Site-specific phosphorylation of TRANSPARENT TESTA GLABRA1 mediates carbon partitioning in Arabidopsis seeds

Seed development is dependent on nutrients, such as a source of carbon, supplied by the parent plant. It remains largely unknown how these nutrients are distributed to zygotic and maternal tissues to coordinate storage of reserve compounds and development of protective tissues like seed coat. Here we show that phosphorylation of TRANSPARENT TESTA GLABRA1 (TTG1) is regulated by SHAGGY-like kinases 11/12 (SK11/12) and that this mediates carbon flow to fatty acid synthesis and seed coat traits in Arabidopsis seeds. SK11/12 phosphorylate TTG1 at serine 215, thus preventing TTG1 interaction with TRANSPARENT TESTA2. This compromises recruitment of TTG1 to the GLABRA2 locus and downregulates GLABRA2 expression, which enhances biosynthesis of fatty acids in the embryo, but reduces production of mucilage and flavonoid pigments in the seed coat. Therefore, site-specific phosphorylation of TTG1 by SK11/SK12 regulates carbon partitioning between zygotic and maternal sinks in seeds.

S eed development is a key process during which seed plants accumulate storage reserves in zygotic tissues that not only support seed germination and post-germinative seedling growth, but also serve as essential dietary nutrients for human beings and animals. The zygotic embryo and endosperm are usually covered by maternally derived tissues, such as seed coat, which protect the seed from external stresses to facilitate the balance between seed dormancy and germination 1 . In the model angiosperm plant Arabidopsis thaliana, reserve compounds in mature seeds are mostly stored in the zygotic embryo and mainly comprise storage proteins and oil in the form of triacylglycerols consisting of esters of glycerol and fatty acids 1,2 . While accumulation of these seed storage reserves is dependent on availability of nutrients supplied by the parent plant, it is highly subject to competition with other metabolic pathways in seeds that share the same nutrients. Sucrose is produced by photosynthetic tissues in the parent plant, and serves as a major carbon source for accumulation of seed storage oil and generation of other important seed components, such as seed coat mucilage and pigments 1,3,4 . Hydrolysis of sucrose produces glucose, which is converted by glycolysis into acetyl-coenzyme A (CoA) that can be further carboxylated into malonyl-CoA. Both acetyl-CoA and malonyl-CoA are key substrates for fatty acid biosynthesis in the embryo 1 , while malonyl-CoA also serves as a major substrate for biosynthesis of flavonoid pigments mainly consisting of proanthocyanidins (PAs) in the seed coat and flavonols in both the seed coat and the embryo in Arabidopsis seeds 5 . Meanwhile, glucose can be converted into rhamnose for further synthesis of rhamnogalaturonan I, which constitutes pectin, a primary component in seed coat mucilage [6][7][8] . Therefore, production of seed oil, pigments, and seed coat mucilage in zygotic and maternal tissues shares the same carbon source, indicating a regulatory demand for coordinating carbon partitioning among these metabolic pathways during seed development.
TRANSPARENT TESTA GLABRA1 (TTG1) encodes a WD40 repeat transcription factor that plays multiple roles in affecting post-embryonic development and seed development in Arabidopsis [9][10][11][12][13][14] . It functions post-embryonically in determining the formation of hairs on leaves, stems, and roots. During seed development, TTG1 simultaneously play distinct roles in affecting several key traits that share the same carbon source. It promotes biosynthesis of flavonoids, including PAs that are mainly present in the seed coat. Consequently, ttg1 loss-of-function mutants exhibit the transparent testa phenotype as a result of low levels of oxidized PAs in the seed coat 5,[9][10][11] . TTG1 is also required for the production of mucilage and the associated columella, both of which are absent in the seed coat of ttg1 mutants 9,15,16 . In contrast, TTG1 plays a negative role in mediating the accumulation of seed storage reserves including fatty acids during the seed maturation process 17 . These findings imply that TTG1 could act as a key regulator that mediates carbon partitioning among various metabolic fluxes in zygotic and maternal tissues during seed development.
Previous studies have indicated that TTG1 could interact with other MYB and basic helix-loop-helix (bHLH) transcription factors to form MYB-bHLH-WD40 ternary complexes that regulate the expression of other downstream genes involved in various developmental processes [18][19][20][21] . For example, it has been suggested that the complex consisting of TRANSPARENT TESTA2 (TT2; MYB123), TRANSPARENT TESTA8 (TT8; bHLH042), and TTG1 regulates the expression of BANYULS that encodes a core enzyme for PA biosynthesis in the seed coat 18 . In addition, a potential target of TTG1, GLABRA2 (GL2) 21 , also affects TTG1-relevant seed characters 3,22-24 . However, as experimental evidence for TTG1 interaction with other partners or its binding to downstream targets in Arabidopsis seeds have so far not yet been obtained, the exact in planta functions of TTG1 in molecular control of seed development remain largely unknown.
Here we show that two glycogen synthase kinase-3 (GSK3)-like kinases, SHAGGY-like kinases 11/12 (SK11/SK12), phosphorylate TTG1 to enhance fatty acid biosynthesis in the embryo while inhibiting production of mucilage and flavonoid pigments in the seed coat. We propose that phosphorylation of TTG1 by SK11/ SK12 represents a molecular framework to modulate carbon partitioning in zygotic and maternal seed tissues.

Results
SK11 and SK12 interact with TTG1. The TTG1 protein contains four typical WD40 repeats 10 , which were predicted to fold into a canonical seven-blade β-propeller 25 (Supplementary Fig. 1) that could play a role in mediating protein-protein interactions through multiple surfaces of the β-propeller as shown in common WD40 proteins 26 . To identify new interacting partners of TTG1 in seed development, we used the recombinant MBP-TTG1 protein to pull down its interacting proteins in total protein extracts from Arabidopsis Col-0 wild-type siliques. After analyzing the co-purified complexes by liquid chromatography coupled with tandem mass spectrometry (LC−MS/MS), we identified a glycogen synthase kinase-3 (GSK3)-like kinase, SHAGGY-like kinases11 (SK11; At5g26751), as a potential interacting protein of TTG1 ( Supplementary Fig. 2). There are a total of 10 GSK3-like kinases that are divided into four subgroups in Arabidopsis, which may play diverse roles in plant developmental signaling pathways 27 . SK11 shares the highest sequence similarity with SK12 (At3g05840) in the subgroup I, implying that they may have overlapping biological functions.
We then performed a detailed analysis of the interaction between TTG1 and SK11 or SK12 using various approaches. Glutathione S-transferase (GST) pull-down assays demonstrated that GST-SK11 and GST-SK12 bound to in vitro-translated MBP-TTG1, but not MBP itself (Fig. 1a). Bimolecular fluorescence complementation (BiFC) assays revealed the yellow fluorescent protein (YFP) fluorescence signal in both the nucleus and cytoplasm of tobacco epidermal cells (Fig. 1b), implying a direct interaction between TTG1 and SK11 or SK12 in living plant cells. Subcellular localization of TTG1-mCherry in tobacco epidermal cells and immunoblot analysis of 4HA-TTG1 ( Supplementary  Fig. 3) in proteins extracted from ttg1-13 g4HA-TTG1 ( Supplementary Fig. 4) Arabidopsis siliques confirmed localization of TTG1 in both the nucleus and cytoplasm. Yeast twohybrid assays also showed the interaction of SK11 and SK12 with TTG1 in yeast cells (Fig. 1c).
SK11 and SK12 are highly expressed in young seeds. To examine the global expression levels of SK11 and SK12, we performed quantitative real-time PCR analysis of total RNA extracted from various aerial tissues of wild-type plants. SK11 was expressed in all the tissues examined with the highest expression in young siliques collected before 7 days after pollination (Fig. 2a), while SK12 expression was much higher in rosette leaves and young siliques than other aerial tissues (Fig. 2b). The expression levels of SK11 and SK12 were generally unaltered in aerial tissues of ttg1-13 or tt2-5 compared to wild-type plants ( Supplementary Fig. 6a-d). SK11 and SK12 exhibited a similar trend of temporal expression in developing wild-type seeds (Fig. 2c, d). Expression of both genes remained at high levels from early to mid-embryo morphogenesis phase (0-4 days after pollination), but progressively decreased at the seed maturation phase. This pattern was comparable to the temporal expression pattern of TTG1, which peaked in seeds 4 days after pollination, but decreased afterwards ( Supplementary Fig. 6e).
To monitor the detailed expression patterns of SK11 and SK12, we transcriptionally fused the 2.6-kb and 2.3-kb 5′ upstream sequences of SK11 and SK12, respectively, to the β-glucuronidase (GUS) reporter gene. Most transgenic lines generated for SK11: GUS or SK12:GUS displayed similar GUS staining patterns. GUS signals in SK11:GUS and SK12:GUS were detectable in most vegetative and reproductive tissues examined, and were relatively strong in some tissues, such as rosette leaves and radicle tips of mature embryos ( Supplementary Fig. 7a). Furthermore, both SK11:GUS and SK12:GUS displayed staining signals in whole developing seeds 2 days after pollination (Fig. 2e, left panels; Supplementary Fig. 7b). In seeds 4 days after pollination, the strongest signals of SK11:GUS and SK12:GUS were found in the chalaza, while less strong, but evident signals were in the integument except the outmost layer that is being specialized for accumulation of mucilage constituents and starch granules (Fig. 2e, right panels). These GUS staining patterns in seeds are generally consistent with the expression data from the Arabidopsis eFP browser [32][33][34] (Fig. 2e and Supplementary Fig. 7c). Notably, robust expression of SK11 and SK12 in maternal tissues of seeds, such as the chalaza and integument, overlapped with strong TTG1 expression in these tissues ( Supplementary Fig. 7c).
TTG1 and SK11/SK12 antagonistically affect seed characters. To investigate the biological function of SK11 and SK12, we isolated two corresponding T-DNA insertional mutants, sk11 (SALK_014382) and sk12 (CS332559), respectively (Supplementary Fig. 8a). There was no detectable expression of SK11 and SK12 in sk11 and sk12, respectively ( Supplementary Fig. 8b), implying that both lines could be null mutants. As sk11 and sk12 single mutants exhibited indistinguishable phenotypes from wild-type plants, we further created sk11 sk12 double mutants since these two genes were highly similar.
Examination of seeds of sk11 sk12 and the SK11 gain-offunction transgenic line SK11:SK11 E292K -GFP (#5-1) revealed several seed coat defects associated with SK activities (Fig. 3a). In the wild-type seed coat, the outer cell layer of the outer integument deposits mucilage mainly consisting of acidic polysaccharides around a polarized secondary cell wall forming a volcano-shaped structure called the columella 16,35 (Fig. 3a). During seed imbibition, the mucilage is extruded and forms a gelatin-like coating surrounding the seed, which may facilitate seed hydration and balance seed dormancy and germination in response to environmental conditions. We found that sk11 sk12 developed larger columellae on the seed coat than wild-type plants, whereas SK11:SK11 E292K -GFP seeds had smaller columellae ( Fig. 3a and Supplementary Fig. 5c). As the formation of columella is associated with mucilage deposition on the seed coat 16 , we also observed denser or sparser mucilage extrusion on the seed coat of sk11 sk12 or SK11:SK11 E292K -GFP, respectively, as compared with wild-type plants ( Fig. 3a and Supplementary  Fig. 5c). In addition, unstained seeds of sk11 sk12 or SK11: SK11 E292K -GFP were darker or paler than wild-type seeds, respectively ( Fig. 3a and Supplementary Fig. 5c). In agreement with this observation, staining of seeds by p-dimethylaminocinnamaldehyde (DMACA) that reacts with the seed coat pigments, such as PA polymers and their precursors 36 , demonstrated higher or lower PA deposition on the seed coat of sk11 sk12 or SK11: SK11 E292K -GFP than wild-type plants ( Fig. 3a and Supplementary  Fig. 5c). As expected, the gain-of-function SK11:SK11 E292K -GFP line suppressed seed coat phenotypes of sk11 sk12 (Supplementary Fig. 5c). Meanwhile, like other ttg mutants 5,9,11,15 , ttg1-13 exhibited several seed coat defects, such as loss of columella and mucilage, transparent testa, and reduced PA deposition (Fig. 3a). Taken together, these results demonstrate that SK11:SK11 E292K -GFP partially phenocopies ttg1-13, whereas sk11 sk12 exhibits opposite seed coat phenotypes. As TTG1 also regulates the accumulation of seed storage reserves including fatty acids 17 , we further examined whether SK11 and SK12 affect fatty acid contents in mature seeds. Although SK11 and SK12 barely affected the oil fatty acid composition in seeds ( Supplementary Fig. 9a), total fatty acid levels in seeds of sk11 and sk12 single mutants were reduced compared to those in wild-type seeds, and the reduction was more remarkable in sk11 sk12 (Fig. 3b). In contrast, the levels of fatty acids in SK11:SK11 E292K -GFP in either the wild-type or sk11 sk12 background were significantly higher (Fig. 3b). These results, together with the negative effect of TTG1 on fatty acid levels in seeds 17 , indicate that TTG1 and SK11/SK12 also play an antagonistic role in controlling fatty acid levels in seeds. Consistently, a group of genes involved in synthesis of fatty acids were downregulated in developing sk11 sk12 seeds ( Fig. 3c and Supplementary Fig. 9b), whereas most of these genes were found to be upregulated in ttg1 developing seeds 17 .
TTG1 and TT2 act downstream of SK11/SK12. As TTG1 (At5g24520) and SK11 (At5g26751) are closely linked on the chromosome 5, we were unable to create their double mutants through genetic crossing. Instead, we applied treatment using bikinin, which specifically inhibits subgroups I and II GSK3-like kinases 37 , to test the genetic interaction between TTG1 and SK11/ SK12. Bikinin treatment of wild-type siliques 4 days after pollination resulted in darker seed coat and less fatty acids than mock treatment (Fig. 3d, g), which are in accordance with sk11 sk12 seed phenotypes (Fig. 3a). In contrast, Bikinin treatment almost had no effect on seed coat color and fatty acid levels of ttg1-13 ( Fig. 3e, g), indicating that TTG1 acts downstream of SK11 and SK12.
We then explored the genetic interaction between SK11/SK12 and two potential TTG1 interacting partners, TT2 and TT8, because they also affect seed coat color and fatty acid synthesis [38][39][40][41] . We first compared temporal expression of TT2 and TT8 in developing seeds and the seed coat phenotypes of their corresponding mutants. TT2 exhibited a similar temporal expression pattern to TTG1 in developing seeds, which peaked at the embryo morphogenesis phase (4 days after pollination), but subsequently decreased ( Supplementary Fig. 6e, f), whereas TT8 peaked at the seed maturation phase (8 days after pollination)  Fig. 6g). In addition, although both tt2-5 and tt8-4 seeds displayed transparent testa and reduced PA deposition, only tt2-5 exhibited the defect in mucilage extrusion (Supplementary Fig. 10) as observed in SK11:SK11 E292K -GFP (Fig. 3a). These results imply that TT2 could be more closely related to TTG1 activity in affecting seed characters. We then created tt2-5 sk11 sk12 triple mutants, and found that tt2-5 completely suppressed sk11 sk12 phenotypes pertaining to seed coat color ( Fig. 3f) and fatty acid levels ( Fig. 3g, left panel). These observations suggest that both TTG1 and its interacting partner TT2 control seed characters genetically downstream of SK11 and SK12.
While SK11 and SK12 affect the phenotypes of seed coat that is a maternal tissue, we also tested the origin of their effects on fatty acid levels in seeds. Reciprocal crosses between wild-type and sk11 sk12 plants (Fig. 3g, middle panel) and between wild-type  Scanning electron microscopy (SEM) of seed coat (first column), seed coat mucilage staining with ruthenium red (second column), seed color (third column), and seed staining with p-dimethylaminocinnamaldehyde (DMACA) (fourth column) are shown in panels from left to right. Scale bars, 25 μm, 200 μm, 1 mm, and 1 mm (from left to right). In SEM panels, "C" indicates columella, while "CC" indicates collapsed columella. SK11 and SK12 phosphorylate TTG1 in vitro and in vivo. Since SK11 and SK12 interacted with TTG1 and affected seed characters upstream of TTG1, we reasoned that TTG1 could serve as a substrate of SK11 and SK12. To this end, we performed in vitro kinase assay using Phos-tag SDS-PAGE, and detected a mobility shift of phosphorylated MBP-TTG1 by GST-SK11 and GST-SK12 in the phosphate-affinity SDS-PAGE (Fig. 4a). More importantly, we also observed a mobility shift of phosphorylated 4HA-TTG1 from protein extracts of ttg1-13 g4HA-TTG1 siliques using Phostag SDS-PAGE, whereas phosphorylation of 4HA-TTG1 was inhibited in protein extracts from bikinin-treated siliques (Fig. 4b), suggesting that TTG1 is indeed phosphorylated in vivo by GSK3-like kinases.
To identify specific phosphorylation sites on TTG1 mediated by SK11 and SK12, we analyzed the phosphorylated MBP-TTG1 products purified from the in vitro kinase assay by LC-MS/MS, and found five potential phosphorylation sites, including Ser 85, Ser 88, threonine (Thr) 169, Ser 215, and Ser 252 ( Supplementary  Fig. 11a-e). We further examined these five sites by in vitro kinase assays, and revealed that only mutation of Ser (S) 215 to alanine (A) greatly abolished in vitro phosphorylation of MBP-TTG1 by both GST-SK11 and GST-SK12 (Fig. 4c, d and Supplementary Fig. 11d, e), indicating that Ser 215 targeted by SK11 and SK12 is essential for phosphorylation of TTG1.
g4HA-TTG1 was able to fully rescue the pleiotropic defects of ttg1-13, including glabrous leaves ( Supplementary Fig. 13a), altered pattern of root hair formation ( Supplementary Fig. 13b), and abnormal seed characters (Fig. 5b, c and Supplementary  Fig. 4), such as loss of columellae and mucilage, transparent testa, reduced PA deposition, and increased fatty acid levels. Like g4HA-TTG1, g4HA-TTG1 S215E and g4HA-TTG1 S215A could equally rescue leaf and root hair phenotypes of ttg1-13 ( Supplementary Fig. 13a, b), indicating that phosphorylation of TTG1 may not affect TTG1 function in mediating epidermal morphogenesis at the post-embryonic stage.
However, g4HA-TTG1 S215E and g4HA-TTG1 S215A had opposite effects on seed characters of ttg1-13 (Fig. 5b, c). ttg1-13 g4HA-TTG1 S215E partially phenocopied SK11:SK11 E292K -GFP, showing loss of columellae and reduced mucilage extrusion, pale seeds, reduced PA deposition, and significantly increased fatty acid levels compared to wild-type seeds, whereas ttg1-13 g4HA-TTG1 S215A displayed opposite seed phenotypes, resembling those of sk11 sk12 (Figs. 3a and 5b, c). These results demonstrate that phosphorylation of TTG1 at Ser 215, which is conserved among TTG1 orthologs in various plant species (Supplementary Fig. 14a,  b), by SK11 and SK12 inhibits formation of seed mucilage and flavonoid pigments, but promotes synthesis of fatty acids, whereas unphosphorylated TTG1 has an opposite effect on these seed characters. As the seed coat defects of ttg1-13 was still slightly rescued by g4HA-TTG1 S215E (Fig. 5b and Supplementary  Fig. 12c), it is possible that TTG1 function in the seed coat is not fully phosphorylation-dependent.
The findings on the similar functions of TTG1 and TT2 in controlling seed characters downstream of SK11 and SK12 (refs. 9,15,17,38,40 ) (Fig. 3e, f, g and Supplementary Fig. 10), together with the reported in vitro protein interaction between TTG1 and TT2 (ref. 18 ) and their similar temporal mRNA expression pattern in developing seeds ( Supplementary Fig. 6e, f), prompted us to investigate whether different effects of TTG1 phosphorylation and unphosphorylation at Ser 215 on seed characters are relevant to altered TTG1 interaction with TT2. Yeast two-hybrid assays using serial dilutions of transformed yeast cells revealed that TTG1 S215E failed to interact with TT2, while TTG1 and TTG1 S215A interacted with TT2 to the same extent (Fig. 6a). In vitro pull-down assays demonstrated that TT2-His had a stronger interaction with MBP-TTG1 S215A than MBP-TTG1, but did not interact with MBP-TTG1 S215E (Fig. 6b). As MBP-TTG1 was not phosphorylated in our bacterial expression system (Fig. 4a), a stronger interaction between TT2-His and MBP-TTG1 S215A in pull-down assays may be due to the effect of S215A mutation on other biochemical characteristics of TTG1 rather than phosphorylation of TTG1. In addition, we also did not detect the interaction between TT2 and TTG1 S215E in BiFC assays (Fig. 6c). These results all indicate that the phosphorylated TTG1 at Ser 215 is unable to interact with TT2. To test this possibility in vivo, we generated 35S:TT2-GFP in wild-type, g4HA-TTG1, g4HA-TTG1 S215A , and g4HA-TTG1 S215E backgrounds, and performed co-immunoprecipitation analysis on protein extracts from their siliques 4 days after pollination. Indeed, TT2-GFP interacted with 4HA-TTG1 S215A stronger than 4HA-TTG1, but not with 4HA-TTG1 S215E (Fig. 6d), confirming that phosphorylation of TTG1 at Ser 215 abolishes the interaction between TTG1 and TT2 in siliques. We further examined the seed phenotypes of tt2-5 ttg1-13 g4HA-TTG1 S215A , and found that this plant exhibited similar seed phenotypes to tt2-5 ( Supplementary  Fig. 15), substantiating that TT2 is important for mediating TTG1 effect on seed phenotypes.
Phosphoserine 215 of TTG1 weakens TTG1 binding to GL2. To find out how interaction between TTG1 and TT2 affects downstream molecules to control seed characters, we selected GL2 as a potential target of TTG1 and TT2 based on the following two reasons. First, loss of GL2 activity results in higher levels of seed oil, deficiency in seed mucilage production, collapsed columella, light seed color and slightly reduced PA deposition 3,22-24 (Supplementary Fig. 16), which are similar to those observed in ttg1 and tt2 mutants 9,15,17,38,40 ( Fig. 3a and Supplementary Fig. 10).  Supplementary Fig. 17 Second, it has been suggested that GL2 could be a direct target of TTG1 in trichome patterning 21 , and also serves as a downstream target of TT2 in affecting fatty acid biosynthesis in seeds 40 . We then performed chromatin immunoprecipitation (ChIP) assays using siliques of ttg1-13 g4HA-TTG1 and 35S:TT2-GFP to test whether TTG1 and TT2 are directly associated with the GL2 locus in seed development, respectively (Fig. 7a-c). Chromatin immunoprecipitation assays showed that both 4HA-TTG1 and TT2-GFP were strongly associated with the transcription start site near the number 8 and 9 fragments with the highest enrichment fold (Fig. 7b, c). This observation, together with the interaction between TTG1 and TT2 (Fig. 6), implies that TTG1 and TT2 interact to physically bind to the same GL2 genomic region. We further found that 4HA-TTG1 S215E , which prevented the interaction between TTG1 and TT2 (Fig. 6), also weakened TTG1 binding to the GL2 locus (Fig. 7b). Consistently, 4HA-TTG1 binding to GL2 was significantly weaker in tt2-5 than in the wild-type background (Fig. 7b). In contrast, TT2-GFP binding to GL2 in ttg1-13 and wild-type backgrounds remained the same (Fig. 7c). Thus, while TT2 binding to GL2 is independent of TTG1, TTG1 binding to GL2 partially relies on its interaction with TT2, which is mediated by the TTG1 phosphorylation status at Ser 215.
Notably, the GL2 genomic region near the fragment 9 contained an AC-rich motif that is typically recognized by MYB transcription factors 42 , such as TT2 (Fig. 7d). To confirm our ChIP assay results, we then carried out the electrophoretic mobility shift assay (EMSA) using the biotinylated probe containing this AC-rich motif or its mutated form. Electrophoretic mobility shift assay showed that MBP-TT2 only bound to the native probe, which was specifically eliminated by native cold competitors, whereas MBP-TTG1 itself did not bind to the native probe (Fig. 7e). However, in the presence of MBP-TT2, MBP-TTG1 was able to bind to the native probe as revealed by observable supershifts because of larger complexes composed of MBP-TT2, MBP-TTG1, and the native probe (Fig. 7f). Taken together, ChIP and EMSA assays suggest that recruitment of TTG1 to the GL2 locus is at least partially dependent on its interaction with TT2.
Quantitative real-time PCR analysis revealed that in developing seeds, GL2 expression gradually increased at the embryo morphogenesis phase, and peaked at 6 days after pollination before seed maturation started ( Supplementary Fig. 6h). Its expression peak was slightly delayed compared to those of TTG1 and TT2 (Supplementary Fig. 6e, f, h). We further measured GL2 expression in siliques in various genetic backgrounds 4 days after pollination, and found that GL2 was downregulated in both ttg1-13 and tt2-5 (Fig. 7g). Its expression was recovered to the wildtype level in ttg1-13 g4HA-TTG1, but remained low in ttg1-13 g4HA-TTG1 S215E , supporting that the TTG1 phosphorylation status at Ser 215 ultimately affects GL2 expression. Consistently, as SK11 and SK12 mediate TTG1 phosphorylation at Ser 215, GL2 was upregulated in sk11 sk12, but downregulated in SK11: SK11 E292K -GFP. Similarly, bikinin treatment, which inhibits SK11 and SK12, upregulated GL2 in wild-type siliques (Fig. 7g). As upregulation of GL2 by bikinin treatment was much stronger than that in sk11 sk12, it is possible that bikini could affect other subgroups I and II GSK3-like kinases, which may share a similar role of SK11 and SK12 in affecting GL2 expression.

Discussion
During seed development, sucrose is the major form of photosynthetically assimilated carbon source that is transported from the parent plant to maternal tissues like integuments and filial tissues, such as embryos and endosperms. Sucrose, together with other nutrients including amino acids and minerals from the parent plant, contributes to both the generation of the protective seed coat and the storage of reserve compounds in filial tissues for seed germination and subsequent seedling growth 1 . As supply of nutrients from the parent plants is usually limited, loading of nutrients into dominant sinks in developing seeds should be dynamically regulated so that the development of zygotic and maternal tissues can be coordinated to well prepare seeds for successful propagation of the species in changing environments. In this study, we have revealed a molecular framework in which site-specific phosphorylation of TTG1 serves as a master switch that specifically controls TTG1-mediated carbon partitioning among metabolic pathways in zygotic and maternal tissues in Arabidopsis seeds.
During seed development, TT2 preferably interacts with unphosphorylated TTG1 and facilitates TTG1 binding to the GL2 locus to promote GL2 expression (Fig. 8). In contrast, two GSK3like kinases, SK11 and SK12, interact with TTG1 and specifically phosphorylate it at Ser 215, which compromises TTG1 interaction with TT2 and the consequent recruitment of TTG1 to the GL2 locus, thus downregulating GL2. In agreement with this molecular framework, the gain-of-function plant SK11: SK11 E292K -GFP, ttg1-13, the phosphorylation-mimicking plant ttg1-13 g4HA-TTG1 S215E , tt2-5, and gl2-3 all exhibit similar seed phenotypes, including decreased production of columellae and mucilage, pale seeds associated with reduced PA deposition, and increased fatty acid accumulation, although the severity of these phenotypes is different in these plants. On the contrary, sk11 sk12 and the dephosphorylation-mimicking plant ttg1-13 g4HA-TTG1 S215A displayed opposite seed coat phenotypes and decreased fatty acid accumulation. Therefore, the site-specific phosphorylation of TTG1 at Ser 215 by SK11/SK12 is a molecular switch that promotes fatty acid synthesis in the embryo, but reduces production of mucilage and flavonoid pigments in the seed coat. A consistent observation in plant materials examined in this study is that changes in fatty acid levels in the embryo are reversely correlated with production of mucilage and flavonoid pigments in the seed coat. This is in agreement with the understanding that the metabolic pathways in producing fatty acids, flavonoid pigments (i.e., PAs and flavonols), and mucilage in seeds compete dynamically for the same carbon source, sucrose, from the parent plant 1,3,4 . Thus, phosphorylation or dephosphorylation of TTG1 at Ser 215 plays an important role in mediating carbon partitioning among these metabolic pathways during seed development.
Notably, reciprocal genetic crosses have revealed that SK11 and SK12 maternally regulate fatty acid levels (Fig. 3g). In addition, previous studies have shown the maternal effects of TTG1 and GL2 on fatty acid levels in seeds 3,17 , and that TT2 is predominantly expressed in the seed coat and controls seed coat development 38,43 . These results support that SK11/12-mediated TTG1 phosphorylation and it interaction with TT2 as well as their effects on GL2 maternally regulate carbon partitioning in seeds. In addition, since all loss-of-function mutants discussed in this study still produce viable seeds, it seems that the defect in carbon partitioning during seed development might not directly affect seed viability under normal growth conditions. By now, there have been very few studies of the signals upstream of SK11 and SK12. It has been reported that SK11 (also known as ASKα) is induced by abiotic stresses, such as salt stress, to regulate stress tolerance by activating glucose-6-phosphate dehydrogenase (G6PD) and affecting reactive oxygen species levels 44 . In addition, another study has shown that SK11 phosphorylates G6PD, thus constituting an immune signaling module in response to pathogenic microbes and linking protein phosphorylation cascades to metabolic regulation 45 . It is noteworthy Values are mean ± s.d. of three biological replicates. Asterisks indicate significant differences in GL2 expression between indicated genotypes and wild-type plants (black bars) or between bikinin-or mock-treated wild-type plants (blue bars) (two-tailed paired Student's t test, P < 0.001). Uncropped original scans of immunoblots are shown in Supplementary Fig. 17 that G6PD catalyzes a key step of the oxidative pentose phosphate pathway that generates NADPH necessary for fatty acid biosynthesis. It will be interesting to investigate whether the link between SK11 and G6PD would influence the role of SK11 in targeting to another substrate TTG1 and the resulting effect on carbon partitioning in seeds. Although TTG1 has been shown to affect various aspects of seed development and post-embryonic development in Arabidopsis 9-14 , the site-specific phosphorylation of TTG1 at Ser 215 by SK11 and SK12 only specifically affects seed characters rather than other post-embryonic traits. g4HA-TTG1 and its phosphorylation-mimicking mutant g4HA-TTG1 S215E or dephosphorylation-mimicking mutant g4HA-TTG1 S215A equally rescue post-embryonic phenotypes of ttg1-13, including glabrous leaves and altered pattern of root hair formation ( Supplementary  Fig. 13a, b), demonstrating that TTG1 function in regulating epidermal morphogenesis at the post-embryonic stage is insensitive to phosphorylation of TTG1 at Ser 215. This is in contrast to the opposite roles of g4HA-TTG1 S215E and g4HA-TTG1 S215A in specifically affecting ttg1-13 seed characters (Fig. 5b, c). Similarly, while sk11 sk12 and the gain-of-function plant SK11:SK11 E292K -GFP barely affect the formation of leaf trichome and root hair ( Supplementary Fig. 13a, b), they display opposite seed characters in terms of production of fatty acids, mucilage, and flavonoid pigments (Fig. 3a, b). These observations, together with the result that TTG1 genetically acts downstream of SK11 and SK12 (Fig. 3d, e, g), strongly suggest that phosphorylation of TTG1 at Ser 215 by SK11 and SK12 is a hitherto unknown mechanism that enables a specific tailoring of TTG1 functions in regulating the carbon flow among the metabolic pathways involved in seed development (Fig. 8). This finding potentially provides a new means of targeted metabolic engineering of metabolites in seeds, such as oils and flavonoids, through post-translational modifications of key transcription factors, such as TTG1, which specifically affects desirable seed characters while maintaining their normal functions in regulating other plant traits.
In addition to the molecular framework described in this study, there is evidence implying that the regulatory hierarchy involving phosphorylation of TTG1 might engage other unknown mechanisms. First, LC-MS/MS analyses of in vitro phosphorylated MBP-TTG1 products have revealed several potential phosphorylation sites for SK11 or SK12 (Supplementary Fig. 11a-c). Although Ser 215 is the dominant and common site responsible for phosphorylation of TTG1 by SK11 and SK12, other sites may also contribute to TTG1 phosphorylation since mutation of other sites slightly alters the overall phosphorylation patterns mediated by SK11 and SK12 (Fig. 4c, d). Thus, it is possible that SK11 and SK12 could modify multiple residues in TTG1 and fine-tune its functions in a subtle manner. Second, despite compromised binding of 4HA-TTG1 to the GL2 locus in ttg1-13 tt2-5 g4HA-TTG1 (Fig. 7b), 4HA-TTG1 binding to GL2 is still detectable near the GL2 transcription start site in this tt2-5 background (Fig. 7b). This suggests that recruitment of TTG1 to the GL2 locus is also dependent on other unknown regulator(s) in addition to TT2. As tt2-5 exhibits relatively weak defects in seed characters than ttg1-13 ( Fig. 3a and Supplementary Fig. 10), it is possible that TT2 acts together with other co-factor(s) to recruit TTG1 at the GL2 locus. Thus, it will be interesting to further investigate whether phosphorylation of TTG1 dynamically influences its interaction with multiple co-regulators to exert its specific activity in regulating seed characters.
Plasmid construction for plant transformation. To construct g4HA-TTG1, the whole-genomic region of TTG1 was amplified into two fragments (−2526 to −1; 1-2136), and inserted with a 4HA tag by overlapping PCR. g4HA-TTG1 S215E and g4HA-TTG1 S215A were then created based on the g4HA-TTG1 fragment by overlapping PCR 46 . All these genomic fragments were cloned into the NT007 entry vector, and then introduced into the binary vector pBGW using the Gateway LR recombination reaction (Invitrogen). To construct SK11:GUS and SK12:GUS, the 2.6-kb and 2.3-kb 5′ upstream sequences of SK11 and SK12 were cloned into pBI101-GUS, respectively. To construct SK11:SK11-GFP and SK11:SK11 E292K -GFP, the coding sequences of the native SK11 and SK11 E292K were cloned into pCAMBIA1302E that was early incorporated with the same SK11 5′ upstream sequence used for SK11:GUS, respectively. To construct 35S:TT2-GFP, the coding sequence of TT2 was cloned into pCAMBIA2302. The primers used for construction of plasmids for plant transformation were listed in Supplementary  Table 1.
Bikinin treatment. For bikinin treatment, 25 μM bikinin (SML0094, Sigma) was dissolved in 1/2 MS pH 5.8 with 0.02% Silwet-77. To examine a long-term effect, bikinin solution was applied onto siliques at 4 days after pollination, and the resulting mature seeds were collected for further investigation. To examine an immediate effect of bikinin treatment in in vivo phosphorylation assay and gene expression analysis, siliques collected at 4 days after pollination were treated in bikinin or mock solution for 2 h.
Expression analysis. Total RNA from various tissues was extracted using FavorPrep Plant Total RNA Mini Kit (Favorgen) and reverse-transcribed using the M-MLV Reverse Transcriptase (Promega) according to the manufacturers' instructions. Quantitative real-time PCR was performed on three biological replicates using the CFX384 real-time PCR detection system with iQ SYBR Green Supermix (Bio-Rad). The expression of U-BOX was used as an internal control. Relative expression levels of genes were calculated by the ΔCt method 47  Microscopy and histochemical analysis. Ruthenium red staining of seed coat mucilage was carried out as follows. Dry seeds were gently washed once with water, and the rehydrated seeds were stained in 0.01% (w/v) ruthenium red solution at room temperature for 1 h for subsequent microscopic observation. For DMACA staining, dry seeds were stained with the DMACA reagent (2% (w/v) DMACA in 3 M HCl and 50% (w/v) methanol) at room temperature under dark conditions for 12 h, and then washed three times with 70% (v/v) ethanol. Treated seeds were observed using Nikon SMZ1500 (Nikon) or Axio Imager 2 (Carl Zeiss).
Pull-down assay. For LC-MS/MS analysis, Col-0 wild-type siliques were collected 4 days after pollination for extraction of total protein. The total protein extract was incubated with the recombinant MBP-TTG1 immobilized onto amylose resin for 3 h at 4°C. After washing, the pull-downed protein mixture was eluted and analyzed by LC-MS/MS using a TripleTOF 5600 System (AB Sciex). For in vitro pull-down assay, the complementary DNAs (cDNAs) encoding TT2, TTG1, and TTG1 mutated forms were cloned into pMAL-c2x vector to generate MBP-tagged proteins. The cDNAs encoding SK11 and SK12 were cloned into pGEX-4T-1 to generate GST-tagged proteins, while the cDNA encoding TT2 was cloned into pET28a to generate the His-tagged protein.
BiFC assay. The coding sequences of TTG1, TTG1 S215A , and TTG1 S215E were fused to cYPF in pXY104, while those of TT2, SK11, and SK12 were fused to nYFP in pXY106. The primers for these constructs were listed in Supplementary Table 1. The plasmids were transformed into A. tumefaciens, which was then infiltrated into Nicotiana benthamiana leaves. Leaves were observed 2 days after infiltration under a confocal microscope.
In vitro kinase assay. Kinases and substrates were expressed by inducing transformed E. coli stain Rossetta (DE3) with IPTG. After purification, they were incubated in reaction buffer (50 mM Tris-HCl pH 7.5, 50 mM NaCl, 12 mM MgCl 2 , 1 mM DTT) with or without 1 mM ATP for 1 h at 37°C. The reaction products were resolved by SDS-polyacrylamide gel electrophoresis with or without 50 μM Phos-tag (Wako Pure Chemical Industries).
Measurement of fatty acids. Fatty acids in seeds were measured by gas chromatography as follows. Acid-catalyzed transmethylation reaction on 10 mg of intact seeds was carried out in 5% H 2 SO 4 in methanol (v/v) at 90°C for 3 h, while 10 μg of heptadecanoic acid was used as an internal control. Fatty acid methyl ester was extracted by hexane and analyzed through a GC-2010 GCMS system (Shimadzu, Japan) equipped with a HP-88 fused silica column (Agilent, USA). The pulsed split mode (250°C) and helium gas (flow rate of 1 ml/min) were used as the injector condition and carrier gas, respectively, while the pulsed pressure was set at 15 psi for 0.5 min. Scan range was m/z 40-500. The oven program was: 150°C for 3 min, ramp of 7°C per min to 240°C, and keep for 5 min. ChemStation software (Shimadzu, Japan) was used for data acquisition and processing, and compounds were identified by comparison of mass spectrum with authentic standards and the NIST/EPA/NIH mass spectral library v2.0.
ChIP assay. ChIP assay was carried out on siliques in various genetic backgrounds 4 days after pollination following the reported protocol 49 with minor modifications. The extracted chromatin was sonicated to produce DNA fragments between 200 and 500 bp. The solubilized chromatin was incubated with anti-HA agarose beads (A2095, Sigma) for 2 h at 4°C. A genomic fragment of TUBLIN2 (TUB2) was amplified as a control. Chromatin immunoprecipitation assays were repeated with three biological replicates. The primers designed for ChIP-PCR were listed in Supplementary Table 1. EMSA assay. Electrophoretic mobility shift assay was performed using LightShift Chemiluminescent EMSA kit (Pierce). Recombinant proteins were produced as described in the section of "pull-down assay." The native probe was produced by PCR using 5′-biotinylated primers, while the mutated probe was produced by sitedirected mutagenesis via overlapping PCR. Competitors were produced by PCR using unlabeled primers. The primers for synthesizing these probes were listed in Supplementary Table 1.
Data availability. The MS proteomics data have been deposited to the Proteo-meXchange Consortium via the PRIDE partner repository with the data set identifier PXD008609. The authors declare that all other data supporting the findings of this study are available within the article and its supplementary information files or from the corresponding author upon reasonable request.