All experiments were performed in Arabidopsis thaliana ecotype Columbia-0 (Col-0). Plants were grown in a controlled environment at 20–22°C either under long day conditions (16 h light/8 h dark) or under short day (8 h light/16 h dark) conditions for 4 weeks after germination and then transferred to long day conditions. The suf4-1 pSUF4:SUF4-GUS seeds were donated by S.D. Michaels. Tobacco plants were germinated and grown at 20–22°C under long day conditions.

To obtain the REM_RNAi construct 252, 232 and 254 base pairs long DNA fragments specific for the coding sequence of each of the genes REM34, REM35, and REM36 were selected (the primers used to amplify the fragments are listed in the Supplementary Table 1 ). The fragments specificity was checked by BLAST against the Arabidopsis genome.

The three selected regions were PCR amplified, adding the BsaI sites to the primers, and cloned in a pENTR™ vector previously modified to function as a Golden Gate acceptor, with a single Golden Gate reaction, producing the pENTR-RNAi_REM vector. The Gateway LR reaction (Invitrogen™ Gateway™ recombination cloning system) was then performed to sub-clone the RNAi_REMs fragments into the pFGC5941 vector and used to transform Arabidopsis. Primers that were used are listed in Supplementary Table 1 .

The EAR motif was added to the C terminus of the REM34 coding sequence (see primer sequences in Supplementary Table 1 ). The fragment was cloned into the pB2GW7 plasmid (35S) passing through the pENTRY-D-TOPO vector (Invitrogen™ Gateway™ recombination cloning system). Arabidopsis plants were transformed using the floral-dip method (Clough and Bent, 1998).

Total RNA was extracted from whole inflorescences. RNA samples were treated with DNase (TURBO DNA-free^®; Ambion, http://www.ambion.com/) and retrotranscribed employing the ImProm-IITM Reverse Transcription System (Promega). Diluted aliquots of the cDNAs or genomic DNA were used as templates in qRT-PCRs, using the iQ SYBR Green Supermix (Bio-Rad) to detect target synthesis. All the experiments were performed with three technical replicates for each of the three biological replicates, with the exception of the expression analysis of REM34, REM35, and REM36 in the T1 REM_RNAi, in the T1 35S:REM34_EAR plants and for T-DNA abundancy evaluation. Primers employed for these analyses are listed in Supplementary Table 1 .

For each line, 10 siliques (dissected from three different plants) were measured, and seed, aborted seed, and non-fertilized ovule numbers were counted. For this purpose, a Leica^® MZ 6 microscope was used.

For the reciprocal crosses between wild-type and REM_RNAi #1 plants, mature siliques as well as open flowers and buds in an advanced stage of development were removed from the inflorescence of the mother plant, along with the meristem and smallest buds. Remaining buds were emasculated by removal of all floral organs except for the ovary. Then, anthers in the correct stage of development were taken from other flowers and used to pollinate the stigma. The numbers of seeds and unfertilized ovules were assessed for at least five pistils for each cross, and three biological replicas of the experiment were performed.

In situ hybridization analysis for REM34, REM35, and REM36 were performed following the same protocol and employing the same probes described by Mantegazza et al. (2014). Evaluation of the expression profile in the inflorescence and flower meristems was used as a positive control.

Yeast two-hybrid assays were performed in the yeast strains PJ69-4A and PJ69-4α (de Folter and Immink, 2011). The coding sequences of REM34, REM35, and REM36 were cloned in the pDEST32 (bait vector, BD; Invitrogen) and pDEST22 (prey vector, AD; Invitrogen) Gateway vector. The bait constructs were tested for autoactivation on selective yeast synthetic dropout medium lacking Leu, Trp, and His supplemented with 1, 3, 5, 10, or 15 mM of 3-aminotriazole, in order to set the screening conditions. After mating, colonies were plated on the proper selective media and grown for 5 days at 20°C.

The same coding sequences were also cloned in the pYFPN43 and pYFPC43 vectors, to perform the BiFC assay. Agrobacterium, transformed with the vectors and the viral suppressor p19 construct, was used to infiltrate tobacco leaves. The abaxial surfaces of infiltrated leaves were imaged 3 days after inoculation. As positive control for the infiltration, the already published VAL-VDD interaction was tested (Mendes et al., 2016). As negative controls, the constructs containing the proteins of interest were co-transformed with the empty pYFN43 and pYFC43 vectors. Furthermore, REM34 homodimerization, which was not observed in the Y2H assays, was also employed as a negative control ( Supplementary Figure 5 ).

Female gametophytes were cleared and analyzed as previously described by Brambilla et al. (2007). Inflorescences were prepared for observation using the following protocol: flowers were emasculated and the next day harvested. The emasculated pistils were left O/N at 4°C in a 1:9 acetic acid:ethanol solution. Samples were rehydrated by subsequent washes with ethanol 90 and 70% and then incubated O/N at 4°C in clearing solution (160 g chloral hydrate, 50 g glycerol, and H₂O to a final volume of 250 ml). Pistils at different developing states were separated from the other floral organs and opened to evaluate the female gametophyte morphology. For these experiments, a Zeiss Axiophot^® microscope equipped with differential interference contrast (DIC) optics was used.

For this experiment, the protocol published by Bou Daher et al. (2009) was followed applying minor modifications.

Pollen grains were plated on small glass plates, containing 2.5 ml of pollen germination medium [PGM:18% sucrose, 0.01% boric acid, 1 mM CaCl₂, 1 mM Ca(NO₃)₂, 1 mM MgSO₄, 0.5% agarose pH = 7]. The plates were incubated overnight at 22°C, with wet paper to maintain humidity. The next day, pollen germination and growth were evaluated with a Zeiss Axiophot^® microscope.

Flowers were emasculated and, after 24 h, pollinated. The pollinated ovaries were collected at two different time points: 5 and 24 h after pollination. Samples were overnight fixed and stained in absolute ethanol/glacial acetic acid 9:1, as previously described by Mori et al. (2006). Subsequently, they were transferred into a 8M NaOH solution for 1 h at 50°C. Finally, the carpels were washed twice with ddH₂O for 10 min. The staining was performed with a modified aniline blue solution (aniline blue 2%, glycerol 1 M ddH₂O) (Takeuchi and Higashiyama, 2016). Samples were stored at 4°C for 3 h or overnight. The observation was done under UV light (350–400 nm) with a Zeiss Axiophot^® microscope.

Pollen was stained according to Park et al. (1998). Mature pollen was obtained by placing 3–4 open flowers in a microcentrifuge tube containing 300 µl of 4′,6-diamidino-2-phenylindole (DAPI) staining solution (0.1 M sodium phosphate (pH 7), 1 mM EDTA, 0.1% Triton X-100, 0.4 µg/ml DAPI high grade, Sigma). After brief vortexing and centrifugation, the pollen pellet was transferred to a microscope slide and observed with a Zeiss Axiophot^® microscope. Pollen at earlier stages of maturation was also analyzed by dissecting single anthers. Anthers were disrupted on microscope slides and squashed in DAPI staining solution (1 µg/ml) under a coverslip.

β-Glucuronidase (GUS) assays were performed as described by Resentini et al. (2017). Pistils at different developmental stages were dissected and fixed in acetone 90% and incubated O/N at 37°C. After staining, they were cleared using the protocol described above.

Staining of pollen grains was performed as described by Peterson et al. (2010). After fixation (performed with 6 alcohol:3 chloroform:1 acetic acid), the anthers were placed on a microscope slide with a few drops of staining solution (10 ml 95% alcohol, 1 ml malachite green (1% solution in 95% alcohol), 50 ml distilled water, 25 ml glycerol, 5 ml acid fuchsin (1% solution in water), 0.5 ml orange G (1% solution in water), 4 ml glacial acetic acid, and distilled water (4.5 ml) to a total of 100 ml). Samples were analyzed with a Zeiss Axiophot^® microscope.

For confocal imaging, the Laser Scanning Confocal Microscope Nikon A1 was used. Inflorescences were fixed as described by Braselton et al. (1996). Samples were then excited using a laser (532 nm), and emission was detected between 570 and 740 nm.

Since REM34, REM35, and REM36 are very similar and in linkage, an RNA interference approach was adopted to investigate their role during reproductive development in Arabidopsis.

Due to sequence divergency, even in the B3 DNA binding domain (Romanel et al., 2009), it was impossible to design a single artificial small interfering RNA fragment that was able to silence the three REM genes simultaneously. Therefore, a multiple RNA interference (RNAi) technology was used to express a single chimeric double stranded RNA that targeted the three REM genes under the control of CaMV35S (Miki et al., 2005; Bucher et al., 2006) ( Figure 1A ).

We selected three regions specific for the coding sequence of REM34, REM35, and REM36. The regions selected for REM34 and REM36 are highly specific for the genes of interest and were expected not to have any off target in the Arabidopsis genome. The RNAi fragment that targets REM35 has a partial complementarity with REM36, and, at a lower level, with REM37, whose expression is almost undetectable in most Arabidopsis tissues (Mantegazza et al., 2014; Klepikova et al., 2016).

Forty REM_RNAi T1 transgenic Arabidopsis lines were obtained. We evaluated the down-regulation of the REM genes in nine different T1 lines ( Figure 1B ), which all showed defects in silique and gametophyte development.

Silencing of the three target genes was confirmed in the T2 generation by qRT-PCRs ( Supplementary Figure 1 ). Furthermore, we showed that the RNAi construct was specific for their targets by testing the expression of REM37 and REM39. The latter was chosen due to the fact that REM39 is highly expressed in the tissues where REM34, REM35, and REM36 are also active (Mantegazza et al., 2014; Supplementary Figure 1 ).

We selected three REM_RNAi lines (#1, #4, and #5), with different levels of silencing of REM36, for further investigations in the T2 generation. In line #1, REM36 showed a downregulation of around 50%, while in lines #4 and #5, REM36 was found to be slightly upregulated compared to the wild-type ( Figure 1B ).

In the T2 generation, silique length and seed number were evaluated for the three selected lines. The REM_RNAi #T2.1 line showed a decrease of 35.3% in the silique length and a 19.4% reduction in total ovule number ( Figures 2A–C ). Furthermore, on average 66% of the ovules failed to be fertilized ( Figures 2C, D ). The other two REM_RNAi lines, #T2.4 and #T2.5, showed a similar phenotype even if the percentage of unfertilized ovules was lower, 35.3 and 45.4%, respectively ( Figure 2C ).

The REM_RNAi #1 line was selected to further investigate the sterility phenotype caused by the downregulation of REM34, REM35, and REM36. This line was propagated to the T3 generation, where plants homozygous for the REM_RNAi construct were selected. Even if the RNAi construct has a dominant effect, we evaluated whether the sterility observed in the REM_RNAi T2 segregating lines was exacerbated in plants homozygous for the construct. For this purpose, the seed set of the REM_RNAi T3.1 homozygous line was evaluated.

Interestingly, comparing both the REM_RNAi #T2.1 and the REM_RNAi #T3.1, we noticed that the percentage of ovule abortion was the same, suggesting that the silencing of REMs is probably acting both at the sporophytic and gametophytic levels.

Since the two lines in which REM36 was not downregulated displayed a milder phenotype compared to the REM_RNAi #T2.1 and REM_RNAi #T3.1 lines, in which all three genes were downregulated, it is possible that REM36 is partially redundant to REM34 and REM35 during gametophyte development. On the contrary, the ovule number was the same in all three REM_RNAi lines ( Figures 2A, C ), indicating that REM36 is not involved in the determination of the ovule primordia number.

To further confirm that no phenotypical differences were detectable between plants homozygous and heterozygous for the T-DNA insertion, we analyzed the silique content of 10 REM_RNAi #T2.4 and 10 REM_RNAi #T2.5 T2 plants in which the construct was still segregating, and we found no significant differences between all the herbicide resistant plants ( Supplementary Figures 2A, C ). For both REM_RNAi #T2.4 and #T2.5 lines, a relative evaluation of T-DNA copies in each of the nine plants considered was performed. The RT-PCR analyses showed a various amount of T-DNA amplicons which is clearly unrelated to the ovule abortions and the overall seed set observed in all the REM_RNAi #T2.4 and #T2.5 analyzed individuals ( Supplementary Figure 2 ). The ACTIN7 amplicon was used as normalizer and the herbicide resistance BAR gene used to estimate the abundancy of T-DNA copies.

These analyses allowed excluding the possibility that the reduced seed set was linked to the presence of a heterozygous T-DNA insertion (Curtis et al., 2009; Clark and Krysan, 2010) and suggests that either the sporphytic silencing of REM34, REM35, and REM36 affects the gametophyte or that the mobile siRNA diffuses from the sporophyte to the gametes (Mlotshwa et al., 2002; Melnyk et al., 2011; Skopelitis et al., 2018).

To understand if the reduced seed set was due to problems in the female or the male gametophyte, we performed reciprocal crosses between REM_RNAi #T3.1 and wild-type plants. As a control, both REM_RNAi #T3.1 (homozygous for the T-DNA triggering the RNAi silencing) and wild-type plants were manually selfed, in order to evaluate if the manipulation of the flower was affecting the fertility of the analyzed plants ( Figure 2E ).

When REM_RNAi #T3.1 pistils were pollinated with REM_RNAi #T3.1 pollen, 73.3% of the ovules failed to be fertilized while wild-type lines manually pollinated with wild-type pollen resulted in 19.5% unfertilized ovules. When the REM_RNAi #T3.1 line pistils were pollinated with wild-type pollen, the percentage of unfertilized ovules was 78.6%, indicating a strong contribution of the female reproductive organ defects to this phenotype. Interestingly, when wild-type pistils were pollinated with REM_RNAi #T3.1 pollen, still 61.0% of the ovules were not fertilized ( Figure 2E ). Moreover, we observed a high variability in the number of unfertilized ovules using REM_RNAi pollen as shown in Figure 2E . Macroscopical inspection revealed a decrease in pollen grain number compared to wild-type anthers and a lack of adherence of the pollen to the wild-type stigma, both observations were further investigated (see below). All these considerations strongly suggest that both female and male reproductive organs are affected in the REM_RNAi lines.

Previously, the expression pattern of the REM genes was characterized in the shoot apex by in situ hybridization analysis, showing that REM34, REM35, and REM36 are expressed from the earliest stages of reproductive development of Arabidopsis in the inflorescence meristem and flower meristem and during the first stages of flower development with the exception of sepals (Franco-Zorrilla et al., 2002; Mantegazza et al., 2014).

In order to analyze the expression profiles in more detail during male and female sporophytic/gametophytic developments, we performed in situ hybridization analysis for REM34, REM35, and REM36 in both female and male reproductive organs. The flower stages are described accordingly to Smyth et al. (1990) and Schneitz et al. (1995).

In Arabidopsis, pollen mother cells differentiate inside the young anther, ovule primordia arise from the placenta in stage 8 of flower development, and differentiation is completed at stage 13.

At stage 8/9 of flower development (Smyth et al., 1990), hybridization signals were detected for all three genes, in the anthers where the pollen mother cells differentiate and within the carpels, although in this case, the signal was stronger in the placenta and ovule primordia ( Figure 3A ).

At stage 10, a strong signal was always detected in developing ovules and pollen ( Figure 3B ). Our analysis revealed that the timing of expression of the three REM genes coincided with male and female sporogenesis.

During subsequent stages of flower development, stages 11–12, both female and male initiate gametophyte development. During these stages, a decrease in the signal was clearly observed in anthers ( Figure 3C ); during these stages, pollen reaches maturity, and the vegetative and generative cells are differentiated after mitosis (Park et al., 1998). In contrast, a strong signal was detected during ovule development when the surviving megaspore undergoes three rounds of mitosis and passes from stages 3I to 3V (Schneitz et al., 1995). Interestingly, when the ovule is at its very last stage of development 3-VI (Schneitz et al., 1995), a strong signal was detected in the funiculus, in the innermost integument and, inside the mature female gametophyte, in the central cell region ( Figure 3D ).

The expression analysis of REM34, REM35, and REM36 highlighted the fact that, also during anther/pollen and carpel/ovule development, these three REMs have a similar pattern of expression.

The analysis of the expression patterns of REM34, REM35, and REM36 combined with the phenotypes observed in the REM_RNAi lines denote an important role for these genes during the development and production of viable male and female structures and gametes.

The expression profile of REM34, REM35, and REM36 suggests that these genes play a role during ovule development. Furthermore, the reciprocal crosses showed that between 73.3 and 78.6% of the ovules in the REM_RNAi #T3.1, which is homozygous for the RNAi cassette, were not fertilized ( Figure 2E ).

Based on this evidence, we hypothesized that the ovule defects in the REM_RNAi lines might be due to an arrest in their development. Therefore, a detailed evaluation of female gametophyte development was carried out in the REM_RNAi #T3.1 homozygous line. In this line, 42.9% (227/529) of the ovules failed to complete their development and showed an arrest in the FG1 stage ( Figure 4A ). These ovules were characterized by an embryo sac containing one large cell, the functional megaspore, with a single nucleus; the rest of the ovules completed their development reaching the FG7 stage ( Figures 4B, C ). The same phenotype was observed in the RNAi #T2.4 and #T2.5 lines which both derived from hemizygous mothers ( Supplementary Figure 3 ).

To confirm that, in the REM_RNAi lines, the defective female gametophytes were arrested in the FG1 stage, after meiosis, we crossed the pSUF4:SUF4-GUS marker line with the REM_RNAi #T3.1 line. In the pSUF4:SUF4-GUS marker line, GUS expression is not detectable during megasporogenesis, but it becomes visible after meiosis, once the functional megaspore is formed, and marks all the nuclei of the embryo sac (Resentini et al., 2017). Observing REM_RNAi #T3.1 pistils, both wild-type like ovules, with more than one nucleus and ovules arrested in the FG1 stage, with the nucleus of the functional megaspore, expressed the GUS reporter ( Figure 4D ) suggesting that the defect in female gametophyte development was post-meiotic.

To investigate in detail the arrest at the FG1 stage, we carried out confocal laser scanning microscopy (CLSM) on REM_RNAi #T3.1 developing ovules. The feulgen staining perfectly marked the cell wall of the ovule integuments and the embryo sac dividing nuclei, allowing the recognition of the gametophytic developmental stages. In the same REM_RNAi #T3.1 pistil, we observed ovules that normally developed until stage FG4 ( Figure 4E ) and those that were arrested in FG1 in which the embryo sac contains the functional megaspore and the three degenerating spores on top of it ( Figure 4F ).

From the analysis of wild-type carpels pollinated with REM_RNAi pollen, we observed that 61.1% of the ovules were not fertilized, suggesting that the male gametophyte in these lines is also defective ( Figure 2E ). To understand the cause of this defect, we first carried out an in vitro pollen germination assay which showed a 30% decrease in the germination rate of the REM_RNAi #T3.1 pollen compared to the wild type ( Figures 5A, B and Supplementary Figure 4 ).

The growth rate of REM_RNAi pollen tubes and their ability to correctly target ovules were also evaluated in vivo by means of aniline blue staining ( Supplementary Figure 4 ). The REM_RNAi pollen tubes did not show any growth defect, they reached the end of the pistil in the same time as the wild-type pollen tubes, and the mature ovules were correctly targeted ( Supplementary Figure 4 ). We noticed that, as mentioned before, the REM_RNAi pollen number appeared to be lower, and it did not adhere well to the stigma papillae, which could explain the high variability observed in the backcrosses between wild-type pistils and REM_RNAi pollen ( Figure 2E ).

To try to understand the cause of the male sterility phenotype, pollen grains were collected from mature anthers and treated with Alexander’s stain, which colors viable pollen red. While in the wild type, all the collected pollen was viable; in the REM_RNAi #T3.1 anthers, 33.9% of the grains were not stained, indicating that those pollen grains were non-viable and did not appear to contain any cytoplasm ( Figures 5C–E ). Interestingly, the percentage of non-viable pollen grains in the REM_RNAi line corresponds to the decreased germination capability observed in vitro, suggesting that the grains which are unable to produce the pollen tube are the degenerated ones.

To investigate the pollen defect in more detail, confocal laser scanning microscopy (CLSM) was used. In Figure 5F , wild-type pollen from a mature anther is shown; intine and exine layers were very well distinguishable and inside the pollen grain, and the sperm cells and the vegetative cell nuclei were stained. On the contrary, in the REM_RNAi #T3.1 mature anthers, a high percentage of pollen grains appeared shrunken and empty; neither sperm nor vegetative cells were identified, although the intine and exine layers looked intact ( Figure 5G ).

To understand when the pollen grains degenerated, we visualized their nuclei with DAPI staining at different developmental stages ( Figures 6A–F and Supplementary Figure 3 ). At the microspore stage, all REM_RNAi #T3.1 grains were characterized by the presence of a single bright nucleus localized at the center of the cell, indicating that the pollen, like wild-type, passed through meiosis correctly ( Figures 6A, D ). After meiosis, in wild-type, the microspores underwent a first mitotic division that produced one vegetative and one sperm nuclei ( Figure 6B ). Subsequently, the second round of mitosis led to the formation of the mature pollen grain, which contained two small sperm cells each with a bright and elongated nucleus and the vegetative cell ( Figure 6C ). Interestingly, in REM_RNAi #T3.1 anthers, some grains were characterized by the lack of nuclei; this phenotype was detectable also at the tricellular stage ( Figures 6E, F and Supplementary Figure 3 ).

Thus, after meiosis, REM_RNAi anthers displayed both viable pollen grains, with two sperm cell nuclei and a distinct vegetative nucleus, and not viable pollen grains, in which no DNA is detectable ( Figures 6E, F ). This is similar to what was observed with the CLSM analysis.

All this evidence suggests that the degeneration of pollen grains observed in the REM_RNAi lines could be due to a post-meiotic block in their development, a similar defect as the one observed in the female gametophytes.

REM transcription factors can form functional heterodimers (Mendes et al., 2016). To understand if also REM34, REM35, and REM36 could function via dimer formation, yeast two-hybrid assays were performed. This approach revealed that REM35 is able to interact strongly with itself and also with REM34, while no interactions were detected with REM36 ( Figure 7A and Supplementary Figure 5 ).

All the interactions observed in the yeast two-hybrid assays were confirmed in vivo with a Bimolecular Fluorescence Complementation (BiFC) assay in Nicotiana benthamiana leaves ( Figures 7B, C and Supplementary Figure 5 ). This finding suggests that REM34 and REM35 could act as heterodimers.

As described above, the downregulation of REM34, REM35, and partially REM36 resulted in a post-meiotic arrest in both female and male gametophytes, suggesting that these transcription factors could be involved in regulating mitosis progression during gametogenesis.

To elucidate the molecular mechanism causing this block, we measured the expression levels of genes that control gametogenesis by q-RTPCR. We focused on genes that, when mutated or overexpressed, cause similar defects to those observed in the REM_RNAi gametophytes. Those genes were divided into three categories based on the biological process in which they are involved in: ribosome biogenesis (MDS, NLE), cell cycle control (RBR, KRP6), and chromatin regulation (HAM1, HAM2).

MDS, which, together with NLE, is involved in the biogenesis of the 60S ribosomal subunit and is essential during megagametogenesis (Chantha et al., 2010), was downregulated in the REM_RNAi #T3.1 lines. KRP6, a CDK inhibitor whose overexpression causes a block in mitosis progression during female and male gametophytic development, was also downregulated in the REM_RNAi lines.

Among the genes involved in chromatin modifications, two histone acetyltransferases (HATs), HAM1 and HAM2, were selected. Only HAM2 was downregulated in the REM_RNAi #T3.1 line ( Figure 8 ).

These results suggest an intricate interconnection among regulators and effectors, which end up in a correct gametogenesis program.

The genes that were downregulated in the REM_RNAi lines might be targets of the REM transcription factors. This suggests that REM34 and REM35 might be transcriptional activators. To investigate whether REM transcription factors work as activators of transcription, we fused REM34 with the dominant EAR repressor domain (known as chimeric repressor silencing technology CRES-T) and transformed wild-type Arabidopsis plants with this construct. Five transgenic lines that overexpressed the REM34_EAR chimeric gene at different levels in the T1 generation were obtained ( Figure 9A ).

In the T2 generation, silique length and ovule number were measured in two independent lines (REM34_EAR#T2.1 and REM34_EAR#T2.7). In both the selected T2 REM34_EAR lines, we observed a decrease in the silique length of 23.1 and 25.0%, respectively, and the presence of 55.0 and 42.6% aborted ovules, similar to what was observed in the REM_RNAi lines ( Figures 9B–E ).

The phenotype of the aborted ovules was further evaluated in cleared mature carpels of the REM34_EAR #T2.1 and #T2.7 lines. We detected both ovules at FG7 stage, with the seven cells clearly distinguishable, and ovules at FG1 stage, characterized by a single cell embryo sac ( Figure 9F ).

To confirm also the post meiotic block in the male gametophyte, mature pollen of both REM34_EAR #T2.1 and #T2.7 lines were stained with DAPI. Similarly to what was observed in the REM_RNAi lines, some pollen grains were able to reach the tricellular stage while others appeared shrunken and degenerated, with no visible nuclei ( Figure 9G ).

The strong similarity between the REM34_EAR and the REM_RNAi phenotypes might suggest that the overexpression of the chimeric REM34_EAR protein was causing co-suppression of other REM genes. To exclude this possibility, we investigated the expression level of REM35, REM36, REM37, and REM39 in the REM34_EAR #T2.1 and #T2.7 lines. The level of expression of the endogenous REM34 was not taken into account, as the perturbation of REM34 expression alone did not cause any evident phenotypical defects ( Supplementary Figure 6 ) (Franco-Zorrilla et al., 2002; Mantegazza et al., 2014). The obtained results showed that the closely related REMs were not affected suggesting that the expression of the REM34_EAR chimeric protein caused the observed phenotypes.

The plant-specific REM family in Arabidopsis is composed of 45 genes, generated through multiple duplication events, which are mostly expressed during flower and ovule development (Romanel et al., 2009). Even if the expression pattern of these genes suggests that they could play an important role in regulating developmental processes such as shoot architecture and flower development, until now, only a few of them have been associated to a function (Levy, 2002; Matias-Hernandez et al., 2010; Mendes et al., 2016). This might be due to their functional redundancy but also because they are often in linkage on the genome.

Here, we investigated the function of the linked duplicated REM34, REM35, and REM36 genes by a multiple RNAi approach and showed that REM34, REM35, and partially REM36, are involved in male and female gametophytic developments during post-meiotic divisions. A similar multiple RNAi approach was previously employed to silence simultaneously up to six target genes in Arabidopsis thaliana (Czarnecki et al., 2016).

The REM_RNAi construct was found to be a very efficient tool: by selecting specific gene sequences, we were able to silence the three target genes with a single construct and a single transformation event. Importantly, the construct showed to be highly specific for the three genes of interest without any obvious off-target activity. Transgenic lines showing silencing of the REM genes under study were all characterized by a reduced seed set and an arrest in female gametophyte development at the earliest stages of gametogenesis. Since REM34, REM35, and REM36 appeared to be mainly expressed in sporophytic tissues throughout Arabidopsis reproductive development, the CaMV35S promoter was chosen to drive the expression of the RNAi fragments. The activity of the CaMV35S promoter seems to be low during female and male gametophyte developments, but it has been shown that such promoter can be successfully employed to silence genes during gametophytic development (Acosta-García and Vielle-Calzada, 2004; Mendes et al., 2016). A valid hypothesis for the observed gametophyte phenotypes might be that it is caused indirectly by the silencing of REM34, REM35, and REM36 in the female and male sporophytic cells. However, it is also important to consider that the RNAi construct is dominant and that it can trigger a non-cell autonomous and systemic silencing signal which might be maintained throughout the different phases of plant development (Mlotshwa et al., 2002; Melnyk et al., 2011; Skopelitis et al., 2018).

Since functional redundancy is a common phenomenon in plants (Briggs et al., 2006), this kind of RNAi approach will be helpful for the functional characterization of members of highly redundant families and especially that are in linkage. Furthermore, since silencing of genes by RNAi is often not complete, this approach could favor the analysis of genes for whom knock-out leads to lethality or complete sterility.

Protein interaction studies revealed that REM34 and REM35 were able to interact with each other; while no interaction was found with REM36, this supports the hypothesis that REM36 might not be able to substitute completely REM34 and REM35 functions. Interactions between REM factors were found before. VDD and VAL, two functionally characterized REM factors involved in synergid degeneration upon fertilization (Mendes et al., 2016), also interact with each other. Furthermore, both VAL and REM35 were also able to make homodimers. These characteristics might well be a common feature for the REM family and, in perspective of the guilt-by-association principle, it would be informative to analyze all possible REM protein interactions. The same approach was shown to be extremely useful for the characterization of MADS domain transcription factor family, for which extensive protein–protein interaction studies effectively guided genetic studies and functional characterization of many of them (de Folter et al., 2005; Gregis et al., 2006; Fornara et al., 2008; Immink et al., 2009).

We discovered that, in the REM_RNAi lines, both the male and female germ lines were able to go through meiosis correctly, but they were not able to pass the FG1 stage, suggesting a role for REM34 and REM35 in the control of gametogenesis in Arabidopsis.

Although the REM gene family was named after the specific meristematic expression of its first member AtREM1, which was named REM34 (Franco-Zorrilla et al., 2002), our data showed that REM34, REM35, and REM36 are also expressed during gametophytic development, and we discovered that they were expressed starting from both carpels and anther primordia specification throughout all the stages of anther and carpel developments. In the carpel, the signal is strongly localized in the placenta and ovule primordia and in the developing ovules.

Indeed, our deep morphological analysis of both female and male gametophytes of the REM_RNAi lines showed that from 35 to 65% of the female gametophytes were unable to undergo mitosis and were arrested at the FG1 stage when the MMC acquires functional megaspore identity.

REM36 seemed to be partially redundant with REM34 and REM35. Indeed, in the two lines in which the level of REM36 expression was higher compared to the wild-type, the penetrance of the embryo sac defect was less. However, in all REM_RNAi lines, we also observed a decrease of around 20% in the total ovules number irrespectively of the expression levels of REM36. Thus, REM36 might be involved in embryo sac development together with REM34 and REM35 but is not controlling ovule primordia specification.

In these lines, also pollen development was affected showing the same post meiotic arrest of the embryo sac. Thus, in Arabidopsis, REM34, REM35, and partially REM36 transcription factors seem to be required post-meiotically for gametophytic development.

Further confirmation for their role during both female and male gametogenesis came from the analysis of different 35S:REM34_EAR lines. These plants, in which REM34 fused to the EAR repressor domain was overexpressed, showed the same postmeiotic arrest both in embryo sac and pollen development, suggesting that a complex formed by REM34 and REM35 could act as a positive transcriptional regulator of gametogenesis.

Because of the redundancy and position in linkage of the three genes of interest in the genome, most of this functional study was conducted using RNAi. This approach was found to be very effective in the silencing of REM34, REM35, and REM36, but the transgenic lines cannot be easily employed for genetic studies, due to the fact that it acts dominantly and because the level of silencing of the target genes can vary between different lines and throughout subsequent generations. Despite these difficulties, the analyses performed on both segregating and homozygous lines suggest that these three genes can influence gametogenesis acting mainly at the sporophytic level. This hypothesis is also supported by the expression pattern of these genes which, as shown by the in situ hybridization analysis, are present in the sporophytic tissues both in pistils and anthers when gametogenesis is taking place. The observation that REM34, REM35, and REM36 appeared to be expressed throughout all stages of gametogenesis in the embryo sac leaves of course the possibility open that they directly play a function in the female gametophyte. The employment of an embryo sac specific promoter could be useful in order to validate this hypothesis and to be able to better distinguish between the sporophytic and gametophytic roles of REM34, REM35, and REM36.

To understand how the REM genes under study act, we tested whether the down-regulations of REM34, REM35, and REM36 perturbed the expression of genes known to be involved in gametogenesis progression. These genes were classified accordingly to their biological function in three categories: cell cycle control, chromatin remodeling, and ribosome biogenesis. Interestingly, we observed that several genes involved in different biological pathways were downregulated in the REM_RNAi lines. This observation suggests that REM34, REM35, and, in some measures, REM36 are involved in the control of a very early steps of gametogenesis. In particular, they regulate the expression of different targets both directly and indirectly along the genetic network that controls gametophytic development in Arabidopsis.

Among the downregulated genes, the one that stands out most is HAM2, a HAT that, together with its homolog HAM1, belongs to the MYST clade of the HAT family and was shown to be involved in post-meiotic control of female and male gametophytic development (Latrasse et al., 2008). In mammals, the MYST protein family was found to be involved in many fundamental cell functions such as cell cycle progression and DNA repair (Pillus, 2008; Sapountzi and Côté, 2011). Furthermore, the human MYST4 acetylase was found to be expressed and involved in the control of gametogenesis as well (McGraw et al., 2007). In Arabidopsis, the ham1 ham2 double mutant is lethal, while keeping one of the two genes heterozygous for the mutant allele resulted in a post-meiotic arrest of both female and male gametophyte developments (Latrasse et al., 2008). This phenotype is similar to the one observed in the REM_RNAi as well as in the 35S:REM34_EAR lines. Interestingly, HAM1 and HAM2 were also found to be involved in the control of flowering time via the epigenetic regulation of FLOWERING LOCUS C (FLC), which is also a target of VRN1, one of the four REM genes for which the function is known so far, suggesting a common mechanism throughout plant development. The artificial silencing of these two acetyltransferases causes an early flowering phenotype (Xiao et al., 2013) which was also noticed in the REM_RNAi lines (data not shown). The downregulation of HAM2 and the phenotypical similarity between the REM_RNAi lines and the HAM downregulation suggest that these genes might be involved in the control of the same biological processes throughout Arabidopsis development. Further analysis will be needed to confirm the possible interaction between REM34, REM35, and the HATs HAM1 and HAM2. The observed downregulation of the other analyzed target genes could be due to the general deregulation of transcription caused by the reduced expression of the chromatin remodeling factor HAM2.

While not much is known about the REM gene family, substantial information is available for other transcription factor families that are characterized by the presence of the B3 DNA binding domain. In particular, the well-characterized auxin response factor (ARF) family, known to play a crucial role in regulating auxin responses, and the related to ABI3/VP1 (RAV) family, which was found to be involved in hormonal regulation during different stages of Arabidopsis development (Swaminathan et al., 2008). The plant hormone auxin was found to be involved in gametogenesis (Pagnussat et al., 2009; Panoli et al., 2015). Indeed, perturbation of auxin transport in the embryo sac causes an arrest in the earliest stages of megagametogenesis (Ceccato et al., 2013). Auxin biosynthesis in the male gametophyte was also recently shown to be essential for the transition from microsporogenesis to microgametogenesis (Yao et al., 2018). Because of the phenotypic similarities between the auxin defective mutant (Pagnussat et al., 2009; Panoli et al., 2015; Ceccato et al., 2013) and the REM_RNAi lines and because of the linkage between transcription factors containing the B3 DNA-binding domain and the regulation of hormonal responses, it is tempting to speculate that the role of REM34, REM35, and REM36 play in gametogenesis is also based on the regulation of a hormonal related processes.

In summary, we gained new information about the expression pattern and function of REM34, REM35, and REM36 during gametophyte development in Arabidopsis; those genes might control post-meiotic divisions in both embryo sac and pollen grains. These findings underline further the importance of REM genes during reproductive development in plants. Although these genes are often highly redundant and physically linked in the genome, slowly on, we start to get a better understanding about their functions in plant development. Of course, we just see the tip of the iceberg and still a huge amount of work has to be done to fully understand in detail the molecular and genetic mechanisms by which REM genes function.