Endosperm stores starch and proteins in cereal crop seeds, providing a major calorie source for human consumption. Rice, given its agricultural importance and rich genetic resources, serves as a model crop for dissecting the molecular basis of endosperm development. Although many genes affecting endosperm development have been functionally characterized, our understanding of storage substance accumulation remains fragmented (He et al., 2021; Huang et al., 2021). Serine hydroxymethyltransferases (SHMTs) catalyse the reversible conversion of glycine to serine in known organisms, including plants (Schirch and Szebenyi, 2005). Genetic evidence from plants suggests the involvement of SHMTs in biotic and/or abiotic stress responses (Liu et al., 2012; Moreno et al., 2005), but their potential functions in endosperm development remain obscure.

To dissect the molecular mechanisms underlying endosperm development, we identified seven allelic floury endosperm mutants flo20-1 to flo20-7 from an ethylmethanesulfonate-mutagenized pool of japonica rice variety Kitaake (Figure S1) and chose flo20-1 for in-depth studies. flo20-1 plants displayed defective endosperm development, as evidenced by floury endosperm appearance and ~23% reduction in 1000-grain weight at maturity (Figure 1a,b). Starch and protein contents were reduced but lipid content was increased in flo20-1, compared with wild type (WT; Figure 1b). To investigate the cytological basis of disrupted storage substance accumulation, we performed scanning electron microscopy (SEM) and light microscopy of endosperm. flo20-1 endosperm was filled with loosely arranged and abnormal starch granules (SGs), compared with tightly packed and sharp-edged SGs in WT (Figure 1c). Coomassie blue staining of developing endosperm at 10 days after flowering (DAF) revealed two types of protein bodies (PBs): lightly stained PBIs and densely stained PBIIs in WT (Figure 1d). Notably, PBII morphology appeared to be amorphous in flo20-1 (Figure 1d), which is similar to a reported mutant with altered storage protein composition (Ashida et al., 2011). Floury grains from reciprocal cross plants phenocopied flo20-1 (Figure S2), suggesting that FLO20 directly regulates endosperm development.

Map-based cloning combined with sequencing revealed that each flo20 mutant harbours a single-nucleotide substitution in Os01g0874900, causing mis-sense, non-sense or splicing mutations (Figures 1e and S3). We introduced genomic or GFP-fused Os01g0874900 into flo20-1 calli to generate transgenic plants for complementation. As predicted, flo20-1 carrying either transgene displayed translucent endosperm (Figure 1f), confirming that Os01g0874900 represents FLO20 and that FLO20-GFP was functional. FLO20 encodes a predicted 64.8-kDa protein harbouring an SHMT domain (https://www.uniprot.org/uniprot/Q8RYY6), named SHMT4. Immunoblotting using anti-SHMT4 antibodies detected SHMT4 in all tissues examined. SHMT4 levels were lower during early endosperm development, peaked at ~12 DAF and decreased thereafter (Figure 1g). SHMT4-GFP localized to nuclei in developing endosperm (Figure 1h).

Given that SHMT4 contains an SHMT domain, we determined whether SHMT4 loss affected the SHMT enzymatic activity, and found that more than 64% of SHMT activity was abolished in flo20-1 (Figure 1i). Unexpectedly, we failed to detect the SHMT4 activity in vitro (Figures 1j and S4). SHMTs usually function as homozygous or heterozygous complexes (Anderson et al., 2012; Schirch and Szebenyi, 2005). As anticipated, we verified SHMT4's interacting with itself and two homologues SHMT3&5 using a combination of in vitro yeast two-hybrid (Y2H) as well as in vivo biomolecular fluorescence complementation (BiFC) and co-immunoprecipitation assays (Figures 1k–m and S5). Using high-performance liquid chromatography, we detected the SHMT enzymatic activity of SHMT4-GFP precipitate (Figure 1n), suggesting that SHMT4 may cooperate with other proteins to execute SHMT function. We indeed detected the SHMT activity of SHMT3 alone in vitro (Figures 1o and S6) and found that SHMT3 mutation slightly but significantly compromised SHMT activities of SHMT4-GST precipitate (Figures 1p and S7). Consistently, the SHMT3 knockout mutant exhibited a 35.3% reduction in SHMT activity, compared with a 64.1% reduction in flo20-1, although SHMT3 loss did not obviously affect endosperm development (Figures 1i and S8). Together, these results suggested that SHMT4 works cooperatively with other proteins such as SHMT3 to execute SHMT activity, in which SHMT4 may be more important than SHMT3.

We specifically detected the S-adenosyl-L-methionine synthetase SAMS2 in the SHMT4-GFP precipitate by mass spectrometry (Figure 1m), and confirmed SHMT4's interaction with SAMS2 via BiFC and co-immunoprecipitation assays in tobacco (Figure 1q). SAMSs catalyse the conversion of methionine to produce SAM, which serves as the methylation donor in transmethylation reactions and an intermediate in polyamine and ethylene biosynthesis (Li et al., 2011). We measured SAM concentrations in WT and flo20-1 endosperm, finding that SAM level was threefold higher in flo20-1 (Figure 1r). We next performed bisulphite sequencing of WT and flo20-1 endosperm DNA to determine whether SHMT4 loss affects genome-wide DNA methylation (Table S1). Compared with WT, CG, CHG and CHH methylations were higher in flo20-1 (Figure 1s, Table S2), suggesting a possible role of SHMT4 in endosperm DNA methylation. Differentially methylated regions (DMRs) analysis identified 25 715 DMRs and 15 888 differentially methylated genes (DMGs) between WT and flo20-1 (Table S3, Data S1), which included genes involved in storage substance biosynthesis, transport or accumulation (Table S4). Additionally, expression levels of several transcript factor genes required for starch and protein accumulation were significantly reduced in flo20-1 (Figure 1t). Together, the SHMT4 mutation causes genome-wide methylation changes in developing endosperm, probably through affecting SAM production.

A leaky SHMT4 allele was recently reported to confer enhanced cadmium tolerance and selenium accumulation by affecting their uptake and assimilation in root and/or shoot (Chen et al., 2020). However, the putative SHMT4 function in endosperm development remains unclear. Here, we identified seven allelic SHMT4 mutants defective in endosperm development. Storage protein composition greatly determines nutritional and functional qualities of cereal crops. We noted that SHMT4 loss caused a disruption in storage protein composition, suggesting the potential value of SHMT4 in breeding rice with special demands. Together, our study provided a functional link between SHMT proteins and endosperm development in plants.

胚乳在谷类作物种子中储存淀粉和蛋白质，为人类消费提供主要热量来源。水稻由于其农业重要性和丰富的遗传资源，可作为剖析胚乳发育分子基础的模式作物。尽管许多影响胚乳发育的基因已经在功能上进行了表征，但我们对储存物质积累的理解仍然是零散的（He et al .,  2021 ; Huang et al .,  2021）。丝氨酸羟甲基转移酶 (SHMT) 在已知生物体（包括植物）中催化甘氨酸可逆地转化为丝氨酸（Schirch 和 Szebenyi，  2005 年））。来自植物的遗传证据表明 SHMT 参与了生物和/或非生物胁迫反应（Liu et al .,  2012 ; Moreno et al .,  2005），但它们在胚乳发育中的潜在功能仍然不清楚。

为了剖析胚乳发育的分子机制，我们从甲磺酸乙酯诱变的粳稻品种 Kitaake库中鉴定了 7 个等位基因粉状胚乳突变体flo20-1至flo20-7 （图 S1），并选择flo20-1进行深入研究。flo20-1植物表现出有缺陷的胚乳发育，如粉状胚乳外观和成熟时 1000 粒重减少约 23% 所证明的（图 1a，b）。flo20-1中淀粉和蛋白质含量降低但脂质含量增加，与野生型相比（WT；图 1b）。为了研究破坏储存物质积累的细胞学基础，我们进行了扫描电子显微镜 (SEM) 和胚乳的光学显微镜检查。与 WT 中紧密包装且边缘锋利的 SG 相比， flo20-1胚乳充满松散排列的异常淀粉颗粒（SG）（图 1c）。开花后10天（DAF）发育胚乳的考马斯蓝染色揭示了两种类型的蛋白质体（PBs）：WT中轻度染色的PBIIs和重度染色的PBIIs（图1d）。值得注意的是，PBII 形态在flo20-1中似乎是无定形的（图 1d），这与报道的具有改变的储存蛋白组成的突变体相似（Ashida等人，  2011）。来自互惠杂交植物的粉状颗粒对 flo20-1进行了表型复制（图 S2），表明FLO20直接调节胚乳发育。

基于图谱的克隆与测序相结合，发现每个flo20突变体在Os01g0874900中都有一个单核苷酸取代，导致错义、无义或剪接突变（图 1e 和 S3）。我们将基因组或 GFP 融合的Os01g0874900引入flo20-1愈伤组织，以产生互补的转基因植物。正如预测的那样，携带任一转基因的flo20-1显示出半透明的胚乳（图 1f），证实Os01g0874900代表FLO20并且 FLO20-GFP 是功能性的。FLO20编码一个预测的 64.8-kDa 蛋白质，该蛋白质含有一个 SHMT 结构域 (https://www.uniprot.org/uniprot/Q8RYY6)，命名为 SHMT4。使用抗 SHMT4 抗体的免疫印迹在所有检查的组织中检测到 SHMT4。SHMT4 水平在早期胚乳发育期间较低，在~12 DAF 达到峰值，此后下降（图 1g）。SHMT4-GFP 定位于发育胚乳中的细胞核（图 1h）。

鉴于 SHMT4 包含 SHMT 结构域，我们确定SHMT4丢失是否影响 SHMT 酶活性，并发现 64% 以上的 SHMT 活性在flo20-1中被消除（图 1i）。出乎意料的是，我们未能在体外检测到 SHMT4 活性（图 1j 和 S4）。SHMT 通常作为纯合子或杂合子复合物发挥作用（Anderson等人，  2012 年；Schirch 和 Szebenyi，  2005 年）。正如预期的那样，我们使用体外酵母双杂交体 (Y2H) 和体内的组合验证了 SHMT4 与自身和两个同源物 SHMT3 和 5 的相互作用生物分子荧光互补（BiFC）和免疫共沉淀测定（图 1k-m 和 S5）。使用高效液相色谱，我们检测到 SHMT4-GFP 沉淀物的 SHMT 酶活性（图 1n），表明 SHMT4 可能与其他蛋白质协同执行 SHMT 功能。我们确实在体外单独检测了 SHMT3 的 SHMT 活性（图 1o 和 S6），并发现SHMT3突变轻微但显着损害了 SHMT4-GST 沉淀物的 SHMT 活性（图 1p 和 S7）。一致地，SHMT3敲除突变体的 SHMT 活性降低了 35.3%，而 flo20-1 降低了 64.1% ，尽管SHMT3损失没有明显影响胚乳发育（图1i和S8）。总之，这些结果表明 SHMT4 与其他蛋白质（如 SHMT3）协同工作以执行 SHMT 活性，其中 SHMT4 可能比 SHMT3 更重要。

我们通过质谱法特异性地检测到SHMT4-GFP 沉淀物中的S-腺苷-L-蛋氨酸合成酶 SAMS2（图 1m），并通过 BiFC 和烟草中的共免疫沉淀测定证实了 SHMT4 与 SAMS2 的相互作用（图 1q）。SAMS 催化甲硫氨酸转化产生 SAM，SAM 充当甲基转移反应中的甲基化供体以及多胺和乙烯生物合成的中间体（Li et al .,  2011）。我们测量了 WT 和flo20-1胚乳中的 SAM 浓度，发现flo20-1中的 SAM 水平高出三倍（图 1r）。我们接下来对 WT 和flo20-1胚乳 DNA 进行亚硫酸氢盐测序以确定SHMT4是否损失影响全基因组DNA甲基化（表S1）。与 WT 相比，flo20-1 中的 CG、CHG 和 CHH 甲基化更高（图 1s，表 S2），表明 SHMT4 在胚乳 DNA 甲基化中可能发挥作用。差异甲基化区域（DMR）分析确定了 WT 和flo20-1之间的 25 715 个 DMR 和 15 888 个差异甲基化基因（DMG） （表 S3，数据 S1），其中包括参与储存物质生物合成、运输或积累的基因（表 S4） . 此外，淀粉和蛋白质积累所需的几种转录因子基因的表达水平在flo20-1中显着降低（图 1t）。一起，SHMT4突变导致发育中胚乳的全基因组甲基化变化，可能是通过影响SAM的产生。

最近有报道称，一个泄漏的SHMT4等位基因通过影响根和/或枝条的吸收和同化来增强镉耐受性和硒积累（Chen等人，  2020 年）。然而，在胚乳发育中推定的 SHMT4 功能仍不清楚。在这里，我们鉴定了七个胚乳发育缺陷的等位基因SHMT4突变体。贮藏蛋白质的组成极大地决定了谷类作物的营养和功能品质。我们注意到SHMT4的丢失导致了储存蛋白组成的破坏，这表明SHMT4的潜在价值育有特殊要求的水稻。总之，我们的研究提供了 SHMT 蛋白与植物胚乳发育之间的功能联系。