Arabidopsis (accession of the Cape Verde Islands, Cvi, N8580) plants were grown in climate cell consisting of 16 h light/8 dark at a temperature of 22°C. After harvest seeds were stored in paper bags under ambient temperature and humidity. Dormancy release was followed by germination assays throughout the after-ripening period, and after 16 months dormancy was released. The imbibed seeds were dissected and the micropylar and chalazal endosperm (MCE) and the radicle and hypocotyl (RAD) were collected for RNA isolation. RNA of the RAD samples was isolated with the Absolutely RNA Nanoprep kit (Agilent Technologies) as we used earlier for our Col-0 seed samples (Dekkers et al., 2013). This method did not work well for Cvi MCE and dry seed samples, so RNA of the MCE and dry seed samples was isolated by an alternative method based on the hot borate isolation protocol (Wan and Wilkins, 1994), and additionally cleaned up with NucleoSpin RNA Clean-Up XS (Macherey-Nagel).

Germination assays were performed to assess dormancy release using the Germinator set-up (Joosen et al., 2010). Samples of approximately 50–100 seeds were sown on two layers of blue germination papers with 50 mL of demineralized water in plastic trays (15 cm × 21 cm). Trays were piled and wrapped in a closed and transparent plastic bag. The bags were incubated in an incubator at 22°C and continuous light. Germination was followed by taking photos daily for up to 10 days.

This temporal and spatial sampling resulted in nine dormant and nine after-ripened samples (Figure 1D) and these 18 samples, with three replicates for each sample, were hybridized to Affymetrix ATH1 gene chips. These 54 chips were normalized together with the 34 chips of near isogenic lines of several DELAY OF GERMINATION (DOG) loci [including samples of D and AR seeds of Ler-0 and near isogenic lines NILDOG1, NILDOG2, NILDOG3, and NILDOG6 (Bentsink et al., 2010) that were imbibed for 24 h] and the previous 116 chips from Dekkers et al. (2013) using Robust Microarray Averaging (RMA; Irizarry et al., 2003), using a custom chip definition file (.cdf) from the CustomCDF project (Dai et al., 2005; Ath1121501_At_TAIRG.cdf v18.0.0, released 23rd January 2014, from http://brainarray.mbni.med.umich.edu/Brainarray/Database/CustomCDF), after removal of the control probes, 21315 genes remain. The microarray data described in this article has been deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus (accession number GSE76907). The quality controls (Supplementary Figure S5; Supplementary Table S1) indicate that this is a robust dataset that allows us to investigate, in seeds with opposing dormancy states, the transcriptional dynamics in two key compartments that regulate seed dormancy and germination. A web tool enabling the querying of this dataset¹, under ‘Dormancy Gene Expression Plots.’

To calculate genes as being differentially expressed at a certain fold level, the difference in means must be at least that large, and they must be statistically significant on a t-test at a p-value of 0.05. In order to reduce bias from genes whose expression is within the noise region, before fold changes are calculated the data means are first clipped at an expression level of 4 (log₂)- replacing anything less than four with four. This may slightly underestimate the number of differentially expressed genes (or their level of fold change), but helps prevent the noise region from heavily influencing the results of the fold changes; this thresholding is reflected in the numbers in Data File S1, and all numbers throughout. The gene lists were analyzed using Genetrial (Keller et al., 2008)² for overrepresented GO categories.

To better understand the early transcriptome changes upon sowing in D and AR seeds, we studied their temporal and spatial gene expression profiles. For this experiment we made use of the highly dormant Arabidopsis Cvi accession which has been frequently used in seed dormancy research (Alonso-Blanco et al., 2003; Ali-Rachedi et al., 2004; Cadman et al., 2006; Finch-Savage et al., 2007). We followed the loss of dormancy by germination assays during AR, and after 16 months of dry storage over 90% of the seeds germinated, and were therefore considered to have lost dormancy (Figures 1A,B). Imbibed seeds were dissected into two distinct compartments, the micropylar and chalazal end of the endosperm (MCE), which acts as a barrier together with the seed coat, and the combined radicle and hypocotyl (RAD; Figure 1C), which has to expand to break through the covering layers to complete seed germination. To capture the transcriptome dynamics, we conducted microarray analysis on RNA isolated from dry seeds and the two compartments at four time points, i.e., 3, 7, 12, and 24 hours after sowing (HAS), of both D and AR seeds (Figures 1C,D). These compartments and time points were chosen to align with the sampling of germinating Col-0 seeds we performed previously (Dekkers et al., 2013). Three HAS was chosen as the earliest time point since at this stage initial transcriptional changes can be clearly observed (Preston et al., 2009; Dekkers et al., 2013; Scheler et al., 2015).

A freely available web tool³ (under ‘Dormancy Gene Expression Plots’) was developed in which the gene expression data of these dormancy samples were combined with the earlier mentioned seed germination samples (Dekkers et al., 2013). The tool also includes transcriptome data of near isogenic lines of DELAY OF GERMINATION (NILDOG) loci, providing gene expression information concerning D and AR seeds of several genotypes with different primary dormancy levels (Leónie Bentsink, personal communication). The expression profile of a gene of interest can be visualized across all these samples, as was previously done for anther and flower microarray datasets (Pearce et al., 2015). The resulting plots show how genes are expressed across these three experiments (Figure 2A), and provides a useful tool for the community, allowing quick determination of the behavior of a gene of interest.

Over the whole time course 12,741 genes were detected to be expressed (>log₂ 5, 59.8% of the 21,315 genes on the chip) in at least one of the 18 samples, while 7,659 genes (60.1% of all expressed genes) were expressed in all 18 samples (Figure 2B). In dry seeds and at the early time points fewer genes were expressed, this number increased during the time course in the two compartments (MCE and RAD) and in both dormancy states (Figure 2C). We detected nearly equal numbers of genes expressed in the MCE and the RAD (11,944 vs. 11,955 respectively) over the time course, but fewer genes were expressed in the D samples (11,820) compared to the AR samples (12,545).

We identified transcript sets that were specifically expressed in either the D or the AR state, by considering genes which were expressed in one state above 6 (on a log₂ scale) and expressed below 5 (below which expression becomes indistinguishable from technical noise on the microarray) in the other state. According to this definition, 9,735 (88.5% of the 11,002 genes expressed over 6) of the genes are shared between both states whilst 40 (0.4%) are specific to D and 362 (3.3%) are specific to AR seeds. The remaining 865 transcripts (7.9%) are expressed over 6 in one state but between 5 and 6 in the other state, and are not classed as being highly specific to either physiological state. Similarly, a number of transcripts were specific to the MCE (299 genes) and the RAD (232 genes) which could be further refined (to 42 genes for the MCE and 48 for the RAD) using the transcript information of the Col-0 seeds (Dekkers et al., 2013, Figure 2D and Data File S1). Subsequently, we investigated genes that were both state (D or AR) and compartment (MCE or RAD) specific, although, these resulted in very small gene sets. We identified 19 transcripts specific for the D MCE while in the D RAD only one such gene was found. Higher numbers of specifically expressed genes were found in the AR state (81 in the MCE and 99 in the RAD, Figure 2D).

We performed a principal component analysis (PCA) to globally investigate the transcriptome time course. In our previous study of AR Col-0 seeds (Dekkers et al., 2013), the PCA analysis separated the samples mainly by time and tissue type. To be able to compare both studies we plotted the D and AR Cvi samples together with these AR Col-0 samples (Figure 3A). Like the AR Col-0 samples, the AR Cvi samples are mainly separated along the first principal component (PC), with that PC appearing to represent time. The AR Cvi samples do ‘move’ slower though, and in particular the later time points (12 and 24 HAS) lag behind those of the AR Col-0, which most likely reflects the slower germination rate compared to the AR Col-0 seeds (germination speed calculated as the time required to reach 50% of germination (T₅₀) was 49 h for Cvi compared to 39 h for Col-0). Intriguingly, the D samples show a very different behavior; although the D samples cluster with the AR samples at 3 and 7 HAS, at later time points (12 and 24 HAS) the samples ‘stall’ around the 7 HAS position in the PCA plot (Figure 3A). This indicates that limited transcriptional changes are occurring at the later time points in the D samples, and is supported by the number of differentially expressed genes detected between the consecutive time points in both compartments (Figure 3B). The number of differentially expressed genes (over threefold up and down) between 3–7, 7–12, and 12–24 HAS in the AR MCE/RAD were 656/740, 614/667 and 760/558 genes, respectively. For the D samples these numbers are considerably lower, most notably between the later time points, i.e., 259/270, 45/27, and 40/30 (Figure 3B). Over the whole time course (3–24 HAS), the total number of differentially expressed genes in the AR state is over 3000 in both tissues (MCE 3018, RAD 3601), which is over four times more than in the D samples (MCE 682, RAD 821; Figure 3C and Data File S1).

To get a more detailed view of the transcriptome similarities as well as the differences (and the gene expression profiles underlying these) in D and AR seeds we employed a two-step approach using the MCE samples. Since there is differential gene expression between D and AR seeds during the time course of imbibition, we have first focused on transcripts with a strong temporal regulation (which are identified in the first step, Figure 4A). We identified 3194 genes which show a strong difference in expression (defined as threefold change) during the germination time course by comparing 3 and 24 HAS in D and/or AR MCE, of which a total of 1890 genes were strongly up-regulated, whereas 1304 genes were strongly down-regulated (Figure 4A). In the second step, we tested whether these strongly differentially regulated genes are differentially expressed between D and AR MCE at the last time point (i.e., the comparison D MCE 24 HAS vs. AR MCE 24 HAS; Figure 4B). We focused on genes that are similarly expressed (an expression difference less than twofold) or differentially expressed (threefold difference) in either D or AR state. A smaller set was identified that contained genes that were neither similarly expressed nor strongly differentially expressed (showing a difference in expression between 2- and 3-fold) and these are denoted as weakly differentially expressed. Since we have selected for strongly temporally regulated genes during germination (step 1, Figure 4A), this suggests that these gene expression profiles are similar, although less pronounced in one or the other state (Figures 4C,D). Based on this analysis these temporally regulated genes were subdivided in five classes: (1) genes that are similarly expressed between D and AR MCE, (2) genes that are weakly differentially expressed in the D MCE, (3) genes that are weakly differentially expressed in the AR MCE, (4) genes that are strongly differentially expressed in D MCE, and (5) genes that are strongly differentially expressed in AR MCE (Figure 4B).

We first investigated genes that are eventually expressed at a similar level (i.e., an expression difference of less than twofold) in D and AR MCE at 24 HAS. These represent transcriptional changes that occur in response to rehydration, independent from the physiological state (D or AR). In total we found 3194 genes that are strongly regulated between 3 and 24 HAS in the MCE (Figure 4A). Of this set 1174 genes (37%, 643 genes up and 531 genes down) become eventually similarly expressed (Figure 4B and Data File S1). In Figures 4C,D, the expression profiles of the categories identified in the second step are indicated. From the germination up-regulated gene set (643 genes), overrepresentation analysis (ORA) of gene ontology (GO) classes showed the presence of genes related to key metabolic processes related to amino acid, organic acid, nucleotide and carbohydrate as well as chromatin organization and translation, among others. The 531 down-regulated genes have functional classes related to transcription, RNA processing and stress overrepresented.

A significant fraction (1337 out of 3194 genes, 42%) of temporally regulated genes become strongly differentially expressed in either D (564 genes) or AR (773) MCE at 24 HAS (Figure 4B). We subdivided the temporally strongly up- and down-regulated genes (from step 1) into those that are strongly up-regulated in the D MCE, D and AR MCE and AR MCE between 3 and 24 HAS and similar subdivision was made for the down-regulated genes (see Figure 5A). Furthermore, these sets were then subdivided based on whether the genes are strongly differentially expressed in the D or AR endosperm (step 2, Figure 5A). These however, do not represent the complete list of genes which are differentially expressed between the D and AR states in the MCE at 24 HAS (as they contain 783 differentially expressed genes in D and 824 differentially expressed genes in AR using a threefold cut-off, Figure 3D, Data File S1). This is explained by the presence of genes that are not strongly regulated during germination, which instead have opposing gene expression profiles between D and AR between 3 and 24 HAS, and these are also included in Figure 5.

A total of 783 genes are strongly differentially expressed in D MCE (vs. AR MCE at 24 HAS, Figure 5A). Four major gene expression profiles underlie this differential gene expression (containing at least 20 genes). The largest gene expression profile (457 genes, 58%) contains genes that are actually down-regulated in AR MCE (Figures 5A,B). ORA of this group identified GO functional classes related to germination/dormancy, lipid storage and stress (Figure 5B). The second largest profile in this D set consists of 221 genes (28% of 783 genes) that are not strongly temporally regulated after imbibition (less than threefold change between 3 and 24 HAS). The strong differential expression of the D set is explained by opposing gene expression profiles between D and AR when comparing 3 and 24 HAS (Figure 5B), with ORA showing the accumulation of stress and hormone related gene classes in D MCE. Less than 10% of the D UP genes are strongly up-regulated during germination in D but not in AR MCE (a set of 74 genes) which is overrepresented for genes whose function is related to response to abiotic and biotic stresses.

Four gene expression profiles underlie the strong activation of 824 genes in the AR MCE (Figures 5A,C). The largest group (649 genes, 79%) is strongly up-regulated during the time course in AR MCE, but not in D MCE (Figures 5A,C); these genes encode functions related to primary and secondary metabolism, growth and cell wall modification and are germination related. The second largest group consists of 101 genes that are strongly up-regulated (threefold) in both D and AR but are even stronger up-regulated in the AR MCE. This gene expression profile resulted in genes differentially expressed in AR MCE whose functions are related to carbohydrate and fatty acid metabolism and stress response. Interestingly, these classes overlap with those identified among the genes which became similarly expressed in AR and D MCE and those that became weakly differentially expressed in the AR MCE (see Figure 4C). Overall these analyses reveal multiple gene expression profiles that are responsible for the differential gene expression between D and AR endosperm, with ORA showing that different patterns regulate the expression of genes related to various processes.

The transcriptional dynamics in D and AR RAD, using this two-step approach, showed similar trends (Supplementary Figure S1). In the RAD a total of 3781 genes were found to be strongly regulated (up and down) between 3 and 24 HAS (step 1), of which 39% became eventually similarly expressed in both D and AR RAD, 26% became weakly and 35% became strongly differentially expressed in either the D or AR RAD (see Supplementary Figure S1 for details). The transcriptional differences (as well as the gene expression profiles underlying these) in D and AR RAD are shown in Supplementary Figure S2. At 24 HAS 780 genes were detected to be higher expressed in the D RAD, with the majority of them being down-regulated over our time course in the AR RAD although not in the D RAD (581 genes, 74%). This is similar compared to the D MCE set, although in the MCE its relative contribution is somewhat smaller (58%, see above) which is explained by a relatively larger fraction of genes with opposing gene expression profiles between the D and AR state (compare Figure 5 and Supplementary Figure S2). We detected a total of 755 genes up-regulated in the AR RAD at 24 HAS, of which the largest group (575 genes, 76%) was strongly up-regulated during our time course in AR RAD but not in the D RAD.

Finally, we employed a strategy using fold change differences to obtain gene sets that are enhanced in either D MCE or D RAD at each individual time point to investigate whether such gene sets reveal functional differences between the both compartments in the D state. The genes selected are at least threefold higher expressed in the D state and threefold higher expressed in either the MCE or the RAD. Using this strategy, we identified gene sets that are differentially expressed in D MCE (Figure 6) and D RAD (Supplementary Figure S3). The gene sets identified in the D MCE contained 7, 45, 59, and 115 genes at the time points 3, 7, 12, and 24 HAS, respectively. Using ORA of these sets identified functional classes related to abiotic and biotic stresses, protein ubiquitination and secondary metabolism (Figure 6). In comparison to the D MCE, relatively small gene sets (ranging between 11 and 30 genes) were identified in the D RAD and only a single overrepresented gene class was found (Supplementary Figure S3).

In our study we identified genes that are strongly differentially expressed (up or down) in both dormancy states in the time course between 3 and 24 HAS in the MCE (3194 genes found) and RAD (3781 genes found). Next, we investigated whether these genes were similarly or differentially expressed between D or AR MCE and RAD at 24 HAS. This revealed that a large proportion of those (37–39%) are expressed in a similar manner in D and AR seed compartments. Seed germination involves re-activation of, and changes in, cellular activity (Fait et al., 2006; Weitbrecht et al., 2011; Rosental et al., 2014), with metabolism resuming in both imbibed dormant and non-dormant seeds (Bewley, 1997). The gene sets that are strongly enhanced in both D and AR seeds (MCE and RAD) are overrepresented for GO classes including key cellular metabolic processes like translation, amino acid, organic acid, nucleotide and carbohydrate metabolism (Figure 4; Supplementary Figure S1). A total of over 500 genes were similarly down-regulated in expression in D and AR seed compartments. Overrepresented GO categories in both MCE and RAD included transcription and response to stress and other environmental cues. These genes likely represent remnants of the dry state or seed maturation that are repressed in rehydrated seeds regardless of their dormancy status.

The remaining genes become differentially expressed (Figure 4B; Supplementary Figure S1). Noteworthy is the finding that an overwhelming majority of the temporally down-regulated transcripts become differentially expressed in the D MCE whereas a vast majority of the temporally up-regulated genes become differentially expressed in the AR MCE with similar observations made in the RAD (Figure 4; Supplementary Figure S1). Among the genes that are higher expressed in AR MCE and RAD the GO classes primary metabolic process, cellular carbohydrate metabolic process, translation, and ribosome biogenesis were overrepresented (Figure 4C). This group of genes is activated in both D and AR seeds, but with a higher relative abundance in the AR state, likely to support the higher cellular activity and metabolic demand to support seed germination and subsequent seedling growth.

Using this transcriptome data, we set out to address three questions. The first question we aimed to answer is which profiles underlie the differential gene expression between D and AR seeds, and we described these patterns, genes and the functional classes they represent (Figure 5; Supplementary Figure S2; Supplemental Data File S1). Although we identified four major gene expression profiles for each, they show striking differences. Most genes that become higher expressed in D seeds are actually down-regulated in AR seeds (both MCE and RAD). As this down-regulation does not occur (or not to the same extent) in the D seeds, these transcripts accumulate eventually at a higher relative abundance in the D samples. This set includes genes related to seed maturation, seed dormancy, lipid storage and response to stress. Only a small proportion of the genes which are up-regulated during the time course in D seeds fail to do so in AR seeds (74 genes in the MCE, 38 in the RAD) and both sets contain genes related to (a)biotic stress resistance. In contrast, the largest number of genes that become higher expressed in AR seeds are strongly up-regulated during the time course in AR seeds, but not (or to a lesser extent) in D seeds. Genes in this set include those involved in (cell) growth and cell wall modification, among others that are important for seed germination. In conclusion, considering all genes that become, either weakly or strongly, differentially expressed within the dormancy states, with the overwhelming majority of temporally down-regulated transcripts becoming higher in the D MCE (740 out of 774, 95.6%) and the vice versa in the AR MCE (1143 out of 1247, 91.7%; Figure 4B), with similar proportions in the RAD. This high directional overlap emphasizes how the genes which are changing in time are changing more in the AR seeds than the D seeds.

Indeed, transcriptome changes within D seed compartments were mainly observed in a transient fashion early after sowing, with a dramatic drop thereafter resulting in a clearly reduced number of changes over the first 24 HAS in D compared to AR seed compartments. However, the genes that increase or decrease in relative abundance over the first 24 HAS in D samples are not very different since these largely overlap with genes differentially expressed in AR samples (Supplementary Figure S4). Thus, although the level of differential gene expression is far less in D samples, the genes that do change upon rehydration are largely similar to those that change in AR samples. In conclusion, it is this huge difference in the number of transcriptional changes (both up and down) in the AR seeds in contrast to D seeds, which mainly drives differential gene expression between D and AR samples.

After-ripening overcomes seed dormancy and allows germination in appropriate conditions (Bewley, 1997; Finch-Savage and Leubner-Metzger, 2006; Holdsworth et al., 2008a). Although linked to dormancy release, the after-ripening response was also observed in stored non-dormant genotypes (Carrera et al., 2008). After-ripening induces changes in seeds during dry storage; the mechanism by which this occurs is unclear, although it may include reactive oxygen species (ROS; Oracz et al., 2007; Bewley et al., 2013; El-Maarouf-Bouteau et al., 2013). We used dry seed after-ripening for 16 months to break dormancy of the D Cvi seeds. In our experiment the transcriptome differences between D and AR dry Cvi seeds are low (only seven genes show a twofold difference), indicating that genome-wide reprogramming of transcript abundances occurs mainly after water uptake (Carrera et al., 2008). This raises the second question of how quickly seeds detect and respond to the changes that occurred during dry storage. Transcriptome differences are already present at 3 HAS (which become progressively pronounced at the subsequent time points, Figure 3D) showing that seeds respond rapidly after rehydration to the changes that occurred during after-ripening in the dry seed suggesting that after-ripening pre-sets the transcriptional response following the initiation of imbibition.

Although the first changes between D and AR samples are visible at 3 HAS, the transcriptomes differ more strongly at 12 and 24 HAS, indicated by the different positions of the D and AR samples in the PCA plot at these time points (Figure 3A). It also indicates that the transcriptome of dormant seeds and that of seeds that progress toward germination start to deviate strongly between 7 and 12 HAS. Investigation of genes differentially expressed at the 7 HAS time point between D and AR samples could reveal potential transcriptional regulators that control these different transcriptional responses. Among the genes which are higher expressed (at a threefold cut-off) in both D MCE and D RAD at 7 HAS are 17 TF (of which 9 are of the ERF/AP2 class, for details see Data File S1). The gene annotations suggest that most of these TFs are related to (a)biotic stress responses indicating that the dormant seed prepares to cope with suboptimal environmental conditions. Similarly, we looked into genes which are higher expressed in the AR MCE and AR RAD at 7 HAS and identified nine TFs (Data File S1). These TFs may include candidates that are responsible for the different gene expression responses of D and AR seeds.

Lastly, we questioned whether we could identify gene sets that are both dormancy and seed compartment specifically expressed. We found very few genes (40, see Figure 2D) that were D specific in our data set. Many dormancy up-regulated transcripts are present in dry (and early imbibed) seeds and are not (or to a lesser degree) down-regulated in D compared to AR MCE and RAD (as discussed above). This explains why so few D specific genes were detected, which was even more obvious when we looked for genes that were both D and seed compartment specific, with only 19 and one genes found for D MCE and RAD, respectively (Figure 2D). Dormancy is regulated by the balance between two hormones ABA (required for dormancy induction) and GA (required for germination; Finch-Savage and Leubner-Metzger, 2006; Finkelstein et al., 2008; Bassel, 2016). Among the 19 genes in the D MCE set are a gene involved in ABA biosynthesis (NCED2) and another involved in GA degradation (GA2Ox6; Data File S1). This suggests that seed compartment specific expression of hormone metabolism genes contribute to the control of ABA/GA balance in dormant seeds.

Thus gene sets which are specific for D MCE and D RAD were identified although these were rather small. Therefore, we aimed to obtain larger gene sets which we could investigate using GO overrepresentation analysis. For this reason we employed a strategy using fold change differences to obtain gene sets that are enhanced in either D MCE or RAD at each individual time point. This resulted in larger gene sets, of which ORA identified functional classes related to abiotic and biotic stresses in the dormant endosperm (Figure 6). Stress related genes and proteins were previously found to be overrepresented in the endosperm of germinating Arabidopsis seeds and in its close relative cress (Lepidium sativum; Muller et al., 2010; Endo et al., 2012; Dekkers et al., 2013). Interesting in this respect, is a recent study using co-expression network analysis of Medicago (Medicago truncatula) seed maturation (Righetti et al., 2015), which revealed that biotic stress related genes were linked to seed longevity. The endosperm is the outermost living cell layer of the seed that is in contact with the environment, so the relatively higher abundance of (a)biotic stress related genes suggests a role for the endosperm in stress signaling and response. Arabidopsis mutants that have disturbed integrity of the testa or the cuticle layer are affected in seed germination, dormancy and longevity (Debeaujon et al., 2000; De Giorgi et al., 2015). Our results show that the expression of stress related genes is further enhanced in the endosperm of dormant compared to non-dormant seeds, the presence of which may support the survival of the dormant seed in its environment.