New roles of NO TRANSMITTING TRACT and SEEDSTICK during medial domain development in Arabidopsis fruits

A large part of our food comes from floral parts, fruits and seeds. Therefore, a deep understanding of the regulatory networks guiding the developmental processes of these structures and tissues is important. Flowering species mostly give rise to the pistil, or so-called gynoecium, in the center of the flower. The gynoecium, from a biological point of view, is essential for plant reproduction. In general, at the apical end it has a stigma to facilitate pollen capture and germination, and the stigma is connected via the style to the ovary where the ovules will be formed. The transmitting tract facilitates pollen tube growth through the style and the ovary, and, upon fertilization inside each ovule, seed development starts. The gynoecium is now called a fruit, which increases rapidly in size owing to hormones produced by the seeds (Roeder and Yanofsky, 2006; Alvarez-Buylla et al., 2010; Ferrándiz et al., 2010; Sotelo-Silveira et al., 2013; Marsch-Martínez and de Folter, 2016).

In Arabidopsis, the correct formation of the medial domain in the gynoecium is a key process for female reproductive competence and seed formation. This domain includes placental tissues and ovules, and the structures that capture the pollen grains and guide pollen tubes to reach the ovules and, therefore, facilitate fertilization. These structures and tissues, including stigma, style, septum and transmitting tract, are also known as the marginal tissues (Fig. 1A). These tissues arise from the carpel margin meristem (CMM) (Bowman et al., 1999; Alvarez and Smyth, 2002; Nole-Wilson et al., 2010; Wynn et al., 2011; Reyes-Olalde et al., 2013), a meristematic tissue that emerges as two internal ridges (termed medial ridges) in the young gynoecium (Fig. 1A), which fuse together when they reach each other in the middle of the gynoecium, thereby forming the septum. This postgenital fusion occurs at stage 9 of gynoecium development (Bowman et al., 1999; Roeder and Yanofsky, 2006). Ovule primordia can be seen at stage 9 (Bowman et al., 1999; Roeder and Yanofsky, 2006; Reyes-Olalde et al., 2013). At stage 11, the gynoecium fully closes and the stigma is then fully developed. During stage 12, the style and the transmitting tract differentiate, and at stage 13 the gynoecium is fully mature (Smyth et al., 1990; Bowman et al., 1999; Roeder and Yanofsky, 2006; Reyes-Olalde et al., 2013).

Over 80 genes have been identified as regulators of medial domain development, mainly participating at stages 9 to 11 (Reyes-Olalde et al., 2013). For instance, in the case of the postgenital fusion of the medial ridges, the basic helix-loop-helix (bHLH) gene SPATULA (SPT) has been found to be an important player (Alvarez and Smyth, 1999, 2002; Heisler et al., 2001; Reyes-Olalde et al., 2017). The formation of the stigma and style is controlled by NGATHA (NGA), STYLISH (STY) and HECATE (HEC) genes (Gremski et al., 2007; Alvarez et al., 2009; Trigueros et al., 2009). SEEDSTICK (STK) directs ovule specification, funiculus development, and seed abscission (Favaro et al., 2003; Pinyopich et al., 2003; Balanzà et al., 2016). Fertilization is a key process for sexual reproduction, and an important point in this process is that the pollen tubes can reach the ovules. The synergid cells, in the embryo sac inside the ovule, produce signals to attract the pollen tube (Mizuta and Higashiyama, 2018). For pollen tubes to reach the ovules, cell wall modifications have to take place (Crawford and Yanofsky, 2008; Dresselhaus and Franklin-Tong, 2013). On the female side, these modifications take place when the transmitting tract forms. Cells in this tissue produce an extracellular matrix (ECM) containing glycoproteins, glycolipids and polysaccharides that facilitates pollen tube growth (Lennon et al., 1998; Crawford and Yanofsky, 2008). A genetic pathway controlling transmitting tract formation includes the three redundant bHLH HEC transcription factors (Gremski et al., 2007), the HALF FILLED (HAF) gene (also known as CESTA, CES), which acts redundantly with the closely related BRASSINOSTEROID ENHANCED EXPRESSION 1 (BEE1) and BEE3 genes (Crawford and Yanofsky, 2011), and the zinc-finger transcription factor NO TRANSMITTING TRACT (NTT), which controls this process in the ovary but not in the style (Crawford and Yanofsky, 2011). All these genes contribute to ECM production and programmed cell death (Crawford et al., 2007; Crawford and Yanofsky, 2011). Furthermore, other genes expressed in the style and transmitting tract encode enzymes that modify cell walls (Dresselhaus and Franklin-Tong, 2013), e.g. beta-1,3-glucanases (Delp and Palva, 1999). On the male side, growing pollen tubes secrete cell wall-degrading enzymes that help pollen tubes on their way through the pistil (Mollet et al., 2013; Hepler et al., 2013), e.g. the pectin methylesterase VANGUARD1 (VGD1) (Jiang et al., 2005).

The NTT transcription factor, besides its role in transmitting tract formation, is also important for root meristem development (Crawford et al., 2015), and during fruit development NTT is involved in valve margin formation (Chung et al., 2013) and replum development (Marsch-Martínez et al., 2014). In the latter report, we detected protein-protein interactions between NTT and other fruit-related transcription factors, including some MADS-box proteins such as SHATTERPROOF1 (SHP1) and SHP2.

The MADS-box transcription factor STK has been well-characterized in ovule and funiculus development, and is involved in ovule identity determination together with SHP1 and SHP2 (Colombo et al., 1995; Favaro et al., 2003; Pinyopich et al., 2003). Furthermore, it has been shown that STK participates in seed development by controlling secondary metabolism (Mizzotti et al., 2012, 2014), cell wall properties (Ezquer et al., 2016) and seed abscission (Balanzà et al., 2016). Here, we report NTT protein interaction with STK and describe novel roles for NTT and STK during medial domain development, further demonstrating that they are important for the reproductive competence of Arabidopsis plants. Our results indicate that NTT and STK are involved in the control of early events of gynoecium development, such as septum fusion, septum cell integrity, impact fertilization efficiency and seed-set, and affect senescence after fertilization. They control genes involved in carbohydrate metabolism and lipid distribution in septum cell walls.

We recently reported that the transcription factor NO TRANSMITTING TRACT (NTT) promotes replum development, and that it interacts in the yeast two-hybrid (Y2H) system with proteins related to fruit development such as FRUITFULL (FUL), REPLUMLESS (RPL), SHP1, SHP2 and SHOOT MERISTEMLESS (STM) (Marsch-Martínez et al., 2014). We expanded this interaction survey with a uni-directional Y2H screen (see Materials and Methods) and found that NTT was able to interact with an additional 24 transcription factors (Fig. 1B; Table S1), suggesting that NTT fulfils various roles by forming part of different protein complexes. Of particular interest is the fact that NTT interacted with all MADS-box proteins tested.

In this work, we focused on the interaction of NTT with the MADS-box protein SEEDSTICK (STK), which is known to provide the D-function for ovule identity (Favaro et al., 2003; Pinyopich et al., 2003). In the Y2H assay, the combination NTT-STK activated all three reporter genes (HIS3, ADE and lacZ), indicating that these proteins are able to interact (Fig. 1C).

To confirm this Y2H result, a bimolecular fluorescence complementation assay (BiFC; Fig. 1D-F) was performed. For this, NTT was fused to the cYFP and STK fused to nYFP region, and fluorescence from the reconstituted YFP was detected in leaf cells (Fig. 1D), indicating that the two proteins interact in planta, confirming the Y2H result. Fluorescence was observed in the nucleus, in agreement with the expected localization of transcription factors.

The Y2H and BiFC assays suggested that NTT and STK could be interacting during gynoecia development in Arabidopsis, as both participate in this process. In order to visualize those regions where these proteins could be acting together, we analyzed transverse thin sections of stage 7 to stage 13 gynoecia of the reporter lines NTT::GUS and STK::GUS (Kooiker et al., 2005).

Activity of the NTT promoter was detected from stage 9 to 13 gynoecia in the medial domain (Fig. 1I-K), as reported before (Crawford et al., 2007; Chung et al., 2013; Marsch-Martínez et al., 2014). The activity of the STK promoter was visible in the medial domain from stage 8 till stage 13 gynoecia (Fig. 1M-P). Blue staining was observed in the medial domain in ovule primordia and later in ovules and the septum (Fig. 1N-P), in agreement with previous reports (Kooiker et al., 2005; Losa et al., 2010). In summary, based on the two promoter activity analyses, NTT and STK are co-expressed during gynoecium development, specifically in medial domain tissues. These results support the possibility of the formation of a dimer or higher-order complex containing NTT and STK in these tissues.

We showed that NTT can physically interact with STK, and that the genes are co-expressed in the medial domain of the gynoecium. Subsequently, we wanted to explore the biological relevance of this putative NTT-STK protein dimer or complex during Arabidopsis flower development. The first approach we took was to generate double constitutive expression plants, assuming that this would increase the accumulation of the NTT-STK protein complex in the plant. For this, we crossed a 35S::NTT line (Marsch-Martínez et al., 2014) with a 35S::STK line (Favaro et al., 2003) and we analyzed the F1 generation. Fertility is affected in the single 35S::NTT line, although this line is still able to produce some seeds (Marsch-Martínez et al., 2014). The 35S::STK line flowers early with respect to wild type and develops small flowers with reduced fertility (Favaro et al., 2003). Interestingly, in double constitutive 35S::NTT 35S::STK plants, reproductive development was severely affected (Fig. S1), and the phenotypic alterations were stronger with respect to those observed in the single constitutive expression lines. In general, plants were very small, and when the first flowers reached around floral stage 10, an arrest of floral development was observed and flowers began to senesce. The formed flowers were male and female sterile and, as a consequence, we never observed fruit development, in contrast to the two single constitutive expression lines (Fig. S1). These results show that increased levels of the possible NTT-STK complex can severely affect flower development, suggesting that they may work together in the plant.

To understand better the biological role of the NTT-STK interaction, and to unravel new putative roles for these transcription factors, we generated an ntt stk double mutant. Fruits of the double mutant presented some phenotypes that were a combination of those observed in the single mutants, such as smaller fruits, fewer seeds, no transmitting tract, larger funiculi, irregular seed spacing, lack of seed abscission and reduced seed size (Fig. 2A-D; Fig. 3A,B,G,H) (Pinyopich et al., 2003; Crawford et al., 2007). Interestingly, new phenotypes were observed in the ntt stk double mutant, all related to septum development. First, septum fusion defects were observed in 16% of the fruits (n=360), a phenotype never observed in either single mutant (ntt n=106, stk n=121), nor in wild-type fruits (n=49) (Fig. 2A-E). These septum fusion defects were observed as holes (up to three holes) in the septum of a fruit. In the most severe cases, the septum fusion defects could be seen along 60% of the length of the fruit (Fig. 2E). Furthermore, alteration in septum fusion was also observed at stage 14 as a longitudinal division line (furrow) in the middle of the septum, which corresponds to the place where the two septum primordia meet and normally fuse during wild-type gynoecium development (Fig. 2F,G). This latter phenotype was observed in most of the ntt stk fruits.

A second phenotype observed was related to the aspect of septum cells. When the septum of stage 14 fruits was inspected using scanning electron microscopy (SEM), septum cell integrity in ntt stk fruits appeared to be preserved (Fig. 2G,I). In contrast, septum cells in wild-type fruits at the same stage presented signs of degeneration and collapse (Fig. 2F,H).

As septum development continues, at stage 15, generalized degradation and holes can be observed in the medial domain of wild-type, ntt and stk single mutant fruits (Fig. 2J-L). Strikingly, in the ntt stk double mutant no cell degradation in this region was observed (Fig. 2M). This lack of septum cell degradation was still visible at late stages of fruit development: at stages 17-18 the integrity of septum cells was still maintained (Fig. 2N,O). Also, the imperfect septum fusion was still visible (Fig. 2O), which probably corresponds to the division line observed in Fig. 2G.

The third phenotype that we noticed was the alteration in septum thickness. In wild-type, ntt, and stk single mutant gynoecia at anthesis (stage 13), septum thickness is around six to seven cells (n=5). At this stage, septa from ntt stk double mutant gynoecia presented no difference in the number of cells. Note, however, that pollen tube growth is affected in ntt stk gynoecia, as discussed in the next paragraph (Fig. S2). However, at stage 17-18, septa thickness in ntt stk fruits increased to ten cells (Fig. 2O). In summary, these results suggest that the simultaneous loss of NTT and STK activity leads to altered septum fusion and delays entry of septum cells to their normal degradation program.

The observed septum defects in the ntt stk double mutant could account for the reduced seed-set and fruit length (Fig. 3). Reduction in seed-set and fruit length has previously been observed for the ntt single mutant (Crawford et al., 2007). For the stk single mutant, a reduction in fruit length has been reported (Pinyopich et al., 2003) as well as a slight reduction in seed-set (Mizzotti et al., 2012). The transmitting tract differentiates at stage 12 and it is functional at the mature gynoecium stage when anthesis occurs (stage 13). In the ntt mutant, no transmitting tract is formed (Fig. 2K) and seed-set is only observed in the apical part of the fruit, owing to reduced pollen tube growth (Crawford et al., 2007). In the stk mutant, Alcian Blue staining of gynoecia suggests that transmitting tract formation is not affected, and the pattern of seed-distribution is similar to wild type (Fig. 2C,L; Fig. S2).

We tested whether the absence of transmitting tract and the absence of dead cells in the septum caused by the ntt stk double mutation could further affect pollen tube growth through the ovary. Therefore, we monitored pollen tube movement in ntt stk gynoecia using Aniline Blue staining. As expected, we observed pollen tubes that reached the ovules along the wild-type and stk ovaries (Fig. 3C,E). By contrast, as reported before, in the ntt mutant, pollen tube growth was mainly observed in the apical part of the ovary (40-50% of total ovary length; Fig. 3D). In the ntt stk double mutant, pollen tube growth was further affected, and observed only in the upper 20% of total ovary length (Fig. 3F).

Our results suggest that NTT and STK together impact cell degradation in the septum and, as a consequence, affect the transmitting tract, fertilization efficiency and final seed-set.

To gain further insight into the biological processes controlled by NTT and STK, we generated an ARACNE-based co-expression network of flowers for both transcription factors (see Materials and Methods). Connections in this network indicate transcriptional correlation between genes. Three gene groups could be identified: those connected to NTT, those connected to STK, and those genes connected to both NTT and STK. The complete list of genes in the network is presented in Table S2 and illustrated in Fig. S7. We focused on the third group (genes connected to both NTT and STK), which we called the core network (Fig. 4A). Interestingly, in the core network two transcription factors belonging to the REPRODUCTIVE MERISTEM (REM) family are present, REM11 and REM13, which are known to be expressed in the developing gynoecium, specifically in the CMM and ovules (Wynn et al., 2011; Mendes et al., 2016), and the transcription factor HAF, known to be involved in transmitting tract development (Crawford and Yanofsky, 2011). Recently, REM11 (also known as VALKYRIE) already has been shown to be a direct target of STK (Mendes et al., 2016), providing evidence that supports this co-expression network. Furthermore, in the core network four enzyme and transporter-encoding genes are present: AT1G28710 (a nucleotide-diphospho-sugar transferase family gene), AT3G26140 (a family 5, subfamily 11 glycosyl hydrolase), AT3G21090 (ABC transporter G family member 15, ABCG15) and AT1G06080 (delta-9 acyl-lipid desaturase 1, ADS1), related to polysaccharide metabolism or membrane lipid transport and synthesis (Kang et al., 2011; Li-Beisson et al., 2013).

In co-expression networks, the connection between two nodes may indicate a possible direct regulation when transcription factors are involved (Serin et al., 2016; van Dam et al., 2017). NTT and STK are both transcription factors, so they might directly regulate the expression of the core genes. A consensus binding site for NTT is not known, so we used the DNA-binding site predictor for C2H2 zinc finger proteins (Persikov and Singh, 2013). For MADS-box proteins the consensus binding site is well-known and is called the CArG-box (de Folter and Angenent, 2006). We analyzed whether putative binding sites for NTT and STK were present in promoter or intron sequences of the core network genes. Interestingly, the regulatory regions of all core network genes have putative binding sites (Fig. 4B).

To confirm experimentally the putative transcriptional regulation of the core network genes by NTT and STK, and to understand better the biochemical processes through which NTT and STK exert their effect in the tissues, we obtained experimental evidence for their regulation of the genes coding for enzymes and a transporter involved in cell wall polysaccharide and lipid metabolism (Fig. 4). Interestingly, recently it has been shown that STK is involved in cell wall architecture of the seed (Ezquer et al., 2016). First, we performed chromatin immunoprecipitation (ChIP) assays using an anti-GFP antibody on wild-type, STK::STK:GFP and gNTT-n2YPET inflorescence tissue, followed by qPCR analysis (Fig. 4C,D). Compared with wild type, ChIP-qPCR results from the STK::STK:GFP line showed a significant enrichment of promoter/intron regions for all four genes tested (Fig. 4D). ChIP-qPCR results from the gNTT-n2YPET line showed a significant enrichment of promoter regions for three genes (Fig. 4C). No enrichment was observed for ADS1, although we cannot exclude the possibility of NTT binding to other sites.

We then reasoned that if NTT and/or STK are transcriptional regulators of the genes present in the core network their expression should be altered in the ntt, stk and ntt stk mutant backgrounds. We analyzed the expression of the four enzyme- and transporter-encoding genes present in the core co-expression network using qRT-PCR (Fig. 4E). The expression of all four genes was reduced in the ntt stk double mutant gynoecia, compared with the wild-type sample (Fig. 4B). The AT3G26140 and AT1G28710 genes presented a roughly similar reduction in expression levels in the single and double mutants, which suggests that these genes could be under the control of an NTT- and STK-containing protein complex, such that single disruption of NTT or STK is enough to impact the regulatory effect of the complex. Note that we could still detect some expression in the double mutant, suggesting they are regulated by more genes. On the other hand, we could not detect a difference in expression level of ABCG15 and ADS1 in whole inflorescence tissue tested in the ntt single mutant background. Furthermore, ADS1 expression was slightly increased in the stk single mutant. Nevertheless, ChIP and expression analyses support a role for NTT and STK as regulators of the genes present in the core co-expression network.

One of the genes present in the core network, AT3G26140, encodes a glycosyl hydrolase (GH5_11). The GH5 Arabidopsis enzymes that have been biochemically characterized are all mannan endo-beta-1,4-mannosidases (mannanase; E.C. 3.2.1.78), which are involved in cell wall remodeling. Although no biochemical evidence is available for GH5_11 enzymes (Aspeborg et al., 2012), it is possible that AT3G26140 could also encode a mannanase involved in septum development, in particular the deposition or remodeling of the transmitting tract polysaccharide matrix. We therefore decided to explore further the role of AT3G26140 in developing gynoecia. For this, we performed in situ hybridization for AT3G26140 in wild-type, ntt, stk and ntt stk genetic backgrounds (Fig. 5A-H; Fig. S4). In wild-type gynoecia, signal was detected from early developmental stages in the CMM, in septa, funiculi and ovules (Fig. 5A,E). In the ntt and stk single mutants, comparable expression patterns were observed (Fig. 5B,C,F,G), although this is not reflected by the qRT-PCR results on whole inflorescence tissue, suggesting that there is tissue-dependent regulation. However, and in accordance to the qRT-PCR results, in the ntt stk double mutant only a weak signal was detected (Fig. 5D,H). This indicates that AT3G26140 is regulated by NTT and STK and suggests that this enzyme participates in cell wall metabolism in the cells of medial domain tissues.

We observed a low AT3G26140 mRNA signal by in situ hybridization in gynoecia of the ntt stk double mutant (Fig. 5), and we wondered if this could be translated into an altered mannan content in septum cell walls. We analyzed mannan polysaccharides distribution in septum cells during gynoecium development by immunofluorescence using the LM21 antibody, which recognizes mannan, glucomannan and galactomannan polysaccharides (Marcus et al., 2010). Significant labeling was detected in septum cell walls, but almost no signal was detected in cells of the transmitting tract of wild-type gynoecia (Fig. 5I). In the ntt mutant, which lacks transmitting tract tissue, a low but detectable mannan signal was present throughout the septum, as expected (Fig. 5J). Surprisingly, in the stk single mutant signal was detected in the septum, but also in the transmitting tract tissue, suggesting that transmitting tract cells in the stk mutant have an altered cell wall polysaccharide composition (Fig. 5K). In the ntt stk double mutant, which as in the ntt mutant also lacks transmitting tract tissue, a continuous signal was observed throughout the gynoecium (Fig. 5L). These results suggest that NTT and STK are both necessary for the correct expression of the putative mannanase-encoding gene AT3G26140 in the medial domain.

The presence of lipid-related genes in the core co-expression network also prompted us to look for possible lipid deposition defects in ntt stk septum cells. Using SEM we observed the presence of wax granules on the septum epidermis cells of mature (stage 19-20) fruits (Fig. 5M-P). These wax granules were scarce on wild-type septum cells, but clearly visible in the ntt or stk single mutants (Fig. 5N,O, arrows). In the ntt stk double mutant a larger number of wax granules was observed (Fig. 5P).

In order to study the individual contribution of the enzyme and transporter-encoding genes to the ntt stk phenotype, we analyzed T-DNA insertion lines for two NTT-STK target genes (Fig. 4). For the putative mannanase-encoding gene AT3G26140, a statistically significant reduction in seed-set and fruit length was detected (Fig. S5). The mild phenotype could be explained by functional redundancy among cell wall regulators.

For ABCG15, we obtained two mutant alleles that showed dramatic and pleiotropic phenotypes, probably related to altered meristematic activity (Fig. 6; Fig. S6). We named this transporter KAWAK (KWK), which comes from Mayan mythology. It is the name of one of the 20 months of the Mayan calendar, and it means ‘storm’ and also ‘monster with two heads’. In these kwk mutant plants, shoot apical meristem (SAM) maintenance was reduced or even absent (Fig. 6A; Fig. S6). The loss of the apical dominance caused growth of secondary shoots. Defects were also observed in inflorescence and floral meristems, causing altered floral bud positioning and number, and alterations in floral organ arrangement (Fig. 6; Fig. S6). Furthermore, we observed floral organ fusion defects, unfused carpels and septa, ectopic formation of ovules and stigmatic tissue or repla. Fruits developed from less-affected gynoecia presented alterations in carpel number, and all fruits had alterations in seed formation and seed arrangement, the latter probably due to funiculi alterations (Fig. 6F-P; Fig. S6). These results highlight the importance of ABCG15 (KAWAK) during Arabidopsis development, making it an interesting target of NTT and STK to study further in future work.

Multiple roles have been reported for the NTT transcription factor during gynoecium development, including transmitting tract formation (Crawford et al., 2007), replum development (Marsch-Martínez et al., 2014) and valve margin specification (Chung et al., 2013). We identified that NTT interacts with a large number of transcription factors belonging to different families, suggesting that NTT participates in many protein complexes during development, possibly performing different, as yet unknown, functions.

In this work, we focused on the interaction with STK, a MADS-box protein that determines ovule identity, correct funiculus development, seed abscission (Pinyopich et al., 2003; Balanzà et al., 2016) and regulation of seed development (Mizzotti et al., 2014; Ezquer et al., 2016). MADS-box transcription factors are able to interact with each other and form functional protein complexes that guide flower development (Honma and Goto, 2001; de Folter et al., 2005; Immink et al., 2009). For STK, protein-interaction partners important for ovule and seed development, such as AG, SEPALLATA3 (SEP3) and ARABIDOPSIS B SISTER (ABS), have been reported (de Folter et al., 2005; Kaufmann et al., 2005; de Folter et al., 2006; Mizzotti et al., 2012). In general, MADS-box proteins interact with MADS family members, and few interactions with members of other transcription factor families have been described to date (Smaczniak et al., 2012; Bemer et al., 2017). Interestingly, we found that NTT, a zinc finger transcription factor, interacts with various MADS-box proteins such as AG, SHP1, SHP2 and STK, which are all paralogs (Marsch-Martínez et al., 2014; this work). Based on the data presented in this work, we suggest that NTT and STK can work cooperatively during gynoecial medial domain development in Arabidopsis.

One of the current challenges in understanding the regulation of flower development is the identification of transcriptional targets of key transcription factors (Wellmer et al., 2014). In order to identify possible target genes of NTT and STK, we generated an ARACNE-based co-expression network, which uses microarray expression data and infers putative transcriptional interactions (Margolin et al., 2006a) (Fig. 4A). Networks inferred using this method are a useful tool in the understanding of biological processes (Yu et al., 2011; Chávez Montes et al., 2014; González-Morales et al., 2016).

Interestingly, by searching in the current literature, we found that many of the inferred interactions in the NTT-STK co-expression network are supported by several reports, indicating that these interactions are biologically relevant. There are examples of functional interactions between STK and SHP2 (Pinyopich et al., 2003; Brambilla et al., 2007) and, in some cases, of direct transcriptional regulation, for instance STK to VERDANDI (Matias-Hernandez et al., 2010), BANYULS (Mizzotti et al., 2014), and REM11 (Mendes et al., 2016). These genes are present in the NTT-STK co-expression network (Table S1). For NTT, a transcriptional relationship with HAF has been reported (Crawford and Yanofsky, 2011), which is also connected in the network to STK. Besides HAF, REM11 and REM13 are also connected to NTT and STK in the network, and they are expressed in young gynoecia in the CMM (Wynn et al., 2011; Mantegazza et al., 2014), which supports a role for NTT and STK in early gynoecium development. Beyond transcription factors, four enzyme and transporter-encoding genes are co-expressed with NTT and STK (Fig. 4). These enzymes and transporter could provide clues about the biochemical processes regulated by NTT and STK.

ChIP experiments indicated binding of STK to CArG-box-containing regions in the promoters/introns of all four genes, and binding of NTT to putative C2H2 zinc finger protein-binding sites of at least three genes, suggesting that they could be direct STK/NTT targets. Moreover, qRT-PCR experiments showed reduced expression of all four enzyme-encoding genes in the ntt stk double mutant; however, this was not the case for all of the single mutants, possibly due to redundancy. But, in general, the results suggest that these enzyme- and transporter-encoding genes are targeted by NTT and/or STK (Fig. 4).

The first enzyme (AT3G26140) we found belongs to the GH5 family for which (1-4)-beta-mannan endohydrolase and cellulase activities have been identified for some members (Aspeborg et al., 2012). The second enzyme is a nucleotide-diphospho-sugar transferase (AT1G28710), a glycosyltransferase involved in the synthesis of polysaccharides. Based on the CAZy database, it belongs to the GT77 family for which α-xylosyltransferase (EC 2.4.2.39), α-1,3-galactosyltransferase (EC 2.4.1.37) and arabinosyltransferase (EC 2.4.2.-) activities have been reported (Lombard et al., 2014). The third enzyme, ADS1 (AT1G06080), has been characterized as a functional fatty acid desaturase (Yao et al., 2003; Heilmann et al., 2004) and is expressed in flowers (Fukuchi-Mizutani et al., 1998). Heterologous expression of this enzyme in Brassica juncea generated decreased levels of total saturated fatty acid in seeds and altered the normal fatty acid profile (Yao et al., 2003). The last gene is ABCG15 (AT3G21090), an ATP-binding cassette (ABC) transporter, which we named KAWAK (KWK; discussed below). Members of this ABCG group are required for lipid deposition and cutin formation (Kang et al., 2011). ABCG15 is phylogenetically close to ABCG12 (also known as CER5), which is required for wax transport to the cuticle (Pighin et al., 2004), and it is also close to ABCG13, which is involved in the transport of cuticular lipids in flowers (Panikashvili et al., 2011), and to ABCG11 (also known as DSO), which is involved in cuticular lipid export (Bird et al., 2007; Luo et al., 2007; Panikashvili et al., 2007; Ukitsu et al., 2007).

So, the four genes found are related to cell wall polysaccharide metabolism or membrane lipid synthesis and transport. Our findings suggest that these two processes are altered in septum cells of the ntt stk mutant. We have recently shown that global changes in cell wall composition take place during gynoecium development (Herrera-Ubaldo and de Folter, 2018), for instance mannan polysaccharide content decreases when the gynoecium matures. Here, mannan polysaccharide content in mutant and wild-type gynoecia was analyzed; an evident alteration in mannan accumulation was observed in the medial region of the single and double mutants, suggesting that the reduction in expression of the mannanase gene AT3G26140 observed by in situ hybridization and qRT-PCR could be related to the increase in mannan accumulation. Reduced fertility is to be expected when this enzyme is affected, which we did indeed observe in a T-DNA insertional mutant for AT3G26140 (Fig. S5). The subtle reduction in fertility observed is probably due to the involvement of other redundant proteins or additional enzymes. This was recently shown for silique dehiscence zone formation, where various cell wall-modifying enzymes participate, such as ARABIDOPSIS DEHISCENCE ZONE POLYGARACTURONASE 1 (ADPG1), ADPG2, CELLULASE 6 (CEL6) and MANNANASE 7 (MAN7) (Ogawa et al., 2009; He et al., 2018). On the other hand, the altered wax deposition in septum cells is a sign of altered lipid metabolism or transport. Alterations in these processes could explain the phenotypes observed in the ntt stk double mutant.

Defects in septum fusion were observed in the ntt stk double mutant. However, these defects are different to those observed in, for example, the spt mutant, which has a reduced cell number in the CMM and reduced growth of the septa primordia (Alvarez and Smyth, 1999, 2002; Heisler et al., 2001). Cell number in the CMM in the ntt stk mutant is similar to that observed in wild type, and septa primordia grow normally and encounter each other to form the septum. This suggests that the observed fusion abnormalities (Fig. 2) are not related to defects in early growth, but might be related to epidermal defects, such as altered cuticle, which is a specialized lipidic modification of the cell wall (Yeats and Rose, 2013). Cuticle seems to be involved in cell-to-cell communication (Tanaka and Machida, 2007), and alterations in cuticle formation cause organ fusion defects (Nawrath, 2006).

The correct formation and composition of the cuticle is important for flower development, as it promotes carpel fusions and prevents ectopic or organ fusions (Lolle and Cheung, 1993; Panikashvili et al., 2010). Mutants such as fiddlehead and hothead have floral organ fusion defects caused by altered cuticle formation (Lolle et al., 1998; Yephremov et al., 1999; Pruitt et al., 2000; Krolikowski et al., 2003). Interestingly, a mutation in the epidermis-expressed ABCG11 transporter (related to ABCG15) affects organ fusion due to altered epicuticular wax on the surface of organs (Luo et al., 2007; Panikashvili et al., 2010). We also observed altered wax deposition on the septum surface of the ntt stk double mutant, which could be related to its septum fusion defects. Furthermore, the expression of ABCG15 was clearly reduced in the double mutant and regulatory regions of ABCG15 were enriched in ChIP assays for STK and NTT (Fig. 4). Interestingly, we identified kwk mutants that showed dramatic phenotypes in meristem development and during reproductive development. Alterations in the ABCG15 transporter function could lead to impaired lipid export and altered cuticle formation. It could also affect membrane structure or block the correct position of membrane proteins. The kwk mutant might be helpful in understanding the role of plant surface lipids and epidermis development, and the role of the epidermis in developmental processes (e.g. Delude et al., 2016; Verger et al., 2018).

Modifications of the cell wall and cell death are important processes during the formation of the transmitting tract and its ECM that allow for pollen tube growth through the ovary, and therefore directly affect fertilization efficiency and seed-set (Crawford et al., 2007; Crawford and Yanofsky, 2008). The ntt mutant lacks a transmitting tract, as indicated by the lack of acidic polysaccharide staining. However, cell degradation does take place in the septum, observed as irregular-shaped, degraded cells and empty spaces (Crawford et al., 2007; Fig. 2). These tissues appear normal in the stk mutant. The ntt stk double mutant, however, lacks cell degradation in the septum. Moreover, the double mutation produces a clear increase in severity of pollen tube growth through the gynoecium. Whereas pollen tube growth is reduced in the single ntt mutant, and does not appear to be affected in the stk mutant, it is severely reduced in the ntt stk double mutant.

Part of the altered cell degradation phenotype could be related to the observed lack of mannanase expression in the medial domain of young ntt stk gynoecia. Furthermore, the nucleotide-diphospho-sugar transferase might be involved in the synthesis of any of the polysaccharides, glycoproteins or glycolipids of the ECM, suggested to provide nutrients and adhesion for correct pollen tube growth (Crawford and Yanofsky, 2008).

Taken together, the data presented indicate that NTT and STK have clear roles in septum fusion and the modification of cell walls, affecting fertilization efficiency and seed-set.

Another important process for which cell wall modifications play a role is during fruit ripening, which is followed by senescence (Gapper et al., 2013; Gómez et al., 2014). Our work also hints that senescence is induced by NTT and STK (Fig. S1). Note that we observed already some induced senescence and cell death by NTT alone, and this could be due to interactions with other proteins, possibly with other related MADS-box proteins (Fig. S3). Though arguably a bit preliminary, this suggests that NTT, enhanced by STK, can promote senescence and induce cell death and, thereby, might regulate fruit maturation. Research in tomato has demonstrated that MADS-box proteins control fruit ripening (Karlova et al., 2014). RIPENING INHIBITOR (RIN), a homolog of the Arabidopsis SEP genes (Vrebalov et al., 2002), regulates the expression of genes involved in cell wall modifications, such as polygaracturonase and B-galactosidase, in addition to proteins controlling shelf life (A-expansin) and fruit softening (B-mannanase) (Fujisawa et al., 2011). The latter is dramatically downregulated in fruit ripening-defective tomato plants (Fujisawa et al., 2014; Shima et al., 2014). Other MADS-box genes involved in tomato fruit ripening are homologs of the AG clade (Itkin et al., 2009; Vrebalov et al., 2009; Pan et al., 2010) and FUL homologs (Bemer et al., 2012). In addition, some zinc finger proteins are also involved in fruit ripening, such as SlZFP2 (Rohrmann et al., 2011; Weng et al., 2015) and MaC2H2-1/2 (Han et al., 2016).

In summary, we found that NTT and STK control genes coding for enzymes and transporters involved in synthesis and degradation of cell wall polysaccharides, and synthesis and transport of fatty acids. It would be particularly interesting to know whether homologous genes could perform similar activities in other fruits, especially those important for food and industry.

Arabidopsis thaliana plants were germinated in soil (3:1:1, peat moss:perlite:vermiculite) in a growth chamber under long-day conditions (16 h light, 22°C; 8 h dark, 20°C) for 10 days and transferred to standard greenhouse conditions (22-27°C, natural light). The following mutants and lines were used in this work: the transposon insertion line ntt-3 is the NASC line N104422 (SM_3.16705) in Col (Tissier et al., 1999); stk-2 (Pinyopich et al., 2003); STK::GUS (Kooiker et al., 2005); 35S::STK (Favaro et al., 2003); 35S::NTT (Marsch-Martínez et al., 2014); gNTT-n2YPET (Crawford et al., 2015); kwk-1 is the line GT_5_99063 in Ler (T-DNA in the third exon); kwk-2 is SALKseq_125172 in Col-0 (T-DNA in the third exon); insertion line for AT3G26140 is SALK_128093 (T-DNA in the third intron); Nicotiana benthamiana and Nicotiana tabacum were used for cell death assays and BiFC, respectively.

For the NTT promoter::GUS fusion, a 1216 bp DNA sequence upstream of the predicted translation start was amplified by PCR from genomic Col-0 DNA, using Pwo DNA polymerase (Roche) and primers S314 and S318 (Table S3). The PCR product was cloned in front of the GUS ORF of the binary vector pANGUS [a derivative of pPAM (GenBank AY027531) described by Stracke et al., 2007] using the restriction endonucleases ClaI and NcoI. A. thaliana Col-0 was transformed by floral dip (Clough and Bent, 1998).

Y2H assays were performed using the GAL4 system (pDEST22 and pDEST32; Invitrogen) as previously described (de Folter et al., 2005; de Folter and Immink, 2011). NTT-BD cloning and yeast autoactivation test was previously described (Marsch-Martínez et al., 2014). The STK-AD clone was derived from recombining the STK Gateway ENTRY clone from the EU-REGIA project (Paz-Ares and The REGIA Consortium, 2002) with the pDEST22 vector (Castrillo et al., 2011). We used the yeast strain PJ69-4 mating type A and α (James et al., 1996). The uni-directional Y2H screen with 45 selected AD clones has been described before (Zúñiga-Mayo et al., 2012; Marsch-Martínez et al., 2014; Lozano-Sotomayor et al., 2016), but, in short, it contains selected transcription factors from the EU-REGIA project (Paz-Ares and The REGIA Consortium, 2002) that are known to be involved in flower and gynoecium development, and meristem activity (Table S1). Interaction assays were performed on SD-GLUC medium lacking Leu, Trp and Ade, and on medium lacking Leu, Trp and His, supplemented with 20 mM 3-AT. Protein-protein interactions were scored after 5 days of growth at 25°C. Positive results (yeast growth) were confirmed by a lacZ assay.

In planta protein interaction assays were performed as previously described (Marsch-Martínez et al., 2014). The cDNA of STK was cloned in pDONR201 (Invitrogen) by the REGIA Consortium (Paz-Ares and The REGIA Consortium, 2002). This clone was recombined with pYFN43 (Belda-Palazón et al., 2012) using an LR Gateway-based reaction to generate N-terminal fusions with the N-terminal part of YFP. NTT entry clone was also recombined with pYFC43 (Belda-Palazón et al., 2012) to generate an N-terminal fusion with the C-terminal part of the YFP protein. The constructs were individually introduced in to Agrobacterium tumefaciens GV2260 and cultured on LB supplemented with 100 µg/ml kanamycin and 25 µg/ml rifampicin. Overnight cultures of Agrobacterium (O.D.: 1.2-1.6) were collected and re-suspended in a similar volume of infiltration medium (10 mM MgCl₂, 10 mM MES pH 5.6, 200 µM acetosyringone), the O.D. was adjusted to 1.0, and the re-suspension was incubated at 25°C during 3 h with weak shaking. Before co-infiltration, Agrobacterium containing pYFC43-NTT was mixed with a similar volume of Agrobacterium with pYFN43-STK. This mixture was introduced in the abaxial air space of young Nicotiana tabacum leaves using a needle-less syringe. YFP fluorescence restoration was assayed 2 days after infiltration using an inverted LSM 510 META confocal laser scanning microscope (Carl Zeiss). YFP was excited using the 488 nm line of an argon laser and emission was filtered using a BP 500-550 nm filter.

For thin tissue section analysis, inflorescences and stage 15, 17-18 fruits of Col-0, ntt-3, stk-2, and ntt stk were collected (according to Smyth et al., 1990), tissue was fixed in FAE solution (3.7% formaldehyde, 5% glacial acetic acid and 50% ethanol) with vacuum (15 min, 4°C) and incubated for 60 min at room temperature. The material was rinsed with 70% ethanol and incubated overnight at 4°C in 70% ethanol, followed by dehydration in a series of ethanol dilutions (70%, 85%, 95% and 100% ethanol) for 60 min each. Inflorescences and stage 17-18 fruits were embedded in Technovit 7100 (Heraeus Kulzer) according to the manufacturer's instructions. Stage 15 fruits were embedded in Paraplast (Sigma-Aldrich) as previously described (Zúñiga-Mayo et al., 2012). Sections (12-15 µm) were obtained on a rotary microtome (Reichert-Jung 2040; Leica). Tissue sections were stained with a solution of 0.5% Alcian Blue and counterstained with 0.5% Neutral Red as previously described (Zúñiga-Mayo et al., 2012), or with Toluidine Blue as previously described (Herrera-Ubaldo and de Folter, 2018). NTT::GUS and STK::GUS inflorescences were collected and stained as previously described (Marsch-Martínez et al., 2014). The GUS-stained inflorescences were fixed, dehydrated as described above and embedded in Technovit 7100; 12-15 µm sections were analyzed. Pictures were taken using a DM6000B microscope (Leica).

For septum epidermis cell observations, fresh fruit samples were dissected and visualized in an EVO 40 scanning electron microscope (Carl Zeiss), using the VPSE G3 detector, with a 15-20 kV beam and at 50 Pa pressure.

Pollen tube growth within the pistil was monitored with Aniline Blue staining. Pistils were collected 24 h after pollination, and tissue fixation and softening were performed as previously described (Jiang et al., 2005). Pistils were washed with distilled water three times and stained with aniline blue solution (0.01% Aniline Blue in 150 mM K₂HPO₄ buffer, pH 11) for 4 h in the dark. Pistils were observed and imaged with a DM6000B fluorescence microscope under UV light (Leica).

A flower co-expression matrix was obtained from microarray data using the ARACNE algorithm (Margolin et al., 2006a,b). Sample data relationship files for ATH1-121501 microarray experiments were downloaded in August 2016 from the ArrayExpress website (http://www.ebi.ac.uk/arrayexpress/; accession numbers E-GEOD-15555, E-GEOD-16056, E-GEOD-2473, E-GEOD-27281, E-GEOD-2848, E-GEOD-30492, E-GEOD-3056, E-GEOD-32193, E-GEOD-40998, E-GEOD-42403, E-GEOD-42841, E-GEOD-46050, E-GEOD-52067, E-GEOD-5526, E-GEOD-55431, E-GEOD-55799, E-GEOD-5632, E-GEOD-66419, E-MEXP-1246, E-MEXP-1592, E-MEXP-1920, E-MEXP-3293, E-MEXP-849 and E-TABM-17) and manually curated to identify 24 experiments that contained 106 high-quality microarray hybridizations for wild-type flower samples. The corresponding CEL files were manually curated and processed as previously described (Chávez Montes et al., 2014; González-Morales et al., 2016), with a single modification: custom CDF version 20 files (http://brainarray.mbni.med.umich.edu/Brainarray/Database/CustomCDF/genomic_curated_CDF.asp; Dai et al., 2005) were used for gcrma normalization. The output of the ARACNE algorithm, which is a mutual information-ranked list of pairs of interactors (i.e. of co-expressed genes), was used to identify genes (both transcription factors and non-transcription factors) co-expressed with NTT and STK.

Genomic regions located between the flanking genes in the co-expression network core were analyzed bioinformatically to identify putative CArG-box regions and predicted NTT-binding sites (Persikov and Singh, 2014). ChIP assays were performed as previously described (Ezquer et al., 2016). One gram of unfertilized flowers from Col-0, STK::STK:GFP (Mizzotti et al., 2014) and gNTT-n2YPET (Crawford et al., 2015) plants were collected. For the immunoprecipitation, we used 2 µl anti-GFP polyclonal antibody per sample (Clontech 632460 for ChIP with STK; Roche 11814460001 for ChIP with NTT). Enrichment of the target regions was calculated by qPCR (iQ_SYBR Green Supermix, Bio-Rad) using a Bio-Rad iCycler iQ optical system. The relative enrichment of the listed targets obtained from STK::STK:GFP and pNTT-n2YPET unfertilized flowers were compared with the enrichment obtained from Col-0 wild-type unfertilized flowers. ACTIN 7 was used for normalization as previously described (Matias-Hernandez et al., 2010). Primers used for ChIP analysis are listed in Table S3.

For the qRT-PCR analysis, gynoecia from stage 7 to 12 were collected under a stereomicroscope (Stemi 2000, Zeiss). Three biological replicates were sampled for each genotype (Col wt, ntt, stk, ntt stk), each containing around 40 gynoecia. Total RNA was extracted using the Quick-RNA MicroPrep Kit (Zymo Research). The samples were treated with DNase I, included in the kit. Reverse transcription and amplification were performed using the KAPA SYBR FAST One-Step qRT-PCR Kit (Kapa Biosystems). The qPCR was performed on a StepOne thermocycler (Applied Biosystems). Target gene expression levels were normalized to ACTIN 2. Data were analyzed using the 2^−ΔΔCt method (Livak and Schmittgen, 2001). Primers used are listed in Table S3.

Inflorescences from Col-0, ntt, stk and ntt stk plants were collected, fixed and embedded in Paraplast as previously described (Sotelo-Silveira et al., 2013). A DNA fragment corresponding to nucleotides 1176-1368 of the AT3G26140 coding sequence was amplified using the At3g26140_probe primers (Table S3) and cloned into pGEM-T easy vector (Promega). The sense and antisense RNA probes were synthesized by an in vitro transcription reaction using SP6 and T7 polymerase (Invitrogen), respectively. Digoxigenin-labeled RNA probes for detection and hybridization were prepared as previously described (Ambrose et al., 2000).

Col-0, ntt, stk and ntt stk inflorescences were fixed overnight at 4°C in 3% paraformaldehyde, 1×PBS 1% pH 7.0. Samples were dehydrated in a series of ethanol dilutions (70%, 85%, 95% and 100% ethanol) and embedded in Technovit 7100 (Heraeus Kulzer) according to the manufacturer's recommendations. Thin sections (14-18 µm) were obtained on a rotary microtome (Reichert-Jung 2040, Leica). Sections were treated with 1 M KOH for 60 min to unmask the manna epitopes (Marcus et al., 2010). Sections were washed three times with wash buffer (2% BSA, 1% PBS pH 7.0) at room temperature before incubation with the LM21 anti-mannan monoclonal antibody (PlantProbes; Marcus et al., 2010) diluted 1:500 in wash buffer for 16 h at 25°C; as described previously (Herrera-Ubaldo and de Folter, 2018). Samples were washed three times with wash buffer and incubated for 4 h at 25°C with the secondary antibody DyLight 488-conjugated goat anti-rat IgM mu chain (ab98368, Abcam) diluted 1:1000 in wash buffer. Samples were washed twice with wash buffer and mounted in 50% glycerol. Photographs of immunolabeled samples were taken using a DM6000B microscope under UV light (Leica).

Three-week-old Nicotiana benthamiana leaves were infiltrated with Agrobacterium cells containing vectors for the transient expression of NTT (pC1300intB-35SnosEX, AY560325; Kuijt et al., 2004; Marsch-Martínez et al., 2014) or GFP (pMOG800; Knoester et al., 1998) in infiltration medium (10 mM MgCl₂, 10 mM MES pH 5.6, 200 µM acetosyringone). Three different Agrobacterium concentrations were used (OD600: 0.2, 0.4 and 0.8). Cell death was monitored with a modified version of the Trypan Blue (TB) staining protocol (Mauch-Mani and Slusarenko, 1994). The TB solution contains: lactic acid:phenol:glycerol:distilled water (1:1:1:1) and Trypan Blue (T-0776, Sigma-Aldrich) in a final concentration of 0.25 mg/ml. Before staining, the TB solution was diluted in 2 volumes of 100% ethanol.

Leaf sections (2 cm diameter) were collected 1, 2 and 3 days after infiltration, boiled in diluted TB solution for 1 min. Samples were de-stained in chloral hydrate solution (8:1:2 w/v/v chloral hydrate:glycerol:water) during 1 h at 65°C, followed by overnight incubation in chloral hydrate solution at 50°C with shaking. The samples were washed with 70% ethanol, mounted in 50% glycerol. Brightfield images were acquired using an Eclipse E-600 microscope with a Digital Sight (DS-Ri1) (Nikon Instruments).

We thank lab members and other colleagues for discussions regarding this work. We also thank Karla L. González-Aguilera and Vincent E. Cerbantez-Bueno for lab logistics and technical support. We thank Brian Crawford for gNTT-n2YPET seeds and the ABRC and NASC for insertion lines.