SvFUL2, an A-class MADS-box transcription factor, is necessary for inflorescence determinacy in model panicoid cereal, Setaria viridis

Inflorescence architecture in cereal crops directly impacts yield potential through regulation of seed number and harvesting ability. Extensive architectural diversity found in inflorescences of grass species is due to spatial and temporal activity and determinacy of meristems, which control the number and arrangement of branches and flowers, and underlie plasticity. Timing of the floral transition is also intimately associated with inflorescence development and architecture, yet little is known about the intersecting pathways and how they are rewired during development. Here, we show that a single mutation in a gene encoding an AP1 A-class MADS-box transcription factor significantly delays flowering time and disrupts multiple levels of meristem determinacy in panicles of the C4 model panicoid grass, Setaria viridis. Previous reports of A-class genes in cereals have revealed extensive functional redundancy, and in panicoid grasses, no associated inflorescence phenotypes have been described. In S. viridis, perturbation of SvFul2, both through chemical mutagenesis and CRISPR/Cas9-based gene editing, converted a normally determinate inflorescence habit to an indeterminate one, and also repressed determinacy in axillary branch and floral meristems. Our analysis of gene networks connected to disruption of SvFul2 identified regulatory hubs at the intersect of floral transition and inflorescence determinacy, providing insights into the optimization of cereal crop architecture.


Introduction
Inflorescence structure determines fruit, seed, and pollen production, which are critical for reproductive success of plants and global food security. During the shift from vegetative to reproductive growth, the indeterminate shoot apical meristem (SAM), which patterns the vegetative organs, transitions to an inflorescence meristem (IM). Like the SAM, the IM continues indeterminate growth but instead, leaf growth is suppressed and axillary meristems (AMs) grow out into reproductive organs on its flanks. In eudicot systems such as Arabidopsis thaliana, the IM directly lays down floral meristems (FMs), which produce flowers. In grasses, FMs are borne from spikelet meristems (SMs) either directly from the IM as in wheat and barley, or after a series of AM branching events such as in maize and sorghum. Eventually AMs acquire SM identity and terminate in a spikelet, the central unit of the grass inflorescence, housing one to several flowers that bear grain. Variation in activity and determinacy of AMs and SMs in grasses allows for the wide diversity of inflorescence branching patterns (Tanaka et al., 2013;Whipple, 2017;Bommert and Whipple, 2018).
Inflorescence architecture is also shaped by the activity and determinacy of the IM. In certain cereals such as rice, barley and maize, the IM is indeterminate and continues meristematic activity, laying down lateral structures until it ceases growth. Alternatively, in wheat and sorghum, the IM takes on a determinate fate and produces a defined number of AMs before terminating in a spikelet. IM determinacy has been linked to flowering time through the action of multiple common regulators, which also affect branching patterns in the inflorescence (Danilevskaya et al., 2010;Li et al., 2019;Liu et al., 2019a). A weak flowering signal tends to delay meristem determinacy in the inflorescence, allowing for increased branch outgrowth and higher order branch initiation (McSteen et al., 2000;Endo-Higashi and Izawa, 2011;Boden et al., 2015).
Much of what we know about the molecular underpinnings of IM determinacy comes from Arabidopsis, which produces an indeterminate inflorescence. In Arabidopsis, indeterminacy in the IM is maintained by the antagonistic relationship between TERMINAL FLOWER 1 (TFL1) and floral identity genes, LEAFY (LFY), APETALA1 (AP1), and CAULIFLOWER (CAL) (Piñeiro and Coupland, 1998;Liljegren et al., 1999;Serrano-Mislata et al., 2017). AP1 and CAL belong to the euAP1 subclade of the AP1/FUL (FRUITFUL)-like MADS box gene family and are key players in controlling flowering time and AM determinacy (Kempin et al., 1995;Alvarez-Buylla et al., 2006). TFL1 expresses in the central region of the IM and prevents it from acquiring FM identity by suppressing floral identity genes (Weigel et al., 1992;Bradley et al., 1997;Benlloch et al., 2007). Loss of TFL1 function results in the mis-expression of AP1 and LFY in the IM, causing a terminal flower(s) to form in place of the indeterminate meristem, early flowering, and enhanced determinacy of lateral branches (Shannon and Meeks-Wagner, 1991;Alvarez et al., 1992). Alternatively, mutations in AP1 and LFY genes result in production of indeterminate lateral 3 shoots, which typically develop determinate FMs and have delayed flowering (Irish and Sussex, 1990;Schultz and Haughn, 1991;Huala and Sussex, 1992;Weigel et al., 1992;Bowman et al., 1993;Schultz and Haughn, 1993).
The regulatory modules that control inflorescence growth habit are somewhat conserved between eudicots and grasses. In maize and rice, TFL1-like genes delay flowering time and prolong the indeterminate status of the developing inflorescence (Nakagawa et al., 2002;Danilevskaya et al., 2010;Kaneko-Suzuki et al., 2018). In rice, AP1/FUL-like genes have overlapping roles in flowering time (Kobayashi et al., 2012). Over-expression of OsMADS14, OsMADS15, or OsMADS18 all result in early flowering phenotypes (Jeon et al., 2000;Fornara et al., 2004;Lu et al., 2012), and in the case of OsMADS15, reduced panicle size and branch number (Lu et al., 2012). In winter wheat and barley varieties, expression of VERNALIZATION 1 (VRN1), an AP1/FUL-like gene, has been well-characterized as an early signal in promoting timely vegetative-to-reproductive transition in response to vernalization (Yan et al., 2003;Preston and Kellogg, 2008;Li et al., 2019). Expression of FUL2 and FUL3 genes in wheat are also induced by vernalization to promote flowering (Chen and Dubcovsky, 2012;Li et al., 2019). A recent study revealed that AP1/FUL-like genes in wheat and the genetic interactions among them contribute to maintenance of IM and SM determinacy, as well as flowering time (Li et al., 2019).
Loss-of-function in both VRN1 and FUL2 genes converted the normally determinate IM of the wheat spike to an indeterminate habit, and also enhanced indeterminacy in primary AMs. Introduction of a single functional copy of either VRN1 or FUL2 reverted the vrn-null; ful1-null mutant IM back to a determinate habit (Li et al., 2019).
While evidence across the plant kingdom supports conserved roles for AP1/FUL-like genes in floral transition and inflorescence architecture, to date there have been no inflorescence phenotypes described for loss-of-function A-class genes in the subfamily Panicoideae, which includes agronomically important crops such as maize, sorghum, and sugarcane. This is likely due to functional redundancy (Litt and Irish, 2003;Preston and Kellogg, 2007). In this study, we show that a single loss-of-function mutation in an AP1/FUL-like gene in model panicoid grass, Setaria viridis (green foxtail), is sufficient to confer both strong flowering time and inflorescence determinacy phenotypes despite its overlapping expression pattern with three closely related paralogs. S. viridis is a weedy, C4 species that has demonstrated promise as a model system for elucidating molecular mechanisms in panicoid crops (Li and Brutnell, 2011;Huang et al., 2017;Yang et al., 2018). It also represents a key evolutionary node between domesticated and undomesticated grasses. Like wheat, S. viridis produces a determinate inflorescence that terminates in a spikelet, but AMs undergo multiple orders of branching (Doust and Kellogg, 2002a;Zhu et al., 2018). We isolated the Svful2 mutant in a genetic screen, which displayed a 'barrel'-like panicle morphology due to enhanced indeterminacy in AMs. The determinate IM was also converted to an 4 indeterminate habit resembling a maize ear. Further investigation of Svful2 loss-of-function at the molecular level using genomics approaches revealed regulatory modules that link floral transition and inflorescence determinacy pathways through interactions among MADS-box TFs and several other developmental regulators. This mutant and the analyses presented here, provide insights into the complex interface of flowering time and inflorescence development, and potential targets for fine-tuning inflorescence ideotypes in cereal crops.

Characterization of the barrel 1 (brl1) mutant in Setaria viridis
In a forward genetics screen of ~3000 N-methylurea (NMU) mutagenized M2 families of S. viridis (Huang et al., 2017;Yang et al., 2018), we isolated the barrel1 (brl1) mutant, named for its abnormal, barrel-shaped panicle. Compared with mature panicles of the wild-type mutagenized reference line (A10.1), mutant panicles were shorter and thicker and appeared more brachy (Fig. 1, A  Previous studies in S. viridis showed that flowering time impacted both plant architecture and biomass (Doust, 2017). Under both LD and SD conditions, plant height and panicle length of brl1 mutants were significantly shorter than wild-type plants at maturity. Under SD conditions, above-ground dry weight was increased in mutants compared to wild-type, largely due to biomass of vegetative tissue (leaves and stems; Fig. S1). In LDs, above-ground dry weight of brl1 mutants was comparable to wildtype, however we still observed a significant increase in dry weight of vegetative tissue (leaves and stems) in the mutant (Fig. S1). Seed shape and size were also different with the mutant seeds being longer and narrower than those of wild-type ( Fig. 1E; Table 1).
Examination of the inflorescence morphology revealed that brl1 mutants displayed various levels of indeterminacy. At the tip of the panicle, the IM appeared indeterminate in mutants, and newly formed branch meristems (BMs) were still visible at maturity (Fig. 1F). At the base of the mutant panicle, rudimentary primary branches were observed, which were not found in wild type (Fig. 1G). Primary 5 branches were longer in brl1 mutants and the panicle rachis was clearly thicker (Fig. 1H;Supplemental Fig. S1A). Bristles, which are modified branches paired with spikelets in Setaria sp., did not elongate to the length of wild-type bristles, and so were largely found buried under spikelets (Fig. 1H). Development of spikelets and flowers was also affected in brl1 mutants, but phenotypes showed low penetrance with varied severities of indeterminacy. For example, approximately 17% of brl1 mutant panicles produced additional flowers, bristles and/or spikelets within spikelets compared to the typical one flower per spikelet in wild-type (Fig. 1, I-K). The lemma and palea of mutant flowers were more elongated in the mutant and were more rigid, which is likely contributing to the elongated seed shape (Supplemental Fig.   S1 B and C).

brl1 mutants show loss of determinacy in various stages of inflorescence development
We used Scanning Electron Microscopy (SEM, Fig. 2) to compare the developmental progression of inflorescence primordia from the brl1 mutant with that of wild-type S. viridis. By 11 DAS, the vegetative SAM of wild-type plants had finished transitioning to the reproductive IM, as the first primary BMs were initiated on its flanks ( Fig. 2A). In the brl1 mutant, the vegetative-to-reproductive transition was delayed to 15 DAS (Fig. 2B), consistent with its late flowering phenotype (Fig. 1C). After the transition, wild-type inflorescences initiated primary branches in a spiral pattern (Fig. 2C), and then secondary and tertiary axillary branches sequentially in a distichous pattern, as previously described ( Fig. 2E) ((Doust and Kellogg, 2002a;Yang et al., 2018;Zhu et al., 2018)). The brl1 IM was elongated compared to that of wild-type (Fig. 2D,Supplemental Fig. S1 D and E), and this appeared to enable capacity for increased initiation of primary and higher order branches (Fig. 2, D and F), consistent with the mature panicle phenotype. By 17 DAS the wild-type IM had become determinate and terminated as a spikelet (Fig. 2G).
BMs then began to differentiate from the tip of the inflorescence primordium into either a spikelet meristem (SM) or a sterile bristle, and this continued basipetally (Fig. 2G). Conversely, the IM of the brl1 mutant remained indeterminate and continued to produce primary BMs at 21 DAS, where SMs and bristles began to differentiate towards the top of the inflorescence primordium (Fig. 2H). By the end of the developmental series analyzed by SEM, the brl1 IM remained indeterminate, which is consistent with its mature phenotype in Fig. 1F.
While differentiation of SMs and bristles appeared normal in the mutant (Fig. 2, I and J), the onset was delayed compared to wild-type and after additional rounds of higher order branching (Fig. 2F).
SMs developed similarly in brl1 mutants and wild-type, initiating glumes and upper and lower FMs; the upper floret typically develops into a perfect flower with lemma, palea, anther, and carpel and the lower floret aborts (Doust and Kellogg, 2002b;Yang et al., 2018). In some cases, we observed aberrant meristematic outgrowths in brl1 FMs (Fig. 2L) which may explain our observations of additional 6 spikelets and bristles within some spikelets (Fig. 1, I and J). Our SEM analysis showed that a determinacy program was delayed in the IM, BMs, and SMs of brl1 mutants. We designed a dCAPs marker specific for this SNP and genotyped over 200 segregating F2 individuals. Our genotyping results showed that this SNP co-segregated with the barrel panicle phenotype at 100% (Supplemental Fig. S3). We also used RT-PCR to test whether expression of Sevir.2G006400 was disrupted in brl1 mutant inflorescence primordia. Results showed that at an early stage of inflorescence development, Sevir.2G006400 was expressed at a lower level in brl1 mutants compared with A10.1 (Supplemental Fig. S4). wheat, maize, and sorghum showed that SvFul2 was located in the FUL2 subclade along with three copies of wheat Ful2s, rice OsMADS15, and maize zap1 and zmm3 (Fig. 3A). SvFul2 is more closely related to SvFul1 (Sevir.9G087300) in the FUL1 subclade, which includes wheat VRN1s, rice OsMADS14, maize zmm15 and zmmads4. SvFul3 (Sevir.2G393300) and SvFul4 (Sevir.3G374401) are located in the FUL3 and FUL4 subclades, respectively. By examining a previously generated transcriptomics resource across six sequential stages of early S. viridis inflorescence development (Zhu et al., 2018), we found that SvFul1, SvFul2, and SvFul3 shared similar spatiotemporal expression patterns, increasing during branching and then decreasing during floral development with a small drop during spikelet specification ( Fig. 3B). SvFul1 was expressed highest at 10 and 12 DAS and SvFul2 expressed more at later stages, which indicate the two may have different functions. Comparatively, SvFul4 was expressed lower throughout inflorescence development, its expression gradually decreasing after the reproductive transition.

Gene editing of SvFul2 validates the mutant phenotype in S. viridis
To validate that Sevir.2G006400 (SvFul2) is responsible for the observed phenotypes of the brl1 mutant, we used genome editing. A CRISPR/Cas9 construct was designed containing two guide (g)RNAs that specifically targeted the first exon of SvFul2 in the highly transformable S. viridis accession, ME034 ( We further tested the allelic relationship between the brl1 mutant in the A10.1 background and SvFul2_KO in ME034 by crossing brl1 with SvFul2_KO. The resulting bi-allelic F1 individuals displayed a barrel phenotype and failed to complement, indicating that they are allelic (Fig. 5). Taken together, our analyses support SvFul2, Sevir.2G006400, as the locus responsible for the brl1 mutant phenotypes in S.

Loss of SvFul2 function alters expression of flowering and meristem determinacy pathways
To determine the molecular mechanisms underlying the complex phenotypes of the Svful2 mutant, we used RNA-seq to profile gene expression in mutant inflorescence primordia across three key developmental transitions and compared them to equivalent stages in wild-type controls: right before (stage 1) and after (stage 2) the floral transition, and during the initiation of spikelet specification (stage 3). Here, we expect to capture transcriptional changes related to both differences in flowering time and meristem determinacy. For each stage, we profiled four biological replicates, each consisting of pooled, 8 hand-dissected inflorescence primordia. Differential expression was determined using DESeq2 (1.22.2).
Our analysis found 382, 2,584, and 2,035 differentially expressed genes (DEGs) at stages 1, 2, and 3, respectively. Based on Principal Components Analysis (PCA), we observed fewer differences in the mutant transcriptome at stage 1, suggesting that the main influence of SvFul2 on inflorescence development begins once the SAM has initiated transition to the IM (Fig. 6A). We also observed dynamic shifts in DEGs among the three stages; only 33 DE genes were shared across all three stages, and 149, 451, and 68 were shared between stage 1 and 2, stage 2 and 3, and stage 1 and 3, respectively (Fig. 6B).
This suggests that SvFul2 potentially modulates different target genes in various spatiotemporal contexts.
As expected, the SvFul2 gene itself was significantly down-regulated in mutant inflorescences at all three stages (Fig. 6C). The other three S. viridis AP1/FUL-like genes were significantly up-regulated in the mutant, suggesting that the four AP1/FUL-like genes may provide some level of functional compensation during inflorescence development. Two B-class genes, SvMads16 (AP3) and SvMads4 (PISTILLATA), which in grasses are typically expressed at low levels prior to floral organ development (Whipple et al., 2004), were up-regulated in Svful2 stage 2 inflorescences (Fig. 6C). In addition, two Eclass genes were differentially expressed in mutant inflorescences: SvMads34 was up-regulated at stages 1 and 2, while SvMads5 was down-regulated at stage 2 ( Fig. 6C). In rice, OsMADS34 coordinates with AP1/FUL-like genes, and physically interacts with some of them, to specify inflorescence meristem identity (Kobayashi et al., 2012). In general, E-class genes play partially redundant roles in specifying floral organ identities via protein-protein interactions with other MADS box proteins (Pelaz et al., 2000;Honma and Goto, 2001;Theißen and Saedler, 2001;Ditta et al., 2004).
As a consequence of increased meristem indeterminacy, Svful2 mutant inflorescences branch more. We also found that genes associated with "anatomical structure formation involved in morphogenesis" (GO:0048646) were overrepresented among DEGs at stages 1 (p.adj = 0.0027) and 2 (p.adj = 3.56e -08 ; Fig. 6E), consistent with enhanced expression of genes involved in organogenesis.

Transcriptional rewiring by perturbation of SvFul2 reveals sub-networks connecting reproductive transition and determinacy pathways
To further investigate how SvFul2 connects within a larger gene network to regulate flowering time and meristem determinacy pathways, we used a computational strategy based on weighted gene co-expression network analysis and a random forest classifier to construct a gene regulatory network (GRN) representing normal inflorescence development in S. viridis (A10.1). Here, we integrated RNA-seq data from a previous study that captured precise stages of A10.1 inflorescence primordia spanning the IM transition to the development of floral organs (Zhu et al., 2018) with the staged wild-type data collected in this study. Using the WGCNA algorithm (Langfelder and Horvath, 2008) we clustered 26,758 genes into 27 co-expression modules ( Fig. 7A; Supplemental Fig S7). Module eigengenes (expression pattern that best fits an individual module) were evaluated for their significant associations with four key developmental events represented in the network: the vegetative-to-reproductive transition (8 and 10 DAS), branching (11, 12, and 14 DAS), meristem determinacy (15-17 DAS), and flower development (18 DAS) (Fig. 7A). Within each module, we tested for enrichment of genes that were differentially expressed in the Svful2 mutant, and found several that showed enrichment at certain stages of development (Fig.   7A). Among these, the module eigengene MEmagenta showed a strong positive correlation with the floral transition and a negative correlation with meristem determinacy (Fig. 7A-B). MEmagenta showed enrichment for DEGs in stages 1 and 2 (Fig. 7A). Alternatively, MEbrown showed a positive correlation with meristem determinacy, but was negatively correlated with the floral transition ( Fig. 7A-B). SvFul2 and SvFul3 were both co-expressed in the brown module, along with 428 DEGs largely at stages 2 and 3.
We also integrated the co-expression network with information derived from regulatory interactions among TFs and their putative targets based on the GENIE3 algorithm (Huynh-Thu et al., 2010). This complementary approach helped us to resolve the directionality and connectivity of important hub genes within the GRN. We selected TFs expressed in our dataset (n = 1,295) based on PlantTFDB (Jin et al., 2017) and used their trajectories in the network to derive points of connection with their potential target genes. Regulatory genes and their predicted targets were restricted based on information from differential expression analysis between wild-type and the Svful2 mutant. We used the resulting regulatory framework to explore functional relationships between SvFUL2, its predicted direct targets, and predicted upstream regulators, particularly in the context of connecting floral transition and meristem determinacy. Our predictions indicated that SvFUL2 controls several co-expressed TFs previously implicated in developmental processes, and localized to modules that positively associated with branching/meristem determinacy and negatively associated with floral transition (Fig. 7D). Among these were SvRA2, and an INDETERMINATE DOMAIN (IDD) TF, several members of which have been involved in both the floral transition and determinacy, including the founding member from maize, indeterminate 1 (id1) (Colasanti et al., 1998;Kozaki et al., 2004)).
Our analyses point to a possible feedback loop mechanism between SvFUL2 and SvRA2, where SvFul2 is also a predicted direct target of SvRA2. We also observed putative feedback regulation between SvFUL2 and TFs encoded by the orthologs of maize knotted 1 (kn1; TALE TF, Sevir.9G107600) and fasciated ear 4 (fea4; bZIP TF, Sevir.4G119100), which promote meristem maintenance and differentiation, respectively (Bolduc et al., 2012;Pautler et al., 2015). Extensive feedback interactions among these developmental TFs could represent endogenous mechanisms for fine-tuning developmental processes during the floral transition and the development of the inflorescence. In maize, KN1 was shown by ChIP-seq to bind a maize paralog of SvFul2, zap1 (Bolduc et al., 2012). Interestingly, several other MADS-box TFs were shown to directly target SvFul2 based on predictions in our GRN: SvMADS37.1 (Sevir.6G230800), SvMADS56 (Sevir.9G347400), SvMADS5 (Sevir.4G060800), SvMADS34/PAP2 (Sevir.9G087100) and SvFUL1 (Fig. 7D). SvMads34/Pap2 was also predicted to be a direct target of inflorescence phenotypes reported for A-class genes in any panicoid species, which include major cereal and energy crops. Therefore, we know little about their specific functions in regulating important agronomic traits such as flowering time and inflorescence determinacy. In this study using S. viridis as a model, we characterized a loss-of-function mutant in an A-class gene, SvFul2, that displayed strong developmental phenotypes, which was unexpected for a single mutant allele. Our morphological and molecular analyses of the Svful2 mutant provide insights into the roles of A-class genes in connecting flowering time and inflorescence determinacy in panicoid grasses, as well as predictions on conserved and novel regulatory interactions underlying the complex phenotypes.

SvFul2 is necessary for proper timing of flowering and determinacy programs.
Phylogenetic studies have reconstructed the evolutionary history of AP1/FUL-like genes in angiosperms (Litt and Irish, 2003;Preston and Kellogg, 2006;Soltis et al., 2007;Wu et al., 2017). The monocot The shift from a determinate to indeterminate fate in the IM of Svful2 mutants, which was also observed in the wheat vrn1-null; ful2-null double mutant, is reminiscent of the tfl1 mutant phenotype in Arabidopsis (Shannon and Meeks-Wagner, 1991). Previous studies that examined spatiotemporal 13 expression of FUL1/VRN1/OsMADS14 and FUL2/OsMADS15 in phylogenetically disparate grasses, showed that they are most abundantly expressed in the tip of the IM (Preston and Kellogg, 2007). In both Svful2 and wheat vrn1-null; ful2-null mutants, significant increases in the expression of TFL1 homologs were detected ( Fig. 6; (Li et al., 2019)). These results suggest that the mechanism for controlling IM determinacy in grasses involves an antagonism between AP1/FUL-like genes and TFL1-like genes, as in eudicots. IM determinacy appears to be very sensitive to the activity of AP1/FUL-like genes. In wheat, complete loss of both VRN1 and FUL2 function leads to an indeterminate IM, while a single functional copy of VRN1 or FUL2 in a heterozygous state was able to recover a determinate IM. It has been proposed that indeterminate growth in the IM was derived from a determinate habit in evolution, which involved the modification and/or loss of an early common TFL1 mechanism (Bradley et al., 1997). This hypothesis could explain this apparent sensitivity.
The strong phenotypes we observed in S. viridis by a single knockout of an AP1/FUL-like gene indicates its central role in controlling multiple developmental processes. Interestingly, the co-expression of closely-related paralogs, SvFul1 and SvFul3, with SvFul2 does not seem to provide much functional compensation, but we did see both genes up-regulated upon SvFul2 perturbation. SvFul2 was expressed at high levels (highest among other AP1/FUL-like genes) at most of the developmental stages we examined.
The functional redundancy of AP1/FUL-like genes in grasses provides an opportunity for diversification of function, and a toolkit for fine-tuning development of desired traits. In our study, the important role of SvFul2 in coordinating flowing time and meristem determinacy is not only supported by its strong pleiotropic phenotypes, but also reflected in our predictions of regulatory relationships between SvFUL2 and its upstream modulators and downstream targets. Several MADS-box TFs, most of which are homologs to those implicated in flowering time, were predicted to directly target SvFul2 (Fig. 7D). Our network analysis also uncovered potential feedback regulation between SvFUL2 and SvRA2, which could point to a conserved mechanism by which flowering links to AM determinacy in grasses. SvRa2 is the ortholog of maize ra2 and barley Vrs4 (Six-rowed spike4) Over-expression of the maize AP1/FUL-like gene, zmm28, enhanced grain yield potential through improved photosynthetic capacity and nitrogen utilization (Wu et al., 2019). In that study, they integrated RNA-seq with ChIP-seq analyses and revealed direct targets of ZMM28, which included genes involved in photosynthesis and carbohydrate metabolism. These analyses were performed in leaf tissue. Homologs of several of these targets were differentially expressed in Svful2 mutants, including photosystem I light harvesting complex gene 6 (Sevir.2G22720), a gene encoding a pyruvate orthophosphate dikinase (Sevir.3G253900), and gene encoding a bZIP TF (Sevir.3G396500). Although SvFul2 encodes a different AP1/FUL-like gene in a different spatiotemporal context, we also observed changes in genes associated with photosynthesis and with sugar and starch metabolism in stage 3 inflorescences where the mutant was highly indeterminate compared to wild-type. There could be common regulatory interactions between AP1/FUL2-like genes associated with photosynthesis, carbon allocation and sugar signals that link flowering time cues from the leaf to inflorescence architecture. We know little about the mechanisms by which sugar signals interface with development, but clear links, for example with trehalose-6-phosphate, The striking phenotype displayed in loss-of-function Svful2 mutants enables us to more clearly define molecular connections between flowering time and various aspects of inflorescence meristem determinacy. One question that comes to mind is why do the pathways regulated by SvFul2 in S. viridis have fewer checks and balances in terms of functional redundancy compared to other grasses? Since S. viridis is an undomesticated weed, one hypothesis is that selection against indeterminacy phenotypes in inflorescences of modern cereal crop species masks the ability to recover individual functions of A-class genes at the phenotypic level. Furthermore, perhaps the phenotypes presented in Svful2 mutants provide plasticity in S. viridis's adaptability to a wide range of environmental conditions. In any case, our analyses of this mutant provide a glimpse into AP1/FUL-like gene function in panicoid grasses and potential regulatory interactions with known players that underlie yield potential across important cereal crops.

Plant materials and growth conditions
The brl1 mutant allele was isolated from an NMU mutagenized M2 population of S. viridis (Huang et al., 2017

Scanning Electron Microscopy analysis
For SEM analysis, brl1 mutant and wild-type inflorescence primordia were harvested from young seedlings to examine the developmental defects of mutants. Samples were fixed, hand-dissected, and dehydrated as described (Hodge and Kellogg, 2014). The dehydrated samples were critical point dried using a Tousimis Samdri-780a and imaged by a Hitachi S2600 SEM at Washington University's Central Institute of the Deaf.

Histology
Wild-type and mutant inflorescence primordia were harvested right after the vegetative-to-reproductive transition at 11 and 15 DAS, respectively. The samples were fixed, embedded, and sectioned as described by (Yang et al., 2018). Sections (10 µm) made with a Microm HM 355S microtome (ThermoFisher, Waltham, MA, USA) were deparaffinized, stained with eosin, and imaged with a Leica M125C LED microscope.

Bulked Segregant Analysis
M3 mutant individuals were crossed to the A10.1 reference line and resulting F1 individuals were selfpollinated to generate segregating F2 families. The F2 individuals with mutant and wild-type phenotypes were identified, and the segregation ratio was tested by a χ2 test. DNA extracted from 30 brl1 mutant individuals was pooled to generate a DNA library. The DNA library was made using the NEBNext Ultra DNA Library Prep Kit for Illumina (NEB), size selected for inserts of 500 to 600 bp, and sequenced using 100bp single-end using standard Illumina protocols on Illumina Hi-Seq 2500 platform at the University of Illinois, Urbana-Champaign W.M. Keck sequencing facility. Read mapping and SNP calling were performed as described (Huang et al., 2017).

CRISPR/Cas9 gene-editing
The genome sequence of SvFul2 (Sevir.2G006400) was obtained from the S. viridis v2.1 genome (https://phytozome.jgi.doe.gov/). CRISPR-P v2.0 (Liu et al., 2017) was used to design guide (g)RNAs to minimize off-targets. Two gRNAs targeting SvFul2 were designed at the first exon and the first intron, 133bp and 395bp downstream of the ATG start codon, respectively. Using a plant genome engineering toolkit (Čermák et al., 2017), gRNAs were combined into a level 0 construct followed by insertion into a plant transformation vector. PCR amplified fragments from pMOD_B_2303 were merged using goldengate cloning with T7 ligase and SapI/BsmBI restriction enzymes back into the pMOD_B_2303 backbone to express the two gRNAa from the CmYLCV promoter, each flanked by a tRNA. This construct, along with pMOD_A1110 (a wheat codon-optimized Cas9 driven by the ZmUbi1 promoter) and pMOD_C_0000 modules, were combined in a subsequent golden-gate cloning reaction with T4 ligase and AarI restriction enzyme into the pTRANS_250d plant transformation backbone. The final construct was cloned into Agrobacterium tumefaciens line AGL1 for callus transformation of S. viridis ME034 at the DDPSC Tissue Culture facility. T0 plantlets were genotyped for the presence of the selectable marker, hygromycin phosphotransferase (HPT) to validate transgenic individuals. In the T1 generation, individual plants with possible mutant phenotypes were selected and the region of the target sites was amplified using PCR and sequenced. A homozygous 540 bp deletion in the 1st exon of SvFul2 was identified. These T1 mutants were self-pollinated to obtain T2 progeny and outcrossed to ME034 and then selfed to select Cas9-free SvFul2_KO plants. Primer sequences used for vector construction and genotyping are listed in Table S2.

RNA-seq library construction, sequencing, and analysis
Poly-A + RNA-seq libraries were generated from pools of hand-dissected inflorescence primordia from wild-type and brl1 mutant seedlings. Wild-type primordia were sampled at 8, 11, and 17 DAS while, accounting for the mutant's developmental progression, brl1 primordia were sampled at 9, 15, and 21 DAS. For each developmental stage, four biological replicates were collected, for a total of 24 data points.
RNA was extracted (PicoPure RNA isolation kit; Thermo Fisher Scientific) and subjected to library preparation from 500ng of total RNA using the NEBNext Ultra Directional RNA Library Prep Kit (Illumina), size-selected for 200bp inserts, and quantified on an Agilent bioanalyzer using a DNA 1000 chip. RNA-seq libraries were processed using an Illumina HiSeq 4000 platform at Novogene with a 150bp paired-end sequencing design. On average, for each data point ~20 million cleaned reads were generated. RNA-seq reads were quality checked and processed using the wrapper tool Trim Galore

Weighted Gene Co-expression Network Analysis
In addition to the samples described above, we included previously described wild-type S. viridis inflorescence primordia samples (Zhu et al., 2018): 23 additional data points from six inflorescence stages (10,12,14,15,16,and 18 DAS). This dataset (GSE118673) was re-processed using the same methods described above and used to build a reference wild-type gene co-expression network spanning S. viridis inflorescence organogenesis, from the transition to reproductive phase to flower development. To reduce samples bias, we first filtered out genes with less than 10 counts (row sum ≤ 10), then we calculated the Euclidean distance and Perason's correlation among samples and removed all replicates with rho coefficient < 0.92 or with an Euclidean score < 0.8. Based on this, two samples were These TFs were used as probes to predict regulatory links between the putative targets and their expression trajectories in our dataset. We ran GENIE3 with the parameters 'treeMethod = "RF", nTrees = 1,000' and putative target genes were selected with a weight cutoff ≥ 0.005. Networks were explored and plotted using the R package iGraph.

Data Accessibility
Data are deposited at NCBI SRA BioProject ID PRJNA649815 and will be released upon publication.            auxin polar transport auxin-activated signaling pathway regulation of auxin mediated signaling pathway gibberellic acid mediated signaling pathway gibberellin biosynthetic process brassinosteroid biosynthetic process salicylic acid catabolic process response to abscisic acid development anatomical structure formation in morphogenesis animal organ development maintenance of inflorescence meristem identity meristem initiation meristem structural organization stem cell development regulation of meristem growth regulation of flower development flower morphogenesis polarity specification of adaxial/abaxial axis determination of bilateral symmetry negative regulation of cell differentiation pattern specification process photomorphogenesis response hyperosmotic response response to sucrose response to cadmium ion response to cold response to red or far red light response to water deprivation sugar and starch pentose-phosphate shunt glucose catabolic process glucosinolate biosynthetic process glycogen biosynthetic process glycolytic process glycoside biosynthetic process glycoside metabolic process maltose metabolic process amylopectin biosynthetic process starch biosynthetic process starch catabolic process
(A) PCA analysis showed that biological replicates were well-correlated with each other and that PC1 (explaining 56% of the variance) was associated with developmental stage. Loss-of-function in Svful2 resulted in fewer transcriptional changes prior to the floral transition with larger changes between genotypes becoming evident after.
(B) Differential expressed genes showed dynamic transcriptional changes in Svful2 mutants at three stages of inflorescence development.
Among differentially expressed genes were several encoding MADS-box TFs (C) and known regulators of flowering time (D).
TPM values were Log2 transformed to generate heatmaps. Yellow and black asterisks indicate up-and down-regulated DEGs (FDR < 0.05), respectively.
(E) Subsets of GO terms that were overrepresented among DEGs at each of the three developmental stages. p.adj < 0.05.