Seeds of the maize inbred line B104 used were originally obtained from the USDA National Plant Germplasm System (Accession no. PI 594047). A single maize B104 seed was placed in a pre-wetted Jiffy-7® pellet and kept in controlled greenhouse conditions (300 μE.m⁻².s⁻¹ light intensity, 16 h light, 26°C and 8 h dark, 22°C). The seedlings were transferred to 10 L pots containing controlled release fertilizer (2.0 kg/m³, Osmocote® Professional potting mixture, Scotts International B.V.) and moved to a larger greenhouse until grown to maturity.

For cloning, DH5α competent cells (Invitrogen) and ccdB Survival™2 cells (Invitrogen) were used. For transformation, hypervirulent disarmed Agrobacterium strains EHA101 (Hood et al., 1986) and EHA105 (Hood et al., 1993) were used.

In the presence of alternative gene models for VYL (Chr.9_ClpP5) in the B73 genome annotation (v4), we targeted exon 4, which is present in all gene models and encodes the start of the conserved ClpP5 domain. We used the B73 sequence of VYL exon 4 and CRISPOR¹ to select a sgRNA with predicted high specificity and efficiency. The targeted region was amplified using wild-type B104 genomic DNA and sequenced. A single nucleotide polymorphism was found, and the spacer was adapted accordingly (TCGGTGCGGAGGCCCTGTTG).

All plasmids used in this study are listed in Supplementary Table S1, primers in Supplementary Table S2 and modules produced by gene synthesis in Supplementary Table S3.

Immature embryos of maize inbred line B104 were transformed using Agrobacterium as previously described (Coussens et al., 2012) with modifications. Briefly, when immature embryos were 1.5–2 mm in size (12–14 days after fertilization), at least three ears from individual B104 plants were collected, stored overnight at 4°C and surface-sterilized the next morning. Immature embryos were isolated from the ears and co-cultivated with Agrobacterium for 3 days on co-cultivation medium (Supplementary Table S4) at 21°C in the dark, with the scutellum side up. After co-cultivation, embryos were transferred to non-selective resting medium (Supplementary Table S4) and incubated for 6 days in the dark at 25°C. Compared to Coussens et al. (2012), co-cultivation and resting media contain 15 μM of the auxin dicamba (Sigma-Aldrich, Saint Louis, Missouri, United States) instead of 2,4-D.

For the experiments, all following steps are identical to Coussens et al. (2012), except for 15 μM dicamba in selection I (Supplementary Table S4) and selection II media (referred to as method 1). For all experiments from 2020 and with morphogenic genes (referred to as method 2), we reduced the time on selection I from 14 to 7 days (Supplementary Tables S4,S5). Next, we transferred embryos on a first maturation medium for 14 days containing cupric sulfate (CuSO₄) and the plant growth regulators IAA, thidiazuron, ABA, BAP, and zeatin (Supplementary Table S4), followed by 14 days on second maturation medium containing CuSO₄, IAA, ABA, and BAP in the light (Supplementary Table S4). During the last tissue culture step (14 days on regeneration II medium, incubated at 25°C), plantlets develop roots. The rooted plantlets are then transferred to soil (a pre-wetted Jiffy-7 pellet) and covered with plastic box to maintain high humidity; this facilitated transition from tissue culture to soil. Transgenic plantlets were kept in controlled growth room conditions (300 μE.m⁻².s⁻¹ light intensity, 16 h light, 26°C and 8 h dark, 22°C). Humidifying cover was removed after 3 days. After 2 weeks on soil, plants were genotyped for presence of T-DNA and, in case of CRISPR/Cas9 mutagenesis, for mutations in the target gene(s). Plants that showed the desired genotype were transferred to 10 L pots containing controlled-release fertilizer (Osmocote, 2.0 kg/m³) and grown in the greenhouse. Resistance to PPT was assayed using AgraStrip® lateral flow strips (Romer Labs). Histochemical 5-bromo-4-chloro-3-indolyl-β-D-glucuronide (X-Gluc) assays were performed as described (Coussens et al., 2012). A VHX-7000 digital microscope (Keyence International, Mechelen, Belgium) was used for imaging of tissue culture at different stages up to regeneration II.

For genomic DNA extraction, a piece of approximately 1 cm² was cut from a leaf and put in a 2 ml reaction tube together with two 4 mm metal grinding beads and snap-frozen in liquid nitrogen. The tissue was then crushed using a Retsch® Mixer Mill MM 400 machine at 20 Hz for 1 min. Genomic DNA (gDNA) was isolated from the crushed, frozen leaf material using the Wizard® Genomic DNA Purification kit (Promega). gDNA samples were subjected to PCR amplification of the targeted region, by using the GoTaq® Flexi kit (Promega). PCR products were subjected to a purification step using magnetic beads (HighPrep™ PCR Clean-up System, Magbio). A Mix2Seq Kit (Eurofins Genomics) was used for sequencing of the purified PCR products and analyzed using ICE (Synthego). For genotyping of the HRA marker, the primers CROPGEN597 and CROPGEN598 (Supplementary Table S2) were used that target HRA and the StPinII terminator, respectively. For genotyping of VYL, CROPGEN6, and CROPGEN56 were used, for Chr.1_ClpP5 CROPGEN452, and CROPGEN453 were used (Supplementary Table S2).

Droplet digital PCR (ddPCR) was used as previously described (Desmet et al., 2020) for gene copy number measurements. Reference primers for the single-copy genes FPGS (Zm00007a00000670, Manoli et al., 2012) and ADH1 (Collier et al., 2017) were used. Additional primers were designed for pZmPLTP::ZmBBM, tStPinII (HRA terminator), and aadA (backbone marker) and are listed in Supplementary Table S2. Genomic DNA of T0 plants was isolated using the Wizard® Genomic DNA Purification kit (Promega), quantified using the Qubit dsDNA HS Assay Kit (Thermo) and digested with CviQI (NEB) for 1 h. Per sample, 15 ng of digested gDNA was used as input for ddPCR with ddPCR EvaGreen supermix (BioRad) and analyzed using the QX200™ (BioRad) in duplicate with a ramp rate of 1°C/s. ddpcRquant (Trypsteen et al., 2015) was used for calculating copy numbers.

ZmBBM (Zm00001eb144510), ZmWUS2 (Zm00001eb148390), ZmPLTP (Zm00001eb406100), ZmAXIG1 (Zm00001eb271530), ZmGLB1 (Zm00001eb052450) Chr.1_ClpP5 (Zm00001eb014750), and Chr.9_ClpP5 (Zm00001eb396050).

A maize transformation pipeline available at our institute has been used widely by the scientific community for ectopic expression of transgenes (Sun et al., 2017), analysis of reporter genes (Bezrutczyk et al., 2021) and recently, gene editing (Gong et al., 2021; Pedroza-Garcia et al., 2021). We use a well-established method for the public B104 inbred line (Coussens et al., 2012) with minor modifications (method 1). In 2016, we performed 35 comparable experiments according to this protocol and with the same vector backbone and selection marker cassette (pBbm42GW7, Karimi et al., 2013) but with varying transgenes of interest. This allowed us to monitor the efficiency of transformation over several independent experiments (Supplementary Dataset S1). For each transformation experiment, we started with on average 675 immature embryos of 1.5–2 mm as explants (12–14 days after pollination). These were isolated from at least three different ears, harvested from individual greenhouse-grown plants. Embryos were co-cultivated with the hypervirulent Agrobacterium strain EHA101, after which callus was induced using the synthetic auxin dicamba and subsequently selected on expression of BAR using PPT. After 2.5 months of tissue culture in the dark, embryogenic calli are transferred to a light regime and T0 plantlets are regenerated. After seven months, T1 seeds can be obtained (Figure 1A). On average 12.6 T0 transgenic plants are obtained in each experiment of which at least 7.2 are independent, as they are derived from discrete immature embryos and cannot be clonal. This results in an overall average transformation efficiency of 1.1%, defined as the number of independent T0 plants obtained per starting immature embryo (Figure 1B). There is, however, a large variability between experiments as the efficiency ranged from 0 to 3.4%. Throughout the whole transformation procedure, the embryos derived from different ears were systematically kept separate and transformation efficiencies are counted for each ear (Figure 1C). Within the same experiment, we observed a large variability in the transformation efficiency when comparing results of independent ears, despite being derived from the same inbred line, grown in the same greenhouse, at the same time (Figure 1C). This ear-to-ear variability is known in the field of research (Masters et al., 2020), but rarely reported. With the advent of gene editing, the demand for maize transformation is increasing quickly and solutions are needed that increase transformation efficiencies and cope with the observed variability.

A highly efficient maize transformation method using Agrobacterium (Lowe et al., 2018) differs from our method 1 on three important points: (a) use of the MRs ZmBBM and ZmWUS2 to boost regeneration (Lowe et al., 2016), (b) use of a ternary vector for Agrobacterium, improving T-DNA delivery (Anand et al., 2018), and (c) optimized tissue culture media (Chu et al., 2019).

We first focused on improving tissue culture, without use of MRs. After 14 days of callus induction on dicamba, we introduced a maturation medium inspired by Chu et al. (2019) that holds ABA, cupric sulfate, and the cytokinins zeatin, thidiazuron, and BAP (method 2, Supplementary Tables S4,S5). We compared this method with method 1 using embryos derived from the same four ears (Supplementary Table S6). Using method 1, regeneration was poor even in the absence of transformation and selection and only nine out of 87 calli produced shoots. In contrast, method 2 was five weeks faster and shoots were obtained from all 45 starting embryos (Supplementary Table S6).

Next, we applied method 2 for stable transformation, adding PPT as selecting agent (Figure 1A; Supplementary Table S4). During 2020, we performed 19 comparable experiments according to this protocol, with vectors only varying in the transgene of interest or in case of CRISPR/Cas9, sgRNAs (Supplementary Dataset S2). For each construct, we started with on average 322 immature embryos isolated from at least three different ears. This resulted in an average transformation efficiency of 4.2% (independent T0/starting IE, excluding any escapes) compared to 1.1% with method 1 in 2016 (Figure 1B). Again, a large variability was observed between experiments ranging from 0 to 9% and ear-to-ear variability was again observed (Figure 1C). A drawback of this method was the number of additional non-transformed regenerated plants that escaped selection as we found two out of three to be escapes (Supplementary Table S7). In conclusion, changing of the tissue culture procedure alone was sufficient to significantly increase transformation efficiency of B104 and shorten the selection and regeneration time of T0 shoots.

Next, we evaluated the combined use of ZmBBM and ZmWUS2 for enhanced plant transformation based on Lowe et al. (2018). We used gene synthesis to create Golden Gate compatible modules for ZmPLTP and ZmAXIG1 promoters and coding sequences of ZmBBM and ZmWUS2. We then combined these in our standard binary vector with a BAR selection marker, and a GUS and mScarlet-NLS reporter (pLAPAU9, Figure 2A). Immature embryos were transformed with pLAPAU9 and a control construct only containing a GUS reporter module and the BAR selection marker (Figure 2A). After five days, control embryos showed a typical pattern of transient GUS expression (Figure 2B). Embryos transformed with pLAPAU9 showed larger multicellular protrusions expressing GUS (Figure 2C, blue arrows). These protrusions are likely somatic proembryos as reported earlier (Lowe et al., 2018). Following method 2, the tissue derived from MR-transformed embryos vigorously produced shoots, in contrast to the control (Figures 2D,E). Interestingly, two main shoot types were observed: (a) wild-type shoots that were identified as escapes (Figure 2E, orange arrows) and transgenic shoots with wider, abnormal leaves (Figure 2E, red arrows). In an independent experiment with pLAPAU6 (Supplementary Figure S1A) containing the MRs but no reporters, only five out of 96 tested regenerants were transgenic, and all transgenics had severe developmental defects and did not grow any further (Supplementary Figure S1B). Hence, the MRs also work in B104 to boost direct somatic embryogenesis and regeneration, but our protocol needs further optimization in selection of transgenics and the expression of MRs genes needs tighter control.

It was reported that the use of MRs combined with a shortened time in tissue culture leads to a high number of escapes when using the selection markers BAR and PHOSPHOMANNOSE ISOMERASE (Lowe et al., 2018). It is assumed that selection markers such as BAR that deactivate the herbicide also detoxify the regions surrounding the transgenic sectors, allowing escapes. In contrast, selection markers such as HRA (HIGHLY-RESISTANT ACETOLACTATE SYNTHASE; Green et al., 2009) encode enzymes that are insensitive to the selection agents and have been used successfully in combination with MRs (Lowe et al., 2018). Another issue with our selection marker might be the use of the CaMV 35S promoter to drive the BAR gene, as it might also influence expression of the neighboring MR-encoding genes, resulting in ectopic expression and unwanted phenotypes later in development (Jopcik et al., 2014). Hence, based on Lowe et al. (2018), we cloned HRA driven by the Sorghum bicolor ACETOLACTATE SYNTHASE (SbALS) promoter as a selection marker (Figure 3A). In addition, we changed the order of the morphogenic genes in the T-DNA, resembling earlier reports more closely (Figure 3A; Lowe et al., 2018).

Transformation of a pZmUBI::GUS control construct with the pSbALS::HRA selection marker yielded no transformants from 212 starting embryos derived from four ears (Supplementary Table S8). However, when we combined pSbALS::HRA with the MRs on the same T-DNA (vector pLAPAU14), independent transgenics were obtained at a transformation efficiency of 6.1% without any escapes as determined by PCR (Supplementary Table S8). While plants initially appeared wild-type (Figures 3B–D), several transformed plants again showed developmental defects after hardening and only four out of 10 were able to produce seed (Figure 3D). We confirm that the selection marker HRA is well-suited to perform rapid selection in combination with MRs, and produces no escapes, while maintaining a transformation efficiency similar to method 2 (~4%) using BAR. However, replacement of the selection marker module was not sufficient to abolish high or ectopic expression of MRs later in development.

To reduce the effect of the MRs later in development, we cloned a Cre/LoxP recombination system similar to Hoerster et al. (2020), in which a monocot codon-optimized Cre recombinase (MoCre) is driven by the promoter of the late-embryogenesis and ABA-inducible maize GLOBULIN-1 gene (pZmGLB1; Duncan et al., 2003). LoxP sites flank the MRs and the pZmGLB1::MoCre module (pLAPAU16; Figure 4). Upon induction of MoCre after somatic embryo formation and/or induction by ABA in tissue culture media (Supplementary Table S4), LoxP sites are recombined, potentially removing the developmental regulators from the genome and thus eliminating the detrimental ectopic expression of ZmBBM and ZmWUS2 (Figure 4; Hoerster et al., 2020).

Similar as above, transformation of the pZmUBI::GUS control with HRA selection marker only yielded two transformants from 216 starting embryos derived from four ears (0.9%, Supplementary Table S8). Using embryos derived from the same four ears, pLAPAU14 (without excision) yielded 15.3% independent transgenics and pLAPAU16 (with excision) 5.1% (Supplementary Table S8). For each construct, we continued with ten random independent transformants. While all ten plants transformed with pLAPAU14 again showed developmental defects and were infertile (Figure 4A), plants transformed with pLAPAU16 appeared wild-type and all but one were fertile (Figure 4B; after crossing to a WT B104 plants, seeds were produced).

To investigate this phenotypic difference between pLAPAU14 and pLAPAU16 transgenic plants in detail, we used ddPCR to examine T-DNA copy number in leaf samples of the ten plants generated with pLAPAU14 and pLAPAU16 (Supplementary Table S9). Primers targeting the pZmPLTP::ZmBBM heterologous fusion or the StPinII HRA terminator were used to amplify only T-DNA encoded genes. pZmPLTP::ZmBBM was present in on average 3.3 copies per T0 plant with pLAPAU14, while with pLAPAU16 the average was 1.3 copies (Supplementary Table S9). Interestingly, the copy number of the HRA resistance marker outside the LoxP sites was reduced from on average 2.9 copies for pLAPAU14 to 1.6 for pLAPAU16. Moreover, while two out of ten plants transformed with pLAPAU14 did not contain the aadA bacterial selection marker gene from the vector backbone, six out of ten transformed with pLAPAU16 lacked the aadA gene. This suggests that Cre-mediated recombination between LoxP sites may also reduce T-DNA complexity by excising T-DNA tandem repeats (De Buck et al., 2007). Finally, the observed non-integer ddPCR values (e.g., 0.5) for pZmPLTP::ZmBBM as compared to HRA, and specifically in plants transformed with pLAPAU16 compared to pLAPAU14 suggest that pLAPAU16-transformed plants are genetic mosaic for excision events. Only one out of ten plants transformed with pLAPAU16 (Plant E) was identified with complete loss of morphogenic genes in T0 leaf material.

In summary, combining MRs with a Cre/LoxP-mediated excision system maintained their functionality and resulted in a high transformation efficiency. Use of the Cre/LoxP-system was associated with a reduced MR copy number, a less complex T-DNA structure, and normal plant development and fertility. Nevertheless, obtained plants in most cases still contained the MR cassettes.

To evaluate gene editing in maize, we aimed to target a gene that leads to a visible knockout phenotype, but without severe developmental and fertility effects. Markers that have been used before such as Zmzb7 lead to albino phenotypes, and are therefore lethal (Feng et al., 2016). The gene VYL (Chr.9_ClpP5) was identified in an ethyl methane sulfonate-mutagenized population derived from the maize inbred line B73 (Xing et al., 2014a). B73vyl homozygous mutant maize plants exhibit a yellow leaf phenotype after emergence but gradually recover and become indistinguishable from wild-type plants after approximately 2 weeks. B73vyl (Zm00007a00050679 in B104) encodes for Chr.9_ClpP5, a proteolytic subunit of the chloroplast Clp protease complex. The B73vyl mutation was found to be a 141 bp insertion in the fourth exon that likely results in a knockout (Xing et al., 2014a). The B73 genome contains a duplicated paralog Chr.1_ClpP5 (Zm00007a00035036 in B104) that explains the lack of more severe phenotypes.

To mimic the B73vyl mutant, we designed a CRISPR/Cas9 construct with a sgRNA targeting exon 4 of VYL (Figures 5A,B). The plasmid was transformed in B104 zygotic embryos using method 1. Already during the regeneration of the plantlets in tissue culture, the expected pale-yellow phenotype was observed in transgenic plants (Figure 5C). Eleven T0 plants, of which five were independent and derived from different zygotic embryos, were analyzed for the VYL locus. Similar to earlier reports with gene editing in B104 (Char et al., 2017), CRISPR/Cas9 gene editing was very efficient with all but one transgenic plant having indel scores of >80% based on Sanger sequencing and subsequent estimation of the efficiency of mutations (Supplementary Table S10). At least three plants had three or more detected alleles, indicating these T0 plants were genetic mosaic, others were homozygous or transheterozygous for edits (Supplementary Table S10). Similar to B73vyl, T0 mutant lines recovered phenotypically as adults, were fertile and edits could be inherited after self-crossing. In T2, we obtained Cas9 null-segregants with homozygous mutations causing a frameshift to show a pale-yellow phenotype compared to WT (Figures 5D,E). Interestingly, we also isolated a weak allele with a three amino acid deletion resulting in an intermediary phenotype (Figure 5F). In conclusion, CRISPR/Cas9 gene editing is efficient and heritable in B104 and VYL is a convenient marker that can be targeted for gene editing in B104.

Continued presence of the MR cassette is undesirable for the generation of stable transgenic lines and downstream phenotypic analysis. However, for gene editing applications, MRs linked to a CRISPR module on a single T-DNA can be removed in the next generation after a genetic cross, creating edited null-segregants. Hence, we constructed the vector pLAPAU17 that contains outside the LoxP recombination sites an additional module with a maize codon-optimized Cas9 variant (Xing et al., 2014b) driven by a ZmUBI promoter and a pOsU3 module that allows direct cloning of the spacer (Figure 6A). We then introduced the same VYL targeting spacer used before (Figure 5) and transformed 222 immature maize embryos. In total, 17 transgenic plants were obtained of which at least nine were independent (4% transformation efficiency). T0 plants were edited in VYL with high efficiency (Figure 6B, green bars; Supplementary Table S10). Some T0 plants unexpectedly showed phenotypic mosaicism with albino stripes, not seen in previous experiments targeting VYL (Figure 6C). The spacer used for VYL (Chr.9_ClpP5) contains three mismatches to the Chr.1_ClpP5 sequence, making the latter a potential off-target (Figure 6D). Indeed, T0 plants also showed editing at the corresponding site in Chr.1_ClpP5 (Figure 6B, red bars; Supplementary Table S10). Remarkably, none of the eleven lines generated with the previous CRISPR/Cas9 construct targeting VYL (Figure 5) showed editing at Chr.1_ClpP5 (Supplementary Table S10), suggesting that one or more factors in the pLAPAU17 construct make CRISPR/Cas9 editing more active at the off-target Chr.1_ClpP5 locus. Line B, generated with pLAPAU17-VYL, scoring −4;+1 for VYL and 0;+1 (23%) for Chr.1_ClpP5, was selected and backcrossed to WT B104. The T-DNA segregated in a Mendelian fashion for a single T-DNA locus and T1 null-segregants were identified using PCR. Confirming inheritance of edits, T1 null-segregants were all heterozygous for a mutated VYL allele, and either WT or heterozygous for a mutated Chr.1_ClpP5 allele (Supplementary Table S10). In conclusion, MRs can be readily combined with CRISPR/Cas9 for efficient transformation and heritable gene editing in B104 maize.