Mutant with defects in ovule development
We obtained a female sterile mutant from an Ethyl methanesulfonate (EMS) mutagenesis screening. The mutant showed increased flower size than the wild-type and shorter siliques with no seed set (Fig.1A-D). Cytological observation showed that compared with the ovule development in wild-type plants (Fig. 2A-E), the outer integument development of mutant was arrested (Fig. 2G). The mutant had defects in integument development, presented embryo sac development partially arrested at stage FG1  in which the functional megaspore either persisted or degenerated after FG1 stage (Fig. 2K,L), the functional megaspore of around 43.70% ovules could still undergo three times mitosis and develop into mature embryo sacs (Fig. 2J). After two rounds of backcrosses to reduce genetic background, the mutant showed very low seed setting rate. The seeds had thin endotesta, but no episperm that should have developed from outer integument to serve as hard dry protective covering (Fig.1E, F). The seed set(11.11%, n=432) was lower than the percentage of the ovules contained normal female gametophytes, indicating the reduction was probably caused by aberrant outer integuments. Despite of the absence of outer integuments, a few of these malformed seeds were still able to germinate in soil.
After backcross to wild-type (WT) plants, the self-pollinated BC1F2 plants were analyzed. The segregation ratio wild type to mutant fits the expected 3:1 ratio (Chi-Square=0.140, df=1, P=0.708), indicating that this mutation is recessive. After backcrossing to WT, BC1 individuals were self-pollinated and DNA of 40 BC1F2 plants of mutant and non-mutant were pooled separately for whole genome sequencing as described by Nordstrom . We compared a causative mutation based on the frequency of the non-reference allele of a SNP (Single Nucleotide Polymorphisms) in the mutant and the non-mutant pools. If the non-reference allele of a SNP is the causal mutation, its frequency in the mutant pool should be 100% and about 33% in the non-mutant pool, and the SNPs associated with the causal gene should also displayed the high frequency of non-reference alleles in the mutant pool . We selected 95 candidate SNPs (0.6% of total SNPs) with the frequency higher than 90% and lower than 50% in the mutant and non-mutant pools respectively (Fig. 3A). Among the 95 candidate SNPs, 81.05% of them were on chromosome I, 0.07% on chromosome Ⅳ,0.06% on chromosomeⅡ,0.02% on each chromosome Ⅲ and chromosomeⅤ. Closer inspection of SNPs on chromosome I, we selected SNPs (30 associated genes) in coding regions caused non-synonymous mutations or located in UTR (Untranslated Regions) for further analysis in the backcross BC2 progeny. As recombination events of the SNPs linked to the causal gene, each SNP was confirmed by PCR and sequencing at least12 mutants separately in BC2 progeny. We found only At1g22730 and At1g23900 had 100% frequency of non-reference allele in mutants, and At1g22410 had frequency of 96%, making At1g22410, At1g22730 and At1g23900 candidate genes of the sterile mutant (Fig. 3B, C).
Confirmation of candidate genes of female sterile mutant
To determine which casual gene was associated to the sterile mutant, we ordered several mutant lines with T-DNA insertion in each candidate gene(Fig .S2). Among the T-DNA insertion lines, two lines with the insertion in AP1G2 (At1g23900) showed phenotype of reduced seed set. The mutant ap1g2-1 (SALK_032500) with T-DNA insertion in exon7, its heterozygote had 51.9% seed set and almost half of the ovules were aborted (Fig .4A-C). For ap1g2-3 (SALK_137129) with T-DNA insertion in 3’UTR, 56bp upstream from poly A tail of the mRNA, homozygous plants (ap1g2-3-/-) was obtained with 23.27% of seed set (Fig .4A-C). To test whether the mutant phenotype could be restored spontaneously, 5 self-pollinated progenies of ap1g2-3-/- was checked. The result showed their seed set had no significant difference to each other (Fig S3).
Reciprocal crosses were carried out to determine whether the ap1g2 mutation affected the female or male gametophyte. ap1g2-1+/- was used to pollinate the wild-type plants, or used as female parent for pollination with wild-type pollens. And the seeds from ap1g2-1+/- were grown in soil. The genotypes of all progeny plants were assessed by PCR and scored (Table 1). The progeny of the self-pollinated ap1g2-1+/- exhibited a 1:1 segregation of the wild type to ap1g2-1+/- plants (Chi-Square=0.342, df=1, P=0.559), and no homozygotes were recovered. When ap1g2-1+/- was used as the female and male parent, the transmission efficiency was 63.63% and 60.47%, respectively. Both female and male transmission were decreased. However, in spite the partial penetrance for the ap1g2-1 allele, homozygotes for the mutation were never identified. The seeds from ap1g2-1+/- and wild-type plants were germinated on MS medium. After 2 weeks, we counted the number of seedlings and seeds failed to germinate. The analysis showed the seed germination rate of ap1g2-1+/- progeny had no significant difference with the wild-type (Pearson Chi-Square=0.668, df=1, P=0.414).
To obtain heteroallelic homozygous mutants, we crossed ap1g2-1+/- as egg donors and ap1g2-3-/- as pollen donors. In the offspring, only ap1g2-1/ap1g2-3 showed the phenotype of reduced seed set with fruits containing 47.75% normal seeds, while heterozygous plants ap1g2-3+/- had no fertility reduced phenotype.(Fig .4B, C). These data supported that the insertion in 7th exon of AP1G2 had stronger defect than that in 3’UTR. Since ap1g2-3 mutation was in untranslated region, and our RNAseq data of ovules at stage FG1 from ap1g2-3, ap1g2-1+/- and ap1g2-1/ap1g2-3 showed AP1G2 transcription level of ap1g2-1/ap1g2-3 was higher than ap1g2-3+/- (unpublished data), and similar to ap1g2-1, indicating other factors might affect AP1G2 transcription of ap1g2-3 allele.
To confirm that ap1g2 was responsible for the fertility reduced phenotype, we carried out complementation test using native promoter (ProAP1G2) driven wild type AP1G2 allele. 5 of 19 independent lines that are heterozygous for ap1g2-1 and carried the transgene showed a higher seed set (70.55%). For ap1g2-3-/-, 26 independent lines were obtained, and 8 lines complemented the ap1g2-3-/- phenotype. The seed set of the ap1g2-3-/- carrying the construct ProAP1G2:AP1G2 was 90. 81%, approaching that of WT (Fig .4B, C). And genetic complementation lines of ap1g2-4-/- also could partially rescue fertility reduced phenotype (Fig .S5). Altogether, these data suggested that the reduced fertility was due to the mutations in the AP1G2 (Fig 4B).
Developmental stage of female gametophyte affected by ap1g2
To understand at which stage the megagametophyte development might be affected in the ap1g2 mutants, around whole-mounted cleared 2000 ovules from WT, ap1g2-3-/- and ap1g2-1+/- at different stages of development were analyzed. The results showed that the outer integuments developed normally in wild-type plants and ap1g2 mutants and these plants were all able to produce normal functional megaspore cell (Fig .5A, F, L). Thereafter at stage FG3, wild-type plants contained a two-nucleate embryo sacs, and continued to develop, producing four-nucleate embryo sacs and then mature embryo sacs (Fig .5B, C, D, E).While in plants ap1g2-3-/-, most of the ovules remained a sigle cell in the nucellus or degenerated gradually (Fig .5G-J). And about 50% of the ovules from ap1g2-1+/- were arrested at stage FG1 with the functional megaspore persisting or degenerating during development. We observed 989 ovules in ap1g2-1+/- mutant, and the number of aborted ovules as described above was half of total number (1:1, Chi-Square=0.171, P=0.679).
As all the impaired embryo sacs observed in EMS-induced mutant (ap1g2-4-/-), ap1g2-3-/- and ap1g2-1+/- were arrested at one-nucleus stage (Table 3), we concluded that these defective female gametophytes were due to the loss AP1G2 function. And ovules in both insertion alleles, ap1g2-1+/- and ap1g2-3-/- were all able to fully develop outer integuments, but complementation lines of ap1g2-4-/- still had the defect of outer integuments, suggesting the defective outer integuments in ap1g2-4-/- were affected by other mutations induced by EMS rather than the mutation in AP1G2.
To confirm the cell that persisted in the abortive ovules was functional megaspore, the ANTI-KEVORKIAN (AKV) cell-identity marker during megagametogenesis were used, . The promoter pAKV is a gametophyte-specific promoter, and pAKV:H2B-YFP marker specifically expressed in the nuclei of the functional megaspore and the developing gametophyte before cellularization . ap1g2-1+/- and ap1g2-3-/- plants were crossed with pAKV:H2B-YFP marker lines. We then analyzed plants with ap1g2-1+/- allele and F2 plants with ap1g2-3-/- allele which were partially sterile. The functional megaspores were formed normally in both ap1g2-1+/- and ap1g2-3-/- (Fig .5L). But after FG1, when the wild-type ovules performed first nuclear division, producing a two-nucleate embryo sac, the defective embryo sacs from two mutants still stayed at stage FG1 instead of completion of mitotic divisions (Fig .5M-O).
The female gametophyte development within a pistil is generally synchronous with a relative narrow range of variation in WT [20, 21]. To investigate the developmental synchrony of female gametophytes in the pistils of ap1g2 mutants, we emasculated the stage 12 flowers, and after 48-72 h, we fixed pistils from flowers of the wild type and mutants at different developmental stages. The pistils from the same inflorescence were sequentially opened, and each ovule in a pistil was examined for their development stages. Compared with wild-type pistils, we observed that the developmental synchrony of female gametophytes in ap1g2-1+/-, ap1g2-3-/- and ap1g2-4-/- mutant was not only disturbed but delayed the progression of nuclear division as shown in table 2-3. In ap1g2-1+/- pistils, about half of the female gametophytes in each mutant pistils (P9-P14) were either persisted at FG1 or degraded and approximately half were wild type. While in the ap1g2-3-/- and ap1g2-4-/-, around 77.75% and 57.3% of the female gametophytes were found failed to undergo nuclear division. The numbers of aborted ovules detected in ap1g2-1+/- and ap1g2-3-/- were very close to the aborted seed rates correspondingly, which suggested that the disruption in megagametogenesis was the main factor of the reduced seed set in ap1g2 mutants.
Developmental stage of male gametophyte affected by ap1g2
To analyze whether the mutation led to additional male abortion phenotype, the viability of pollens was tested using Alexander staining. 46.04% (n = 2096) non-viable pollen was detected in mature anthers in ap1g2-1+/- plants (Fig .7F, H), and 49.71% non-viable pollen was obtained for ap1g2-3-/- plants (n = 1750). However, in ap1g2-1+/- carrying the construct proAP1G2:GUS, the viability rate was 71.22% (n=300), and in ap1g2-3-/- plants carrying the transgene, viability rate reached similar level as the wild type, resulting in 1.64% (n=600) aborted pollen, respectively.
In Arabidopsis, the development of the male gametophyte begins with the expansion of the microspore (Fig .6A, B) and a large vacuole produced, accompanied by the microspore nucleus moving to a peripheral location against the cell wall. The microspore then undergoes the first asymmetric pollen mitosis(PMI) which results a bicellular pollen gain with a large vegetative cell engulfing a small germ cell in the cytoplasm (Fig .6C). After PMI, the smaller germ cell undergoes the second mitosis (PMⅡ) to produce twin sperm cells (Fig .6D). Therefore, a mature pollen grain consists of a vegetative cell and two sperm cells [22-24].
In order to understand how the ap1g2 mutation affected pollen viability, 4’,6-diamidino-2-phenylindole (DAPI) staining was used to analyze pollen development in wild-type plants and ap1g2-1+/-. The normal mature pollen grains from wild type and ap1g2-1+/- showed three nuclei, including one vegetative nucleus and two generative nuclei (Fig .6I, J). While nearly half of the pollens from ap1g2-1+/- could not detect nuclear fluorescence signal in abnormal pollens showing shriveled shape (Fig .6J).Though at microspore stage, pollens in both WT and ap1g2-1+/- showed normal single nucleus fluorescence (Fig .6K, L), nearly half microspores of ap1g2-1+/- were not observed nuclear polarization before pollen mitosis but still showed unicellular and shriveled microspores (Fig .6O, P) at stage 12 when tricellular pollens had formed in the wild type (Fig .6M, N).
Half the pollens from ap1g2-3-/- under scanning electron microscopy (SEM), showed to be wrinkled shaped (Fig .7B, D), in contrast to those of wild type (Fig .7A, C). Besides, solid pollen germination medium  were used with the pollens from WT and ap1g2-3-/-, and we obtained 72.22% (n=180) and 30% (n=180) germination respectively (Fig .7E, F).
AP1G2 expression pattern
Analysis of mutation in AP1G2 showed that AP1G2 is of importance for the development of both the female and male gametophyte. To characterize AP1G2 expression in plants, we analyzed AP1G2 expression using qRT-PCR and reporter gene expression experiments.
Total RNA was isolated from different organs. And specific primers were used to detect AP1G2 mRNA, Actin ( Act2, At3g18780 ) as an internal control. The qRT-PCR analysis revealed that AP1G2 expression was present in each organ selected from wild-type plants, including roots, leaves, stems and flowers, but the relative expression in flowers was the highest, followed by stems and leaves that were about half of the level of AP1G2 expression in flowers (Fig .8). And for ap1g2-3-/- mutants with the T-DNA insertion in 3’UTR, the expression levels were significantly down-regulated compared with the wild-type using t-test (P<0.01).
Expression pattern was analyzed in transgenic plants to study the temporal and spatial profiles of AP1G2 gene expression. A construct in which 2 kb upstream of AP1G2 of the start codon was fused with the GUS reporter gene was transformed into the wild-type plants. 23 independent lines of T2 generation were analyzed, of which 5 showed GUS expression in the female gametophyte and GUS expression was detected after the big vacuole formed (Fig .9C) and remained until embryogenesis began, after which GUS staining reduced (Fig .9C-F). And it seemed to show the same pattern in anthers, ProAP1G2:GUS notably expressed in the male gametophyte at maturation in all independent lines we observed (Fig .9G-I). Additionally, proAP1G2:GUS was expressed in the 8-10 days seedling stage, and expression was also noted in hypocotyle (young shoot), leaves and flowers including expression in anthers, filament, pedicles, leaf primordial and shoot apical meristem. GUS expression was also observed in root tips, strong GUS staining was noted in trichomes (Fig .S4).