Arabidopsis thaliana ecotype Col-0 and fis2-8 mutant (Wang et al., 2006) plants were housed in a walk-in Environmental Room (Norlake Scientific, Hudson, WI, United States) at 22°C under 16-h light/8-h dark long-day condition. Seeds were stratified for 4 days at 4°C before germination. Plants were grown in Pro-Mix BX soil (Premier Horticulture, Quakertown, PA, United States) and fertilized with Peters 20-20-20 (Scotts-Sierra Horticultural Products Company, Marysville, OH, United States).

Microarray data was downloaded from Gene Expression Omnibus database (accession no. GSE12404). The expression values were normalized with the GeneChip Robust Multiarray Averaging method (GC-RMA) implemented as a Bioconductor package under the R platform¹ (Wu et al., 2004). Hierarchical cluster analysis (average linkage and Euclidean distance as similarity measure) was performed using Cluster 3.0 on log2-transformed expression values (de Hoon et al., 2004). The result of cluster analysis was visualized using Java TreeView (Saldanha, 2004).

The gene IDs for eight acid invertases (Glycoside Hydrolase Family 32) and nine neutral/alkaline invertases (Glycoside Hydrolase Family 100) were obtained from Carbohydrate-Active enZYmes Database² (Sturm et al., 1999; Lammens et al., 2009). The gene IDs for 125 members of the plant InvINH/PMEI superfamily were obtained from SUPERFAMILY database³ (Wilson et al., 2009). Cluster analysis was conducted on 107 genes that are present on Affymetrix Arabidopsis ATH1 Genome Array, including 17 invertase genes and 90 InvINH/PMEI genes.

The following online programs were used to predict the subcellular localizations of InvINH1 and InvINH2: PSORT⁴ (Nakai and Kanehisa, 1991), MultiLoc2⁵ (Blum et al., 2009), and YLoc⁶ (Briesemeister et al., 2010). All three programs were run with the default setting for plant proteins.

Genevestigator⁷ (Hruz et al., 2008) was used to analyze the expression pattern of InvINH1 and InvINH2. The data selection includes a compendium of 5,825 wild-type A. thaliana samples profiled on the Affymetrix Arabidopsis ATH1 Genome Array platform. The Anatomy tool and Perturbations tool from the CONDITION SEARCH toolset were used to analyze the expression level of InvINH1 and InvINH2, which are represented by the same Affymetrix probe (248823_s_at).

Total RNAs were extracted from roots, stems, rosette leaves, closed floral buds from stage 0 to 12 (Smyth et al., 1990), and young siliques at 3 and 5 days after pollination (dap) following a modified hot borate extraction method (Wan and Wilkins, 1994). The first-strand cDNAs were synthesized with the RETROscript kit (Ambion, Inc., Austin, TX, United States). Quantitative PCR (qPCR) was performed with the SYBR Select Master Mix and the StepOnePlus real-time PCR system (Applied Biosystems, Foster City, CA, United States). The primers used for qPCR are as follows: InvINH1, forward 5′-ctgagtgctgctttggatgta-3′, reverse 5′-gttctcgttggtaatcggagac-3′; InvINH2 forward 5′-aagacccgcaatcgtcatac-3′, reverse 5′-gtcgatgctagggccaaatc-3′; and Actin2 forward 5′-tccctcagcacattccagcagat-3′, reverse 5′-aacgattcctggacctgcctcatc-3′. C_T values were normalized against Actin2. The use of Actin2 as the reference gene for qPCR analysis has been previously described (Wang et al., 2010). Most tissues were analyzed with three biological replicates except for silique tissues, which were analyzed with two biological replicates and two technical replicates. The mRNA level of InvINH1 or InvINH2 in 3-dap wild-type siliques was set as the reference point (100%) to calculate the relative mRNA levels in other tissues following the -ΔΔC_T method (Livak and Schmittgen, 2001).

The 5′ flanking region of InvINH1 or InvINH2 was cloned into the binary vector pBN-GFP (Wang et al., 2006) to create promoter fusion constructs. In brief, the 5′ flanking regions of InvINH1 (1172 bp) and InvINH2 (2238 bp), including the entire 5′ intergenic region and the coding region encoding the first seven amino acids, were amplified from Col-0 genomic DNA using Phusion Polymerase (Thermo Fisher, Waltham, MA, United States). The 5′ UTR region and the first seven amino acids were included in the promoter fusion to ensure proper translation of the GFP gene. Since the N-terminal signal peptide for InvINH1 and InvINH2 was predicated to be 20-amino-acid long, the inclusion of the first seven amino acids is unlikely to change the subcellular localization of the GFP protein. The primers used for amplification are as follows: InvINH1 forward primer (5′-aatgtctagagctgaaatgaaactacatgtgc-3′), InvINH2 forward primer (5′-cgtttctagacgtctccgattaccaacga-3′), and a common reverse primer (5′-gagaaggatcccaatgaaaccaagaacttcat-3′). The amplified fragments were cloned in frame into the pBN-GFP vector between the XbaI and BamHI sites.

The ectopic expression construct for InvINH1 was generated by cloning the ZOU promoter and InvINH1 coding region into the binary vector pBN (Wang et al., 2006). Since InvINH1 has no intron, both InvINH1 coding region and Zou promoter were amplified from Col-0 genomic DNA using Phusion Polymerase (Thermo Fisher). The Zou promoter region (-2009 bp to +18 bp) was amplified with forward primer (5′-tgattacgccaagcttgtgttacgttgtaacgaattt-3′) and reverse primer (5′-tgctcaccatggatccctcttgagcattagtcatattg-3′), then cloned into pBN vector between the HindIII and BamHI site, resulting in construct pBN-pZOU. Next, the InvINH1 coding region (525 bp) was amplified with forward primer (5′-attaggatccatgaagttcttggtttcattggt-3′) and reverse primer (5′-gataggtaccttacaacatattagtaaaagccaaagga-3′), then cloned into pBN-pZOU in between the BamHI and KpnI sites, resulting in construct pBN-pZou-InvINH1. All constructs were verified by sequencing.

Arabidopsis plant transformation was carried out as described previously (Wang et al., 2006). In brief, Agrobacterium tumefaciens strain GV3101 pMP90 (Koncz and Schell, 1986) carrying the appropriate binary vector was used to perform the standard floral dip method (Clough and Bent, 1998). Transgenic seedlings were selected on 0.5x Murashige and Skoog (MS) media containing 35 μg/ml Kanamycin. The presence of the transgene in T1 plants was confirmed using PCR.

Seeds were dissected out of the silique as described previously (Wang et al., 2010). To isolate the embryos, seeds were punctured with a dissecting needle and then gently pressed to release the embryos. GFP expression pattern in whole-mount seeds were obtained with a Zeiss LSM 700 inverted confocal microscope (Carl Zeiss, Oberkochen, Germany). Dissected embryos were imaged with Nikon C-DS stereoscopic microscope (Nikon, Tokyo, Japan) equipped with an AxioCam Icc1 digital camera imaging kit (Carl Zeiss). Image processing was performed with Adobe Photoshop CS (Adobe Systems Inc., San Jose, United States).

In Arabidopsis, embryo growth rate accelerates after endosperm cellularization (Goldberg et al., 1994; Baud et al., 2008; Hehenberger et al., 2012). To investigate whether the change in embryo growth rate is correlated with any change in invertase activity, we analyzed the temporal and spatial expression pattern of invertase and InvINHs by performing hierarchical cluster analysis on a published seed microarray dataset (Belmonte et al., 2013). The dataset includes six developmental time points from pre-globular stage to mature green embryo stage (Figure 1). Each stage contains five to six distinct seed compartments captured with laser capture microdissection (Figure 1). The list of genes included in our analysis was described in Section “Materials and Methods.” Since it is difficult to distinguish InvINHs from PMEIs based on sequence conservation alone (Hothorn et al., 2004), we included all members of the InvINH/PMEI superfamily in our analysis.

Hierarchical cluster analysis revealed seven co-regulated gene clusters that share distinct spatial and temporal expression patterns during seed developments (Figure 1). Some of the co-regulated clusters were specific to seed compartments located at the interface of nutrient transfer, such as chalazal endosperm (Figure 1B, Cluster IV), chalazal seed coat (Figure 1B, Cluster VII), general seed coat (Figure 1B, Cluster VI), and micropylar endosperm (Figure 1B, Cluster II). Among these four seed compartments, chalazal endosperm, chalazal seed coat, and general seed coat are located at the seed coat/endosperm interface, suggesting that clusters IV, VI, and VII genes might be involved in nutrient transfer from the maternal tissues to the endosperm during seed development (Li and Berger, 2012). Since we were interested in the nutrient transfer mechanism that regulates embryo growth rate, we focused on Cluster II genes that were specifically expressed in the micropylar endosperm surrounding the embryo (Figure 1B). Cluster II contains nine members of the plant InvINH/PMEI superfamily represented by eight Affymetrix probes (Figure 1C). After the transition at heart-stage (Figure 1), cluster II genes were down-regulated in the endosperm (Figure 1B), which is expected to cause an increase in invertase or pectin methylesterase activity after endosperm cellularization.

To validate the microarray data, two genes from Cluster II were selected for additional expression analysis. These two genes were tentatively named as InvINH1 (At5g46960) and InvINH2 (At5g46950). InvINH1 and InvINH2 are represented by the same Affymetrix probe due to 93.5% DNA sequence identity between the two genes. Both InvINH1 and InvINH2 proteins are 174 amino acids long and are 88.5% identical to each other. We used three different methods to predict the subcellular location of InvINH1 and InvINH2. The probability of InvINH1 and InvINH2 to be localized to the extracellular space is 75.2, 99.9, and 68% according to PSORT (Nakai and Kanehisa, 1991), Yloc (Briesemeister et al., 2010), and MultiLoc2 (Blum et al., 2009), respectively. The PSORT program predicted that both InvINH1 and InvINH2 have a 20-amino-acid-long N-terminal signal peptide. In addition, InvINH1 has been experimentally confirmed to be a secreted protein associated with the plant cell wall (Irshad et al., 2008), which supports the possibility of InvINH1 to act as an inhibitor of cell-wall-associated enzymes.

To determine whether InvINH1 and InvINH2 are specifically expressed in the seed, we used qRT-PCR to analyze the mRNA level of InvINH1 and InvINH2 in both vegetative tissues (roots, stems, and rosette leaves) and reproductive tissues (closed floral buds, 3-dap siliques, and 5-dap siliques). Among the reproductive tissues, the 3-dap siliques contained globular-stage embryo and syncytial endosperm, while the 5-dap siliques contained early-torpedo stage embryo and cellularized endosperm. Since InvINH1 and InvINH2 are 93.5% identical in DNA sequence, gene-specific qRT-PCR primers were first designed to distinguish InvINH1 from InvINH2. In wild-type plants, the expression level of InvINH1 and InvINH2 were dramatically higher in reproductive tissues than in vegetative tissues (Figure 2). InvINH1 was primarily expressed in closed floral buds and 3-dap siliques, while InvINH2 was primarily expressed in 3-dap siliques (Figure 2). Both InvINH1 and InvINH2 were strongly down-regulated in 5-dap siliques compared to 3-dap siliques (Figure 2). Therefore, both the qRT-PCR and the microarray data suggested that InvINH1 and InvINH2 were expressed in the seeds during the syncytial phase and are down-regulated after endosperm cellularization.

In support of our qRT-PCR data, the silique was identified as the structure with the highest expression level for InvINH1 and InvINH2 among the 87 anatomical structures annotated by Genevestigator (Supplementary Figure S1A). Genevestigator analysis also revealed that the expression of InvINH1 and InvINH2 were regulated by additional developmental cues and environmental stimuli (Supplementary Figure S1B). For example, InvINH1 and InvINH2 were up-regulated during germination (Narsai et al., 2011) and upon Pseudomonas syringae inoculation (GEO accession GSE5520, GSE18978). The down-regulation of InvINH1 and InvINH2 was observed in the endosperm/seed coat fraction from germinating seeds upon ABA treatment (GEO accession GSE5751), and in seedlings upon exposure to sucrose (Stokes et al., 2013). Moderate expression was also detected in senescent leaves (Supplementary Figure S1A). In general, InvINH1 and InvINH2 are expressed in tissues that undergo active sugar reallocation, such as the endosperm, germinating seeds, and senescent leaves. Collectively, theses data suggested that InvINH1 and InvINH2 are potentially involved in other biological processes in addition to their function during early endosperm development.

To investigate the spatial and temporal expression pattern of InvINH1 and InvINH2 during seed development, we generated promoter-GFP fusions for these two genes and analyzed GFP expression pattern in stable transgenic plants. The InvINH1 promoter-GFP signal was detected in both the female gametophyte and the syncytial endosperm (Figures 3A–K). Briefly, the InvINH1 promoter-GFP activity was detectable in stage FG4 female gametophyte (4-nucleate, Christensen et al., 1997, Figure 3A). In the mature female gametophyte at 1 day after emasculation, the promoter activity was only present in the central cell (Figure 3B). After fertilization, the promoter-GFP signal was more prevalent in the syncytial endosperm from endosperm stage I (one nucleus) to stage VIII (∼100 nuclei, Boisnard-Lorig et al., 2001, Figures 3C–J). After endosperm cellularization at stage IX (Figure 3K), the GFP signal decreased dramatically and was no longer detectable in seeds containing heart stage embryos (Figure 3L). In seeds that contained globular-stage embryo and syncytial endosperm, the GFP signal was primarily detected in the micropylar endosperm, with weak expression in the periphery endosperm and no expression in the chalazal endosperm (Figure 3J). Promoter-GFP expression pattern similar to that of InvINH1 was also observed for InvINH2 (Figures 3M–X), except that InvINH2 promoter-GFP signal was not detectable in the female gametophyte at 1 day after emasculation (Figure 3N). The observed GFP expression pattern was consistent among the 13 analyzed T1 lines for pInvINH1-GFP, as well as among the 11 analyzed T1 lines for pInvINH2-GFP. Therefore, both InvINH1 and InvINH2 were preferentially expressed in the micropylar endosperm prior to endosperm cellularization, which was in good agreement with the microarray and qRT-PCR data.

Based on the specific expression pattern of InvINH1 and InvINH2, we hypothesized that their function is to inhibit cell-wall-bound enzymes, such as invertase or pectin methylesterase, that are located at the embryo-endosperm interphase prior to endosperm cellularization. To investigate whether the down-regulation of InvINH1 after endosperm cellularization is connected to the acceleration in embryo growth rate, we ectopically expressed InvINH1 after endosperm cellularization using the ZOU promoter, which is preferentially active in the micropylar endosperm at both the syncytial stage and the cellularized stage (Yang et al., 2008). Since InvINH1 and InvINH2 share similar expression pattern (Figure 3) and 93.5% similarity in DNA sequence, there is a possibility that these two genes are functionally redundant. Moreover, it is difficult to generate a double mutant to address the redundancy issue, because these two genes are only 2 kb apart. Therefore, we decided to use the ectopic expression approach to investigate the function of InvINH1 during seed development. We generated transgenic plants carrying pZOU-InvINH1 and analyzed seed morphology in 21 independent T1 lines. Embryo morphology was analyzed at the late-bent-cotyledon stage. Six out of the 21 T1 lines had more than 25% delayed embryos within a given silique (Figure 4A). Within the same silique from a T1 hemizygous transgenic plant, the normal embryos were at the late-bent-cotyledon stage, while the delayed embryos ranged from early torpedo to mid-bent-cotyledon stage (Figure 4C). The delay in embryo growth was transient. At maturation stage, most of the seeds contained normal size embryos. Since the delay in embryo growth was not observed in all the lines, we next investigated if the expression level of the transgene is variable among the T1 lines. We analyzed the mRNA level of InvINH1 at 5dap in six selected T1 lines that represented different degrees of delayed embryo phenotype (Figure 4D). In general, there was a good correlation between the expression level of InvINH1 and the severity of the phenotype. As the control, we generated and analyzed 20 T1 lines carrying the promoter-only pZOU transgene (Figure 4B). Seeds containing delayed embryos were observed occasionally. However, none of the pZOU lines contained more than 25% delayed seeds (Figure 4B). We also observed semi-sterility lines containing close to 50% undeveloped ovules for both pZou and pZou-InvINH1 transgene (Figures 4A,B). Semi-sterility is caused by chromosomal rearrangement during T-DNA mediated transformation (Nacry et al., 1998). Since semi-sterility was observed for both pZou and pZou-InvINH1, this phenotype was not associated with the ectopic expression of InvINH1. Collectively, our data indicated that prolonged InvINH1 expression is sufficient to delay embryo growth.

Since embryo growth was severely suppressed in mea, fis2, fie, and msi1 mutant (Ohad et al., 1996; Chaudhury et al., 1997; Kiyosue et al., 1999; Köhler et al., 2003), we next investigated whether the embryo abortion phenotype in these mutants could be attributed to any changes in InvINH1 and InvINH2 expression level. We used qRT-PCR to compare the mRNA level of InvINH1 and InvINH2 between the wild-type and fis2-8 mutant siliques (Figure 2). In wild-type plants, InvINH1 and InvINH2 were expressed in 3-dap siliques containing syncytial endosperm, and down-regulated in 5-dap siliques containing cellularized endosperm (Figure 2). In fis2 mutant plants, the expression level of InvINHs1 and InvINH2 were up-regulated at both 3 and 5 dap (Figure 2). More specifically, roughly 2-fold up-regulation was detected at 3 dap, while at least 80-fold up-regulation was detected at 5 dap. The elevation in InvINH1 and InvINH2 expression level at 5 dap was not simply the consequence of extended syncytial stage in fis2 mutant, because elevated expression was detected as early as 3 dap when there was no morphological difference between the wild-type and the mutant seeds. Our data indicated that FIS2-PRC2 complex was required to directly or indirectly silence the expression of InvINH1 and InvINH2, which is expected to increase the activity of cell-wall-bound enzymes, such as invertase or pectin methylesterase after endosperm cellularization.