Recruitment of an ancient branching program to suppress carpel development in maize flowers

To identify genes that regulate carpel suppression with gt1, we conducted an ethyl methanesulfonate (EMS) enhancer screen, looking for mutants where carpel growth was derepressed in tassels. Maize flowers typically initiate three stamens and three carpel primordia, with stamen development suppressed in ear flowers and carpel development suppressed in tassel flowers (Fig. 1 A and D). When carpel growth is not suppressed, gynoecia, each consisting of three carpels, develop in tassel flowers, resulting in silks emerging from tassel spikelets (Fig. 1C). In gt1 single mutants, only 7% of scored tassel flowers developed long silks that emerged from spikelets (>0.2 cm, Fig. 1 B and G). However, in the double mutants we identified in our screen, 100% of scored tassel flowers developed long silks (Fig. 1 C and G). We called this enhanced mutant phenotype rapunzel (rzl), after the Grimm brothers’ fairy tale character with long hair. Two of these rzl mutants, gt1;rzl-3 and gt1;rzl-4, phenocopied one another and did not complement each other, indicating that rzl-3 and rzl-4 are allelic (Fig. 1 C and I; SI Appendix, Fig. S1). Therefore, we focused many of our downstream analyses on either rzl-3 or rzl-4.

The gt1;rzl-3/4 floral phenotype was specific to carpels and manifested in both tassels and ears. In contrast to most ts mutants (9, 10, 12, 13), stamen development in gt1;rzl-3 and gt1;rzl-4 tassel flowers was not suppressed (Fig. 1; SI Appendix, Fig. S1 and Table S1). In normal ear spikelets, two flowers initiate, but only the upper flower completes development; the lower flower is suppressed (7). In gt1;rzl-3 ears, carpel growth was derepressed in lower flowers, resulting in two fertile flowers per spikelet (Fig. 1 F and H; SI Appendix, Fig. S2). Thus, rzl-3/4 interacts with gt1 to disrupt carpel suppression while leaving other floral and inflorescence traits unaffected.

To identify the rzl-3/4 gene, we used both bulked segregant analysis coupled with whole genome shotgun sequencing (BSA-seq) and fine mapping (19). Our BSA-seq results revealed a broad peak between 160 and 180 Mbp on chromosome seven that represented the rzl-4 mapping interval (Fig. 2A). The narrow peaks on chromosomes three and nine are likely because of differences between B73 laboratory stocks and the reference genome (19, 20). The broad region of homozygosity on chromosome one results from the introgression of gt1, which arose in A619 (17), into B73. Using fine mapping (19), we reduced the chromosome seven mapping interval to a ∼230-kbp region containing nine genes (Fig. 2B), only one of which contained a canonical EMS single nucleotide polymorphism (SNP) predicted to negatively affect gene function (21).

This EMS SNP was predicted to change a single amino acid (Ala337Thr) in the trehalose-6-phosphate phosphatase (TPP) encoded by RAMOSA3 (RA3) (18). Ala337 is deeply conserved in TPP paralogs and contacts the active site in homology models (SI Appendix, Fig. S3). In addition, the substitution of a homologous amino acid in an RA3 paralog negatively affects protein function (22), suggesting that the rzl-4 SNP would impact RA3 function similarly. Because rzl-4 and rzl-3 were allelic, we sequenced RA3 in rzl-3 mutants and found an EMS mutation at a likely splice acceptor site 5′ of exon 4. Transcripts from this ra3 allele had an in-frame deletion of two active-site amino acid codons (SI Appendix, Fig. S2). Thus, both gt1;rzl-3 and gt1;rzl-4 double mutants harbor alleles of ra3 predicted to negatively affect gene function, indicating that rzl-3 and rzl-4 encode alleles of ra3. From now on, we will refer to these alleles as ra3-rzl3 and ra3-rzl4.

Usually, ra3 mutants have branched ears and increased tassel branching due to indeterminate meristems that produce many spikelets on long branches (18, 22). Therefore, we were surprised to find unbranched ears in gt1;ra3-rzl3/4 double mutants and in ra3-rzl3/4 single mutants heterozygous at gt1 (Fig. 1F). Given that gt1 is semidominant, the lack of ear branching in these mutants could have been caused by a second genetic interaction between gt1 and ra3, regulating meristem determinacy. The lack of ear branching may also have been the result of the genetic background (A619) used for EMS mutagenesis. Most characterization of ra3 has been in the B73 genetic background (18, 22), and background modifiers affect the ra3 ear determinacy phenotype (18).

To test for a genetic interaction between gt1 and ra3 independent of any A619 modifiers, we made a gt1;ra3 double mutant with a third, well-characterized allele of ra3 (ra3-fea1) in the B73 genetic background (18, 22) (Fig. 2). gt1;ra3-fea1 double mutants in B73 recapitulated the rzl phenotype, with silks in tassel flowers, as in our gt1;ra3-rzl3 and gt1;ra3-rzl4 mutants (Fig. 2F). ra3-fea1 single mutants had branched ears and increased tassel branching, as expected (18). However, most gt1;ra3-fea1 double mutants lacked ear branches and had fewer tassel branches than ra3 single mutants (Fig. 2 C–H), indicating a second genetic interaction between gt1 and ra3, regulating meristem determinacy. Thus, RA3 and GT1 act to regulate both meristem determinacy in inflorescences and carpel suppression in flowers.

ra3-rzl3 mutants (ra3-rzl3;gt1/+) had significantly more derepressed carpels in tassel flowers than A619 individuals (Fig. 1G, P value ≪ 0.001, χ² test, two degrees of freedom). This could be because of gt1 semidominance (17) or because ra3 single mutants have a weak floral phenotype. Indeed, there have been hints of likely background-dependent ra3 floral phenotypes (23). To investigate the ra3 floral phenotype further, we followed tassel flower development in mutants using scanning electron microscopy.

In wild-type flowers, the two silk carpels are first visible as a raised line of tissue called the gynoecial ridge. In ear flowers, these silk carpels grow to enclose the third, ovulate carpel and fuse to form the stigma, called the silk in maize (7). In tassel flowers, carpels initiate but undergo programmed cell death shortly after the initiation of the gynoecial ridge (8, 24) (Fig. 3 A–D). In contrast, in gt1 single mutants, silk carpels continued to grow past the gynoecial ridge stage and formed a peak of tissue over the developing ovulate carpel (Fig. 3 E–H). The tassel flowers of ra3 single mutants had carpels that resembled those of gt1 mutants: The two silk carpels formed a peak of tissue over the ovulate carpel but did not fuse laterally to enclose the ovulate carpel (Fig. 3 I–L). These data indicate that carpel suppression is partially disrupted in both ra3 and gt1 single mutants.

The carpels in gt1;ra3 double mutant flowers continued to grow well past the gynoecial ridge stage, eventually fusing to form silks (Fig. 3 M–P; SI Appendix, Fig. S2). In addition, the gynoecia of gt1;ra3 double mutants closely resembled the gynoecia of wild-type ear florets, with two silk carpels and an ovulate carpel (Fig. 3 O and P; SI Appendix, Fig. S2). In contrast, other floral meristem determinacy mutants, such as Z. mays agamous1 (zag1), bearded ear (bde), Tasselseed6/indeterminate spikelet1 (Ts6/ids1), drooping leaf1 (drl1), and drooping leaf2 (drl2), have indeterminate floral meristems, leading to disorganized gynoecia with extra primordia, many of which do not fuse to form silks (12, 25–29). Thus, our results indicate that the gt1;ra3 phenotype arises not because the floral meristems had become indeterminate but because floral organ suppression was disrupted.

The floral phenotypes of both single and double mutants led us to assess GT1 and RA3 localization in developing flowers, using immunofluorescence with custom antibodies. GT1 and RA3 both localized to the nuclei of carpel primordia in 1.5 cm tassels, prior to carpel growth suppression (Fig. 4A). We did not detect GT1 protein in gt1 (B73) carpels, nor did we detect RA3 in ra3-fea1 carpels, which we used because ra3-fea1 is a known protein null allele (18). Importantly, both GT1 and RA3 were predominantly localized to the nuclei of subepidermal cells, where programmed cell death initiates in maize and its relatives in the Andropogoneae (8, 30). While RA3 likely acts non–cell autonomously in regulating meristem determinacy (18, 31), the GT1 and RA3 localization patterns in tassel flowers suggest that both act cell autonomously in regulating carpel suppression.

To identify the effectors of carpel suppression downstream of GT1 and RA3, we examined the transcriptional profiles of wild-type (A619), gt1, ra3, and gt1;ra3 tassels at two developmental timepoints: 1) prior to carpel suppression (presuppression, 0.8–1 cm tassels) and 2) during active carpel suppression (midsuppression, 1–2 cm tassels). We reasoned that genes that were 1) misexpressed in gt1;ra3 mutant flowers and 2) changed over wild-type carpel development represented the best carpel suppression gene candidates. We performed differential expression analyses on our sequencing data (32) and focused on the 73 genes that satisfied both of these conditions and were highly differentially expressed at the midsuppression stage (|fold change| > 2, Fig. 4 B and C; Dataset S1). Most genes in this carpel suppression gene set that were down-regulated in mutants were up-regulated over the course of development in A619 or vice versa (69/73 genes, or 95%; Fig. 4C). This pattern of regulation is consistent with a role for these genes in mediating the development of the rzl phenotype. Notably, 62/73 genes (85%) were misexpressed in gt1 or ra3 single mutants, and 44/73 (60%) were misexpressed in gt1 and ra3 single mutants. This suggests that the carpel suppression phenotype in gt1;ra3 double mutants is a matter of degree, not the result of an entirely different set of genes being misregulated in double versus single mutants.

Although the carpel suppression gene set was not enriched for specific gene ontology (GO) terms, it did contain several genes whose homologs in Arabidopsis thaliana (Arabidopsis) have roles in reactive oxygen species signaling, callose deposition, cell wall remodeling, and programmed cell death (Fig. 4C and Table 1). For example, homologs of the Arabidopsis NAM, ATAF and CUC (NAC)-family transcription factor genes KIRA1 (GRMZM2G081930, NACTF36) and ANAC087 (GRMZM2G181605, NACTF2) were up-regulated as carpel suppression commenced in A619 but expressed at lower levels in mutant tassel primordia versus A619 at the midsuppression stage (Fig. 4D) (33, 34). KIRA1 acts to regulate programmed cell death in stigmatic papillae (33). ANAC087, with ANAC046, initiates programmed cell death in the root cap and regulates chromatin degradation following programmed cell death (34). Two protease genes show a similar pattern of regulation: a cysteine protease gene (GRMZM2G456217) with an Arabidopsis homolog (CEP1) with roles in programmed cell death of xylem elements and the tapetum (35, 36) and an uncharacterized serine protease gene (GRMZM2G313321; see ref. 37). One of the genes that was down-regulated over development in A619 but highly up-regulated in mutant versus A619 tassels (GRMZM2G161233) likely encodes a pectin biosynthetic enzyme involved in cell wall remodeling (38). Cell wall remodeling genes are also differentially expressed between upper and lower floral meristems in maize ears, and pectin modification differs between upper and lower floral ear meristems and between B73 and gt1;ra3 ear primordia (39). These and other examples (Table 1) are consistent with roles for GT1 and RA3 in carpel suppression and further point to genes that may be directly or indirectly regulated by GT1 and RA3.

Two ra3 paralogs, TPP7 (GRMZM2G080354) and TPP10 (GRMZM2G055150), were highly up-regulated during carpel suppression in A619 and strongly down-regulated in gt1;ra3 mutants (Fig. 4C and Table 1). Although RA3’s enzymatic activity is not essential in regulating meristem determinacy, it is a catalytically active TPP enzyme (22). TPP enzymes catalyze the second (and last step) of the only trehalose biosynthesis pathway in plants; they dephosphorylate T6P to produce trehalose and a free phosphate group (22, 40). The misexpression of multiple TPP genes led us to measure sugar and metabolite levels at the carpel suppression stage in tassel primordia (1–1.5 cm in length) (Fig. 4D; SI Appendix, Fig. S4). Consistent with the down-regulation of TPP genes, trehalose was substantially lower in gt1;ra3 mutants (Fig. 4D). However, T6P levels were not concomitantly higher in mutants versus A619. T6P is derived from the hexose phosphates glucose 6-phosphate (G6P) and uridine phosphate-glucose (UDPG) and likely signals sucrose status during plant growth and development (40–42). Similar to T6P and most metabolites, sucrose, G6P, and UDPG levels were also not significantly different between A619 and mutant tassels (Fig. 4D; SI Appendix, Fig. S4). This indicates that while trehalose biosynthesis is impacted in gt1;ra3 double mutants, sucrose status, as signaled by T6P (40, 42), is not affected, at least not at the level of entire tassel primordia. We reasoned that T6P levels may be unchanged because of the homeostatic control of sucrose and T6P levels (41). In support, four T6P synthase genes (TPSs), including at least one encoding a likely catalytically active TPS (43, 44), were also down-regulated in gt1;ra3 double mutants (Dataset S1). Similarly, trehalose metabolism gene expression and sugar accumulation differ between the upper and lower floral meristems in ear primordia (39). Taken together, these data suggest that the trehalose pathway is important in the regulation of carpel suppression.

T6P signaling has roles in regulating bud outgrowth in eudicots (45, 46) and is associated with bud outgrowth in maize (47). Notably, 24 of the 73 carpel suppression genes (∼31%), including TPP7 and TPP10, were differentially expressed in gt1 versus wild-type tiller buds (47). A gene set enrichment analysis also revealed that genes differentially expressed in gt1 versus wild-type tiller buds (47) were enriched in the differentially expressed genes in our study (SI Appendix, Table S2), and RA3 and GT1 are bound and regulated by TEOSINTE BRANCHED1 (TB1), a conserved regulator of axillary bud suppression (47–49). These associations led us to ask whether ra3 is also involved in regulating axillary bud outgrowth. We counted tillers and measured their lengths in A619, gt1, ra3, and gt1;ra3 plants and found that gt1;ra3 double mutants produced more and longer tillers than gt1 single mutants (Fig. 5). In addition, these double mutants produced more ears than single gt1 mutants (SI Appendix, Fig. S5). Thus, RA3 also regulates vegetative branching in concert with GT1 and adds another developmental context in which both genes act to suppress growth.

We used the gt1-1 allele (17) for all our experiments with gt1. gt1-1 arose in A619 and was backcrossed five times with B73 to generate the B73 introgression lines used for BSA-seq and for phenotypic characterizations of the gt1;ra3 double mutants in B73. The ra3 alleles used here were either those that arose in our enhancer screen (ra3-rzl3, ra3-rzl4) in A619, or ra3-fea1 in B73 (18).

Plants for the ra3-rzl4 BSA-seq, gt1-1;ra3-rzl3 tassel and ear mature phenotypes, and gt1-1;ra3 tiller quantification were grown at the University of Massachusetts Amherst Crop and Animal Research and Education Farm in South Deerfield, MA (∼42°29′N, 72°35′W). Plants for the gt1-1;ra3-fea1 tassel and ear branch counts were grown over two seasons at the UMass Amherst Crop and Animal Research and Education Farm and near Valle de Banderas, Mexico (∼20°47′N, 105°15′W). Plants for scanning electron microscopy (SEM), metabolite measurements, and RNA sequencing (RNA-seq) were grown in the College of Natural Sciences and Education Greenhouse on the UMass Amherst campus under long day conditions (16 h light, 8 h dark) at 28 °C.

Flowers and spikelets for phenotyping were selected from the central sections of ears or from the central sections of the tassel central spike. For tassel flower phenotyping, 20 flowers from each of four individuals per genotype were examined. For ear spikelet phenotyping, 50 spikelets from each of five individuals per genotype were examined. For measuring silk length, silks from 12 flowers from each of three individuals per genotype were measured. Tassel branches from 68 B73, 67 gt1, 59 ra3, and 66 gt1;ra3 individuals and ear branches from 26 B73, 30 gt1, 54 ra3, and 65 gt1;ra3 individuals were counted over two field seasons. Tillers were measured from 57 gt1 and 64 gt1;ra3 individuals in three planting blocks over the course of two field seasons. To account for differences in plant height, we normalized tiller length measurements to the height of the main culm. Statistical significance was assessed using an ANOVA and Tukey’s test, χ² test, or Student’s t test, as appropriate (72, 73).

We performed an EMS mutagenesis screen as in (74). Briefly, gt1-1 (A619) pollen was mutagenized with EMS and crossed onto gt1-1 (B73) ears. M1 progeny were selfed to generate M2 families that were screened for enhanced silk growth in tassels. We identified rzl-4 and rzl-3 in this screen.

To map rzl-4, we crossed gt1-1;rzl-4 (A619) individuals to gt1-1 (B73) individuals and selfed the F1 progeny to generate an F2 mapping population. In the F2 population, leaves from 238 gt1-1;rzl-4 mutant individuals were selected for a pooled DNA extraction for BSA-seq, as described (19). Extracted genomic DNA was sequenced on an Illumina HiSEq. 2500 (paired-end reads, 150 bp) at Brigham Young University.

Bioinformatic tools on the Galaxy platform were used to assess read quality, align reads, and call SNPs and indels. We used FastQC (version 0.69) to assess read quality, and we used a PHRED cutoff of 20 (75). Reads were aligned to version 3 of the B73 reference genome using Bowtie2 (version 2.3.2.2) (76, 77). Variants were called using Samtools Mpileup (version 2.1.3) and filtered using Varscan (version 0.1) and Samtools filter pileup for SNPs and indels (78). Homozygous SNP variants were plotted using the R package ggplot2 (79). Candidate SNPs were identified using SnpEff (version 4.3a) (80), and SNPs of moderate effect were filtered using Provean (21).

gt1-1 was genotyped with a cleaved amplified polymorphic sequences BsaJI restriction site that overlaps the gt1-1 lesion (17, 81), resulting in cleavage of the wild-type but not mutant allele. A 287-bp fragment surrounding the gt1-1 lesion was amplified with NEB Phusion High-Fidelity DNA polymerase with GC buffer (New England Biolabs). PCR conditions were 98 °C for 30 s, 72 °C for 15 s, and 72 °C for 30 s (40 cycles). BsaJI cuts the 287-bp fragment in the wild-type allele to 33 base pairs and 254 base pairs but does not cut the mutant gt1-1 allele.

ra3-rzl3 was genotyped with a derived cleaved amplified polymorphic sequences (dCAPS) Sau96I restriction site designed with dCAPS Finder 2.0 (82) (SI Appendix, Table S3). A 119-base pair fragment surrounding the ra3-rzl3 lesion was amplified with NEB Taq Polymerase with standard NEB buffer. PCR conditions were 95 °C for 45 s, 60 °C for 30 s, and 68 °C for 30 s (40 cycles). Sau96I cuts the 119-base pair fragment in the wild-type allele to 24 base pairs and 95 base pairs but does not cut the mutant ra3-rzl3 allele. PCR products were digested with BsaJI at 55 °C for 1 h (gt1-1) or Sau96I at 37 °C for 1 h (ra3-rzl3) and run on a 3.5% agarose gels (120 V, for 60 min) to separate the digested bands.

Protein homology modeling was performed with SWISS-MODEL (83). A sequence of 363 amino acids encoded by the primary RA3 transcript was used as input into the homology search. RA3 had highest homology to the structure “Aspergillus fumigatus trehalose-6-phosphate phosphatase crystal form 1” (33.47%, Protein Data Bank code: 5dxl, X-ray, 1.6 Å) (84). POLYVIEW-3d was used to visualize RA3 homology models with ra3-rzl4 (85). ConSurf was used to assess amino acid conservation (86).

SEM was performed with a JEOL JCM-6000Plus Neoscope Benchtop scanning electron microscope with fresh tissue samples. Tassels and ears were dissected from the main culm and their bases painted with colloidal silver paint (Ted Pella, Inc.) and placed onto prechilled SEM stubs with electron microscope-conductive carbon double-sided tape (Nisshin). Stubs were prechilled on ice for at least 15 min. Images were recorded under high vacuum and 5-kV voltage within 15 min of being in the SEM.

We prepared a total of 32 pools of three tassels per pool at two developmental stages (1–1.1 cm and 1.3–2.0 cm) for A619, gt1, ra3-rzl3, and gt1 ra3-rzl3 plants for RNA-seq. Pools for gt1 and ra3-rzl3 samples contained a mix of individuals: either homozygous wild-type or heterozygous at ra3-rzl3 or gt1, respectively. Total RNA was extracted from 96 individual tassels using TRIzol and the Qiagen RNeasy Plant Mini kit protocol. Tissue was collected in 1.7 mL RNase-free safe-lock tubes with ceramic grinding beads and immediately placed into liquid nitrogen. Samples were ground in a Qiagen TissueLyser II for 30 seconds, and 1.0 mL of TRIzol was added to each sample and mixed well by vortexing until all sample powder was thoroughly mixed. Samples were incubated for five minutes at room temperature (RT), with frequent vortexing. Then, 0.2 mL of chloroform was added to each tube and vortexed for 15 s, followed by a 1-min incubation at RT and another 15-s vortex. Samples were centrifuged at 15,000 × g for 10 min to separate phases. After centrifugation, 200 μL of the aqueous layer was removed and added to 700 μL of Qiagen RLT buffer in a new tube. RLT buffer was prepared by adding 10 μL of β-mercaptoethanol per 1 mL RLT buffer. Then, 500 μL of 96–100% (vol/vol) ethanol was added to the 200 μL of sample now combined with 700 μL RLT buffer. Samples were mixed well by vortexing, and half of the sample (∼700 μL) was applied to a Qiagen MinElute pink spin column. From this point, samples were processed using the Qiagen RNeasy Plant Mini kit protocol, according to the manufacturer’s instructions. RNA quality was assessed on a 1% agarose gel run for 20 min at 120 V and quantified using a Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific). Four pools were generated for each genotype and developmental stage by combining 1 μg of RNA from each of the three tassels. Total RNA from each sample was treated on-column with DNase I (NEB) for 15 min. RNA pools were sent to Novogene for library construction and 20 Mbp Illumina paired-end 150-bp sequencing for each pool.

RNA-seq sequencing libraries were trimmed using Trimmomatic (version 0.36.3) (87). These were then aligned to the Z. mays B73 v3 genome (76) and processed into BAM files using STAR (version 2.7.0) (88). Read counts were generated with the Rsubread package function featureCounts in R (89, 90). edgeR was used to construct principal component analysis plots of libraries (32). Because two of these libraries did not cluster with their corresponding replicates, we removed these libraries from further analysis (SI Appendix, Table S4).

Pairwise differential expression was calculated between wild-type and each mutant using edgeR (32). GO term enrichment was called using topGO (91) and maize-GAMER gene annotation (92). Enrichment of differentially expressed genes between wild-type and gt1 from Dong et al. (47) in this study was assessed via gene set enrichment analysis, with 1,000 permutations of the gene set in all cases except for the gt1 versus A619 postcomparison, which used 10,000 permutations (93).

GT1 (guinea pig) and RA3 (rabbit) antisera were used for antibody affinity purification to the carboxyl-terminal region of GT1 or to the N-terminal region of RA3 recombinant protein, using magnetic beads (Invitrogen), as described in (94). Validation of antibody was carried out by immunoblot following the protocol in (95), using total protein extract from wild-type and gt1 or ra3 mutants as negative controls.

To perform triple whole-mount immunolocalizations, we used previously published protocols with minor modifications (50, 96). Tassels between 1 and 1.5 cm from ra3-fea1 (B73), gt1-1 (B73), and B73 were collected and prefixed at 4 °C in 4% (wt/vol) paraformaldehyde and 2% (vol/vol) Tween-20 in phosphate-buffered saline (PBS) for 1 h, embedded in 6% agarose, sectioned at 75 μm using a vibratome (Leica), and collected in fixative solution for 2 h. Sections were washed and permeabilized by cell wall digestion (1% Driselase, Sigma-Aldrich; 0.5% cellulase, Sigma-Aldrich; 0.75% Pectolyase Y-23, Duchefa Biochemie) for 12 min at RT. Tissue was rinsed and incubated for 2 h in PBS and 2% Tween-20, rinsed two times, and blocked with 4% (wt/vol) bovine serum albumin (Sigma-Aldrich) for 1 h. The blocking solution was removed, and the tissue sections were incubated overnight at 4 °C with anti-RA3 (1:200), anti-GT1 (1:75), and anti-YSPTSPS repeat S2Pho (RNA Pol II, B1 subunit; 1:200; Diagenode). The samples were washed for 8 h at 4 °C with gentle agitation with PBS (0.2% vol/vol Tween-20), replacing the buffer every 2 h, and incubated overnight at 4 °C with the appropriate secondary antibodies (Thermo Fisher Scientific)—anti-rabbit Alexa 488 (RA3), anti-mouse Alexa 568 (RNA POL II), and anti-guinea pig Alexa 647 (GT1)—counterstained with DAPI (Sigma), and mounted with ProLong Gold (Thermo Fisher Scientific). Immunolocalizations were repeated at least three times for each genotype. Images were acquired using a Zeiss LSM 780 confocal microscope.

Tassels between 1–1.4 cm were dissected from A619, gt1-1, ra3-rzl3;gt1-1/+, and gt1-1;ra3-rzl3 individuals and immediately frozen in liquid nitrogen. Eight tassels of each genotype were sampled in four waves over a 4-wk period. All individuals in a wave were sampled at 15 h 00 min to minimize variability. T6P and sugar metabolites were extracted from single tassels using a protocol from (41). T6P, other phosphorylated intermediates, and organic acids were measured by anion-exchange high performance liquid chromatography coupled to tandem mass spectrometry as described by (41) with modifications as described by (97). Sugars and sugar alcohols were measured as described by (98). Statistical significance was assessed using an ANOVA and Tukey’s test.