Double haploid (DH) technology can be used to rapidly develop homozygous lines (Geiger and Gordillo, 2009). As the fundamental component of DH technology, the traditional inducer lines were developed through a process of recurrent selection over multiple generations, a method that was inherently time-consuming. The advent of gene editing technology has facilitated the creation of inducer lines in an efficient manner (Kelliher et al., 2017; Zhong et al., 2019). However, these inducer lines lack sort markers for sorting haploid, and the introduction of genetic markers is achieved through hybridization (Yu and Birchler, 2016). Though anthocyanin marker or oil content has been primarily used for sorting haploid (Qu et al., 2021), there is a notable discrepancy in the false discrimination rate for manual or automated sorting due to the influence of anthocyanin expression. The NMR system can enhance the haploid correct discrimination rate (CDR), but the equipment is expensive. Fluorescent markers represent another type of genetic markers for the sorting of haploids; however, the fluorescent is not visible to the naked eyes (Dong et al., 2018). Consequently, the current genetic markers exhibit delayed coloration (Chen et al., 2022; Wang et al., 2023), which limits the application of DH technology.

In this study, we developed a Cas9 system for breeding inducer and sorting haploid in maize, with three advantages: (i) we innovatively employed a promoter pOsBBM1 to drive Cas9 in maize, which does not yield new edits in haploid progeny, (ii) this technique integrates the promoters pOsBBM1, DsRed2 and elements capable of targeted editing of two induction genes (ZmPLA1 + ZmDMP) at the same time into the same vector. This approach facilitates the efficient generation of inducer lines without Cas9 and with the DsRed2 marker through a single genetic transformation step. Furthermore, it improves the breeding efficiency of haploid inducer lines in different maize backgrounds and reduces cost and (iii) the DsRed2 protein exhibits specific expression in the embryo which is visible to the naked eye. This allows for the efficient sorting of haploid at various stages of seed development, which is independent of the genetic background.

We selected a promoter to drive Cas9 expression highly only in rice callus (Figure S1), while the embryo-specific promoter pZmESP was utilized to drive maize codon-optimized DsRed2 (MoDsRed2) expression, supplemented with the CaMV35S enhancer for visible to the naked eye in natural light (Xu et al., 2021). During the experimental process, we observed that the promoter pOsBBM1 activity in the callus tissue exclusively (Figure 1k). Therefore, we proposed to use this vector to generate an inducer line with red fluorescent tags through a single transformation.

Ultimately, ZmPLA1 and ZmDMP single-gene mutations were obtained in KN5585 background (Figure 1b,c). The homozygous events were named KN5585-PLA1 and KN5585-DMP, respectively. Subsequently, we investigated the agronomic traits and haploid induction rate (HIR). The results demonstrated that, with the exception of plant height, no significant differences were observed in the agronomic traits when compared to the wild type (Figure 1d, Figure S2, Table S1).

For HIR evaluation, the hybrid Zhengdan958, inbred lines Zheng58 and Chang7-2 were used as maternal materials for in vivo haploid induction. To ensure the accuracy of haploid sorting, the MoDsRed2, R1-nj and flow cytometry were employed (Figure 1d–f). The results indicated that the HIR for KN5585-PLA1 and KN5585-DMP was 5.37% and 1.71%, respectively (Figure 1h). Furthermore, we also conducted sequencing and comparative analysis of the haploid induction genes in 859 haploid progenies. No mutations were identified in the haploid progenies. Meanwhile, the expression of Cas9 was measured in immature embryos of inducer line, as well as in haploid and diploid immature embryos, and the results demonstrated that Cas9 was ineffective in dividing vigorous tissues (Figure S3).

We next compared the timing differences in haploid identification utilizing MoDsRed2 and R1-nj. Observations commenced 3 days post-pollination, followed by assessments at 4–5 days interval until grain maturity. The anthocyanins were observed to be expressed in the endosperm aleurone layer and embryo at 23–35 and 33 days after pollination, respectively (Figure S4). However, the red fluorescence can be observed under external excitation light, enabling precise identification of haploid embryos from 7 to 11 days after pollination (Figure 1i), with a CDR as high as 100%.

Moreover, at the mature grain stage, under different genetic backgrounds, the haploid CDR exceeded 99% in various materials through the MoDsRed2 marker under both excitation light and natural light conditions (Figure 1g,j). Conversely, the anthocyanin marker was impeded by genetic background variations in tropical germplasm, resulting in an accuracy of only 83.3% for materials from different backgrounds (Table S2).

To enable high throughput for sorting haploid kernels, we evaluated the efficiency of an automatic sorting equipment (Figure 1l). Initially, the equipment employed the R1-nj for sorting and underwent testing on different materials. The results demonstrated that the haploid CDR was 94.9% and a diploid correct rejection rate (CRR) was 98.2% (Table S3). Subsequently, the MoDsRed2 marker was utilized to conduct a sorting test on Zheng58, the results of which demonstrated a haploid CDR of 99.7% and a diploid CRR of 99.8% (Table S4).

In conclusion, we have devised a genetic editing system that accelerates the breeding of inducer lines, which can be utilized in a variety of crops. The naked eye-visible embryo colour marker facilitates rapid and precise sorting of haploid at various stages of seed development, which benefits early embryo identification in tissue culture doubling techniques, and enables the studies on haploid formation mechanisms (Figure 1o). The implementation of automated sorting equipment has resulted in a significant increase in sorting efficiency, from 20 000 kernels per day to 20 000 kernels per hour. The portable fluorescence excitation light equipment was employed in a darkroom screening for sorting haploid, with a haploid CDR of 99.7%. Upon completion of the hardware development, the CDR is expected to reach 100%, with an anticipated sorting efficiency over 24 000 grains per hour.

双单倍体 （DH） 技术可用于快速开发纯合系 （Geiger 和 Gordillo， 2009）。作为 DH 技术的基本组成部分，传统的诱导系是通过多代递归选择过程开发的，这种方法本身就很耗时。基因编辑技术的出现促进了以有效方式创建诱导系（Kelliher等 人，2017 年;Zhong et al.， 2019）。然而，这些诱导系缺乏用于对单倍体进行分类的排序标记，遗传标记的引入是通过杂交实现的（Yu 和 Birchler，2016）。尽管花青素标记物或油含量主要用于对单倍体进行分类 （Qu et al.， 2021），但由于花青素表达的影响，手动或自动分类的错误判别率存在显着差异。NMR 系统可以提高单倍体正确鉴别率 （CDR），但设备价格昂贵。荧光标记代表了另一种用于单倍体排序的遗传标记;然而，肉眼看不到荧光（Dong et al.， 2018）。因此，目前的遗传标记表现出延迟着色（Chen等人 ，2022 年;Wang et al.， 2023），这限制了 DH 技术的应用。

在这项研究中，我们开发了一种用于玉米育种诱导子和分选单倍体的 Cas9 系统，具有三个优点：（i） 我们创新性地使用启动子 pOsBBM1 来驱动玉米中的 Cas9，这在单倍体后代中没有产生新的编辑，（ii） 该技术整合了启动子 pOsBBM1、DsRed2 和能够靶向编辑两个诱导基因 （ZmPLA1 + ZmDMP） 同时转换为同一向量。这种方法有助于通过单个遗传转化步骤高效生成不含 Cas9 和具有 DsRed2 标记物的诱导子系。此外，它提高了不同玉米背景下单倍体诱导系的育种效率并降低了成本，并且 （iii） DsRed2 蛋白在胚胎中表现出肉眼可见的特异性表达。这允许在种子发育的各个阶段对单倍体进行有效分选，这与遗传背景无关。

我们选择了仅在水稻愈伤组织中高度驱动 Cas9 表达的启动子（图 S1），而胚胎特异性启动子 pZmESP 用于驱动玉米密码子优化的 DsRed2 （MoDsRed2） 表达，并辅以在自然光下肉眼可见的 CaMV35S 增强子（Xu等人 ，2021 年）。在实验过程中，我们观察到启动子 pOsBBM1 活性仅在愈伤组织中（图 1k）。因此，我们建议使用该载体通过单次转化生成带有红色荧光标签的诱导线。

最终，在 KN5585 背景中获得 ZmPLA1 和 ZmDMP 单基因突变 （图 1b，c）。纯合事件分别命名为 KN5585-PLA1 和 KN5585-DMP。随后，我们研究了农艺性状和单倍体诱导率 （HIR）。结果表明，与野生型相比，除株高外，农艺性状没有观察到显著差异（图 1d、图 S2、表 S1）。

对于 HIR 评价，使用杂交种 Zhengdan958、自交系 Zheng58 和 Chang7-2 作为体内单倍体诱导的母体材料。为了确保单倍体分选的准确性，采用了 MoDsRed2、R1-nj 和流式细胞术（图 1d-f）。结果表明，KN5585-PLA1 和 KN5585-DMP 的 HIR 分别为 5.37% 和 1.71%（图 1h）。此外，我们还对 859 个单倍体后代的单倍体诱导基因进行了测序和比较分析。在单倍体后代中未发现突变。同时，在诱导系的未成熟胚胎以及单倍体和二倍体未成熟胚胎中测量 Cas9 的表达，结果表明 Cas9 在分裂有活力的组织方面无效（图 S3）。

接下来，我们比较了使用 MoDsRed2 和 R1-nj 进行单倍体识别的时间差异。授粉后 3 天开始观察，然后每隔 4-5 天进行评估，直到谷物成熟。观察到花色苷分别在授粉后 23-35 天和胚胎中表达（图 S4）。然而，可以在外部激发光下观察到红色荧光，从而能够在授粉后 7 至 11 天精确识别单倍体胚胎（图 1i），CDR 高达 100%。

此外，在成熟籽粒阶段，在不同的遗传背景下，在激发光和自然光条件下，通过 MoDsRed2 标记在各种材料中单倍体 CDR 都超过 99%（图 1g，j）。相反，花青素标记受到热带种质中遗传背景变异的阻碍，导致来自不同背景的材料的准确性仅为 83.3%（表 S2）。

为了实现单倍体内核分选的高通量，我们评估了自动分选设备的效率（图 1l）。最初，该设备使用 R1-nj 进行分拣，并对不同的材料进行了测试。结果表明，单倍体 CDR 为 94.9%，二倍体正确排斥率 （CRR） 为 98.2%（表 S3）。随后，利用 MoDsRed2 标记对 Zheng58 进行分选测试，结果显示单倍体 CDR 为 99.7%，二倍体 CRR 为 99.8%（表 S4）。

总之，我们设计了一种基因编辑系统，可以加速诱导系的育种，其可用于多种作物。肉眼可见的胚胎颜色标记有助于在种子发育的各个阶段快速、精确地对单倍体进行分类，这有利于组织培养倍增技术中的早期胚胎识别，并能够研究单倍体形成机制（图 1o）。自动分拣设备的实施使分拣效率显著提高，从每天 20 000 粒增加到每小时 20 000 粒。便携式荧光激发光设备用于暗室筛选单倍体，单倍体 CDR 为 99.7%。硬件开发完成后，CDR 预计将达到 100%，预计分拣效率将超过每小时 24,000 粒。