DELAY OF GERMINATION1 requires PP2C phosphatases of the ABA signalling pathway to control seed dormancy

The time of seed germination is a major decision point in the life of plants determining future growth and development. This timing is controlled by seed dormancy, which prevents germination under favourable conditions. The plant hormone abscisic acid (ABA) and the protein DELAY OF GERMINATION 1 (DOG1) are essential regulators of dormancy. The function of ABA in dormancy is rather well understood, but the role of DOG1 is still unknown. Here, we describe four phosphatases that interact with DOG1 in seeds. Two of them belong to clade A of type 2C protein phosphatases: ABA-HYPERSENSITIVE GERMINATION 1 (AHG1) and AHG3. These phosphatases have redundant but essential roles in the release of seed dormancy epistatic to DOG1. We propose that the ABA and DOG1 dormancy pathways converge at clade A of type 2C protein phosphatases. The DOG1 protein is a major regulator of seed dormancy in Arabidopsis. Here, Née et al. provide evidence that DOG1 can interact with the type 2C protein phosphatases AHG1 and AHG3 and that this represents the convergence point of the DOG1-regulated dormancy pathway and signalling by the plant hormone abscisic acid.

A ccurate timing of seed germination is important for the adaptation of plants to their environment. Seeds shed from plants with a life cycle adapted to seasonal changes are usually not able to germinate directly due to dormancy, which is defined as the incapacity of an intact viable seed to complete germination under favourable conditions 1 . Seeds will only germinate after dormancy has been released. Seed dormancy impacts on agricultural production and was under negative selection during domestication. Low dormancy ensures fast and uniform germination of crop seeds, but it can also lead to an unwanted early germination on the mother plant (pre-harvest sprouting) and reduced seed quality 2 . A good understanding of dormancy control will benefit both ecological understanding and crop management. The induction and release of dormancy are regulated by developmental and environmental factors. Dormancy is induced during seed maturation. Regulators of seed maturation and environmental conditions during seed development affect the strength of seed dormancy. Dormancy is released by imbibition at low temperatures (stratification) or extended dry storage of seeds (after ripening) 3,4 .
The role of hormones in dormancy and germination has been intensively studied. Abscisic acid (ABA) regulates seed maturation and is required for the induction of dormancy. Gibberellins (GA) are needed for germination. It is in particular the balance between ABA and GA that determines germination potential 5,6 . ABA regulates dormancy by reducing the activity of protein phosphatase 2C (PP2C) clade A proteins like ABA INSENSITIVE 1 (ABI1) and ABI2 7,8 . As a consequence these phosphatases lose their ability to inhibit the activity of class II SNF1-related protein kinase 2 (SnRK2) by dephosphorylation [9][10][11] . Several of these SnRKs positively control dormancy and the triple mutant snrk2.2 snrk2.3 snrk2.6 shows a loss of seed dormancy 12 .
Two major dormancy genes, DELAY OF GERMINATION 1 (DOG1) and REDUCED DORMANCY 5 (RDO5), have been identified that seem to function independent from the plant hormones. RDO5 is a member of the PP2C family of protein phosphatases that does not show phosphatase activity [13][14][15] . The molecular function of DOG1 has not been solved and its protein lacks domains with a known function. Both genes control natural variation in seed dormancy between Arabidopsis accessions 13,15,16 . Mutations in DOG1 and RDO5 completely abolish or reduce seed dormancy, respectively. DOG1 is highly conserved in the plant kingdom and homologues in various crop species have been shown to control seed dormancy 17,18 . The amount of DOG1 protein in seeds determines the time they have to be stored to release dormancy and we have shown that the protein loses its function during this after-ripening process 19 . The regulation of DOG1 is complex and involves polyadenylation, alternative splicing and self-binding of its splicing forms 20,21 . In addition, DOG1 transcription is influenced by environmental factors occurring during maturation and in the seed bank such as low temperature, which is associated with enhanced DOG1 transcript and protein levels 19,[22][23][24] . DOG1 is predominantly located in the nucleus suggesting that it might function as a transcriptional regulator 19 . DOG1 has been proposed to act by a temperaturedependent alteration of the GA metabolism, leading to weakening of the endosperm 25 . In addition, it was recently shown that DOG1 influences transcript levels of genes involved in miRNA processing causing altered accumulation of miRNAs that control developmental phase transitions in Arabidopsis and lettuce 18 . These two potential functions of DOG1 are not mutually exclusive, but none of them has been conclusively proven to be part of the primary mechanism by which DOG1 regulates dormancy. Genetic and transcriptomic analyses suggested that DOG1 is likely to function independent from ABA. However, both ABA and DOG1 have to be present to induce seed dormancy as absence of a single one of these two regulators results in complete lack of dormancy even when the other regulator is highly accumulated 16,19,24,25 . DOG1 was also shown to be required for multiple aspects of seed maturation, partially by interfering with ABA signalling components 26 . This suggests that the two pathways converge at downstream steps.
In this work, we identify four phosphatases that interact with DOG1 in dry and imbibed seeds, including the PP2C phosphatases ABA-HYPERSENSITIVE GERMINATION 1 (AHG1), AHG3 and RDO5. Genetic analysis indicates that AHG1 and AHG3 act downstream of DOG1 and are essential for its function. This suggests a model in which DOG1 controls seed dormancy by suppressing the action of specific PP2C phosphatases, which function as a convergence point of the ABA and DOG1 pathways.

DOG1 forms complexes with protein phosphatases in seeds.
The DOG1 gene is essential for dormancy in Arabidopsis and seeds lacking functional DOG1 protein are completely non-dormant 16,19 . The molecular mechanism by which DOG1 controls dormancy is not obvious because the primary amino acid sequence of DOG1 lacks domains with a known function. To obtain an insight into its function, we set out to identify proteins that interact with DOG1 in vivo using pull-down experiments. The antibody that we used for DOG1 detection in previous experiments 19 is not suitable for pull-down experiments.  Fig. 1 The YFP-DOG1 transgene complements the dog1-1 non-dormant phenotype. a Germination of freshly harvested seeds on 0.05% ethanol (control) or on 0.05% ethanol plus the indicated concentrations of fluridone. Radicle emergence was scored 7 days after sowing. Shown are averages ± s.d. of three independent batches of seeds for each genotype. b YFP-DOG1 protein accumulation in the complementation lines. (top) Ponceau S staining. DOG1 protein was detected using GFP antibody. After stripping, the same membrane was blotted with H3 antibody as loading control. Cdog1#1-4 represent four independent dog1-1 transformants containing the YFP-DOG1 transgene Therefore, transgenic dog1-1 plants were generated containing a yellow fluorescent protein (YFP) fused with DOG1 at the N-terminus, expressed from a strong DOG1 promoter derived from the Cape Verde Islands (Cvi) accession (pDOG1:YFP-DOG1). Independent homozygous single-insertion lines were selected and analysed for their seed-dormancy level. Seeds from the dog1-1 mutant germinated 100% directly after harvest whereas its wild-type background, NIL DOG1, only showed low germination (Fig. 1a). The transgenic lines (named Cdog1 #1, #2, #3 and #4) complemented the dog1-1 phenotype and their dormancy levels correlated with YFP-DOG1 protein accumulation ( Fig. 1). This indicated that our YFP-DOG1 fusion protein is functional. Seeds from the dormant NIL DOG1 and the complemented lines fully germinate on the ABA biosynthesis inhibitor fluridone confirming the essential role of ABA biosynthesis in dormancy even in seeds with high levels of DOG1 protein 19 . The line with the highest dormancy level and protein accumulation (Cdog1 #4) required 11 weeks of dry storage to fully release dormancy ( Supplementary Fig. 1a) and was selected for pull-down experiments together with the dog1-1 mutant as a control.
Dormant and non-dormant seed samples were taken from the same batch after different storage durations ( Supplementary  Fig. 1a). Proteins were isolated in their native state from dry and 24 h imbibed seeds in three biological replicates to pull-down YFP-DOG1 protein complexes using green fluorescent protein (GFP) -binding agarose beads. Gel analysis of the pulldown procedure revealed a high specificity and efficiency for YFP-DOG1 protein enrichment ( Supplementary Fig. 1b). Pulled-down proteins were analysed by quantitative mass spectrometry (Supplementary Data 1), and stringent data processing (see Methods) identified 184 protein groups, which were reproducibly quantified and exclusive for the Cdog1#4 pull-downs in all three YFP-DOG1 replicates of a given condition but in none of the corresponding control pull-downs (Fig. 2). To identify proteins present in DOG1 complexes that can be relevant for its function in seed physiology we selected targets annotated with a gene ontology (GO) biological process related to seeds. We obtained 46 matches for 17 unique proteins ( Fig. 2a and  Dormant seeds showed the highest number of proteins co-purifying with YFP-DOG1 (138 proteins in dry and 80 proteins in imbibed dormant seeds; Fig. 2b). Loss of seed dormancy induced a drastic decrease in the number of interacting protein (six proteins in dry and two proteins in imbibed non-dormant seeds). The decrease in number of interacting proteins upon dormancy alleviation might be a direct effect of the loss of DOG1 activity during after-ripening 19 . Two proteins, RDO5 and AHG1, were found to interact under all tested conditions (Fig. 2b). RDO5 belongs to the PP2C phosphatase family and has been described as a positive regulator of dormancy with seed-specific expression 14 . AHG1 is a PP2C clade A phosphatase 27 and members of this clade are negative regulators in the ABA signalling pathway 28,29 . AHG1 has a seed-specific expression pattern and shows the highest transcript levels in dry seeds among all PP2C clade A phosphatases 27 (Supplementary Fig. 2). Furthermore, among the identified interacting proteins, AHG3 and PROTEIN PHOSPHATASE 2A SUBUNIT A2 (PP2AA/PDF1) were identified in dry as well as in imbibed dormant seeds. AHG3 belongs to the same clade of PP2Cs as AHG1, while PDF1 is a scaffolding subunit A of PP2A protein phosphatase. We focussed our further studies on AHG1/3, RDO5 and PDF1 because they share relevant characteristics. They are all expressed in seeds 14,27,30,31 , interact with YFP-DOG1 in dormant seeds and are involved in the same molecular process: phosphoserine/ threonine dephosphorylation. In addition, three of them, AHG1, AHG3 and RDO5, have been implicated in either ABA signalling or dormancy 14,27,30 .
To confirm the interaction of DOG1 with the identified PP2C phosphatases and the PP2A phosphatase subunit, we performed a yeast two-hybrid GAL4 assay. Co-transformation of pACT2: DOG1 with pAS2:RDO5/AHG1/AHG3 restored yeast growth on selective medium and β-galactosidase activity. This restoration was also observed for pAS2:PDF1, although at a weaker level ( Supplementary Fig. 3). We subsequently analysed the interaction of DOG1 with these proteins in planta using bimolecular fluorescence complementation (BiFC) in epidermis cells of Nicotiana benthamiana leaves. Interaction of DOG1 with AHG1, AHG3 and RDO5 was observed in the nucleus (Fig. 3). The specificity of DOG1 binding to AHG1 and AHG3 within the PP2C clade A phosphatases was analysed by testing its interaction with another phosphatase of this clade, ABI2, which had not been identified in the DOG1 pull-down assay. Restoration of YFP fluorescence was not observed in leaves co-transformed with DOG1 and ABI2 fusion proteins despite the presence of both 4   The BiFC assay showed interaction of DOG1 with PDF1 in the cytosol. Interaction of PDF1 and DOG1 in the cytosol might explain the weaker interaction observed in Y2H since this system requires translocation of the reconstituted complex into the nucleus to activate histidine autotrophy. Overall, these results confirmed and validated the selective interaction of DOG1 with RDO5, AHG1/3 and PDF1.
DOG1-interacting phosphatases control seed dormancy. DOG1 is a key seed dormancy gene and interacting proteins that are important for its function are likely to influence dormancy. Dormancy phenotypes have not been described for pdf1 and thus we analysed its germination during seed storage. An insertion mutant in Columbia (Col) background, pdf1, showed enhanced seed dormancy ( Fig. 4a and Supplementary Fig. 5a) compared to wild-type Col. The pdf1 and dog1-2 mutants were combined by crossing and selection. This double mutant completely lacked dormancy, similar to the dog1-2 single mutant (Fig. 4a). Therefore, dog1-2 appears to be epistatic to pdf1, suggesting that DOG1 functions downstream of PDF1. Interestingly, both ahg1 and ahg3 mutants have been described to show enhanced sensitivity to ABA during germination and reduced germination speed 27,30 , but their dormancy loss during after-ripening has not been previously analysed. We obtained homozygous insertion mutants for ahg1 and ahg3 in the Col background, which were named ahg1-5 and ahg3-2 and which lacked full-length transcripts of AHG1 and AHG3, respectively ( Supplementary Fig. 5b,c). Freshly harvested seeds of wild-type Col germinated~70%, whereas ahg1-5 and ahg3-2 seeds germinated 10 and 20%, respectively (Fig. 4b). The wild-type seeds fully after-ripened after 2 weeks of storage, whereas the ahg single mutants needed 5-6 weeks of after-ripening to reach a germination percentage close to 100%. This demonstrated their enhanced dormancy phenotypes. Because AHG1 and AHG3 are closely related and the two highest expressed PP2C phosphatases of clade A in seeds ( Supplementary Fig. 2a), we tested their redundancy by constructing the double-mutant ahg1-5 ahg3-2. It has been described that this double mutant shows enhanced sensitivity to ABA compared to the single mutants 27 . We observed a very strong dormancy phenotype of the ahg1-5 ahg3-2 double mutant. Freshly harvested seeds were fully dormant and after extended storage of 22 weeks only 30% of the seeds germinated (Fig. 4b). These seeds still had the capacity to germinate because imbibition of the seeds in 100 µM GA 4+7 led to 100% germination (Fig. 4c). These results indicated that AHG1 and AHG3 have a largely redundant function in the negative regulation of seed dormancy. Since DOG1 interacts with AHG1 and AHG3 whose mutations have been shown to confer ABA hypersensitivity 27,30 , we also analysed the ABA sensitivity of dog1 mutants. A small reduction in ABA germination sensitivity had previously been shown for the dog1-1 mutant in the NIL DOG1 background 16 . Analysis of fully after-ripened seeds in the Col background confirmed the ABA hypersensitive germination phenotype of ahg1-5 and ahg3-2 single mutations and showed a decrease in ABA sensitivity for the dog1-2 mutant (Fig. 4d).
In contrast to the other three interacting phosphatases, mutations in RDO5 lead to strongly reduced dormancy 14 indicating an opposite role. This could be related with the lack of phosphatase activity of the RDO5 protein 13,15 .
AHG1 and AHG3 phosphatases are required for DOG1 function. The opposite extreme dormancy phenotypes of dog1-2 and the ahg1-5 ahg3-2 double mutant provide an ideal background to study their genetic relation by combining their mutations ( Supplementary Fig. 6). The double mutants dog1-2 ahg1-5 and dog1-2 ahg3-2 were completely non-dormant. In contrast, the triple mutant dog1-2 ahg1-5 ahg3-2 showed a very strong dormancy phenotype similar to the ahg1-5 ahg3-2 double mutant (Fig. 4b). The triple mutant was able to germinate 100% after imbibition on 100 µM GA 4+7 (Fig. 4c). Thus, the genetic analysis demonstrated that DOG1 requires the phosphatases AHG1 and AHG3 for its function and that both phosphatases function redundantly downstream of DOG1. Seeds with high DOG1 protein levels or mutations in AHG1 or AHG3 show enhanced dormancy, suggesting that DOG1 negatively influences the action of AHG1 and AHG3. Strikingly, as for the double-mutant ahg1 ahg3, the triple mutant with dog1-2 did not gradually release seed dormancy, but remained at~10-30% germination. This suggests that AHG1 and AHG3 are essential factors in the release of dormancy by after-ripening controlled by DOG1. In addition, using stratified seeds, we could show that the triplemutant ahg1-5 ahg3-2 dog1-2 had a similar sensitivity to ABA as the double-mutant ahg1-5 ahg3-2, indicating that the ahg1 ahg3 double mutant is epistatic to dog1 for both dormancy and ABA sensitivity (Fig. 4d).
The phosphatase activity of most of clade A PP2Cs is inhibited in the presence of ABA by the formation of a stable complex consisting of ABA, a member of the PYR/PYL/RCAR family of ABA receptors, and PP2C. In this complex, the receptor bound to ABA hinders the active site of the phosphatase by mimicking the SnRK class II activation loop 32 . Interestingly, the seed-specific PP2CA AHG1 was shown to be resistant to inhibition by the ABA-PYR/PYL/RCAR complexes 29 , which raised the question of its way of inhibition in seeds to allow downstream signalling required for dormancy. Our genetic analysis indicates that this function depends on DOG1. We therefore investigated whether DOG1 could directly influence the catalytic activity of AHG1 or AHG3 using recombinant protein and in vitro phosphatase activity assays (Supplementary Fig. 7). Incubation of DOG1 with AHG1 or AHG3 did not lead to any significant change in their phosphatase activities even in the presence of ABA and/or the pseudophosphatase RDO5 (Supplementary Fig. 7b). It is likely that our in vitro experimental system, using purified components from Escherichia coli, cannot mimic the in vivo situation in seeds. A missing factor (for instance other proteins, metabolites or post-translational modifications) or a specific context (limitation in water and/or oxygen availability) putatively prevent a reliable in vitro investigation of the relationship between DOG1 and PP2Cs.

Discussion
DOG1 has been identified as an essential gene for seed dormancy in Arabidopsis. DOG1 homologues regulating dormancy have been found in various species across the plant kingdom 17,18,33 .
The lack of homology of DOG1 with proteins that have a known function hindered attempts to understand its molecular function. Several laboratories proposed different hypotheses for a role of DOG1 in dormancy, which were mainly based on observations at the transcript level. These included the inhibition of expression of GA-regulated genes encoding cell wall remodelling proteins and alterations in seed GA metabolism 25 , reduced expression of genes associated with microRNA processing leading to reduced levels of miR156 18 , and a role in seed maturation by interference with ABA signalling 26 . In this work we identified two protein phosphatases (AHG1 and AHG3) that interact with DOG1 in the seed and that are essential for its function. This provides an important direct insight into the mechanism by which DOG1 controls seed dormancy and we propose that previously identified mechanisms might function downstream of this direct mechanism. The phosphatases AHG1, AHG3, RDO5 and PDF1 were found among the proteins that interact with DOG1 in seeds. Plants with mutations in the corresponding genes showed altered seed dormancy. Three of these phosphatases belong to the PP2C family while PDF1 encodes one of the three scaffolding subunits of the PP2A family. Relatively little is known about these scaffolding units. One of them (RCN1) has a major role in the regulation of phosphatase activity and its mutant showed several defects, mainly in relation with hormone signalling 34,35 . In contrast, mutants in the other two units (pdf1/pp2aa2 and pdf2/pp2aa3) did not show obvious phenotypes 31 . Here, we identified a negative role for PDF1 in seed dormancy. Our genetic analysis suggests that PDF1 acts upstream of DOG1 because the dog1 mutant phenotype is epistatic to pdf1. RDO5 belongs to the PP2C phosphatase family but lacks phosphatase activity 13,15 , therefore it might act as a pseudophosphatase. In agreement with a role as pseudophosphatase, the rdo5 mutant showed reduced dormancy in contrast to the other three identified phosphatases interacting with DOG1 that are functional and which mutants all showed enhanced dormancy.
Most importantly, DOG1 interacts in seeds with the PP2C phosphatases AHG1 and AHG3, which have been previously identified based on their ABA hypersensitive mutant phenotypes 27,30 . In this work, we demonstrated that AHG1 and AHG3 have redundant roles in seed dormancy because the phenotype of the double mutant is much more severe than that of the single mutants (Fig. 4b). The double mutant shows extreme dormancy and did not germinate more than 20-30% after 6 months of storage. These low germination rates are not due to a loss of viability of the seeds because stored ahg1 ahg3 doublemutant seeds could germinate 100% when imbibed in GA 4+7 . Overall, this indicates a crucial role for AHG1 and AHG3 in the We further demonstrated that these two phosphatases are essential for DOG1-dependent seed dormancy control. Our genetic data indicate that the interaction of DOG1 with AHG1 and AHG3 can negatively affect the function of these PP2Cs rather than that these phosphatases control DOG1 activity by phosphorylation. The triple mutant dog1 ahg1 ahg3 showed the same dormancy and ABA sensitivity phenotypes as the ahg1 ahg3 double mutant, indicating that DOG1 functions upstream of these PP2Cs. Enhanced DOG1 protein amounts 19 and absence of AHG1 and AHG3 (in the ahg1 ahg3 double mutant) both lead to increased dormancy. Therefore, our genetic analysis suggests that DOG1 acts as a suppressor of AHG1 and AHG3 action in dormancy release.
ABA and DOG1 are both required for seed dormancy. Genetic and physiological experiments suggested that they act independently although they are likely to have common downstream targets 19 . The mechanistic basis of the crosstalk between ABA and DOG1 has been mentioned as one of the outstanding open questions in seed biology 36 . In this work, we provide data indicating that the DOG1 and ABA pathways are connected at the clade A PP2Cs.
The clade A PP2Cs function as key negative regulators of the ABA signalling pathway. In accordance with their role in ABA signalling and their expression pattern, mutants in the clade A PP2Cs show varying seed-dormancy phenotypes. Most of the nine phosphatases belonging to this clade are inhibited by the PYR/PYL/RCAR family of ABA receptors in the presence of ABA 7,37 . Interestingly, the seed-specific clade A PP2C AHG1 was shown to be resistant to inhibition by ABA 29 , although it does negatively control seed dormancy. Our observation of AHG1 and AHG3 being complexed with DOG1 in seeds and the fully epistatic behaviour of the ahg1 ahg3 double mutant over dog1 indicate that AHG1 and AHG3 actions are suppressed by DOG1. However, the in vitro assays did not reveal the exact molecular mechanism by which DOG1 represses phosphatase activity in seeds. It is possible that DOG1 controls PP2C action by a mechanism different than catalytic activity inhibition upon binding. The inhibition of AHG1 is likely to be controlled by DOG1 and that of AHG3 by both DOG1 and ABA (Fig. 5). Other PP2Cs are inhibited by the ABA-receptor complex and were not identified as interactors of DOG1 (Figs. 2b, 3 and Supplementary Data 1).
We propose that the ABA and DOG1 pathways converge at the level of the clade A PP2C phosphatases. The inhibition of the clade A PP2Cs by both ABA and DOG1 is supported by the reduced sensitivity to ABA of the dog1 mutant (Fig. 4d). The functional redundancy of AHG1 and AHG3 in seed dormancy and their common inhibition by DOG1 but different inhibition by ABA could explain previous observations about the relation between DOG1 and ABA. Enhanced expression of DOG1 in ABA biosynthesis mutants cannot induce dormancy 19 . This is probably due to the activity of PP2Cs that are not or only poorly inhibited by DOG1 and that are still able to promote germination. Similarly, the lack of dormancy of mutants with enhanced ABA levels in a dog1 mutant background can be explained by the activity of AHG1, which is probably not inhibited by the conventional ABA pathway 29 .
ABA has multiple functions in plant development and stress resistance, including seed maturation and dormancy. The main function of DOG1 is to promote seed dormancy, although a role in seed maturation has recently also been demonstrated 26 . The inhibition of germination by ABA and DOG1 through both common and separate PP2Cs would enable a flexible but robust way to shape dormancy.
The role of DOG1 in seed dormancy was shown to be conserved between several dicot and monocot species. Therefore, it is likely that the mechanism of DOG1 action that we propose for Arabidopsis will be conserved within the plant kingdom and could have evolved as an early adaptation of seed plants to survive seasonal conditions.

Methods
Plant material and growth conditions. The dog1-1 mutant is in the NIL DOG1 background, which is a near isogenic line that contains the DOG1 allele from Cvi in a Landsberg erecta (Ler) background 16 . The dog1-2 mutant is in the Col background 19 . AHG1, AHG3 and PDF1 insertion lines were obtained from the NASC collection with the following seed stock numbers: ahg1-5, SALK_049885C; ahg3-2, SALK_028132; pdf1, SALK_037095. The gene-specific primer sequences were obtained from the SALK SIGnAL database. PCR with reverse transcription with DNA and RNA isolated from dry or imbibed seeds was performed to confirm the homozygous mutant lines ( Supplementary Fig. 5 and Supplementary Table 2). Double mutants were constructed by crosses between the single mutants dog1-2, ahg1-5 and ahg3-2 and selected from the self-pollinated progeny of F2 plants using PCR to confirm their homozygosity. The triple-mutant dog1-2 ahg1-5 ahg3-2 was obtained by crossing the double mutants dog1-2 ahg1-5 and dog1-2 ahg3-2 ( Supplementary Fig. 6).
Seed batches were obtained from plants cultivated on soil in a growth chamber with a 16-h-light/8-h-dark cycle (22/16°C) with the exception of the seed batches used in Fig. 4a that were obtained from plants cultivated at lower temperature (16/14°C), conditions that enhance dormancy 4 . Freshly harvested seeds were immediately used for experiments or stored under constant conditions (21°C, 50% humidity, in the dark) for after-ripening treatment. All comparative germination assays presented in this work use seed batches obtained at the same time from the same environment.
Germination assays. About 50 seeds were plated onto a filter paper moistened with demineralized water, 100 µM GA 4+7 , 0.2-3 µM ABA or 5-10 µM fluridone in Petri dishes and incubated in a growth chamber (12 h light/12 h dark, 25/20°C cycle). Stock solutions of ABA, fluridone and GA were dissolved in ethanol (final concentrations in the assays were 0.05%). When imbibed on chemicals, the control corresponds to 0.05% ethanol in water. Radicle emergence was scored after 7 days.
Construction of YFP-DOG1 transgenic lines. Binary constructs were prepared using the Gateway Technology (Invitrogen). A chimeric DNA fragment of YFP and a genomic DOG1 fragment from Cvi (ATG to 1.1 kb downstream of the stop codon of alpha/delta splicing variants) was generated by fusion PCR using the adapter sequence 5′-ccgcagcagcccccttcacc-3′ and cloned into the pENTR/D-TOPO vector. The DOG1 promoter region from Cvi of 2.17 kb (corresponding to the upstream region of Ler genomic fragment 19,21 ) was inserted in the entry clone pENTR:YFP-DOG1g_Cvi using the unique NotI site. The binary construct pGWB1:ProDOG1: YFP-DOG1g_Cvi was produced from the above mentioned entry clone and the destination vector pGWB1 38 by LR reaction. The resulting pDOG1_Cvi:YFP-DOG1 Cvi construct was introduced by electroporation into Agrobacterium tumefaciens strain GV3101, which was subsequently used to transform dog1-1 mutant ABA DOG1 AHG3 PP2CAs AHG1 Release of seed dormancy Gene constructs. Full-length coding DNA sequence (CDS) for AHG1, AHG3 and PDF1 were amplified from cDNA of dry Ler seeds using gene-specific primers including or not the stop codon. The PCR product was extended using the attB adapter sequences and cloned into pDONR201 or 207 (Invitrogen) using BP reactions. The pDONR207: RDO5 plasmid containing the CDS (including stop) of Ler RDO5 as well as the cloning of N-terminus truncated AHG3 (starting at amino acid position N88) was already described 14,15 . To obtain the DOG1 expression clone for recombinant protein production, the full-length CDS of β-DOG1 from Cvi was amplified from the entry clone 21 using gene-specific primers including a BamHI site extension in the reverse primer. A synthetic double-strand DNA fragment containing an NcoI restriction site, a STREP-tag and a linker sequence was generated by denaturation and annealing (from 98°C to room temperature) of complementary primers. DOG1 PCR product and the reconstituted tag sequence were digested with BamHI and NcoI respectively and the restricted fragment was purified from agarose gel. The chimeric STREP:DOG1 fragment was generated by blunt end ligation of digested products and cloned into pET16b (Novagen) using NcoI and BamHI sites. Sequences of all constructs were verified by Sanger sequencing.
Yeast two-hybrid assay. The CDS (including stop) of AHG1, AHG3, RDO5 and PDF1 were recombined from entry clones in the pAS2-gateway (GAL4 BD fusion) vectors (modified from Clontech) using LR reactions. The pACT2-gateway (GAL4 AD fusion; modified from Clontech) pACT2:DOG1 and pAS2:DOG1 containing the β-isoforms of DOG1 from Cvi were previously described 21 . Gal4-DNA-binding domain fusion proteins (prey) and GAL4 activation domain fusions protein (bait) vectors were transformed in the yeast strain PJ69-4alpha using a LiAc/SS carrier DNA/PEG method 40 . Co-transformed colonies were selected on selective medium (-LW) lacking Leu (L) and Trp (W). Interaction tests were performed on -LWH medium (-LWH) lacking L, W and His (H) with 5 mM 3-aminotriazole. Yeast was grown at 30°C for 7 days. β-galactosidase assay was performed by pressing filter papers on the plates with yeast colonies. The filters were frozen in liquid nitrogen and subsequently incubated at 30°C in a solution containing 0.1% X-gal.
Bimolecular fluorescence complementation. The full-length AHG1, AHG3, ABI2, PDF1 and RDO5 coding sequences (without stop codon) were recombined from entry clones in the pBatTL-B-sYFPc gateway vector (gift from Joachim Uhrig; MPIPZ Cologne, Germany) using LR reactions. The pBatTL-B-sYFPn containing DOG1 beta was already described 21 . Constructs were transformed into the Agrobacterium strain GV3101. Overnight cultures were diluted to an OD 600 of 0.5 in resuspension buffer (10 mM MgCl 2 , 10 mM MES (pH 5.7) and 100 μM acetosyringone and injected into 4-6-week-old N. benthamiana leaves with a syringe. Fluorescence within the infiltrated regions was visualized after 2 days using a confocal laser scanning microscope CLSM Zeiss LSM 700. The exact same leaves used for imaging were frozen in liquid nitrogen and used to extract total protein for western blotting.
Production and purification of recombinant protein. Full-length CDS of AHG1, or N terminus truncated CDS of AHG3 were recombined from entry clones in the pDEST-HIS-MBP or pDEST17 gateway vector, respectively (Addgene). Expression constructs were introduced in BL21(DE3)pLysS and expression of fusion proteins was induced by 1 mM IPTG when bacterial culture reached an OD 600 of 0.6. After induction cells were grown at 23°C overnight, harvested by centrifugation and stored at −80°C until protein extraction. Cell pellets were resuspended in either HIS lysis buffer: 50 mM HEPES pH 7.5, 500 mM NaCl, 10% (v/v) glycerol, 5 mM DTT, 1 mM PMSF, and 25 mM imidazole or in STREP lysis buffer: 50 mM HEPES pH 7.5, 150 mM NaCl, 5 mM DTT and 1 mM PMSF. Cells were disrupted by sonication. The soluble protein fractions were separated from the insoluble fractions by centrifugation. Soluble protein fractions containing 6xHIS-MBP-AHG1, 6xHIS-N88AHG3 or STREP-DOG1 were applied on HisTrap (GE Healthcare, USA) or STREP-Tactin Superflow (IBA) 1 ml column. The purification process was performed and monitored using AKTAprime and Primeview software (GE Healthcare). Washes were performed with resuspension buffer and elution with the resuspension buffer supplemented with 500 mM imidazole or 5 mM desthiobiotin for HIS and STREP purification, respectively. Excess of salts and elution agents were removed from purified proteins by desalting against 50 mM HEPES pH 7.5, 150 mM NaCl, 10% (v/v) glycerol using PD-10 columns (GE Healthcare). Desalted purified protein were concentrated tõ 2 mg ml −1 using Amicon Ultra-15 Centrifugal Filter Unit with Ultracel-10 membrane (Millipore) and stored at −80°C. N-truncated recombinant PP2C or MBP-PP2C fusion has previously been described to be active and suitable for PP2C inhibition assays 29,41 . Production and purification procedures of 6xHIS-RDO5 were already described 15 . All purification steps were monitored by SDS-PAGE and all recombinant proteins could be successfully purified to full homogeneity ( Supplementary Fig. 7).
In vitro phosphatase assays. In vitro phosphatase activity of recombinant proteins was performed with the Serine/Threonine Phosphatase Assay system (Promega, USA) using the synthetic RRA(pT)VA phosphorylated peptide as a substrate in a final volume of 50 μl in half-area, flat-bottom 96 wells plate. Assays were performed as follows: 0.5 μM of purified recombinant 6xHIS-MBP-AHG1 or 6xHIS-N88AHG3 were incubated for 20 min at room temperature in 50 mM HEPES buffer pH 7.5; 10 mM MgCl 2 in the absence or presence of purified recombinant STREP-DOG1 or/and purified 6 × HIS-RDO5 or/and 50 µM ABA, before starting the phosphatase activity assay by addition of the synthetic substrate at a final concentration of 50 µM. After 10 min of reaction time the activities were stopped by addition of 50 μl of molybdate dye solution. Dye was developed for 30 min, and absorbance was read at 630 nm on a multi scanspektrum (Thermo) 96 well plate reader. Specific activities were calculated using serial dilution of free phosphate as standard. In this experimental setup our recombinant purified PP2Cs, when incubated alone, displayed specific activities of 45.77 ± 4.29 and 132.35 ± 9.6 pmol of Pi min −1 μg −1 for HIS-MBP-AHG1 and HIS-N88AHG3 respectively. Full length 6 × HIS-RDO5 was already described as a PP2C without phosphatase activity 15 .
Protein extractions. One hundred milligram of seeds (dry weight) were pulverized in liquid nitrogen with mortar and pestle. Soluble or total proteins were directly extracted from the powder in native or denaturing conditions. For native extracts, soluble proteins were resuspended for 2 h at 4°C under constant rotation in a native buffer containing 50 mM HEPES pH 7.5; 150 mM NaCl, 1 mM EDTA; 1 mM EGTA; 0.25% (w/v) Triton X-100; 5 mM DTT; 50 U ml −1 DNase I (Roche); 5 U ml −1 RNase A (Machery Nagel); 1 U ml −1 macerozyme (Boehringer-Mannheim). For total extracts, proteins were solubilized in a buffer containing 50 mM HEPES pH 7.5; 2.5% (w/v) SDS; 5 mM DTT. Both soluble and total protein extraction buffers were supplemented with 1% (v/v) of protease inhibitor cocktail special plant (Sigma) and 1% (v/v) phosphatase inhibitor cocktail 2 and 3 (Sigma). Protein fractions from seed extracts were purified by repeated centrifugation until lipid free samples were obtained. Protein concentrations of extracts were quantified by BCA protein assays (Thermo) against BSA standard dilution curve and directly used for downstream applications.
Protein gel and immuno-detection methods. Proteins (10 µg of total extract unless otherwise stated in the figures legends) were separated using NuPAGE ® Bis-Tris gels (Thermofischer) and MES buffer. Protein gels were stained using either Oriole™ fluorescent gel stain (Biorad), or colloidal Coomassie Brilliant Blue. For Western blot, proteins were transferred on immoblion-P PVDF membrane (Milipore) using semidry transfer. After transfer proteins were stained with Ponceau S before blocking with 3% BSA. Detection of YFP-tagged DOG1 in seed extracts was performed using anti-GFP (7.1 and 13.1 mixture) monoclonal antibody produced in mouse (Roche #11814460001) at a dilution of 1:1,000. Detection of C-terminal HA-cYFP and MYC-nYFP fusion protein in tobacco leaf extracts was performed using HA tag antibody produced in rabbit (Abcam ab9110) and MYC tag antibody raised in mouse (MA1 980 Thermofischer) at a dilution of 1:7,500 and 1:1,500, respectively.
Detection of histone H3 was performed using anti-H3 antibody produced in rabbit (Abcam ab1791) at a dilution of 1:12,000. Horse radish peroxidase conjugated secondary antibodies were goat anti-mouse IgG (Sigma A3562) or goat anti-Rabbit IgG (Sigma A0545) both at a dilution of 1:30,000. Amersham ECL Prime (GE Healthcare Life Sciences) was used for detection of peroxidase activity. Images were recorded using Chemidoc XP imager (Biorad). Full scan images are available in Supplementary Fig. 9.
YFP-DOG1 pull-down procedure. Native seed protein extracts from 5 and 29 weeks stored Cdog1#4 and dog1-1 dry and 24 h imbibed seed were used for the pull-down assays. Pull-downs were performed from three independent biological replicates each consisting of a mix of seeds from three different plants. For each pull-down, 25 µl of agarose beads coupled to a GFP antibody (GFP-Trap_A Chromoteck gta-10) corresponding to 10 µg GFP-binding capacity was equilibrated extensively in native extraction buffer. For each replicate, an adjusted quantity of 4 mg of total protein (in 1 ml final volume) was incubated with the beads under constant rotation for 2 h at 4°C. After incubation, beads were separated from the non-bound fraction by centrifugation and washed three times using 500 µl of native extraction buffer without Triton X-100, DNase I and RNase A. Bound proteins were eluted by incubation for 5 min with 0.1% (v/v) trifluoroacetic acid. Elutions were recovered from the beads by centrifugation and immediately neutralized by an equal volume of neutralization buffer (8 M urea; 100 mM Tris-HCl pH 8). All steps of the enrichment procedure were monitored by SDS-PAGE analysis and western blotting (Supplementary Fig. 1b).
Mass spectrometry sample preparation and analysis. Eluted proteins for pulldown assays were reduced, alkylated and digested in solution 42 . Cysteines were reduced by adding DTT to a final concentration of 5 mM and incubation for 30 min. Subsequently, alkylation was performed by adding chloroacetamide to a final concentration of 14 mM and incubation for 30 min. The reaction was quenched by addition of DTT. Urea concentration was adjusted to 2 M by dilution with 0.1 M Tris-HCl pH 8.0, 1 mM CaCl 2 . Trypsin digestion (1:100 enzyme-to-protein ratio) was performed over night at 37°C and stopped by addition of 1% formic acid. Peptides were desalted with StageTips 43 (Empore C18, 3 M), dried and directly subjected to MS measurement.
Mass spectrometry data processing. Raw data were processed using MaxQuant software (version 1.5.1.2, http://www.maxquant.org/) with label-free quantification (LFQ) enabled 44 . Tandem mass spectrometry spectra were searched by the Andromeda search engine against the Arabidopsis TAIR10_pep_20101214 database (ftp://ftp.arabidopsis.org/home/tair/Proteins/TAIR10_protein_lists/). In addition, the protein sequences of the three putative splicing variants 21 of the YFP-DOG1 (Cvi allele) fusion proteins were added to avoid bias due to DOG1 polymorphisms between Cvi and Col. Sequences of 248 common contaminant proteins and decoy sequences were automatically added during the search. Trypsin specificity was required and a maximum of two missed cleavages was allowed. The minimal peptide length was set to seven amino acids. Carbamidomethylation of cysteine residues was set as fixed, oxidation of methionine and protein N-terminus acetylation as variable modifications. Peptide-spectrum-matches and proteins were retained at a false discovery rate of < 1%.
Analysis of the pull-down samples. Subsequent statistical analyses of the MaxLFQ values were performed in Perseus (version 1.5.2.6, http://www.maxquant. org/). A high reproducibility between replicates for a given genotype and condition was generally observed (Supplementary Fig. 8). Quantified proteins were only retained in the final analysis if they had valid values in at least two out of the three biological replicates. MaxLFQ values intensities were log 2 transformed. Twosample t-tests were performed with a P-value cutoff of 5%. The Perseus output of the pull-down data allowed us to quantify in total 364 protein groups (Supplementary Data 1). The 364 identified protein groups were further filtered in Microsoft Excel to select proteins that were exclusively identified in all the three replicates of the YFP-DOG1 pull-downs (absent in dog1-1 background pull-down). This filtering highlighted 184 proteins.