We obtained the plasmid pGFPGUSPlus containing eGFP (Vickers et al., 2007) from Addgene (Plasmid #64401). XbaI and SacI sequences were added by PCR amplification to the eGFP coding region using eGFP-Fw-XbaI and eGFP-Rv-SacI primers. The amplicon was digested with XbaI and SacI and the resulting fragment was ligated to the pTF101 vector under the control of the 35S promoter as previously described (Aboobucker et al., 2021). The new plasmid is named “pTF101–35S-eGFP” and the vector contains a bar-selectable marker under the 35S promoter. mRFP1 coding sequence was amplified from pRU1144 plasmid (Addgene plasmid #14474; Karunakaran et al., 2005) with BamHI and SacI sites using primers mRFP1-Fw-BamHI and mRFP1-Rv-SacI. Sub-cloning mRFP1 in the pTF101 vector was similar to the process described for eGFP, which resulted in “pTF101–35S-mRFP1”. Selected clones were verified by Sanger sequencing at the ISU DNA Facility. Plasmids were introduced into Agrobacterium tumefaciens strain C58C1 by the freeze-thaw method (Holsters et al., 1978). For Arabidopsis transformation, the floral dip method was used (Clough and Bent, 1998). Putative transformants were those resistant to 60 mg/L glufosinate ammonium sprayed on seedlings seven days after sowing. Further molecular confirmation for eGFP, mRFP1, and bar sequences was carried out on DNA isolated from the putative transformants.

Arabidopsis thaliana Columbia ecotype (CS70000) used as genetic background, was obtained from the Arabidopsis Biological Resource Center. Transgenic seeds were sterilized, planted, sexually crossed, and callus was initiated and subcultured, all of which is described in Cook et al. (2024). Seeds were stratified for four to eight days.

Protoplasts for spike-in experiments were isolated from callus tissue that was subcultured up to 39 days before isolation, using the method outlined previously (Cook et al., 2024). A 10µL protoplast sample was placed on a Marienfeld 0.0025mm² hemacytometer to estimate the cell concentration. The hemacytometer measurement was repeated multiple times for each sample and the average was used to determine the protoplast concentration.

For the spike-in sample preparations, known concentrations of cells from the RFP line were mixed with a known concentration of cells from the GFP/RFP genotype and mixed thoroughly. Total concentrations of RFP were 0.74%, 1%, 1.5%, and 5%.

Protoplasts were isolated and analyzed as described previously (Cook et al., 2024). Callus-derived protoplasts were analyzed using consistent settings for six different channels ( Supplementary Table S2 ).

For root imaging, we used a fluorescent Olympus SZH10 stereo microscope equipped with a camera and Leica software. A Texas red filter (excitation: 562/40, emission: 624/40) and an L5 filter (excitation: 480/40, 527/30) was used for visualization.

The statistical model and estimation method for estimating RFP rates after accounting for misclassifications by the gating scheme are published in more detail in a previous article from our group (Cook et al., 2024). We perform a simple linear regression analysis to investigate the relationship between the estimated and the true spike-in percentages and how closely the estimated spike-ins mimic the true value of spike-ins. To that end, let, s_il and s ^ i l be the true and estimated l t h spike-in rates in the i t h set of experiments. We use the simple linear regression model:

for i = 1 , … , K   and l = 3 , … , L i , with e i l being normal random error for each observation. Here, under lossless detection of spike-in rates β 1 should be 1. However, in practice we expect β 1 to be less than 1 because some cells are destroyed in the process.

Two Agrobacterium-mediated transgenic lines were identified and chosen for further development of a high-throughput in vitro system for meiosis induction detection. Supplementary Figures S2 , S3 identify two RFP and two GFP-expressing Arabidopsis lines analyzed for phenotypic analysis. We selected 35s::RFP 2–3 and 35s::GFP 1–3 lines (now referred to as RFP and GFP, respectively) for additional use, based on the fact these lines displayed high fluorescence expression in roots ( Figure 1 ). Moreover, more than 50% of the protoplasts generated from the leaves of these lines showed expression of the fluorescent markers ( Supplementary Figures S1 , S3 ). To ensure that the transgenes conferring either RFP or GFP were located at a single locus, the inheritance of a linked BAR-resistance gene was evaluated among the selfed progeny from a heterozygous parent. The data presented in Supplementary Table S3 are consistent with the RFP and GFP transgenes being at a single genetic locus.

Bi-fluorescent plants were generated by crossing the RFP line and the GFP line ( Figure 1F ). Subsequently, fluorescence was evaluated in protoplasts isolated from roots, leaves, and callus cultures from these F1 plants and their parental lines. Pairwise comparisons of the normalized FITCA (GFP expression) and PECF594A (RFP expression) fluorescence intensity values are presented in Figure 2 . There was evidence for a higher GFP (FITCA) fluorescence in callus-derived protoplasts than in leaf protoplasts isolated from the GFP/RFP and GFP genotypes. These callus protoplasts also showed higher GFP (FITCA) fluorescence than root protoplasts from the GFP/RFP genotype ( Figure 2A ). Similarly, RFP (PECF594A) fluorescence was higher in callus-derived protoplasts as compared to leaf protoplasts from the GFP/RFP genotype and RFP line ( Figure 2B ). Furthermore, RFP (PECF594A) fluorescence was also higher in callus protoplasts than in root-derived protoplasts from the RFP line ( Figure 2B ).

Thresholding was used to classify callus-derived protoplasts from the GFP, RFP, and GFP/RFP genotypes. Fluorescence detection rates for each genotype, as classified by thresholding, are presented in Table 1 . Callus-derived protoplasts from the RFP line were correctly identified 66.5% of the time while those isolated from the GFP line were correctly identified 40.3%. The majority of misclassifications were identified as wild-type or non-fluorescing cells. Callus-derived protoplasts isolated from the GFP/RFP genotype were correctly identified 11.8% of the time while GFP false-identification rates exceeded 55% and RFP false-identification was less than 0.1%. Additionally, when callus-derived protoplasts from the RFP line were analyzed, cells that were falsely detected as being GFP or GFP/RFP were at rates below 0.1%. When callus-derived protoplasts from the GFP line were analyzed, cells that were falsely detected as being RFP or GFP/RFP were detected at rates below 3.5%.

Based on the findings in Experiment 2, we decided to focus on RFP protoplasts, as they could be much better discriminated from the protoplasts classified as GFP/RFP as compared to what GFP protoplasts could. Because of the inability to differentiate cells classified as GFP and those classified as GFP/RFP; GFP, and GFP/RFP cell classifications were combined in a new merged class for a multinomial statistical model (see materials and methods). To determine if minute changes in fluorescent cell population proportions could be detected even with the inability to separate GFP and GFP/RFP classified cells, callus-derived protoplasts from the GFP/RFP genotype ( Figure 1F and Table 1 ) were spiked with known amounts of callus-derived protoplasts from the RFP line ( Figure 1E and Table 1 ), both haploid and diploid, and detected via flow cytometry ( Figure 3 ). Figure 3A provides the overlapping population distributions for the haploid and diploid cells isolated from the RFP line. Figure 3D provides an overlapping population distribution for a collation of individual 1% RFP spike-in experiments that were developed from cells contained in Figures 3B, C . Classification rates are shown in Table 2 and Figure 4 . The estimated spike-in rates, from the multinomial model analysis, are plotted against the true spike-in rates in Figure 4 . Collated classification rates from multiple spike-in experiments detected mean values of 0.88%, 1.3%, and 4.3% for 1%, 1.5%, and 5% spike-ins, respectively ( Table 2 ), and the confidence interval for the spike-in rate of 0.74% is (0.58%, 0.86%). Thus, these data indicate that spike-in percentages as low as 0.74% and as large as 5% can accurately be detected with this biological system and statistical model. The recovery rate of RFP cells mixed into GFP/RFP cells was 86% ( Figure 4 : βˆ 1 = 0.86, SE = 0.025, R2 = 99.2%, p-value< 0.0001). The inability to differentiate GFP and GFP/RFP classified cell populations decreases the detection sensitivity but given the large number of cells capable of being analyzed by flow cytometry and the low misclassification rates of fluorescent cells with the RFP line, detection of small changes in fluorescent cell populations can be detected when observing RFP using our current system.

We developed an Arabidopsis thaliana bi-fluorescent F1 genotype to create a single-cell meiosis-like induction evaluation system using flow cytometry. We validated our system with spike-in experiments using flow cytometry and a multinomial statistical model. Borges et al. (2012) used a similar bi-fluorescent setup to purify sperm cells and vegetative nuclei using fluorescence-activated cell sorting (FACS) based on the presence of mRFP and eGFP. In our study, fluorescent genotypes underwent callus induction to obtain rapidly dividing cells that can easily be isolated as protoplasts. Thresholding enabled the classification of RFP, GFP, and GFP/RFP cell populations using red and green fluorescence properties. Spike-in experiments analyzed with a multinomial model allowed for the detection of uni-fluorescent cells added to bi-fluorescent cells at rates as low as 1%. Further, these spike-in experiments helped to develop and test a statistical framework for future screening efforts of factors triggering meiosis-like induction. Based on our tests, to detect cell population changes of 0.25% (a 1% meiosis induction rate with the current system), it is necessary to have a sample size of around 200,000 cells (for 80% power), which is still within the capabilities of flow cytometry. Further, the recovery rate is predicted to be 0.2399% with SE 0.05610%.

In our tests, none of the detection rates of fluorophores in cells reached 100%. For instance, protoplasts isolated from our GFP line had cells classified as a majority GFP and WT, whereas a minority of cells were classified as RFP and GFP/RFP. We also observed that RFP fluorescence is reduced in our GFP/RFP genotype. As a consequence, data from the GFP and GFP/RFP cells overlapped substantially and could not be accurately differentiated. The overlapping cell populations resulted in a high-false positive rate for cells classified as GFP in the GFP/RFP genotype. Because of this, we combined the cell classes of GFP and GFP/RFP into one class for statistical analysis (See Cook et al., 2024). The decreased RFP expression in the GFP/RFP genotype can potentially be explained by mutations, gene silencing due to two TDNA constructs being present (Matzke et al., 1989), and homology with multiple 35S promoters (Daxinger et al., 2008; Mlotshwa et al., 2010). High transcript numbers due to multiple gene copies led to silencing as well in previous studies (Schubert et al., 2004), but we are uncertain this is a major factor in our system as there were only two copies of the BAR resistance gene, one copy of RFP and one copy of GFP in our GFP/RFP plants. Schubert et al. (2004) found that six copies of GFP under the 35S promoter led to silencing while silencing was seen with only three copies of GUS. Further testing will be required to fully understand the expression characteristics in our samples. In contrast to the overlapping GFP and GFP/RFP classified cells, RFP cells were easily separated from GFP/RFP cells. We were able to efficiently detect low percentages of RFP protoplasts when observing RFP-only cells in our spike-in tests. Since induced haploid cells are expected in equal numbers of RFP and GFP-expressing cells, our assay can be efficiently used by focusing on the detection of cells classified as RFP. Potential improvements to increase the efficiency of our assay include: (i) using different promoters to decrease the risk of homology-mediated silencing, (ii) increasing fluorescence expression with more efficient RFP proteins, and (iii) development of allelic fluorescent markers. As similarities in promoters derived from viruses can result in gene silencing (Daxinger et al., 2008; Mlotshwa et al., 2010), increasing the diversity of constitutive promoters will help to mitigate the risk of silencing due to homology. Promoters such as the nopaline synthase and ubiquitin promoters may be potential candidates, especially for resistance genes. Further, constructs without resistance genes may be an option as genotyping can take place instead, thus eliminating the need for multiple constitutive promoters. A more effective red fluorophore such as mCherry, which has brighter fluorescence and is more photostable than mRFP1 (Shaner et al., 2004; Fink et al., 2010), may help to increase the fluorescence intensity when the GFP construct is present in the same cell. Using allelic markers would also substantially increase the efficiency of the current assay (Cook et al., 2023; Figure 5 ). As shown above, uni-fluorescent cells can successfully be identified with RFP at low rates in the system described here, but the efficiency would be doubled with an allelic approach in the current system and quadrupled if GFP-expressing cells could be identified apart from the GFP/RFP cells ( Figure 5 ; Equation 1, Equation 2, Equation 3). Efficiency improvements would increase detection limits while decreasing the needed sample size. Using a binomial formula (Lübberstedt and Frei, 2012), efficiency increases using non-allelic and allelic single-cell systems can be estimated. For example, assuming a 100% identification rate, we have calculated the different efficiencies below.

Detecting a simulated 1% meiosis-like induction rate using the current non-allelic system:

Detecting a simulated 1% meiosis-like induction rate using an allelic system looking only at RFP:

Detecting a simulated 1% meiosis-like induction rate using an allelic system looking at RFP and GFP:

We were able to detect spike-ins of 1% uni-fluorescent cells in our current non-allelic system (which reflects a meiosis induction rate of 4%). To improve the detection capabilities of the system below 1%, it is necessary to (i) increase the sample sizes, or (ii) minimize misclassification rates. While in vitro meiosis induction rates are unknown in plants, we believe it is necessary to accurately detect haploid cells well below 1% due to the work of Medrano et al. (2016).

Overcoming the in vitro meiosis induction bottleneck will allow advanced plant breeding ideas such as the IVN (De La Fuente et al., 2013) or the COGIV method (Murray et al., 2013) to be further explored. The biological materials and statistical framework validated in this study provide a scalable system capable of detecting meiosis-like induction in callus cells using fluorescent markers. With this investigation, we provide a resource to researchers to test large numbers of meiosis induction candidates quantitatively and efficiently. Callus was selected as the tissue of choice because of the high rates of cell division, fluorescent expression compared to other tissues ( Figure 2 ), and the ability to survive months with proper care. We did not consider cell types such as microspores as they require a plant in reproductive phases which would still present the bottleneck of timing similar to flowering, we tested vegetative tissues (leaf and root) and callus derived from leaf. Using the current callus based system,several induction tests could be run in a short time using this method as multiple samples can easily be analyzed, especially with flow cytometers designed to use 96-well plates. Genetic treatments may also be possible through transformation with Agrobacterium co-cultivation or transient transfection, but there are many challenges associated with genetic testing using our system that further feasibility assessments need to take place. While this system provides an efficient and quantifiable method to assess a factor’s ability to induce a meiosis-like fate, further testing of promising candidates will need to take place to ensure that recombination has occurred.

The current system provides an important resource to initially screen a multitude of factors for their ability to induce meiosis. Successful factors can then be used to develop a library that will be tested in crop species. The use of Arabidopsis provides a model organism with vast genetic resources and short generational cycles. Indeed, however, there will most likely be changes to dosage concentrations when testing factors in other crop species, but the conservation of meiotic factors across plants provides evidence that a successful factor in one species has promise in another (see Mieulet et al., 2016). The in vitro induction of meiosis and future utilization in crop breeding is a difficult road, what we provide here is the first step to enable a systematic approach for step changes in breeding.