Rapid generation of a tomato male sterility system and its feasible application in hybrid seed production

A practical approach for the rapid generation and feasible application of green hypocotyl male-sterile (GHMS) tm6 dfr lines in tomato hybrid breeding was established. Male sterility enables reduced cost and high seed purity during hybrid seed production. However, progress toward its commercial application has been slow in tomato due to the disadvantages of most natural male-sterile mutants. Here, we developed a practical method for efficient tomato hybrid seed production using a male-sterile system with visible marker, which was rapidly generated by CRISPR/Cas9-mediated gene editing. Two closely linked genes, TM6 and DFR, which were reported to be candidates of ms15 (male sterile-15) and aw (anthocyanin without) locus, respectively, were knocked out simultaneously in two elite tomato inbred lines. Mutagenesis of both genes generated green hypocotyl male-sterile (GHMS) lines. The GHMS lines exhibited male sterility across different genetic backgrounds and environmental conditions. They also showed green hypocotyl due to defective anthocyanin accumulation, which serves as a reliable visible marker for selecting male-sterile plants at the seedling stage. We further proposed a strategy for multiplying the GHMS system and verified its high efficiency in stable male sterility propagation. Moreover, elite hybrid seeds were produced using GHMS system for potential side effects evaluation, and no adverse influences were found on seed yield, seed quality as well as important agronomic traits. This study provides a practical approach for the rapid generation and feasible application of male sterility in tomato hybrid breeding.


Introduction
Male sterility is an extensively used strategy during hybrid seed production in crops, as it can enable reduced costs and high seed purity (Chen and Liu 2014;Kim and Zhang 2018). Tomato (Solanum lycopersicum) is one of the most widely cultivated and commonly consumed vegetables all over the world. Although many recessive genic malesterile mutants have been reported in tomato (Cheng et al. 2022;Wang et al. 2022), commercial application of male sterility in tomato hybrid breeding is still limited. Many tomato sterility mutants have been discovered and are divided into three categories: functional sterility, structural sterility and sporogenous sterility, as represented by positional sterile-2 (ps-2) (Atanassova 1999), stamenless (sl) (Gomez et al. 1999), and male sterile-10 (ms10) (Jeong et al. 2014), respectively. To date, several male sterility genes have been cloned. The functional sterility mutant ps-2 conferred nondehiscent anthers due to a single-nucleotide polymorphism (SNP) in the polygalacturonate gene Communicated by Esther van der Knaap.
1 3 197 Page 2 of 13 PS-2 (Gorguet, 2009). ms10 was characterized by failed pollen production and an exerted stigma owing to a deletion mutation in a basic-helix-loop-helix (bHLH) transcription factor (Jeong et al. 2014). Another spontaneous sterility mutant, ms32, presented completely impaired pollen and tapetum development and was also found to be controlled by a putative bHLH transcription factor . Tomato male sterile-15 (ms15) bears flowers with deformed anthers and exerted stigmas, for which the B-class MADS-box gene TM6 (Tomato MADS-box gene 6) was identified as a candidate . Mutation in another B-class MADS-box gene, GLO2, resulted in male sterility 7B-1 (Pucci et al. 2017).
In contrast to the many theoretical studies on male-sterile genes, rare successful application has been reported due to disadvantages of most natural male-sterile mutants. Apart from unstable sterility or unwanted side effects, there are two limitations in common. One limitation lies in the difficulty of male-sterile seed propagation. All of the discovered male sterility traits are controlled by recessive genes, which are usually spread through cross-pollinating homozygous male-sterile plants with heterozygous malefertile plants. Thus, progeny from the cross is indeed a mixture of 50% male-sterile and 50% male-fertile plants which cannot be distinguished from each other without costly genotyping before transplanting to the field (Atanassova 2004). Accordingly, the usage of linked earlydevelopment-stage morphological markers for male sterility has been proposed to address this problem. This was effectively demonstrated by linkage of ms10 and ms15 with aa (anthocyanin absent) and aw (anthocyanin without), which are both responsible for the impaired anthocyanin production and led to green hypocotyl of seedlings (Clayberg 1965;Mutschler et al. 1987;Zhang et al. 2016). The other limitation is the time-consuming and laborious efforts for transferring recessive male sterility into elite lines. To complete the whole directional transfer process, at least five generations of backcrossing are needed. Moreover, some other challenging problems, such as false selection and linkage drag, seem to be irradicable.
As a powerful genome editing technology (Aglawe et al. 2018;Gao 2021), the CRISPR/Cas9 system provides a promising tool to overcome the above limitations to efficiently create male sterility. In this study, by engineering two closely linked genes TM6 and DFR (Dihydroflavonol 4-reductase) Goldsbrough et al 1994), we rapidly generated a tomato male-sterile system with male sterility and a visible marker (green hypocotyls) which was allowed to identify male sterility at the seedling stage. We further proposed a strategy for propagating the "green hypocotyl" male-sterile plants and verified its stability and efficiency in practice.

Plant materials and growth conditions
All of the tomato inbred lines used were originally developed by our lab. Two inbred lines, TB0993 and TB0249, were subjected to CRISPR/Cas9-mediated gene editing targeting TM6 and DFR, respectively, and gave rise to tm6 dfr double mutants in the T0 generation. Crossing tm6 dfr mutants with corresponding WT plants to obtain T1 and further self-pollination of the T1 plants allowed the generation of T2 offspring. In the T2 population, genotyping for foreign fragments and TM6 as well as the DFR genes was performed to identify transgene-free homozygous tm6 dfr lines (i.e., GHMS/TB0993 and GHMS/TB0249). Homozygous GHMS plants were cross-pollinated with WT pollen to generate a hemizygous maintainer line.
Three F2 populations, SP1, SP2 and SP3, were generated following crosses GHMS/TB0993 × TB0993 (WT), GHMS/TB0249 × TB0249 (WT) and GHMS/ TB0993 × TB0748 (pink-cherry-fruited inbred line). For male sterility stability assessment, the three populations were grown in Tongzhou farm in Beijing (March-May, 2021) and Sanya farm in Hainan province (October-December, 2021), where temperature and relative humidity were automatically measured using recorders at an interval of 30 min. During the flowering stage, plant self-pollination was assisted via a "shaking plants" method. For color marker reliability assessment, seedlings of the three F2 populations at the one-leaf stage were cultured in optimal culture conditions (26 °C, 150 microeinsteins m -2 s -1 , 16 L/8 D) or extreme culture conditions (32 °C, 50 microeinsteins m -2 s -1 , 8 L/16 D) in an artificial climate chamber until the fifth leaf developed.
Two BC1F1 populations, BC1F1/TB0993 and BC1F1/ TB0249, were generated by crossing a homozygous GHMS line (green, mmaa) with a hemizygous maintainer line (purple, MmAa). The plants were subjected to linkage analysis of the "green hypocotyl" and male sterility together with the above three F2 populations (SP1, SP2 and SP3) through individual genotyping of TM6 and DFR genes.
The GHMS line (green, mmaa) was maintained through a backcross with heterozygous maintainer line (purple, MmAa). To assess the stability and efficiency of the GHMS propagation system, GHMS lines were propagated at Tongzhou farm in Beijing (March 2020 to August 2021) and Sanya farm in Hainan province (September 2019 to February 2021) for four successive generations. During every round of crossing, green and purple plants were visually sorted out at the seedling stage, and recombinants (fertile green plants) were removed based on fruit set after spontaneous self-pollination on first flower cluster or one retained lateral branch. The first generation (G1) and the fourth generation (G4) were subjected to analysis in terms of genotype composition, seed yield and seed germination.
Two GHMS-derived F1 hybrids JF101-S and JF501-S were generated following crosses GHMS/ TB0993 × TB0994 and GHMS/TB0249 × TB0244 to assess the side effects of GHMS lines on hybrid production. Moreover, the WT-derived F1 hybrids JF101 and JF501 were also generated following crosses TB0993 × TB0994 and TB0249 × TB0244 and used as the controls. For each cross, 200 female plants were used. Evaluation of side effects on agronomic traits of GHMS was performed in tunnel greenhouses at Tongzhou farm (March to August, 2021) and Sanya farm in Hainan province (September 2021-February 2022). Female parents for cross and hybrids for side effects evaluation were decapitated when the sixth inflorescence developed, and flower and fruit thinning was performed with 4-5 fruits per inflorescence.

CRISPR/Cas9-mediated mutagenesis
The knockout vector was constructed using the binary vector PTX041 as described previously (Deng et al. 2018). Two synthesized sgRNAs targeting the first exon of TM6 (Solyc02g084630) and DFR (Solyc02g085020) were cloned into the PTX041 vector and then introduced into Agrobacterium tumefaciens EHA105 for subsequent transformation into tomato inbred lines TB0993 and TB0249, respectively. PCR was performed for generated T0 plants using primers designed based on the flanking sequence of on-targets as well as predicted off-targets, and the amplicon(s) from each line was then cloned into a high-copy vector for sequencing (Tianyi Huiyuan Bioscience & Technology, Inc. (Beijing, China)).
The primers and sgRNA sequences used are listed in Table S1.

Gene expression analysis
Gene expression analysis was conducted in GHMS/TB0993 and GHMS/TB0249 via quantitative real-time polymerase chain reaction (qRT-PCR) using the corresponding WT plants as controls. Transcripts were examined in hypocotyls and cotyledons at the seedling stage for DFR and in flower buds at the anthesis stage for TM6. Gene expression levels were normalized against those of tomato ACTIN2 calculated using the ΔΔCt method. All experiments were performed with four biological replicates.
The primers used are listed in Table S1.

Pollen vitality analysis
Pollen vitality was determined using the FDA staining method (Li et al., 2011) combined with the germination in vitro method (Song and Tachibana, 1999). For the FDA staining method, fresh stamens taken from flowers at the anthesis stage were squashed to release pollen into 20 µL FDA solution (2 µg ml −1 ) on a glass slide. A Zeiss AX10 fluorescence microscope was used for microscopic observations under fluorescent light conditions. For the germination in vitro method, fresh pollen harvested at the anthesis stage was spread on germination medium consisting of 1% agar, 13% sucrose and 15 mg/l boric acid and then observed using a Zeiss AX10 microscope in a bright field. All experiments were performed with three biological replicates for CRISPR/Cas9-derived mutants (tm6 dfr/TB0993 and tm6 dfr/TB0249) and with thirty biological replicates for three alternative genotypes of TM6 from SP1, SP2 and SP3 populations.

Anthocyanin content determination
The anthocyanin content was determined as described previously (Sun et al. 2020). Approximately 100 mg fresh weight (FW) of collected hypocotyl or young leaf tissues was ground to powder in liquid nitrogen and then put in extraction buffer (1% HCl, 18% 1-propanol and 81% water) overnight incubation at 4 °C. The absorbance values of the collected supernatant after centrifugation were determined at wavelengths A535 and A650, and the anthocyanin content was calculated as (A535-A650)/mg FW. All experiments were performed with three biological replicates.

Linkage analysis
Individuals from two BC1F1 populations (BC1F1/TB0993 and BC1F1/TB0249) were genotyped for TM6 and DFR genes, and the recombination fraction was calculated as the percentage of recombinants in all of the plants. Before genotyping, green plants were visually selected, and the selection efficiency of morphological markers for male sterility was expressed as the percentage of mm seedlings in all of the seedlings.

Assessment of side effects on seed yield, seed quality and agronomic traits
At the flowering stage, flowers at anthesis in the second inflorescences were tagged. Seeds were collected from fruits at the red stage and were subjected to yield and quality analysis. Seed yield was calculated as the average seed number per fruit, which was determined by collecting seeds from 10 fruits of the second inflorescence with 5 replicates. Seed weight was calculated as the average weight per 1000 seeds represented by 5 replicates. Seed germination was determined after 3 days of cultivation in an artificial climate chamber at 30 °C. For each assay, 200 seeds were used with three replicates. Seed purity was determined using a male parent-specific molecular marker Ty1 (Table S1) for four F1 hybrids (JF101, JF101-S, JF501 and JF501-S) or using the color marker (green hypocotyls) for JF101-S and JF501-S. For seed germination and seed purity assay, 384 seeds were used and three biological replicates were performed.
Hybrid fruits were collected at the breaker stage and weighed. Their longitudinal diameters (LD) and transverse diameters (TD) were determined using a Vernier caliper, and fruit shape index was calculated as the ratio of TD to LD. At the B + 7 stage (7 days after the breaker stage), fruits were harvested to measure fruit firmness, soluble solid content (SSC), lycopene content and β-carotene content. Flesh firmness was determined using a hardness tester HPE II Fff (Bareiss, Germany), with the results recorded in N. Then, each fruit was cut in half. One half was used to determine the SSC of the juice using a refractometer, with the results recorded as °Brix, and the other half was subjected to lycopene and β-carotene content measurements by HPLC with the method described previously (Zhou et al. 2022). All experiments above were performed with fifteen biological replicates.

Statistical analyses
All data represent the mean value ± standard deviation (SD) of biological replicates. Statistical significance was determined using Student's t test at the 0.05 (*) level.

Rapid generation of green hypocotyl male-sterile lines by the CRISPR/Cas9 system
Most tomato inbred lines develop purple hypocotyls due to anthocyanin accumulation, thus green hypocotyl conferred by aw could serve as an efficient seedling morphological marker for identifying male sterility plants of ms15, which is tightly linked to the aw. TM6 and DFR (Dihydroflavonol 4-reductase) were reported to be the underlying genes for the ms15 and aw locus, respectively Goldsbrough et al. 1994), although direct genetic evidence are absent. We therefore proposed to employ a modified CRISPR/Cas9 gene-editing system to edit TM6 and DFR simultaneously in two elite inbred lines to generate "green hypocotyl" male-sterile lines and to evaluate their usage in our breeding program.
For this purpose, two synthetic gRNAs targeting the first exon of each gene were designed and cloned into the pTX041 vector (Fig. 1), as described in our previous study (Deng et al. 2018). The resulting construct was transformed Fig. 1 Knocking out TM6 and DFR in tomato using the CRISPR/ Cas9 system. A Schematic diagram of chromosomal location of TM6 and DFR. The male fertility gene TM6 is located ~ 0.34 Mb upstream of DFR on Chr02, which is responsible for anthocyanin accumulation. B and C PCR and sequence-based genotyping of T0 plants. Two gRNAs were designed to target the first exon (E1) of TM6 (B) and DFR (C), and the target sites are indicated by red arrows. Mutations in T0 regeneration plants in the background of elite inbred lines TB0993 and TB0249 are shown below the target sequences, respectively, with deletions and insertions indicated by red dashes and blue letters. Number of base pairs (bp) deleted (-) is indicated on the right-hand side. The letters on the left-hand side represent different genotypes. Ho, homozygote; Bi, biallelic; He, heterozygote. D and E Summary of mutagenesis of TM6 (B) and DFR (C) in the two elite inbred lines TB0993 and TB0249 (color figure online) into one pink-fruited tomato inbred line TB0993 and one red-fruited tomato inbred line TB0249 through Agrobacterium-mediated transformation.
After the recovery of transgenic plants, 10 T0 plants for TB0993 and 13 T0 plants for TB0249 were analyzed for mutations detection within the target regions ( Fig. 1B-E). Genome-editing event analysis by sequencing indicated that seven TB0993 transgenic plants carried homozygous or biallelic double mutations with small indels (≤ 12 bp) at both target sites. Likewise, among 13 TB0249 T0 plants, six plants carried homozygous or biallelic double mutations. Moreover, we sequenced the potential off-target sites predicted by Cas-OF Finder ) in all identified mutated T0 plants, and no off-target mutations were found at these sites. These results highlight the high editing efficiency and specificity of the CRISPR/Cas9 system used in this study.
After removing the transgene insertions by backcrossing, we selected two homozygous double-mutated lines tm6 dfr/TB0993 #1 and tm6 dfr/TB0249 #1 for phenotypic analysis. Mutations in TM6 and DFR in these lines were predicted to introduce premature stop codons and cause undetectable expression of each gene probably owing to nonsense-mediated mRNA decay (Fig. S1). There was no difference in growth and development between wild-type (WT) and tm6 dfr plants until the flowering stage. At flowering, tm6 dfr lines were found to produce obviously twisted stamens and failed to set fruit ( Fig. 2A, D and Fig. S2A, D). The exhibiting male sterility was further supported by the fluorescein diacetate (FDA) staining assay and pollen germination assay, which did not detect viable pollen in the tm6 dfr flowers (Fig. 2B, C and Fig. S2B, C). In contrast, tm6 dfr plants produce normal pistils, and they set comparable fruits and seeds when pollinated with WT pollen (Fig. 2E, F and Fig. S2E, F). These results demonstrated that the tm6 dfr plants are fully male sterile and have application prospects for hybrid seed production in tomato.
Although tm6 dfr plants did not have any aberrant phenotypes during seedling growth and development as expected, we observed obvious defects in anthocyanin accumulation in the tm6 dfr plants when compared with the WT plants ( Fig. 2 G, H and Fig. S2G, H). The differences in anthocyanin accumulation, as represented by seedling color (i.e., purple in WT versus green in the mutants), could be seen as early as 2-3 days after germination. The "green hypocotyls" in tm6 dfr plants were stable during the whole seedling stage. In addition, various color differences between the WT and tm6 dfr plants were also observed in both vegetative and floral tissues throughout the entire life cycle (Fig. 2I and Fig. S2I). These results confirmed previous findings that DFR functions in anthocyanin accumulation in tomato and indicated that the "green hypocotyls" in the tm6 dfr plants can serve as a visible marker. We therefore renamed the tm6 dfr plants GHMS (green hypocotyl male sterile) for further analysis.

Phenotypic stability of GHMS across genetic backgrounds and environmental conditions
To evaluate the application potential of the GHMS lines in tomato hybrid breeding, we sought to examine the phenotypic stability across various genetic backgrounds and environmental conditions. Three GHMS-derived F2 segregating populations were generated: (i) SP1, derived from a cross between GHMS/TB0993 and its parental line TB0993, (ii) SP2, derived from a cross between GHMS/TB0249 and its parental line TB0249 and (iii) SP3, derived from a cross between GHMS/TB0993 and a pink-cherry-fruited inbred line TB0748.
The three populations were grown for male sterility assessment on Tongzhou farm in Beijing (March-May, 2021) and Sanya farm in Hainan province (October-December, 2021), which represented contrasting environmental conditions, as shown in Fig. 3A. The male fertility of flowers from the first three inflorescences was examined in terms of pollen vitality and fruit set. Under both growth conditions, all homozygous tm6 (mm) plants from the three F2 populations, as genotyped by mutation-specific molecular markers, displayed male sterility; they failed to produce viable pollen grains and set seeded fruits, in contrast to normally developed pollens and seeded fruits of heterozygous tm6 (Mm) or homozygous Tm6 (MM) plants (Fig. 3A). These results demonstrated that the male sterility of GHMS is stable across genetic backgrounds and environmental conditions. Anthocyanin production was reported to be affected by genetic, environmental and nutritional cues (Cominelli et al. 2008;Butelli et al. 2012;Das et al. 2012;Hodges and Nozzolillo 1995). Thus, we carefully analyzed the correlation between the dfr mutations and anthocyanin accumulation (seedling color) in the three F2 populations under different growth conditions ( Fig. 3B and Fig. S3). When grown under optimal conditions (26 °C, 150 microincisions m -2 s -1 , 16 L/8 D), all homozygous dfr (aa) seedlings identified from the three populations displayed anthocyanin-deficient green color, with conspicuous color differences from the heterozygous dfr (Aa) seedlings. Although a wide range of variation in anthocyanin accumulation was observed in Aa individuals, apparent discrepancies between purple and green plants allow them to be easily distinguished as aa plants (Fig. S4A). These results indicate that, under optimal conditions, seedling color can serve as a reliable marker to select aa plants. To explore environmental effects, F2 seedlings were subjected to extreme conditions with weak light and high temperature (32 °C, 50 microincisions m -2 s -1 , 8 L/16 D). Under this growth condition, anthocyanin accumulation was obviously reduced in Aa plants, which attenuated color differences between aa and Aa plants and led to an ~ 10%-15% decrease of the efficiency for selecting aa plants (Fig. S4B). This result implied that adverse conditions should be avoided during the seedling stage in practice to maintain highly effective screening of aa plants by color markers.

Linkage analysis of the "green hypocotyl" and male sterility
We next discussed the use of the GHMS line in tomato hybrid breeding. The classic approach for the use of recessive genic male sterility is to cross-pollinate homozygous male-sterile plants with heterozygous male-fertile plants. Progeny from the cross is a mixture of segregating male sterile (50%) and fertile (50%) plants. The use of linked genetic markers expressed at the seedling stage, such as the "green hypocotyls" of our GHMS line, allows for the visual identification of male-sterile plants from the mixture population before transplanting to the field. The selection efficiency for male sterility by "green hypocotyls" largely depends on the recombination frequency between the TM6 and DFR genes. Thus, we used two GHMS-derived BC1F1 populations, BC1F1-TB0993 and BC1F1-TB0249 (Table 1), to perform the linkage analysis of TM6 and DFR. A total of 8 recombinants were detected among 384 BC1F1-TB0993 individuals, while 7 recombinants were detected among 384 BC1F1-TB0249 individuals. The recombination fractions in the two populations were 2.08 and 1.82%, respectively. Accordingly, the selection efficiency for male sterility by "green hypocotyls" in the two populations was calculated to be approximately 97.50 and 97.87%, respectively. These analyses suggested that the "green hypocotyl" of our GHMS provides a reliable visible marker for the selection of male sterility.

Propagation strategy of the GHMS seeds
We next proposed a strategy to propagate and identify GHMS plants for hybrid seed production without costly genotyping (Fig. 4A). Homozygous GHMS plants (green, mmaa) were cross-pollinated with WT pollen to generate a hemizygous maintainer line (purple, MmAa). The cross between mmaa and MmAa produced the generation 1 (G1) progeny population (hereafter named "the propagation population"), which was predicted to segregate half green (consisting of 98% mmaa and 2% Mmaa) and half purple plants (consisting of 98% MmAa and 2% mmAa). The two groups of plants could be easily selected by seedling color and were grown separately. At the flowering stage, the first flower clusters of the green plants were allowed to be selfpollinated, and the male-fertile plants with self-pollinating fruits (the 2% Mmaa) were removed. The remaining green plants (98% mmaa) were then cross-pollinated with pollen from the purple plants (please note that only the MmAa purple plants can produce viable pollen), which gave rise to the new generation of the propagation population. Theoretically, the above-mentioned procedures can be operated iteratively, and the genetic composition of the propagation population should be similar over generations. To assess the stability and efficiency of this propagation strategy, we performed the propagation procedure for 4 successive generations and genotyped the individuals in G1 and G4. As expected, the composition of genotypes in G4 resembled that in G1 (Fig. 4B). This result indicated that the GHMS seeds can be stably and effectively propagated using our proposed strategy, and this strategy can be used to produce hybrid seeds at a commercial scale.

Side effects analysis of using GHMS lines in hybrid production
To further evaluate the application of our GHMS lines in hybrid tomato breeding, we used the GHMS lines (GHMS/ TB0993 and GHMS/TB0249) to produce hybrid seeds of two elite F1 varieties, Jingfan101 (JF101, TB0993 × TB0994) and Jingfan501 (JF501, TB0249 × TB0244), which were developed by our group. The GHMS/TB0993 and GHMS/ TB0249 lines were propagated, and seedlings with "green hypocotyls" were selected for subsequent hybrid seed production. At the flowering stage, the male-fertile plants with self-pollinating fruits were removed. The remaining plants were then cross-pollinated with the male parent inbred lines (i.e., TB0994 for JF101 and TB0244 for JF501) to generate the F1 hybrid seeds. To compare the performances with those of the original hybrids, the GHMS-derived F1 hybrids were designated JF101-S and JF501-S, respectively.
Seed yield and quality are critical for hybrid seeds. As shown in Fig. 5 and Fig. S5, the seed yield, weight and germination rate were comparable in the WT-and GHMSderived F1 hybrids. However, as expected, the GHMSderived F1 hybrids had obviously higher seed purity than the WT-derived F1 hybrids ( Fig. 5A and Fig. S5A). Other important agronomic traits were also tested (Fig. 5B, C and Fig. S5B, C). The fruit ripening time of JF101-S and JF501-S was similar to that of WT-derived F1 hybrids. In addition, no significant difference was observed in single fruit weight, yield per plant, as well as fruit shape. Fruit quality was Fig. 2 Phenotyping for male sterility and anthocyanin deficiency of CRISPR/Cas9-derived mutants. A tm6 dfr/TB0993 produced normal pistil but carpelloid stamen. B and C No viable pollen grains of tm6 dfr/TB0993 were detected by the FDA assay (B) and pollen germination assay (C). D tm6 dfr/TB0993 failed to set seeded fruits indicating its male sterility. E and F tm6 dfr/TB0993 gave rise to normal fruits successfully when pollinated with WT pollen, suggesting its female fertility. G and H tm6 dfr/TB0993 developed green hypocotyls and cotyledons owing to anthocyanin deficiency during the seedling stage, which could be served as a reliable morphological marker. I Obvious color differences between tm6 dfr/TB0993 and WT plants were observed in young leaves, petioles, peduncles, sepals and axillary buds during the adult stage (color figure online) ◂ Fig. 3 Phenotypic stability of the GHMS system across genetic backgrounds and environmental conditions. A Stability analysis of the male sterility of GHMS lines. Three GHMS-derived F2 populations, SP1, SP2 and SP3, were generated following crosses GHMS/ TB0993 × TB0993 (WT), GHMS/TB0249 × TB0249 (WT) and GHMS/TB0993 × TB0748, which represented multiple genetic backgrounds. F2 populations were grown in Tongzhou farm in Beijing (March to May, 2021) and Sanya farm in Hainan province (October to December, 2021), which represented contrasting environmental conditions for stability assessment of male sterility. Fruit-set rate and pollen vitality frequency of the first three flower clusters from three groups of genotyped plants (mm, Mm, MM) were examined. Pollen vitality frequency was determined through the FDA method. Fruit set was determined upon self-pollination assisted via a "shaking plants" method. The data represent the mean ± SD from thirty biological replicates. ND: not detected. B Reliability analysis of color marker of GHMS system. SP3 seedlings were cultured under optimal (26 °C, 150 microeinsteins m -2 s -1 , 16 L/8 D) or extreme environmental conditions (32 °C, 50 microeinsteins m -2 s -1 , 8 L/16 D) for visual color comparison. And anthocyanin contents were determined, which were expressed as (A535-A650)/g. −1 FW. Values are the means of three biological replicates ± standard deviation (SD). Asterisks denote significance by Student's t test compared with data between aa and Aa genotypes (P* < 0.05) (color figure online) further examined. The main quality indices of ripe fruits, including fruit firmness, levels of lycopene, β-carotene and total soluble solids content (°Brix), did not show any significant difference between the GHMS-and WT-derived F1 hybrids. These results suggest that GHMS-produced hybrids do not have any negative effects on variety performance and can meet the quality requirements of commercial tomato production. Thus, the GHMS system can be well applied to the commercial production of hybrid seeds.

Discussion
As one of the most consumed vegetables all over the world, tomato provides a classical model system for studying crop domestication, molecular breeding and fruit biology (Du et al. 2017;Xia et al. 2021). Most commercial varieties of tomato are F1 hybrids due to their superior over parents in terms of yield, disease resistance and environmental fitness (Perez-Prat and van Lookeren Campagne, 2002). During the process of hybrid seed production, artificial emasculation results in increased labor expenses. Utilizing male sterility to produce hybrids is an efficient approach to reduce the cost and ensure high varietal purity (Chen and Liu 2014;Kim and Zhang 2018;Perez-Prat and van Lookeren Campagne, 2002). Many spontaneous male-sterile mutants have been  reported in tomato (Wang et al. 2022;Jeong et al. 2014;Liu et al. 2019;Cao et al. 2019;Pucci et al. 2017). However, some inherent disadvantages prevent their application in hybrid seed production. Apart from unstable sterility or unwanted side effects, there are two limitations in common.
One limitation is the time-consuming and laborious efforts for transferring recessive male sterility into an elite inbred line. To complete the whole directional transfer process, at least five generations of backcrossing are needed. Moreover, some other challenging problems, such as false selection and linkage drag, seem to be irradicable. CRISPR/ Cas9 genome editing technology has provided alternative opportunities for breeders to introduce male sterility into elite varieties within one generation at low costs and with no linkage drag. Until now, several CRISPR/Cas9-derived male-sterile mutants have been successfully generated (Liu et al. 2021;Tiwari et al. 2023). For example, we previously generated novel male-sterile tomato lines by CRISPR/Cas9mediated mutagenesis of the stamen-specific gene SlSTR1 in several elite inbred backgrounds . Similarly, rapid generation of male sterility in elite inbred lines was also achieved in this study by knocking out the TM6 which was reported to be a candidate gene for ms15 . The tm6 lines generated in the present study developed deformed stamens, with no viable pollen observed at the anthesis stage. These phenotypes mimic ms15, indicating that TM6 indeed underlies ms15 locus. Moreover, we found that the male sterility caused by the tm6 mutation was very stable under various genetic backgrounds and environmental conditions, highlighting its valuable and extensive application potential in tomato hybrid seed production.
The other limitation for applying genic male sterility to hybrid seed production lies in the difficulty of propagating pure male-sterile seeds at a large scale. Usually, homozygous male-sterile (ms/ms) seeds are propagated by crosspollinating ms/ms plants with heterozygous male-fertile (Ms/ms) plants. The resulting F1 progeny is mixture of 50% ms/ms plants and 50% Ms/ms plants. To produce F1 hybrid seeds, all of the Ms/ms plants must be removed based on fertility identification during flowering stage, which is labor-intensive and resource-consuming. Two strategies have been proposed to overcome this difficulty. The first strategy is to create transgenic maintainers by transforming the fertility restoration gene linked with a pollen-lethality gene or a seed-color gene into the male-sterile plant (Perez-Prat and van Lookeren Campagne, 2002). When cross-pollinating the male-sterile line (ms/ms, − / −) with hemizygous maintainer (Ms/ms, Maintainer/ −) pollens, it would generate only the male-sterile seeds (in the case of pollen-lethality maintainer) or 50% male-sterile seeds and 50% color maintainer seeds which can be visually sorted out according to seedcolor differences. Using such kinds of transgenic maintainers to multiply male sterility have been recently reported in maize and rice (Kim and Zhang 2018;Chang et al. 2016;Wu et al. 2016;Zhang et al. 2018;Wan et al. 2019;Cai et al. 2022). Using a similar strategy, we recently developed a transgenic maintainer with a seedling color marker to propagate our CRISPR/Cas9-generated tomato male-sterile line . The second strategy to obtain a large quantity of recessive genic male-sterile plants is the use of closely linked morphological markers expressed in seedling that can help remove fertile plants before transplantation. This approach was effectively demonstrated by linking ms10 with aa or linking ms15 with aw in tomato (Clayberg 1965;Mutschler et al. 1987;Zhang et al. 2016). In this study, the candidate gene DFR for aw locus (Goldsbrough et al. 1994) was knocked out using CRISPR/Cas9 system and it led to no anthocyanin accumulation in hypocotyls (green hypocotyls), which serves as a visible marker for selecting malesterile plants of tm6 at the seedling stage at high accuracy (~ 98%). On the other hand, this provides genetic evidence to support the idea that DFR underlies aw locus. We must note that, using a similar approach, a recent study generated ms10 male-sterile lines with two different seedling markers (Liu et al. 2021). However, our CRISPR/Cas9-derived GHMS system is more efficient in male sterility selection as the recombination fraction between the marker gene and male sterility gene in GHMS system (about 2%) is obviously lower than that (about 5%) in ms10 system. Moreover, in that study, the authors did not provide convincing data in terms of the propagation method and application of these sterile males in hybrid seed production. In contrast, we proposed a strategy to propagate green hypocotyl male-sterile seeds without costly genotyping and verified its feasible application. Our data demonstrated that the proposed propagation method is stable and efficient, and the developed GHMS lines are suitable for tomato hybrid seed production. Especially, it had no visible negative influence on key agronomic traits of elite hybrids. However, although the "green hypocotyls" can help select the male-sterile plants at the seedling stage, the selection efficiency of "male sterility" by the "green hypocotyl" in our GHMS system was not 100% due to the incomplete linkage between the TM6 and DFR (Table 1).

Fig. 5
Performance of the GHMS-and WT-derived elite F1 hybrids. A Seed yield and seed quality of the GHMS-derived F1 hybrid JF101-S. Seed number per fruit, 1000-seed weight and seed germination of JF101-S were comparable to those of the WT-derived hybrid JF101. But seed purity of JF101-S was obviously higher. The data represent the mean ± SD from different biological replicates. Asterisks denote significance by Student's t test. P* < 0.05. B Performance of important agronomic traits of GHMS-derived F1 hybrid JF101-S. Important agronomic traits of JF101-S were all equivalent to those of WT-derived F1 hybrid JF101. The data represent the mean ± SD from different biological replicates. Statistical significance was determined by Student's t test at P* < 0.05. C Typical pictures of the GHMS-derived hybrid JF101-S and WT-derived hybrid JF101 during the ripening stage ◂ Further searches for co-segregated genetic markers of male sterility are needed to improve this system.
The stigma protrusion in tomato is important for crossing as it allows the avoidance of emasculation. Exerted stigmas in tomato ms15 (tm6) mutants were reported previously ). However, this trait seemed to be unstable in both of tm6/TB0993 and tm6/TB0249, and it was only observed occasionally under optimal environmental conditions. This suggests that TM6 disruption gave rise to exerted stigma in a genetic background-dependent way. Nevertheless, the tm6 mutants make the crossing become easier as it is not necessary to remove the stamen cone completely any more but just need to expose the stigma.
Collectively, we developed a green hypocotyl male-sterile system by using the advantages of powerful CRISPR/Cas9 technology. More importantly, we further present a practical proposal for feasible application of this system to tomato hybrid production. With stable male sterility across different genetic backgrounds and environmental conditions and no adverse side effects on seed quality and agronomic traits, widespread applications of this system can be anticipated. The practical study of our work could be extended to other vegetable crops for rapid male sterility creation and its feasible application in hybrid seed production.