Gene expression profiles of Arabidopsis Cvi seeds during dormancy cycling indicate a common underlying dormancy control mechanism



This article is corrected by:

  1. Errata: Corrigendum Volume 47, Issue 1, 164, Article first published online: 12 June 2006

(fax +44 024 7657 4500; e-mail


Physiologically dormant seeds, like those of Arabidopsis, will cycle through dormant states as seasons change until the environment is favourable for seedling establishment. This phenomenon is widespread in the plant kingdom, but has not been studied at the molecular level. Full-genome microarrays were used for a global transcript analysis of Arabidopsis thaliana (accession Cvi) seeds in a range of dormant and dry after-ripened states during cycling. Principal component analysis of the expression patterns observed showed that they differed in newly imbibed primary dormant seeds, as commonly used in experimental studies, compared with those in the maintained primary and secondary dormant states that exist during cycling. Dormant and after-ripened seeds appear to have equally active although distinct gene expression programmes, dormant seeds having greatly reduced gene expression associated with protein synthesis, potentially controlling the completion of germination. A core set of 442 genes were identified that had higher expression in all dormant states compared with after-ripened states. Abscisic acid (ABA) responsive elements were significantly over-represented in this set of genes the expression of which was enhanced when multiple copies of the elements were present. ABA regulation of dormancy was further supported by expression patterns of key genes in ABA synthesis/catabolism, and dormancy loss in the presence of fluridone. The data support an ABA–gibberelic acid hormone balance mechanism controlling cycling through dormant states that depends on synthetic and catabolic pathways of both hormones. Many of the most highly expressed genes in dormant states were stress-related even in the absence of abiotic stress, indicating that ABA, stress and dormancy responses overlap significantly at the transcriptome level.


The seed phase is arguably the most important stage of the higher plant life cycle, ensuring species survival. Most seeds are well equipped to survive extended periods of unfavourable conditions before germination and to establish plants under the most favourable conditions. At least in part, this ability stems from the evolution of a wide range of dormancy mechanisms, one of which is physiological dormancy. In this state seeds temporarily fail to germinate in conditions that might be adequate for germination. Essentially, the breaking of physiological dormancy is a reversible process by which seeds can become fully dormant again (secondary dormancy, SD) when the conditions for germination are not favourable (Figure 1). A reversible block to progress enables cycling through different depths of dormancy until the environment is favourable for seedling establishment (Baskin and Baskin, 1998; Vleeshouwers et al., 1995).

Figure 1.

 Current model for transitions between dormant and after-ripened states in physiologically dormant Arabidopsis accession Cvi seeds.
Primary dormancy (PD) is acquired during seed development. A set of environmental conditions (E1), e.g. dry after-ripening or moist chilling, is required to reduce dormancy. The after-ripened (DL) seeds will germinate in the right set of environmental conditions (E2), e.g. light and optimal temperature. When E2 is not (fully) present, seeds will enter secondary dormancy (SD). SD may be broken in the right set of conditions to enter DL again. This cycle may be repeated. Note: since E1≠ E2, the breaking of PD can be studied separately from the processes of germination by withholding light (i.e. the DL state).

Physiological seed dormancy is present throughout the plant kingdom and has a profound impact on the structure and development of plant communities across all major climatic regions. As a consequence the induction and loss of dormancy are triggered by divergent environmental cues activated through many different physiological mechanisms. Even within the same species, several cues and physiological mechanisms may operate with differing importance in distinct accessions. Yet, the function of dormancy is remarkably similar both within and across species, i.e. to spread germination across time, but in synchrony with seasonal cycles to avoid unfavourable weather, to maximize competitive advantages and to ensure the establishment of plants (Baskin and Baskin, 1998). However, whether there is a common underlying molecular mechanism for this well-conserved behaviour is a question yet to be answered.

To date, the emphasis of dormancy research has been almost entirely on primary dormancy (PD; Figure 1) which is acquired during seed development (Hilhorst, 1995). The intensity of PD in a species seems to depend on the environment during seed development, in addition to physiological characteristics imposed by the mother plant (Bewley and Black, 1994). For this reason, SD of seeds in the soil seed bank may differ from PD. On the other hand, differences may be purely chronological. The question of whether PD and SD differ physiologically or at the molecular level remains unresolved.

There is extensive disagreement in the seed biology community as to both how dormancy should be defined and which factors control dormancy induction, maintenance and breaking (Vleeshouwers et al., 1995). The effects of, for example, light and temperature on dormancy are difficult to separate from their effects on the process of germination itself, as dormancy can only be defined post-facto by germination (Murdoch and Ellis, 2000). This apparent continuum between dormancy and germination that exists in many species is a major constraint to studying dormancy (Cohn, 1996). This is particularly true in crop species where dormancy is minimal as a result of intensive breeding and may occur only under stress conditions or certain growing conditions. Thus, it is essential to have a well-characterized and unambiguous experimental system. It has been argued that wild species, in particular annual species which have seeds that persist in the soil seed bank, are better subjects than domesticated crops and have particular features that make them suitable as model species for physiological studies of seed dormancy (see Cohn, 1996; Hilhorst, 1997). For example, seeds of Sisymbrium officinale have an absolute dependence on light and nitrate to terminate dormancy and initiate germination, and this provides a tool to separate the state of dormancy from germination (reviewed by Hilhorst, 1997).

Arabidopsis thaliana is another annual soil seed bank species (Baskin and Baskin, 1972) with well-known advantages for molecular studies. However, the large variation in (primary) dormancy level of the seeds and the relatively high rate of dry after-ripening (henceforth termed after-ripening), leading to rapid loss of dormancy in commonly studied accessions (Derkx and Karssen, 1993), result in a similar lack of distinction between dormancy loss and germination to that exhibited by crop plants. In contrast, the Arabidopsis accession Cvi originating from the Cape Verdi Islands has deeper seed dormancy than many other accessions (Ali-Rachedi et al., 2004; Koornneef et al., 2000) and has been used in quantitative trait locus (QTL) mapping analysis to identify loci associated with seed dormancy (Alonso-Blanco et al., 2003).

We show that seeds of Cvi also have a dependence on light to terminate dormancy and initiate germination like that of S. officinale, and cycle through different depths of dormancy unless exposed to light (Figure 1). As different dormant states can be achieved uniformly in the whole seed population, this accession provides an ideal experimental system for transcriptional analysis of dormancy maintenance and cycling to address the questions raised above.

In the present work we identify these dormant states in A. thaliana Cvi, induced by different conditions, and contrast them to fully after-ripened seeds both with and without a short (4 h) exposure to red light. The dormant states include primary dormancy (PD), secondary dormancy following after-ripening (SD1), and secondary dormancy after a subsequent cycle of dormancy breaking and induction under different environmental conditions (SD2). A distinction is also drawn between PD seeds imbibed for short periods after harvest, as in almost all published molecular studies of dormancy, and those remaining imbibed in the dormant state for a prolonged period (30 days). These situations arguably represent the establishment and maintenance, respectively, of the PD state. Transcriptional arrays are used to develop a global view of gene expression across dormant states and to question whether these dormant states are fundamentally different or whether there is evidence of a common underlying molecular mechanism.

Results and discussion

Seed dormancy in Cvi can be relieved and induced by a range of different environmental conditions during dormancy cycling

Seasonal flushes of seedling emergence are a common phenomenon in temperate regions, for example, species that originate from climates with hot dry summers and cool humid winters, such as A. thaliana, mainly germinate in autumn. Such flushes are strongly promoted by disturbance of the soil which terminates dormancy and stimulates germination of seeds by exposing them to light, but the timing of flushes is largely governed by changes in the depth of seed dormancy. Dose–response analysis of factors involved in germination and dormancy of A. thaliana seeds shows that annual dormancy cycling is controlled by temperature (Probert, 1992) and associated with changes in sensitivity to a range of environmental factors that terminate dormancy and therefore initiate germination (e.g. light and nitrate, Hilhorst et al., 1996). Experiments were carried out in the present work to characterize the response of A. thaliana Cvi seeds to different treatments known from previous work to influence dormancy status and germination in other accessions. The purpose was to identify treatments that would produce the most uniform populations of seeds in contrasting dormant states for subsequent transcriptome analysis. Seed dormancy status was assessed by germination at 20°C in the presence or absence of light and nitrate. A number of these experiments were repeated to confirm the selected treatments (Table 1), and for brevity only these data are presented (Figure 2).

Table 1.   Summary of treatments (states of dormancy) used to produce samples for transcriptome analysis. Experimental treatments were conducted at 20°C unless otherwise stated. Percentage germination shown was recorded in germination tests carried out at 20°C in white light [with and without KNO3 (10 mm)] on the samples used for transcriptome analysis. Germination in the dark was not more than 5% following any treatment except LIG
AbbreviationTreatmentPercentage germination (SE)
+ light+ light + KNO3
PD24hPrimary dormant: seeds imbibed for 24 h in the dark12 (3.2)91 (3.4)
PD48hPrimary dormant: seeds imbibed for 48 h in the dark12 (3.2)91 (3.4)
PD30dPrimary dormant: seeds imbibed for 30 days in the dark2 (1.2)100 (0)
DLLight requiring: seeds dry after-ripened for 120 days and then imbibed for 24 h in the dark (Figure 2a)94 (0.3)98 (1.2)
LIGLight induced to germinate: seeds dry after-ripened for 120 days, imbibed for 20 h in the dark and then for 4 h in red light to terminate dormancy and induce germination94 (1.7)97 (1.3)
SD1Secondary dormant: DL seeds imbibed in the dark for a further 24 days (sensitive to nitrate) (Figure 2b)3 (2.0)100 (0)
SD2Secondary dormant: SD1 seeds imbibed at 3°C in the dark for 20 days (insensitive to nitrate) (Figure 2c)1 (0.5)0 (0)
Figure 2.

 Germination of A. thaliana accession Cvi seeds, with (filled squares) and without (filled circles) nitrate (10-mM KNO3) in continuous white light at 20°C, as an indication of dormancy status during treatment.
(a) The ability to germinate in the light without nitrate increases in primary dormant (PD) seeds as they are after-ripening (dark storage at 33% equilibrium relative humidity and 20°C). Here and in panels (b) and (c), vertical bars are standard errors.
(b) Moist incubation of after-ripened seeds in the dark at 20°C removes the ability to germinate subsequently in the light as they become secondary dormant (SD1, one cycle through DL); sensitivity to light + nitrate is lost more slowly.
(c) The ability of a proportion of SD1 seeds to germinate in light and in light + nitrate is regained (i.e. they become DL) during moist incubation in the dark at 4°C, but the ability to germinate in both conditions is then lost as the seeds become dormant (SD2, two cycles through DL) with continued exposure to 4°C.
(d) A cartoon summarizing the relationship between the physiological states sampled for transcriptome analysis (Table 1). Germination in the dark remained less than 5% during all treatments.

Seeds would not germinate in the absence of light in any of the treatments studied. In general, primary dormant (PD) seeds would not germinate in the presence of light or nitrate individually, but would germinate slowly when both light and nitrate were present (Figure 2a). During after-ripening of the PD seeds at 20°C an increasing proportion of the seed population became able to germinate in light without nitrate, reaching a maximum at c. 100 days. These seeds were considered to be fully after-ripened (DL). Loss of dormancy followed a similar time-scale to that shown by Ali-Rachedi et al. (2004) and Alonso-Blanco et al. (2003) for the accession Cvi.

After-ripened seeds that were imbibed at 20°C in the dark, and therefore not stimulated to germinate by light, progressively lost the ability to germinate when subsequently transferred into the light in the absence of nitrate, so that by c. 20 days they became secondary dormant (SD1; Figure 2b). In this treatment the seed response to light plus nitrate decreased at a lower rate so that by c. 80 days seeds could not germinate in the presence of both light and nitrate (Figure 2b).

When SD1 seeds were kept moist at 4°C in the dark, germination of a proportion of the seed population became possible in the absence of nitrate when stimulated by light (Figure 2c). The proportion of the population that became able to germinate differed in experimental repeats, but generally reached a maximum of less than 50% after 2 days. However, in all experimental repeats further exposure to 4°C decreased sensitivity to both light and nitrate so that by 6 days, germination would not take place even when both were present. These seeds were considered to be secondary dormant (SD2), but in contrast to SD1 would not germinate in the presence of nitrate.

From these experiments, treatments were identified to provide seeds in six contrasting uniform states during dormancy cycling for comparison (Table 1). Further experiments were carried out to develop a seventh treatment in which dormancy was terminated by light (LIG), and, although germination was initiated, seeds would not be undergoing downstream gene expression patterns in preparation for radicle emergence. Dark imbibition of DL seeds was interrupted by exposure to red light for different periods after different lengths of time and then germination in the dark was recorded (data not shown). The treatment chosen was a 4-h exposure to red light after 20-h imbibition in the dark. First radicle emergence then occurred after a further 36 h at 20°C in the dark. Percentage germination on nitrate in the dark after this treatment was 91 ± 3.8% compared with 98 ± 0.8% in continuous white light. Samples taken immediately after this treatment (LIG; 20-h dark plus 4-h red light imbibition) could therefore be compared directly to PD and DL seeds imbibed for 24 h in the dark (Table 1). The relationship between dormant (PD24h, PD48h, PD30d, SD1 and SD2) and after-ripened (DL, LIG) states (Table 1) is summarized in Figure 2(d).

Secondary dormant SD2 seeds that had lost sensitivity to both light and nitrate were kept moist at both 20°C and 4°C for more than 120 days. Germination was tested at approximately 30-day intervals, but germination was not stimulated by any combination of light, nitrate or alternating temperatures (5:25°C; 12:12 h). However, the seeds were fully viable throughout the range of treatments described above and were stimulated to germinate at 118 days by gibberelic acid (GA)4/7 (100% germination) and fluridone (50% germination), and a mixture of both GA4/7 and fluridone (100% germination). Fluridone is an inhibitor of carotenoid biosynthesis (Bartels and Watson, 1978) and therefore blocks the production of ABA. This result therefore indicates that synthesis of ABA is necessary for the maintenance of dormancy, an observation that is discussed further below. Germination began earlier on fluridone alone and GA4/7 + fluridone (<8 days) than on GA4/7 alone (>30 days).

The pattern of transcription differs between newly imbibed primary dormant seeds and those maintained in an imbibed state

To question whether different dormant states are fundamentally similar, microarray analyses were carried out to characterize gene expression of different physiological states. Principal component analysis (PCA) was applied to the expression of all the genes represented on the microarray, over the seven different physiological states (Figure 3a). The analysis revealed clear similarities and differences between the states. The first dimension of the analysis (X-component) grouped all the dormant states (PD24h, PD48h, PD30d, SD1, SD2) together and separate from the after-ripened (DL, LIG) states in line with the breaking of dormancy. The second dimension (Y-component) identified similarities between PD24h and PD48h (Group 1) and between PD30d, SD1 and SD2 (Group 2), but differences between these two groups of dormant states suggesting different depths of dormancy. These first two components of the analysis explain 57.5% of the variance in gene expression and show a clear distinction between three groups of expression patterns: Group 1 and Group 2 represent dormant states and Group 3 the after-ripened states. Group 1 states contain the genes related to PD as a developmental event and probably represent the establishment of dormancy in the newly imbibed seeds which still retain transcripts stored while on the mother plant. Nakabayashi et al. (2005) show that stored transcripts from more than 12 000 genes, including all ontological categories, were present in dry Arabidopsis seeds [accession Columbia (Col)]; however, embryogenic genes were thought to be down-regulated within 6 h of imbibition. Group 2 states contain the genes associated with forms of dormancy resulting from prolonged incubation of seeds under conditions that induce or maintain dormancy. Group 3 represents expression of genes in after-ripened seeds when only light is required to terminate dormancy and at the initiation of germination following exposure to light. These three groups of physiological states were adopted in the subsequent comparisons of gene expression below to question how these dormant states differed and how they differed from the after-ripened state.

Figure 3.

 The results of PCA applied to the expression of all the genes represented on the microarray.
(a) The results of PCA over the seven different physiological states.
(b) The results of PCA with the further addition of unstratified dry Arabidopsis accession Col seeds (ColDry) and the same seeds imbibed for 24 h in the light (Col24h). The latter data for the Col accession were published by Nakabayashi et al. (2005).

The grouping of states in the PCA provides a first indication that the state of dormancy used in many molecular studies (i.e. PD seeds imbibed for short periods, e.g. 24 and 48 h) is dissimilar to the maintained (prolonged imbibition of PD seeds) and SD states that exist in natural conditions, which may have a common underlying mechanism. Although the short-imbibed and long-imbibed PD states are grouped separately, they follow directly in time. Large numbers of genes are differentially transcribed (3790 genes) between 24-h-imbibed PD seeds and 30-day-imbibed PD seeds of which 1833 genes are more highly expressed in the latter. These changes in expression indicate that the dormant state is far from transcriptionally quiescent and that the transcripts present are not merely those accumulated during development on the mother plant.

A further PCA analysis was carried out to include data for unstratified dry Arabidopsis seeds of the relatively non-dormant accession Col (Coldry) and the same seeds imbibed for 24 h in the light (Col24h) published by Nakabayashi et al. (2005) (Figure 3b). In the first dimension (X-component) of the analysis, which accounted for 29% of the variance, Coldry grouped with the dormant states of accession Cvi and Col24h grouped with the after-ripened states of Cvi. The latter result indicates the importance of using more dormant accessions for the specific study of dormancy. The grouping of Coldry with dormant Cvi states is likely to have resulted from the high proportion of stored transcripts remaining from development on the mother plant where developmental arrest is required to avoid viviparous germination. The second dimension (Y-component) of this analysis, which accounted for a further 23% of the variance, separated the two Col data sets from the Cvi data reflecting genetic differences between the two accessions that make direct comparisons of data from different accessions more difficult.

Genes related to metabolism and energy are more highly expressed during the establishment of dormancy compared with maintained dormant states

Similarities and differences in gene expression between the two groups of dormant states (Group 1, PD24h and 48 h; Group 2, PD30d, SD1 and SD2) were investigated. The expression of 11 914 genes did not differ between all five dormant states in both groups, whereas 113 genes were at least twofold (P < 0.05) more highly expressed in Group 1 compared with Group 2 (Figure 4). In the reverse case, 153 genes were more highly expressed. The classification of these differentially expressed genes into functional groups according to the Munich Information centre for Protein Sequences (MIPS) was investigated. In general, there was a greater representation of the genes more highly expressed in Group 1 (establishment of dormancy) in the metabolism and energy functional groups than in Group 2 (maintained dormancy). In contrast, genes more highly expressed in Group 2 had greater representation in the transcription and cell cycle and DNA processing functional groups than those more highly expressed in Group 1. Other more detailed differences concerning hormone synthesis and catabolism are considered in a later section.

Figure 4.

 Similarities and differences in gene expression between the two groups of dormant states (Group 1, PD24h, PD48h; Group 2, PD30d, SD1, SD2) in Arabidopsis accession Cvi seeds.
The numbers of genes with more than twofold higher expression in all states of one group compared with all those of the other group are shown. Normalized expression levels of these genes and the percentage of them associated with functional groups defined by the MIPS classification are also shown.

In Figure 4 and subsequent figures, normalized expression values are given on the Y-axis. Following the normalization procedure used, normalized values greater than 1.4 behave like a log scale in which a difference in normalized expression of 0.69 is equivalent to a twofold difference (P < 0.05 for comparison between any pair of treatments) in expression level. In these same graphs, dormant states and after-ripened states are presented in a sequence on the X-axis and have connected data points to show the expression pattern across states. Although the dormant states do follow in the time sequence implied, the after-ripened states do not follow directly from imbibed dormant seeds (Table 1) and so the connection between SD2 and DL and LIG is not direct.

Dormant and after-ripened seeds appear to have equally active, but different, gene expression programmes

A large number of genes were found to be associated with the dormant state despite the stringent selection method used, i.e. that these genes must be more highly expressed (twofold or greater) in all five dormant states compared with both after-ripened states (Figure 5). A core set of 442 genes were more highly expressed in dormant states, and in the reverse case 779 genes were more highly expressed (Table S1). The expression of 7545 genes did not differ between all seven physiological states. When looking at the functional groups to which differentially expressed genes were assigned, the most striking difference is the fourfold greater representation of genes associated with protein synthesis in after-ripened states and the twofold greater representation of genes associated with stress responses in dormant states. These two patterns are elaborated in the next sections below.

Figure 5.

 Similarities and differences in gene expression between all five dormant states (PD24h, PD48h, PD30d, SD1, SD2) and the two after-ripened states (DL, LIG) in Arabidopsis accession Cvi seeds.
The numbers of genes with more than twofold higher expression in all dormant states compared with those of both after-ripened states and the reverse case are shown. Normalized expression levels of these genes and the percentage of them associated with functional groups defined by the MIPS classification are also shown. Functional categories are as shown in Figure 4.

A large number of genes encoding transcription factors were differentially expressed in dormant and after-ripened states (dormant, 57; after-ripened, 40; Table S2). In addition, genes encoding several histones, involved in transcriptional regulation, were also differentially expressed (more highly expressed in dormant states: At1g08179, At1g48620, At2g18050; more highly expressed in after-ripened states: At1g51060, At2g38810, At3g54560) and one histone deacetylase (At2g27840) was more highly expressed in after-ripened states. These data indicate that a change in chromatin structure may have occurred, resulting in a complete switch in the gene expression programme during loss of dormancy.

There was no significant difference in the expression levels of genes encoding nuclear RNA polymerases between states, suggesting that RNA polymerase transcript abundance does not limit transcription. However, two genes for nuclear-encoded plastid-localized sigma cofactors SIGA and SIGD (At1g64860 and At5g13730 respectively) are more highly expressed in the after-ripened states compared with dormant states. Upon transcription and translation these sigma cofactors are transported into the plastid where they become active (Isono et al., 1997; Tiller et al., 1991). Sigma cofactors form a complex together with α, β and β’ subunits, which binds to ‘-35’ and ‘-10’ promoter elements in the plastid genome, upon which transcription is initiated. Higher expression of these sigma cofactors implies a mechanism for controlling transcription in the plastid, associated with the after-ripened state.

Gene expression associated with the assembly of translation machinery is repressed in dormant seeds potentially regulating progress towards germination

It has recently been reported that transcription is not the restricting step in completion of germination per se (Rajjou et al., 2004). Instead, translation was strongly correlated with radicle protrusion and inhibitor studies suggested that, although this event was significantly delayed, de novo transcription was not an absolute requirement. The data presented here are consistent with a hypothesis that translation is associated with seeds that are ready to germinate in the presence of light, and its absence is integral to deeper dormancy (not responsive to light) since a much greater proportion of the after-ripened gene set are associated with protein synthesis than those in the dormant gene set (Figure 5). The after-ripened gene set includes genes encoding the large majority of proteins forming the ribosomal complex and a translation initiation factor (see below). Therefore it is plausible that translation runs at a lower rate in more dormant seeds, simply because of lower availability of the components that form an essential part of the translation machinery. Moreover, our data indicate that initiation of the assembly of translation machinery is not associated with more dormant states; yet it is also a reversible process, since SD seeds which have passed through the after-ripened state are similar in this respect to PD seeds.

This hypothesis is in agreement with Bove et al. (2005), who suggest that inhibition of protein translation is controlled at the RNA level in dormant Nicotiana plumbaginifolia seeds, which show reduced transcription of genes associated with protein synthesis compared with non-dormant seeds. Elsewhere, it has been reported that two ribosomal protein transcripts were barely detected in dormant dry seeds, but transcript abundance increased during imbibition and was associated with conditions that promote germination (Toorop et al., 2005). However, ribosomes are present in the dry seed and initial translation does not depend on de novo transcription (Bewley and Black, 1994) as shown by the results of Rajjou et al. (2004), but de novo assembly of ribosomes appears to be required for normal rapid germination and preparation for seedling growth. Given the general higher expression of ribosomal protein genes and rDNA transcription genes in the after-ripened states, it is reasonable to conclude that the induction of the translation machinery is avoided in more dormant seeds, either to prevent the completion of germination or as part of a strategy to prevent wasting energy.

The total number of ribosomal protein genes and genes related to translation in the after-ripened gene set greatly exceeds the same class of genes in the dormant gene set (102 versus eight; Table S3). Of the 102 genes in the after-ripened gene set, 81 genes encode ribosomal proteins, whereas no ribosomal protein genes were found in the eight genes in the dormant gene set. In addition, there are more genes involved in tDNA transcription (four versus two) and rDNA transcription (seven versus zero) in the after-ripened gene set than the dormant gene set (Table S4). Moreover, a translation initiation factor EIF4E1 (At4g18040) and an elongation factor EF-Tu (At4g02930) are more highly expressed in after-ripened states. Initiation factors are components of the EIF4F translation initiation complex in eukaryotes. EIF4E is expressed in most plant tissues, and has been described as a crucial element in translational control (Rodriguez et al., 1998). Pumilio (At3g20250), a specific translational repressor of EIF4E (Menon et al., 2004), is expressed fourfold higher in all dormant states. The putative translational regulation by Pumilio concurs with the transcriptional regulation of EIF4E1 and emphasizes the potential importance of translation as a master switch in the loss of dormancy.

Stress-related gene expression can be promoted by dormancy-inducing conditions in the absence of abiotic stress

There was a twofold greater representation of genes associated with stress responses in the dormant gene set than in the after-ripened set (Figure 5). Moreover, the list of the 20 genes with the highest overall expression in the dormant set contained 11 stress-related genes (Figure 6), including those encoding for late embryogenesis abundant (LEA) and Em-like proteins, peroxiredoxin, and small heat shock proteins (sHSP).

Figure 6.

 The pattern of expression across different physiological states for 11 stress-related genes in Arabidopsis accession Cvi seeds.
Normalized expression levels from microarrays are shown. Gene description: At2g40170, EM6; At5g12030, HSP17. 6A; At1g60740, PER2; At3g56350, putative MNSOD; At3g22500, ECP31 (LEA group 5); At1g52560, HSP26. 5P; At2g41260, M17 (LEA); At4g24220, hypothetical protein similar to VEP1 (wound induced); At1g56280, hypothetical protein similar to Dl19 (drought induced); At3g53040, putative LEA group 3; At3g51810, EM1.

Many putative ‘dormancy genes’ that have been identified so far appear to be in one of the classes of stress-inducible genes. For example, several cDNA clones that were expressed in dormant but not in non-dormant seeds were isolated in Avena fatua. However, when the non-dormant seeds were stressed by exposure to elevated temperatures the ‘dormancy genes’ were expressed (Li and Foley, 1995, 1996). The majority of these dormancy-associated transcripts appeared to encode members of the family of LEA proteins. Figure 6 shows more than twofold higher expression of genes encoding for LEAs from Group 3 and Group 5, as well as the Em-like proteins (Em1 and Em6), in all the dormant states compared with after-ripened states. Vicient et al. (2000) showed that expression of Em1 and Em6 increased towards the completion of maturation in Arabidopsis seeds; but, despite their similarity, these genes were differentially expressed in different parts of the embryo. Both LEA and Em-like proteins are thought to play a protective role during seed desiccation (Philips et al., 2002), which occurs naturally in most species.

The 17.7-kDa sHSP (HSP17.6A) belongs to class I HSP and is located in the cytosol (Sun et al., 2001), whereas the 26.5-kDa sHSP (HSP26.5-I) is class P related and is plastid-located (Hong and Vierling, 2000). Both genes showed higher expression in all the dormant states (Figure 6). HSP17. 6A has been shown to accumulate during seed maturation, whereas its expression decreased during germination (Sun et al., 2001). Moreover, the gene appeared to be under control of the ABI3 gene (Wehmeyer et al., 1996). Over-expression of HSP17. 6A enhanced osmo- and salt tolerance of transformants, which is a clear indication of its role in the stress response. However, thermotolerance was not affected (Sun et al., 2001).

Other studies have also yielded stress-related genes, such as the Per1 genes from barley and Arabidopsis, which encode peroxiredoxins (Prx) (Aalen, 1999). Peroxiredoxins are enzymes that putatively play a role in protecting seeds from desiccation damage by exposure to reactive oxygen species. Per1 gene expression can be induced by osmotic stress and ABA in immature barley embryos (Aalen et al., 1994), and the closely related pBS128 transcript was upregulated in both dormant and non-dormant imbibed mature embryos of Bromus secalinus (Goldmark et al., 1992). Based on the expression patterns of pBS128 in B. secalinus and PER1 in Arabidopsis, it was suggested that peroxiredoxins play a role in dormancy (Goldmark et al., 1992; Haslekas et al., 1998). However, PER1 expression in the Arabidopsis aba1 mutant, which develops no PD, was similar to wild-type seeds, which suggests that the presence of the PER1 transcript alone is not sufficient to maintain dormancy (Haslekas et al., 1998). Furthermore, transgenic tobacco plants, over-expressing the rice 1Cys-peroxiredoxin (R1C-Prx) gene, showed germination similar to control plants (Lee et al., 2000). More conclusive evidence that argues against a role for PER1 in the regulation of dormancy was shown in transgenic Arabidopsis lines over-expressing the barley Per1 protein, as well as transgenic lines with reduced levels of PER1 (Haslekas et al., 2003). No correlation was found between Prx levels and the maintenance of dormancy. However, our results show higher expression of PER2 in all dormant states compared with after-ripened states (Figure 6). Furthermore, PER2 expression was lower in PD24h and PD48h than in PD30d, SD1 and SD2, indicating a possible relationship between expression level and dormancy maintenance rather than the establishment of PD.

The expression patterns of the 11 stress-related genes were very similar over the different dormant states. Interestingly, all of them showed a moderate increase in expression in the LIG seeds compared with DL seeds. This was clearly not the result of thermal or osmotic stress; however, the 4-h irradiation with red light (LIG) may have elicited a stress response. This hypothesis was tested by measuring the expression of HSP26. 5-I by quantitative RT-PCR. Relative expression levels were 23.1 ± 2.5 and 7.4 ± 0.8 for PD24h, with and without irradiation respectively; the equivalent levels for DL seeds were 1.3 ± 0.8 and 1.0 ± 0.6. Therefore, irradiation of dormant seeds, which does not terminate dormancy and initiate germination, resulted in a substantial increase in gene expression. The same, but smaller, responses were evident in DL seeds. These results suggest that, although PD seeds are insensitive to light with respect to induction of germination, they do respond to light, probably by perceiving light as a stress factor only.

Stress-related genes and proteins in seeds, particularly the sHSPs and LEA or LEA-like, are generally related to drought- and/or osmotic stress, which are naturally perceived by seeds undergoing maturation drying. In this sense, expression of these genes and proteins is part of the seed developmental programme. In general, these proteins are thought to play a role as chaperones, protecting proteins against (heat) denaturation or assisting in the correct folding of proteins (Wang et al., 2004). However, the data presented in Figure 6 show that stress-related gene expression in seeds is not necessarily the result of abiotic stress. For example, dormancy state SD1 is induced in the dark at constant 20°C in fully imbibed seeds. Apparently, conditions that are conducive to the induction of dormancy also trigger a stress response. ABA is the common denominator in these signalling networks (Finkelstein et al., 2002; Xiong et al., 2002). Indeed, Ali-Rachedi et al. (2004) show that ABA synthesis is required for the induction and maintenance of dormancy in Arabidopsis Cvi seeds, a view that is supported by the results of this study (see below).

ABA-responsive elements are over-represented in genes within the dormant gene set

Genes in the dormant and after-ripened sets (Figure 5) were searched to identify potential transcriptional regulatory elements. Sequences upstream (10–1000 bases) from the translated region of the genes were searched for 6- to 10-base sequences that were over-represented (P < 0.05) compared with the genome average. Five 6-base, three 7-base and two 8-base sequences were significantly over-represented in the dormant gene set (Table 2a), and two 6-base sequences were over-represented in the after-ripened gene set (Table 2b). In each case the occurrence of these sequences in the other gene set was not significantly different from the genome mean. Not all 6-, 7- and 8-base sequences in Table 2(a) are uniquely identified in the search because the 6-base search will pick up the 7- and 8-base sequences as they contain common sequence.

Table 2.   Sequences over-represented in the 1-kb upstream region of (a) dormant and (b) after-ripened gene sets. ‘Observed’ and ‘Expected’ are either the numbers of genes in the set containing the sequence or the number of genes predicted from the whole genome mean respectively
SequenceDormant (442 genes)After-ripened (779 genes)
 ACGTGT122711.38 × 10−51341241
 CACGTG109669.92 × 10−41211151
 GGCCCA576511721155.77 × 10−4

Of the 442 genes in the dormant set, 281 (64%) contain one or more of the 6-base sequences in Table 2(a) compared with 42% in the after-ripened set. Nine of the 10 sequences (Table 2a) contain the core motif ACGT, which is characteristic of abscisic acid responsive elements (ABRE). Single copies of ABREs require a second cis-acting element for the gene to be induced by ABA (Shen and Ho, 1995). The remaining 6-base motif (CGTGTC) has a sequence characteristic of a CE3-like element (Hobo et al., 1999). These elements can interact with ABREs to produce a functional ABA response complex (reviewed by Shen and Ho, 1997). However, both types of elements (ABRE and CE3) are thought to be functionally equivalent (Hobo et al., 1999). It is therefore likely that gene response to ABA has a complex regulation requiring the interaction of two or more of these cis-acting elements. Indeed, multiple ABRE copies are thought to enhance ABA responsiveness (Nakabayashi et al., 2005; Shen and Ho, 1995). In the dormant gene set 43% of the genes contain two or more (up to 6) cis-acting elements compared with only 20% of the after-ripened gene set (indicated by 6-base sequences only in Table 2a). Similarly, five (45%) of the 11 stress genes that were more highly expressed in dormant states (Figure 6) had multiple copies of these elements.

Within the dormant gene set, t-tests showed that in all seven physiological states expression levels were significantly (P < 0.009) higher from genes containing multiple copies compared with single and even two copies of the different ABREs, but not multiple copies of the same ABRE. The different levels of expression of these genes between physiological states will also be influenced by the different endogenous concentration of ABA present in seeds to bring about the patterns shown in Figure 6. These results suggest that ABREs directly regulate gene expression levels in response to endogenous ABA, the regulation of which is considered in the next section.

In the Arabidopsis Col accession, there was a high occurrence of ABREs in genes with high mRNA levels in dry seeds, but not in 24-h-imbibed seeds, in line with the reduced endogenous levels of ABA in the latter (Nakabayashi et al., 2005). This indicates a role for ABA-regulated gene expression during development that is lost during imbibition. Re-analysing these data from the Col accession shows that 40% of the 500 most highly expressed genes in dry seed contain one or more copies of the motif CACGTG, compared with 18% in 24-h-imbibed seed. In 24-h-imbibed DL seed of Cvi we found a similar percentage (21%) of the 500 most highly expressed genes containing this motif, whereas, in the 24-h-imbibed dormant (PD24h) seed, there was a higher percentage (30%). Interestingly, the percentages in other dormant states PD30d, SD1 and SD2 were essentially the same (31, 30 and 31% respectively).

Other motifs, CE1 (CACCG) and RY (e.g. CATGCAT), have been described as cis-acting elements (Baümlein et al., 1992; Niu et al., 2002; Shen and Ho, 1995) and were associated with enhanced gene expression when in the presence of ABREs in the Col accession (Nakabayashi et al., 2005). In the Cvi dormant gene set we found no similar relationship, suggesting that CE1 and RY motifs do not play a role in seed dormancy.

In the after-ripened gene set, 38% of the genes contain one of the two sequences shown in Table 2(b), compared with 28% in the dormant gene set. Although these sequences could not be assigned to any known promoter motif, gene expression levels in all physiological states were significantly (P < 0.01) enhanced by the presence of multiple copies of the sequences in the after-ripened gene set. In contrast, the presence of multiple copies of these sequences had no significant effect on the expression of genes in the dormant gene set.

ABA synthesis is required for the maintenance of dormancy

Control of dormancy in Arabidopsis is thought to result from the antagonistic effect of ABA and GAs on germination (Debeaujon and Koornneef, 2000), and this is apparent in the genes differentially expressed in the dormant and after-ripened states (Figure 7 and Table S5). Those genes more highly expressed in dormant states can be linked to high-ABA and low-GA endogenous concentrations, and the reverse in after-ripened states. Endogenous ABA concentrations are believed to be the result of a dynamic balance between continuous synthesis and catabolism (Cutler and Krochko, 1999), and thus the active hormone levels that control seed physiological processes may result from either changes in rate of synthesis or catabolism in the seed (Nambara and Marion-Poll, 2005).

Figure 7.

 The pattern of expression across different physiological states of representative genes encoding enzymes involved in (a) the synthesis and catabolism of ABA, (b) genes responsive to ABA, and (c) the synthesis and catabolism of GA in Arabidopsis accession Cvi seeds.
Normalized expression levels from microarrays are shown. The accompanying tables show relative expression levels of a selection of these genes assayed by QRT-PCR. The expression patterns of other members of gene families are provided in Table S5. Gene numbers: AAO3, At2g27150; ZEP, At5g67030; SDR1, At1g52340; NCED6, At3g24220; NCED9; At1g78390; CYP707A1, At4g19230; CYP707A2, At2g29090; ABI2, At5g57050; ABI3, At3g24650; ABI4, At2g40220; ABI5, At2g36270; ABA3, At1g16540; AREB2, At3g19290; ABF1, At1g49720; KAO1, At1g05160; GA20ox1, At5g07200; GA2ox1, At1g30040; GA3ox2, At1g15550.

Abscisic acid accumulation in seeds is likely to rely on regulation of the 9-cis-epoxycarotenoid dioxygenase (NCED) gene family (Nambara and Marion-Poll, 2005) encoding carotenoid cleavage enzymes that carry out the first committed step to ABA synthesis (Cutler and Krochko, 1999; Nambara and Marion-Poll, 2005). Two members of this gene family NCED6 and NCED9 (Figure 7a) and the closely related CCD4 (At4G19170; Tan et al., 2003) are expressed at high levels across all seven physiological states, and differ between dormant and after-ripened states. NCED2 has the same pattern as NCED6 and NCED9, but is expressed at lower levels in all states (Table S5). Array data show that the NCED genes are more highly expressed in Group 2 dormant states than Group 1 or after-ripened states, a pattern confirmed by QRT-PCR (Figure 7). The lower expression in Group 1 dormant states (24- and 48-h imbibition of PD seeds) is consistent with measurements of endogenous ABA concentration by Ali-Rachedi et al. (2004). They show that ABA concentration in Cvi decreased in the first 2 days of imbibition of PD seeds and then increased from day 3. Within the ABA synthesis pathway it is interesting to note that there are also differential expression patterns in genes encoding enzymes in the steps preceding [zeaxanthin epoxidase (ABA1/ZEP)] and following [abscisic acid aldehyde oxidase (AAO3)] NCED (Figure 7a), which is also consistent with higher endogenous ABA concentrations in dormant compared with after-ripened states.

The major catabolic route for ABA is via the cytochrome P450 enzyme ABA 8′-hydroxylase (Cutler and Krochko, 1999) encoded by the CYP707A family in Arabidopsis (Saito et al., 2004). Two genes in this family showed differential expression patterns between dormant and after-ripened states (Figure 7a). The expression of one gene (CYP707A1) was greatly reduced in LIG compared with other states and the other gene (CYP707A2) showed the reverse pattern, which was confirmed by QRT-PCR (Figure 7). Kushiro et al. (2004) showed that the expression of both of these genes was upregulated during re-hydration of seeds. However, CYP707A2 was responsible for the decrease in ABA during seed imbibition, and mutants in this gene exhibited hyperdormancy and accumulated sixfold greater endogenous ABA concentrations than wild-type Arabidopsis. Ali-Rachedi et al. (2004) suggest that greater loss of ABA in DL Cvi seeds exposed to light compared with PD on imbibition resulted from a more intense catabolism, and this is supported here by the greatly (c. fourfold) increased expression of CYP707A2 in LIG seeds. Such differences in ABA catabolism in DL and PD seeds have also been shown in other species (Jacobsen et al., 2002; Le Page-Degivry et al., 1997; Sondheimer et al., 1974). Expression levels were similar in all dormant and after-ripened seeds kept in the dark which prevented germination (Figure 7a).

In the experiments reported here fluridone, especially in conjunction with GA4/7, was effective at breaking dormancy in the dormant states. Fluridone is an inhibitor of carotenoid synthesis (Bartels and Watson, 1978) and ABA synthesis (Gamble and Mullet, 1986). This result provides physiological evidence in support of the gene expression data above, which indicates a role for ABA synthesis in the maintenance of dormancy. In other studies, ABA synthesis has also been associated with the maintenance of PD (N. plumbaginifolia, Grappin et al., 2000; A. thaliana Cvi, Ali-Rachedi et al., 2004), and the induction and maintenance of stress-induced SD in Brassica napus (Gulden et al., 2004), lettuce and a range of annual weed species (Yoshioka et al., 1998).

A number of genes known to be involved in ABA-mediated signalling, or that are responsive to ABA in the mature seed and likely to influence radicle emergence and dormancy (Finkelstein and Rock, 2002; Finkelstein et al., 2002; Himmelbach et al., 2003; Holdsworth et al., 1999), were differentially expressed in dormant and after-ripened states (Figure 7b). The expression of these genes was generally lower in the after-ripened states than the dormant states, and in some cases higher in Group 2 dormant states compared with Group 1 dormant states in line with likely differences in endogenous ABA concentrations. Differences in the expression of these genes between different physiological states are more limited than may have been anticipated from the literature, and this may because the present study is concerned more with dormancy maintenance than with developmental dormancy like most other reports. In contrast to these genes, the expression of ABI4 is relatively low in the dormant states, but fourfold higher in the after-ripened states (Figure 7b). Therefore this contrasting ABI4 expression pattern cannot be related to the inhibition of germination, but may be regarded as a preparation for seedling growth. This is consistent with the absence of any dormancy phenotypes for ABI4 (Finkelstein, 1994). Our results support the conclusion of Söderman et al. (2000) that ABI4 may be regulated by an as yet unidentified seed-specific regulator other than ABA.

A change in the dynamic balance of catabolism and synthesis of active GAs is required to promote germination

The GA response pathway that regulates a wide range of plant processes is tightly linked to GA biosynthesis and catabolism (Olszewski et al., 2002). Recent work has shown how this balance is linked to germination (Ogawa et al., 2003) and loss of PD during low-temperature treatment (Yamauchi et al., 2004). The gene encoding GA 3-beta-dioxygenase (GA3ox2), which converts inactive precursors to the active GA1 and GA4, is expressed in all states, but expression is c. 40-fold higher in the after-ripened seeds exposed to light (Figure 7c). In contrast to this, the gene encoding GA 2-oxidase (GA2ox1) which de-activates these active GAs and their immediate precursors is again highly expressed in all states, but highest in PD30d and SD2. These results are consistent with a continued dynamic state of synthesis and catabolism of GA which maintains lower endogenous concentrations in dormant states compared with after-ripened states.

A number of GA-responsive genes are differentially expressed in the seven physiological states (Table S6). Where a clear pattern exists across physiological states, genes tend to be either more highly expressed in dormant states, e.g. GASA2 (At4g09610), or more highly expressed in after-ripened states, e.g. gibberellin-responsive protein (putative GASA4; At1G74670). The latter gene was expressed c. 120- and 30-fold higher than all dormant states in the DL and LIG states respectively, a result confirmed by QRT-PCR (data not shown).

Higher expression of genes associated with reserve mobilization and cell wall modification is linked to dormancy loss rather than the induction of germination

A set of 47 genes involved in storage mobilization and cell wall modification were more highly (most >fourfold) expressed in after-ripened states compared with dormant states (Table S7). Figure 8 shows representative examples of these classes of genes. In contrast, only four genes encoding a pectinesterase (At1G53830), an endoxylanase (At1G10050), a glycosyl hydrolase (At4G22100) and an endoglucosidase (At4G34480) were more highly expressed in dormant seeds than after-ripened seeds.

Figure 8.

 The normalized levels of expressions for genes encoding proteins involved in reserve mobilization and cell wall modification and extension that are more highly expressed in after-ripened states in Arabidopsis accession Cvi seeds.
A more complete list is given in Table S7. The table accompanying this figure shows the relative expression level of At5g05290 assayed by QRT-PCR. Gene description: At4g30810, putative serene carboxypeptidase S10 family protein; At5g05290, putative expansin; At1g68560, alpha-xylosidase; At5g63810, putative beta-galactosidase/lactase; At4g36880, putative cysteine proteinase; At5g13870, endo-xyloglucan transferase; At3g14310, putative pectinesterase family protein.

Of the 47 genes, five genes encode proteins involved in the mobilization of lipid reserves (Table S7a) and 10 encode proteins involved in the hydrolysis of storage proteins in seeds (Table S7b), which have been intimately linked to germination (Bewley and Black, 1994). The gene set also contains cell wall hydrolysing enzymes (Table S7c) that are thought to play a role in seed germination, either by weakening the endosperm that surrounds the embryo or by mobilizing carbohydrate reserves (e.g. Bewley, 1997). Further genes encode other cell wall modifying enzymes that allow for the growth of cells required for embryo elongation and the completion of germination, by loosening the cell wall matrix and allowing turgor-driven cell extension (e.g. Cosgrove, 1999). These results support the involvement of these well-described cell wall enzymes in seed germination; surprisingly, they also show that the expression of genes encoding them is associated with the loss of dormancy rather than the induction to germination.

ABA, stress and dormancy responses significantly overlap at the transcriptome level

This study has demonstrated that dormancy is not a quiescent state and is characterized by significant transcriptional activity. Transitions among different intensities and cycles of dormancy are accompanied by relatively small changes in the expression of a large number of genes, whereas dormant versus after-ripened states show clearly distinct transcriptomes. For example, many genes involved in protein synthesis are expressed at a much lower level in more dormant seeds, compared with after-ripened seeds, and this may limit progress to germination. Long-term imbibition of PD seeds at constant temperature results in a transition to a more intense dormancy (PD30d), which is associated with a downregulation of metabolism- and energy-related genes and a significant change of transcription-related gene expression. Also the transition from SD1 to SD2 is associated with a shift of the dormancy-related gene expression to higher values in the SD2 state (Table S1). The SD2 state is characterized by a more intense dormancy that could not be broken by light and nitrate (Table 1).

Comparison of dormant states during cycling determined by different environmental conditions has shown significant similarities supporting the existence of a common underlying molecular mechanism relating to the long-standing hormone balance theory (Karssen and Laçka, 1986). In this theory, ABA and GA act at different times in the seed life, ABA inducing (primary) dormancy during maturation and GAs playing a key role in dormancy release and the promotion of germination. However, as discussed above, more recent work has shown that ABA synthesis and action occur during imbibition and are required to maintain dormancy. Further work in maize has shown that inhibition of GA biosynthesis mimics the effect of exogenous ABA in suppressing vivipary, suggesting that it is the GA/ABA balance, rather than absolute hormone amounts, which controls germination during maturation (White and Rivin, 2000; White et al., 2000). Figure 9 shows a simple model extending the hormone balance theory to changes in gene expression during dormancy cycling indicated in the current study. These data suggest that the prevailing environment determines the relative activity of key elements in the ongoing synthesis and catabolism of ABA and GA to influence dormancy status by regulation of transcription. This model has the benefit of being flexible and reversible and could therefore form a control for dormancy cycling. Clearly many other factors will have some further influence on dormancy status, especially downstream from transcription. The mechanism does not provide for a crisp interface between dormant, after-ripened and germinating states, but this is what is observed in practice, since not all seeds in the population will have the same response. Whether a seed will germinate or remain dormant may thus depend on the intrinsic balance of GA and ABA biosynthesis and catabolism, which will determine the dominance of either of the hormones.

Figure 9.

 Model describing the influence of environment on the ABA–GA balance and resulting state of dormancy or germination.
The model assumes an intimate interplay between ABA- and GA-biosynthesis and catabolism to determine hormone levels that is influenced by the prevailing environmental conditions. Those conditions that induce dormancy will tip the balance to the left and those breaking dormancy will tip the balance to the right. It is suggested that the resulting dominance of either hormone determines dominance of signalling and, ultimately, induction or breaking of dormancy, or germination. According to this model, seasonal cycling will cause shuttling between environmental extremes that result in dormancy cycling. Physiological states from the current study and some of the key target genes are shown in parentheses. The shaded arrows at the bottom represent intensity of gene expression in several functional gene groups, as well as the involvement of ABREs.

The results presented suggest that ABA, stress and dormancy responses significantly overlap at the transcriptome level, as suggested by Hilhorst (1995). Many of the genes more highly expressed in the dormant states appeared to be related to stress. This may be linked to the synthesis of ABA, which appears essential to the maintenance of dormancy. Indeed, ABA-responsive elements were over-represented in the dormant gene set, suggesting that they are under ABA control. Under prolonged conditions that are non-permissive to germination, there is an increase of ABA content (Ali-Rachedi et al., 2004). These conditions will therefore always lead to co-expression of dormant- and stress-related genes that are controlled by ABA. There is an evolutionary advantage in this co-expression since dormancy is a mechanism to survive prolonged periods of environmental stress that are unfavourable for growth.

Experimental procedures

Plant material, seed production and germination conditions

Plants of the A. thaliana accession Cape Verdi Islands (Cvi; accession number N8580) were grown in media containing peat/sand/vermiculite from seeds donated by Maarten Koornneef (Wageningen University, the Netherlands). For the first 4 weeks, four replicate blocks of 24 seedlings were grown in 2.5 cm square plastic modules at 20°C under daylight fluorescent tubes (16-h photoperiod). Seedlings were then transplanted to 5 cm square plastic modules and placed in a glasshouse that was vented and heated and had supplementary lighting to maintain the same mean temperature and photoperiod. When the seeds were mature the plants were no longer watered. Seeds were harvested when siliquae were dry on the plant. Seeds were bulked from each replicate block of plants. The four replicate seed lots were then equilibrated with ambient relative humidity (rh) (c. 33% rh; 20°C), placed in separate sealed moisture-proof containers and stored at 20°C. Initial PD samples (3 × 25 mg per replicate) were taken and then six further samples were taken when seeds had uniformly entered different physiological states induced by treatments described in the text (Table 1). Physiological status was determined in subsamples of seeds from each replicate in germination tests described below. The very low number of seeds that germinated in the dark during preparation of the samples were removed, and the remainder of each seed sample was immediately frozen and stored at −80°C until used for RNA extraction.

Germination (radicle protrusion) was recorded at 20 ± 0.5°C from either 50 (during after-ripening) or 100 seeds (all other treatments) from each replicate on two layers of filter paper (Whatman International Ltd, Maidstone, UK), kept moist throughout treatment with either water or a 10-mm KNO3 solution in clear polystyrene boxes. These boxes were placed in either the dark or the light [daylight fluorescent tubes; 60 μmol m−2 sec−1 (photosynthetic photon flux density)] both with and without prior exposure to 4 h of red light after 20-h imbibition. Germination was monitored for a minimum of 28 days to record all seed capable of germination. Seeds were not sterilized to avoid influencing their dormancy status; however, fungal infection was minimal in the seed lot used. Viability of seeds before and during prolonged treatments was determined by placing seeds on a mixture of 100-μm GA4/7 (ICI Ltd, Bracknel, UK) and 38-μm fluridone (Dow Chemical Co., Hitchen, UK). Both fluridone and GA4/7 were initially dissolved in ethanol and then diluted. The final concentration of ethanol was less than the 0.03% which was found to have no significant influence on germination in preliminary studies.

Total RNA isolation from seeds

Following treatment, seeds were stored in 2-ml tubes at −80°C. Total RNA was extracted using a hot borate method (modified from Wan and Wilkins, 1994). Seeds were homogenized with single steel cone balls (Retsch GmbH & Co. KG, Haan, Germany) in 2-ml tubes using a Retsch Mixer Mill MM200 (Retsch GmbH & Co. KG) for 3 × 2 min at 30 Hz, submersing tubes and tube adapter blocks in liquid nitrogen between each 2-min session. Ground seeds were kept submersed in liquid nitrogen prior to RNA extraction. Ground seeds were re-suspended in 700-μl extraction buffer [(0.2-m sodium tetraborate decahydrate, 30-mm EGTA, 1% w/v SDS, 1% w/v sodium deoxycholate, pH 9.0) containing 14 mg polyvinylpyrilidone and 1.1 mg dithiothreitol], which had been heated to 80°C. The seed suspension was added to 0.35 mg proteinase K in a 2-ml round-bottomed tube and incubated at 42°C for 90 min. 56 μl 2-m KCl was then added and samples incubated on ice for a further 2–3 h. Samples were centrifuged at 16 000 g at 4°C, after which the supernatant was removed to a new tube and its volume determined. A one-third volume of 8-m LiCl (to give a final 2-m LiCl concentration) was added and samples incubated at 4°C overnight. Samples were centrifuged, and pellets washed by vortexing in 2-m LiCl until the supernatant remained colourless. Pellets were re-suspended in 300-μl diethylpolycarbonate (DEPC)-treated water, and RNA precipitated by the addition of 30 μl 3-m sodium acetate and 990 μl ice-cold 100% ethanol, and incubation at −80°C for a minimum of 3 h. RNA was pelleted by centrifugation at 16 000 g for 20 min at 4°C, followed by a wash with 70% ethanol. Partially dried pellets were re-suspended in 100-μl DEPC-treated water and purified with RNeasy columns (Qiagen, Hilden, Germany). All kits were used according to the manufacturers’ instructions. RNA concentration was determined using a NanoDrop ND-1000 spectrophotometer (Nanodrop® Technologies, Wilmington, DE, USA), and RNA quality was checked using formaldehyde gel electrophoresis.

Affymetrix micorarrays

20 μg total RNA per biological replicate were submitted to Nottingham Arabidopsis Stock Centre (Nottingham, UK) for synthesis of labelled cRNA, hybridization to the Affymetrix ATH1 Arabidopsis GeneChip (Affymetrix, Santa Clara, CA, USA), chip scanning and data accumulation (using Affymetrix Microarray Suite version 5.0 software) (Craigon et al., 2004). Four independent biological replicates were submitted for each of seven physiological states (Table 1), and the resulting expression data sets are available at

Microarray data handling and analysis

Data were normalized in bioconductor (Gentleman et al., 2004) using the vsn algorithm (Huber et al., 2003). The algorithm transforms raw data from each array with an affine transformation, and then transforms the data with a generalized logarithm to stabilize the variance. To visualize the normalized gene expression patterns, data were imported into GeneSpring version 7.2 (Silicon Genetics, RedWood City, CA, USA) and Pymood software (Allometra, Davis, CA, USA). PCA was performed in GeneMaths (version 2.01: Applied Maths, Kortrijk, Belgium). PCA was based on the expression patterns of all genes over the seven different physiological states (PD24h, PD48h, PD30d, SD1, SD2, DL, LIG). The first two principal components divided the physiological states into three groups and accounted for 57.5% of the variation in gene expression patterns. These groups of physiological states were adopted in the subsequent comparison of gene expression in dormant and after-ripened states.

The normalized data were subjected to analysis of variance using GenStat for Windows (version 8.1; VSN International Ltd., Hemel Hempsted, UK). Variance estimates were pooled across the different genes. Comparisons were made on the transformed scale, but, to simplify the manuscript, are all reported in terms of equivalent fold changes for genes with transcription levels high enough that the generalized log behaves like log. Gene transcription levels between pairs of physiological states were tested using the Benjamini–Hochberg method (Benjamini and Hochberg, 1995) to control the false discovery rate to 5%. For different pairwise contrasts, the fold-changes required for significance varied from 1.48 to 1.83. To study differences in gene expression related to the three groups of physiological states identified earlier, genes were identified that had twofold or greater difference in all states of one group compared with all those of the other group to identify characteristic gene sets. A twofold difference was selected as a commonly used criterion for comparing gene expression levels, which would be statistically conservative for all the comparisons.

Functional classification and categorization of genes were carried out according to MIPS (Schoof et al., 2002). Gene annotation was determined using The Arabidopsis Information Resource (TAIR) (Rhee et al., 2003).

Motif analysis

Potential regulatory motifs were found using the ‘Find potential regulatory sequences’ tool in the GeneSpring version 7.2 software. Six- to 10-base sequences were searched for within 10–1000 bases upstream of the translational start of each gene in the dormant or after-ripened gene lists relative to sequence upstream of other genes. No discrepancies or gaps were allowed, and the probability cut-off was 0.05. The search for multiple copies of the sequences within a 1-kb sequence upstream of translational start sites was carried out using a program developed by Dr L. Cadman and using data from TAIR (

Quantitative PCR (QRT-PCR) – validation of microarray data

The expression of selected genes was quantified by QRT-PCR. 2 μg seed total RNA from four biological replicates per dormancy stage was treated with 2 U RNase-free DNase (Promega, Madison, WI, USA). cDNA was synthesized using TaqMan® Reverse Transcription Reagents (Applied Biosystems, Foster City, CA, USA) according to the manufacturer's instructions. Primers for real-time PCR were generated using the Primer3 web interface (, Rozen and Skaletsky, 2000) using sequence data from TAIR ( Gene-specific primer sequences are given in Table S8. Genes chosen for validation were normalized to 18S rRNA. The absence of genomic DNA was confirmed by analysis of dissociation curves and agarose gel electrophoresis of the PCR products where applicable. PCR reactions were performed in 384-well plates using a SYBR® Green PCR Master Mix (Applied Biosystems) and an ABI Prism® 7900HT Sequence Detection System (Applied Biosystems) to monitor cDNA amplification. Reactions were performed in triplicate for each biological replicate (four biological replicates for each of the seven dormancy stages) in a mixture containing 1 μ l of first-strand cDNA in a volume of 15 μl (corresponding to approximately 20 ng cDNA). The standard thermal profile used was: 2 min at 50°C and 10 min at 95°C, followed by 40 cycles of 15 sec at 95°C and 60 sec at 60°C. Average threshold cycle (Ct) and relative quantities were calculated using sds software v2.1 (Applied Biosystems) and used for subsequent analysis.


We thank J. R. Lynn for statistical advice; J. R. Stephen and K. Thompson for discussion during the development of the project; K. C. Dent, H. A. Clay and F. Christodoulou for technical assistance with germination experiments; Jan Kodde for PCA analysis, and the UK Department for Environment, Food and Rural Affairs (Defra) for funding.