Developing a robust multiplex CRISPR/Cas12i3‐5M system for trait stacking in soybean

The high-efficiency multiplex gene editing technology based CRISPR-Cas12i3-5M is capable of simultaneously editing 13 target sites in soybean, and was used to generate germplasm with high oleic acid content and no beany flavor. The CRISPR-Cas12 enzyme system, Cas12i3, offers a promising tool for crop genome editing due to its relatively flexible TTN PAM requirement and compact protein size (Lv et al., 2024; Lin et al., 2025). Its engineered high-activity variant, Cas12i3-5M, provides a powerful platform for gene editing (Duan et al., 2024; Wang et al., 2025a, 2025b). However, achieving efficient multiplex editing in crops remains challenging, largely due to the difficulty in precisely processing multiple crRNA arrays (Lv et al., 2024; Xie et al., 2024). Endogenous processing elements, such as tRNAs and the HDV ribozyme, have proven effective for precise RNA processing and enhancing the multiplex editing capacity of Cas12i3-5M in rice and wheat (Wang et al., 2025a). Soybean (Glycine max L.) is a globally important crop, providing oil and protein to human diets and animal feeds. Although Cas12i3-5M has proven effective for editing in soybean with a single crRNA, its application for efficient multiplex editing remains constrained (Duan et al., 2024; Xie et al., 2024). To address this gap, we developed an integrated triple-component expression system specifically for soybean, designated “pSoC12i-V1” (Figure 1A). In this system, the expression of Cas12i3-5M is driven by the strong endogenous promoter GmM4pro (Class II), along with an optimized Kozak sequence, to ensure robust expression across various tissues in soybean plants (Xie et al., 2023). To augment transcript stability and abundance, dual 3′ UTR insulators were incorporated. A FTO module was also introduced, which increases chromatin accessibility and improves editing outcomes (Bai et al., 2024). For precise crRNA maturation, a self-processing HDV ribozyme system was implemented (Figure 1A). Given the limited pre-crRNA processing capacity of Cas12i3, which necessitates specific tandem architectures (e.g., TUTU) for multiplex editing (Lv et al., 2024), we integrated HH- (pSoC12i-V2) and tRNA-based (pSoC12i-V3) dual-ribozyme systems to enhance the precision and efficiency of individual crRNA release (Figure 1A; Table S2). Optimization of a Cas12i3-5M-mediated multiplex gene editing system in soybean and generation of an ultra-low beany flavor variant (A) Schematic of pSoC12i-V1, pSoC12i-V2, and pSoC12i-V3. (B) Assessment of editing efficiencies for Cas-SF01, pSoC12i-V1, pSoC12i-V2, and pSoC12i-V3 at a single-crRNA vector in soybean hairy roots. (C) Editing efficiencies of 2× crRNA for Cas-SF01, pSoC12i-V1, pSoC12i-V2, and pSoC12i-V3 systems compared in hairy roots. (D) 4× crRNA editing efficiency using the pSoC12i-V3 system in hairy roots. (E) 8× crRNA editing efficiency using the pSoC12i-V3 system in hairy roots. Data from (B) to (E) (means ± SD) are from 16 fluorescent-positive roots per construct across three biological replicates. Columns sharing the same letter are not significantly different. (F) Mutation profile induced by pSoC12i-V3 in hairy root assays. Each data point represents the editing ratio (determined by Hi-TOM sequencing) from a single, randomly selected positive root (means ± SD, n = 8). (G) Workflow for generating ultra-low beany flavor soybeans using pSoC12i-V3, with schematic diagrams of target sites within respective candidate genes. (H) Genotyping of homozygous DS89-5m lines with quintuple gene edits. (I) Phenotypes of DS89-WT and DS89-5m plants grown in the field (Bars = 10 cm, upper), and seed size comparison (Bars = 1 cm, down). (J) Comparison of agronomic traits between the DS89-5m mutant and DS89-WT plants at harvest, including the plant height, node number, pods per plant, yield per plant, and 100-seed weight, n = 15 (P-values were determined by the Welch's t-test). (K) Comparison of fatty acid (FA) content in mature seeds of the DS89-5m mutant and DS89-WT, with a focus on oleic and linoleic acids, n = 3 (P-values were determined by the Welch's t-test). (L) Comparative analysis of beany flavor-related volatile compounds in mature seeds of DS89-5m, GmLox-28, and DS89-WT, including hexanal, hexanol, and pentanal, n = 3 (P-values were determined by the one-way ANOVA analysis). To benchmark the performance of pSoC12i-V1/pSoC12i-V2/pSoC12i-V3, we compared their editing efficiency with the previously reported soybean editing system, Cas-SF01, in single crRNA and double crRNA editing (Duan et al., 2024; Figure 1B, C; Table S1). Hi-TOM sequencing at six endogenous target sites revealed that the editing efficiency of pSoC12i-V1 ranged from 68.75% ± 0% to 91.67% ± 14.43%, representing a consistent and moderate improvement over Cas-SF01 (50.00% ± 6.25%–89.58% ± 13.01%) (Figure 1B). Moreover, the parallel comparison demonstrated a significant editing efficiency advantage of pSoC12i-V3 over pSoC12i-V2, as well as over pSoC12i-V1 (Figure 1B). We next compared the multiplex editing efficiency by assembling tandem arrays of 2× crRNA, and compared it with Cas-SF01 and pSoC12i-V1 as controls. To minimize sequence redundancy and avoid element duplication, we developed a compact processing strategy utilizing a single terminal HDV ribozyme. Across all target loci, the pSoC12i-V3 architecture demonstrated superior editing efficiency, achieving 55.28%–75.42%. This performance exceeded that of pSoC12i-V2 (38.33% ± 1.44%–53.33% ± 5.77%), pSoC12i-V1 (8.33% ± 7.22%–31.25% ± 6.25%), and Cas-SF01 (16.67% ± 7.22%–31.33% ± 6.25%) (Figure 1C). Together, these results point to the (tRNA-crRNA)n-HDV configuration in pSoC12i-V3 as a strategy for achieving more robust multiplex editing efficiency. Given the high performance of pSoC12i-V3, we next tested its editing capacity with 3× crRNA, 4× crRNA, and 8× crRNA arrays, respectively. In a 3× crRNA test, pSoC12i-V3 exhibited effective editing (> 55.42%) at all sites except crBADH-1 (Figures S1, S2). Relocating crBADH-1 in redesigned arrays (Arrays 2 and 3) did not improve its efficiency (2.08% ± 3.61%), while other targets remained largely unaffected (Figure S2). This suggests that the low editing at crBADH-1 is due to its poor inherent editability. In the 4× crRNA array, the editing efficiencies of each gene ranged from 59.72% ± 8.68% to 68.06% ± 6.36% across replicates, with simultaneous editing observed at all six target sites in approximately 50%–62.5% of transformed hairy roots (Figure 1D). In the 8× crRNA array, which efficiently edited all 13 target sites in hairy roots, with editing efficiencies ranging from 30.65% ± 5.38% to 85.42% ± 3.61% (Figure 1E). Analysis of the editing types induced by the 4× crRNA construct in hairy roots revealed that pSoC12i-V3 predominantly generates deletions (indels) of 6–15 bp across the targeted sites (Figure 1F). No off-target mutations were detected by Sanger sequencing of predicted sites, confirming high specificity (Figure S3; Table S3). These data demonstrate that pSoC12i-V3 is a robust, stable, and scalable system for multiplex editing in soybean. DongSheng 89 (DS89) is a high-yielding soybean variety developed in China (Fang et al., 2026). Here, we aimed to improve its seed quality by using pSoC12i-V3 to edit five genes—GmLOX1/2/3 (to eliminate beany flavor), and GmFAD2-1A/1B (to increase oleic acid). We designed a 4× crRNA array targeting all five genes and tested it with Cas-SF01, pSoC12i-V1, and pSoC12i-V3 (Figure 1G). In hairy root assays, pSoC12i-V3 achieved substantially higher editing efficiencies (27.5% ± 9.01%–68.97% ± 5.92%) compared to Cas-SF01 and pSoC12i-V1 (0%–10.42% ± 3.61%) (Figure S4). Following Agrobacterium‑mediated stable transformation into DS89, we obtained 42 transgenic positive T0 plants from 200 explants screened. Sanger sequencing of these plants revealed editing efficiencies ranging from 19.04% to 54.76% across all target sites. Notably, eight out of the 42 T0 plants (19.04%) were heterozygous at all five target loci (Figure S5). In the T1 generation of pSoC12i-V3, we screened the progenies of one line and identified quintuple homozygous mutations (designated as DS89-5m) without T-DNA transgene and off-target editing (Figures 1H, S6). In the T2 generation, DS89-5m plants exhibited morphology and seed size comparable to the wild type, with no notable phenotypic alterations (Figure 1I). Analysis of yield-related traits, including plant height (65–70 cm), node number (13–14), pods per plant (45–55), yield per plant (18–20 g), and hundred-seed weight (14.61–15.56 g), confirmed similarity to the wild type (Figure 1J). To evaluate the seed quality traits, we profiled the fatty acid composition of DS89-5m seeds. As anticipated, the mutant line displayed a substantial increase in oleic acid, which rose from 22.12% in DS89 to 76.67%, while linoleic acid content decreased dramatically to 4.42% (Figure 1K). Using headspace solid-phase microextraction coupled with gas chromatography–mass spectrometry (HS-SPME–GC-MS), we quantified key beany flavor compounds (hexanal, hexanol, pentanol) in DS89-5m, wild-type DS89, and the lipoxygenase triple mutant GmLox-28 (Wang et al., 2020). Compared to wild type, GmLox-28 seeds showed a 45.85%–48.64% reduction in these compounds (Figure 1L). Remarkably, DS89-5m seeds completely eliminated them, with levels below the detection threshold. This may be explained by the high oleic and low linoleic acid content, which reduces precursors for unsaturated fatty acid auto-oxidation, thereby leading to reduced beany flavor. Thus, DS89-5m combines ultra-low beany flavor with elevated oleic acid. In summary, we have developed a robust multiplex gene editing system for soybean, designated pSoC12i-V3, to facilitate trait stacking and germplasm innovation. This system consists optimized soybean Cas12i3-5M expression cassette and utilizes a compact (tRNA-crRNA)n-HDV architecture to drive efficient crRNA array expression. By simultaneously knocking out several lipoxygenase and oleic acid synthesis-related genes, we generated soybean germplasm with ultra-low beany flavor and high oleic acid content in the elite cultivar DS89. This work provides an efficient tool for soybean precision breeding with intellectual property independence and technical flexibility. This work was funded by the National Key Research and Development Program of China (2023YFF1000203) to M.L. and Guangzhou Science Grant (2024A03J0010) awarded to Y.G. Guangzhou University is listed as the applicant on Chinese patent application No. 2025120125474, with Y.G., W.L., H.W., and M.B. listed as co-inventors. The remaining authors declare no conflicts of interest. Y.G., W.L., F.K., and B.L. designed the experiments; W.L., H.W., H.K., X.F., F.H., and R.L. performed the experiments; W.L., H.W., M.B., Y.Z., F.K., B.L., M.L., and Y.G. analyzed data; Y.G., B.L., F.K., W.L., and H.W. wrote the paper. All authors agreed to the final version of the manuscript. Additional Supporting Information may be found online in the supporting information tab for this article: http://onlinelibrary.wiley.com/doi/10.1111/jipb.70233/suppinfo Figure S1. Schematic diagram of three distinct crRNA array architectures for 3× crRNA editing. Figure S2. 3× crRNA editing efficiency achieved with the pSoC12i-V3 system in hairy roots Figure S3. Off-target analysis of 4× crRNA editing by pSoC12i-V3 in positive hairy roots. Figure S4. Comparison of the editing efficiency of 3x crRNA for creating soybean-free and high-oleic acid varieties based on Cas-SF01, pSoC12i-V1, and pSoC12i-V3 in hairy roots. Figure S5. Editing efficiency at different target sites during stable transformation of high-oleic, low-beany-flavor soybean based on pSoC12i-V3. Figure S6. Off-target assessment in homozygous DS89-5m T1 generation. Table S1. List of all target sites used in the study Table S2. DNA sequences of genetic elements used in this study Table S3. Predicted off-target sites analyzed in this study Table S4. Primer pairs used for off-target detection in this study Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.