated with dormancy and germination, we analysed lists of differentially expressed genes using a novel approach that re-annotates gene lists according to defined functions. This approach, using the TAGGIT workflow, replaces the commonly used gene ontology (GO) annotation, because it provides more useful information related to seed biology. This provides a ‘developmental signature’, tagging that part of the transcriptome with known functions in relation to dormancy and germination (Figure 2). This analysis showed clear differences between D or AR upregulated genes in both Ler and Cvi WT accessions, and demonstrated that characteristic ‘D’ and ‘AR’ signatures were easily distinguishable (Figure 2a, derived from Table S1). Using this approach, it was possible to show that signatures of genes upregulated in F compared with S seeds (F > S) of both aba1 and abi1 more closely resembled those of the WT D > AR differentially upregulated gene group, and that signatures of aba1 and abi1 S > F resembled those of WT AR > D (Figure 2b).

9Cohorts of genes that were AR-associated or expressed independently of AR were obtained from comparisons of differentially regulated genes (Figure 3). Comparison of genes from F, D or S AR differentially expressed gene sets (Table S1) identified 15 of which expression was AR-independent and associated solely with germination potential (‘Germination’, Figure 3a; Table S2), 19 genes common to all S-AR sets (AR upregulated, Figure 3b, which includes the gene LPP2 that removes sensitivity to ABA during germination) and 103 common to F-D sets (AR downregulated, Figure 3c, which includes ABA1, involved in ABA synthesis). The expression patterns of representative members of each set was measured in WT and mutant seeds from 0–48 h of imbibition using quantitative (Q)RT-PCR (Figure S2). For each gene, similar biases in expression in F and S seeds were seen to those observed in transcriptome experiments.

10In order to further examine the robustness of expression characteristics for these gene sets, we also analysed global gene expression in WT accession Cvi. Comparison of transcriptomes of AR and D Cvi seeds showed large changes in expression between the two physiological states at 24 h of imbibition (Table S1). To compare expression properties of the AR-associated and AR-independent gene sets, we derived the average ratio of expression levels of each gene set, comparing either different physiological states or different genetic backgrounds (Figure 4a, Figure S4). Comparison of the D with the AR state in the WT accessions Ler and Cvi revealed very similar average ratios of expression of each gene set. Ratios of expression for the genes ABA1 and LPP2 were consistent with their placement in ‘AR-downregulated’ and ‘AR-upregulated’ sets, respectively.

11Treatment of AR WT seeds with exogenous ABA greatly reduced germination potential over the time period taken by untreated seeds to complete germination (Figure 1a) (although ABA-treated seeds do eventually germinate; ). However, if AR and dormancy are independent developmental processes, manipulation of the germination potential of AR seeds by application of ABA should not influence AR-related gene expression. Transcriptome analysis showed that many genes were differentially regulated between ABA-treated and untreated WT AR seeds at 24 h of imbibition in comparison with WT D and AR seeds (Table S1). Twice as many genes were differentially expressed when comparing ABA-treated AR seeds with D seeds, than when comparing AR ABA-treated seeds with untreated AR seeds, suggesting that ABA-treated AR seeds more closely resemble AR seeds than D imbibed seeds. PCA analysis demonstrated that the transcriptome of ABA-treated AR seeds more closely resembled untreated AR than D seeds (Figure 1c). Assay of functional representation by developmental signature analysis demonstrated that the (AR + ABA) > D differentially regulated gene set resembled the AR > D set, including augmented representation of genes associated with protein degradation, cell cycle and translation (Figure 2c). In comparison, the signature of (AR + ABA) > AR set more closely resembled the D > AR set, suggesting that ABA does induce transcripts associated with dormancy, although less than 10% of specific genes were shared. Analysis of the average ratios of expression of the three gene sets demonstrated that, in comparison with expression in the D state, addition of ABA to AR seeds did not significantly alter expression of any of the gene sets (Figure 4a). Quantitative RT-PCR examining the dynamics of expression of members of these gene sets confirmed that ABA treatment of AR seeds did not result in D-like expression (Figure 4b). Expression of RD29B (AR downregulated) was upregulated by ABA at 48 h of imbibition, as has previously been shown for this ABA-inducible gene.

12In addition to effects of exogenous application of hormones, environmental conditions of the imbibed seed, in particular light, can radically influence germination potential, but would not be expected to influence AR-associated gene expression, as this is a post-imbibition associated phenomenon. Treatment with red light is a common mechanism for breaking some types of dormancy. showed that imbibed AR Arabidopsis Cvi accession seeds would not germinate in the dark (DL, light requiring), but would do so following a 4-h red light treatment (LIG, light induced to germinate). Therefore, DL and LIG seeds have completed AR, but have opposite germination potential. We used the transcriptome datasets of to analyse the behaviour of the gene sets identified here. As both studies assayed transcriptomes of unchilled seeds imbibed on water–agarose at 24 h (mid phase II), direct comparison of datasets was possible. Analysis of developmental signatures showed that the LIG > PD24 (PD24, 24-h imbibed primary D seeds) and DL > PD24 differentially regulated gene sets were highly similar (Figure 2d), and virtually indistinguishable from the CviAR > D set (Figure 2a), despite exhibiting opposite germination potential. In addition, average ratios of expression of ‘AR-upregulated’ and ‘AR-downregulated’ gene sets were similar in both PD24/DL and PD24/LIG comparisons (Figure 5).

13Manipulation of the environment of imbibed AR seeds can lead to the induction of secondary dormancy. Induction of secondary dormancy would be predicted to reset the expression characteristics of AR-regulated genes, because in the absence of dormancy breakage in the imbibed state, these seeds would require dry AR to complete germination. The transcriptomes of two secondary dormant states in Cvi have been reported by. We analysed these datasets to test whether the expression characteristics of AR gene sets identified in the present study are indeed altered by secondary dormancy states (Figure 5). Expression in different D states (30 days imbibed primary D seeds, PD30; secondary dormancy states 1 and 2, SD1 and SD2) was compared with expression in the two physiologically-different AR states (DL and LIG). Analysis of the average ratio of expression of each gene set showed that, for the ‘AR-upregulated’ set, expression was reduced in D seeds, whereas expression of the ‘AR-downregulated’ set was enhanced (Figure 5, Figure S4b). This shows that gene sets behaved in a similar way in all physiological states. These results show that, following AR of the dry seed, the expression of these gene sets can be ‘reset’ if the SD status of the imbibed seed is changed. Analysis of ratios of expression of ABA1 and LPP2 showed that these maintained an AR-related pattern.

14At present, little is known about the molecular mechanisms that regulate the process of AR, or about possible downstream target genes in imbibed seeds. More generally, nothing is known about AR as a discrete process in isolation from dormancy and germination potential. In this paper we used a targeted transcriptomics approach to define dry seed AR as a distinct developmental process, and to identify gene sets that were associated with AR or in which expression was dependent only on germination potential. A model describing the regulation of AR, AR-regulated gene expression and the relationship of AR to ABA is presented (Figure 6a,b).

15Expression characteristics of gene sets have allowed us to make and test specific predictions relating to the regulation of AR, and the relationship of AR to seed ABA biology. One important feature of AR-regulated gene expression is that it should not simply be related to germination potential. Using aba1-1 and abi1-1 mutant seeds, which have a constant high germination potential regardless of dry seed AR time, we showed that genome expression in imbibed seeds changed with dry seed storage time. PCA and analysis of developmental signatures both indicated that gene expression changes resulting from dry storage in each mutant were similar to those in WT seeds, suggesting a common cohort of AR-regulated genes in imbibed seeds. Three categories of gene sets were defined: those that showed positive or negative AR-associated expression, and those in which expression appeared to be independent of AR but dependent on germination potential. Exogenously applied ABA greatly reduced germination potential of AR WT seeds, but did not appreciably alter expression of AR-related gene sets. However, 80% of the germination-associated gene set was present in the AR > AR + ABA list, demonstrating the dependence of this gene set on germination. In particular, the ‘AR-downregulated’ set was not induced in ABA-treated AR seeds, even though a developmental signature similar to that of D seeds was evident. This indicates that ABA-treated AR seeds are in a different physiological state to D seeds (Figure 6). Previous work has suggested that metabolic and physiological processes in ABA-treated seeds are not equivalent to those of D seeds, and therefore we conclude that AR seeds treated with exogenous ABA should not be used as a method to study the relationship between ABA biology and dormancy or AR. The observation that aba1-1 and abi1-1 mutant seeds do after-ripen demonstrates that AR is a process that can occur independently of ABA (Figure 6a), although this does not rule out the existence of a parallel ABA-dependent pathway. Two recent studies also defined D- and AR-related gene sets, based on expression profiles of genes in Arabidopsis WT accession Cvi subjected to different environmental stimuli. Those studies are different from that presented here, as manipulation of WT dormancy levels does not divorce AR from the potential of the seeds to germinate. Nevertheless, approximately half of the ‘AR-downregulated’ gene set are represented in the D-related gene sets, and one third of the ‘AR-upregulated’ gene set are represented in the AR-related gene sets, respectively (data not shown).

16Our hypothesis predicts that subjecting AR seeds to environmental conditions that alter germination potential should not influence the expression of AR-related gene sets. In agreement with this, analysis of the transcriptome datasets of using the TAGGIT methodology showed that the AR-associated gene sets maintained predictable expression patterns, independent of germination potential (Figure 6). It is also predicted that imposition of secondary dormancy should ‘reimpose’ an AR requirement on unimbibed seeds, and thereby ‘reset’ expression of these gene sets. Analysis of the transcriptome datasets of showed that this was the case (Figure 5), and induction of two secondary dormancy states led to the restoration of AR-related gene expression characteristics (i.e. reduced expression for the ‘AR-upregulated’ set, and enhanced expression for the ‘AR-downregulated’ set).

17As our approach focused on only one time point of analysis, and one approach to the identification of AR-regulated genes, it is likely that the gene sets identified here are only subsets of potential AR-associated genes. In addition, subtle differences in timing of germination/imbibition would alter observed gene expression levels at the one time point assayed, and this may partially explain why only a few genes are represented in the AR-related gene sets, and many others are excluded (Figure 3). The observation that expression of many genes conforms to an AR-regulated expression pattern provides the conceptual basis to design molecular tools with which to dissect the AR process, independent of the effects of D or germination potential. Several reports have defined quantitative trait loci (QTL) in Arabidopsis that may be candidates for loci that regulate AR, and introgression of these QTLs into suitable mutant backgrounds (such as aba1 and abi1) may provide materials to analyse the molecular mechanisms regulating AR. It is also important to note that transcriptional changes have also been observed in ‘dry’ seeds. The changes observed here in imbibed seeds at 24 h do not necessarily reflect changes in gene expression in dry seeds over storage. More analysis is required to understand the relationships between changes in gene expression in dry and imbibed seed states.

18Using previously published datasets it was possible to determine several additional expression properties of genes within sets, including tissue localization within the seed. Members of the ‘AR-upregulated’ gene set showed enhanced expression in the embryo, whereas members of the ‘AR-downregulated’ and AR-independent ‘germination’ gene sets showed enhanced expression in the endosperm (Figure S3; ). These observations may indicate different functions of these subcomponents of the seed in relation to the regulation of AR or germination potential-responsive genes.

19The behaviour of the different gene functional classes observed in WT and mutant seeds in response to AR agree with previously published seed transcriptomics studies. For example, before AR, many genes associated with late embryogenesis (including heat-shock proteins and storage proteins) are evident, whereas afterwards genes associated with functions including RNA translation, cell-wall modification and protein degradation predominate (Figure 2a). Although our data indicate that AR can occur independently of ABA, they are nevertheless consistent with a role for ABA in AR-mediated processes, as the ‘AR downregulated’ and ‘AR upregulated’ gene sets included members known to control mature seed dormancy and ABA responsiveness. Specifically, the ‘AR-downregulated’ set includes ABA1, involved in ABA synthesis, and the ‘AR-upregulated’ set includes LPP2, which negatively regulates ABA responsiveness. It is possible, therefore, that this provides a molecular mechanism for AR-mediated processes, such that in imbibed seeds that have not undergone AR, accumulation of ABA1 transcripts results in the maintenance of dormancy through increased ABA synthesis, and in AR seeds, accumulation of LPP2 removes sensitivity of the seed to ABA, allowing completion of germination (Figure 6b). The 121 genes identified in the D > aba1-1 F gene set (Table S1) represent downstream genes that are ABA responsive and may contribute to dormancy maintenance. Interestingly the ‘ABA-responsive’ gene RD29B is also part of the ‘AR-downregulated’ set, which may suggest that it is an early target of increased ABA1 function (Figure 6b). Although the results presented here indicate that ABA is not a requirement for dry AR regulation of gene expression (Figures 1c and 2b), other ABA-associated processes (including seed dormancy and drought responses) are known to have both ABA-dependent and ABA-independent components. Therefore, we cannot rule out the possibility that there may be a small influence of ABA on AR, but the observation of similarities in transcriptomes between F and D WT and mutant seeds, and similarities in types of functional groups observed in these transcriptome datasets, argues strongly that ABA at most must have a very small role in dry seed AR, as opposed to its major role in dormancy of the imbibed seed.

20The uncovering of AR-regulated gene sets has implications for the study of dormancy cycling in the soil seed bank, and for mechanisms influencing the domestication of crop plants via the removal of seed dormancy. Arabidopsis is a winter annual that germinates mainly in autumn, and exposure to high summer temperatures leads to a loss of dormancy. If germination is prevented, seeds subsequently enter a dormancy cycling process that can last several years. As seeds in the soil can exist in degrees of hydrated states, those entering dormancy cycling in natural environments would need AR to occur in the subsequent unimbibed state to remove dormancy and allow germination to be completed. We have shown that secondary dormancy resets the expression of AR-regulated gene sets, and so it is possible that this resetting provides the mechanism to allow dormancy cycling to continue through multiple iterations in the soil, by preventing the completion of germination via the action of the ‘AR-downregulated’ gene set, which includes ABA1 as well as genes encoding stress-related proteins [including several heat-shock proteins, ERD1, RD29B] and several dormancy-associated genes (including DRM1; ). It is possible, therefore, that this provides a molecular explanation for AR-related mechanisms that regulate dormancy maintenance in seed populations cycling in the soil bank, as ABA synthesis and perception have been shown to be important in the maintenance of the imbibed dormant state.

21Loss of the capacity for seed dormancy is one of the major traits acted on by evolutionary processes during the introduction of wild plant species into agriculture (the ‘domestication syndrome’). There are two possible mechanisms that could lead to reduced seed dormancy in domesticated crop species: removal of the capacity for dormancy (for example by crippling the functions of genes such as ABA1) or reduction in the AR dependence of seeds. These are two distinct molecular processes occurring in two different developmental pathways. By defining gene sets that are AR regulated it may now be possible to distinguish between the contributions of these two mechanisms, for example through the identification of molecular markers or transcriptome developmental signatures that can be used to differentiate crop varieties with differing D/AR characteristics. As a lack of dormancy can be a major agronomic problem (for example in the disorder pre-harvest sprouting in wheat and barley), such approaches may provide the tools to deliver improved AR and dormancy characteristics.

22Wild-type accession Cvi and mutant seeds fus3-8, rdo2 and abi3-4 were provided by Prof. M. Koornneef (MPI, http://www.mpiz-koeln.mpg.de); Ler, abi1-1, aba1-1 were obtained from NASC (http://arabidopsis.info). All mutants are in the Ler background.

23Arabidopsis WT and mutant plants were grown as previously described. Dry seeds were harvested from main bolt siliques and maintained under constant storage conditions (24°C in the dark). Germination assays were carried out on water agarose under constant light; germination was assessed as endosperm rupture by the elongating radicle over 7 days. The AR status of dry seed lots was determined by an assay of the germination potential of appropriate WT accessions. AR was considered to be complete following a time period that allowed WT accessions to achieve 100% germination after 7 days of imbibition. Freshly harvested seed lots were assayed when siliques were yellow, but not completely dehisced; stored seed lots of mutants were assayed following storage for time periods that at least matched those required for complete AR of WT accessions. Under the storage conditions used here, Ler seeds completely after-ripen within 60 days (Figure 1a). All seed lots were dry stored in the same cabinet in the dark at 24°C in sealed glassine bags. Where appropriate, exogenous (+/–) ABA (Sigma-Aldrich, http://www.sigmaaldrich.com), was included in media at a concentration of 10 μm.

24All experiments analysing RNA expression levels were carried out using three replicates of seed materials, obtained from three independent seed lots, from plants grown under controlled-environment conditions. Total RNA was extracted from seed material and real-time quantitative PCR was carried out using the Universal ProbeLibrary system (Roche Diagnostics, http://www.roche.com), with primers and probes for each gene assay designed using the Universal ProbeLibrary Assay Design Center (Roche Applied Science) as previously described. Transcriptome analyses were carried out using the NASC Affymetrix ATH1 genechip service. All array information reported here is available online from NASC arrays (http://affymetrix.arabidopsis.info/narrays/experimentbrowse.pl). Gene expression levels were measured using the microarray suite 5.0 software (Affymetrix, http://www.affymetrix.com), and further analysis was performed with genespring 7.3 software (Agilent, http://www.agilent.com) using a twofold differential expression between samples (P value < 0.05 in a one-way anova) to define differentially regulated genes. Normalized expression values from microarray datasets were used to obtain ratios of expression. Developmental signatures were obtained using the TAGGIT workflow within Microsoft Excel (http://www.microsoft.com).