A genome-wide association study reveals a novel regulator of ovule number and fertility in Arabidopsis thaliana

Ovules contain the female gametophytes which are fertilized during pollination to initiate seed development. Thus, the number of ovules that are produced during flower development is an important determinant of seed crop yield and plant fitness. Mutants with pleiotropic effects on development often alter the number of ovules, but specific regulators of ovule number have been difficult to identify in traditional mutant screens. We used natural variation in Arabidopsis accessions to identify new genes involved in the regulation of ovule number. The ovule numbers per flower of 189 Arabidopsis accessions were determined and found to have broad phenotypic variation that ranged from 39 ovules to 84 ovules per pistil. Genome-Wide Association tests revealed several genomic regions that are associated with ovule number. T-DNA insertion lines in candidate genes from the most significantly associated loci were screened for ovule number phenotypes. The NEW ENHANCER of ROOT DWARFISM (NERD1) gene was found to have pleiotropic effects on plant fertility that include regulation of ovule number and both male and female gametophyte development. Overexpression of NERD1 increased ovule number per fruit in a background-dependent manner and more than doubled the total number of flowers produced in all backgrounds tested, indicating that manipulation of NERD1 levels can be used to increase plant productivity.

During plant reproduction, pollen tubes deliver two sperm cells to female gametophytes 53 contained within ovules. This allows double fertilization to occur in order to produce the 54 embryo and endosperm in the developing seed. Angiosperms with all kinds of 55 pollination syndromes (insect-, wind-, and self-pollinated) produce much more pollen 56 than ovules in order to ensure successful pollination. For example, most soybean 57 varieties produce only 2 ovules per flower, but more than 3,000 pollen grains, a 1,500-58 fold difference(1). Wind-pollinated plants such as maize have an even more extreme 59 difference in pollen production vs. ovule production per plant, with more than 1 million 60 pollen grains versus an average of 250 ovules per plant (a 4000-fold difference (2)). 61 Arabidopsis thaliana, which is a self-pollinating plant, also produces an excess of pollen, The ability to manipulate ovule number to increase the reproductive potential of plants 67 requires an understanding of the molecular pathways that control ovule initiation. 68 The model plant Arabidopsis thaliana produces flowers with four whorls of organs: 69 sepals, petals, stamens, and carpels. The inner whorls (3 and 4) are responsible for 70 sexual reproduction, with pollen (the male gametophytes) produced in the whorl 3 71 stamens and the female gametophytes (also known as the embryo sacs), produced in 72 ovules contained within the whorl 4 carpels. Specification of the 4 whorls is controlled by 73 the "ABC" genes, with the C-class gene AGAMOUS (AG) a major regulator of carpel 74 development(4). 75

76
In Arabidopsis, ovules are initiated from the carpel margin meristem (CMM) at stage 9 77 of floral development(5). The Arabidopsis gynoecium comprises two carpels that are 78 fused vertically at their margins (6). The CMM develops on the adaxial face of the 79 carpels (inside the fused carpel cylinder) and will give rise to the placenta, ovules, 80 septum, transmitting tract, style, and stigma. Once the placenta is specified, all of the 81 ovule primordia are initiated at the same time (7). Subsequently, each primordium will be 82 patterned into three different regions: the funiculus, which connects the ovule to the 83 septum; the chalaza, which gives rise to the integuments; and the nucellus, which gives 84 rise to the embryo sac. Ovule development concludes with the specification of the 85 megaspore mother cell within the nucellus which undergoes meiosis followed by three 86 rounds of mitosis to form the mature haploid embryo sac (reviewed in (8)). 87

88
In Arabidopsis, CMM development requires coordination of transcriptional regulators 89 involved in meristem function with hormone signaling (reviewed in (6)

Natural Variation in Ovule Number 153
We set out to identify new regulators of ovule number in Arabidopsis by taking 154 advantage of phenotypic variation in naturally occurring accessions. We obtained 189 155 Arabidopsis accessions from the ABRC and assayed them for variation in ovule number 156 (Table S1) In contrast to flowering time variation which has been shown to correlate with latitude of 165 origin in Arabidopsis accessions(21), ovule number was not strongly correlated with 166 location of origin in the accessions analyzed (Fig 1C, S1). Mapping ovule number data 167 onto a cladogram of the accessions used in our study revealed a cluster of low ovule 168 number accessions in one specific clade, indicating that these closely-related 169 accessions may have similar genetic control of ovule number (Fig 1D). Interestingly, 170 several clades were made up of accessions with high, medium, and low ovule numbers. 171 This suggests that the ovule number trait may be regulated by different loci that have 172 been selected for in some lineages. 173

GWAS reveals SNPs linked to natural variation in ovule number 175
In order to identify genomic regions linked to variation in ovule number, we assessed 176 whether the average ovule number per flower from 148 accessions was predicted by 177 Single Nucleotide Polymorphisms (SNP) available in the 1001 Full-sequence dataset; 178 TAIR 9. Logarithmic transformation was applied to the ovule number data to make the 179 results more reliable for parametric tests. Associations were tested for each SNP using 180 a linear regression model (LM) and the results were analyzed using GWAPP(22) ( Fig  181   2A). A significance cutoff value of -log10(p values) ≥ 6.2 identified at least 9 genomic 182 regions that are associated with variation in ovule number, while a higher cutoff of -183 log10(p values) ≥ 7.5 identifies only four significant genomic regions. 184

185
We next determined if known ovule number regulators (9, 15, 17, 23) colocalized with 186 our GWAS loci. Of the previously described loci, only BIN2 mapped close to a 187 significant association (Fig S2). This indicates that our GWAS has identified novel 188 functions for loci in the regulation of ovule number. For further analysis, we focused on 189 the two loci with the lowest p-values which are both located on the long arm of 190 chromosome 3 (Fig 2B-C). Genes containing the most significantly associated SNPs as 191 well as the 10 surrounding genes around the highest peak were considered as 192 candidates for regulating ovule number. These genes were further prioritized based on 193 whether they are expressed in developing pistils, by examining publicly available 194 transcriptome data in ePlant(24) (Fig S3). Based on this prioritization, 35 candidate 195 genes were selected, and of these 26 insertion mutants were available from the 196 Arabidopsis Biological Resource Center. All 26 mutants were evaluated for changes in 197 ovule number compared to the wild-type background, Col-0 (Table S2). Of these, two 198 had significantly reduced ovule number compared to the Col-0 control (Fig 3A, S4). The 199 strongest ovule number phenotype was found in insertion mutants in At3g51050, a gene We focused our analysis on NERD1 since it had the strongest effect on ovule number. 208 The fruits of homozygous mutants in two T-DNA insertion alleles, nerd1-2 (described in 209 (25) and nerd1-4, had fewer ovules than wild-type plants ( Fig 3A). To confirm that this 210 resulted from the disruption of NERD1, we generated transgenic plants expressing a 211 translational fusion of the NERD1 protein with the red-fluorescent protein tdTomato 212 under the control of the native NERD1 promoter. In nerd1-2 and nerd1-4 mutants, this 213 transgene fully rescued the reduced ovule phenotype and resulted in plants with ovule 214 numbers indistinguishable from Col-0 ( Fig 3A). This demonstrates that NERD1 is a 215 positive regulator of ovule number. performed reciprocal crosses between heterozygous nerd1-2 mutants and Col-0 to 224 determine if the fertility defects were gametophytic or sporophytic. When heterozygous 225 nerd1-2/NERD1 was used as the female, there was no transmission defect, 226 demonstrating that the reduced female fertility in nerd1-2 mutants was not 227 gametophytic. When nerd1-2/NERD1 was used as the pollen donor, the transmission 228 efficiency of the mutant allele was reduced to 45%, indicating a male gametophytic 229 transmission defect (Table 1). 230 231 232

NERD1 encodes an integral membrane protein 233
Phylogenetic analyses indicate that NERD1 is a member of a low-copy number, highly 234 conserved gene family that is found throughout the plant kingdom and in cyanobacteria 235 (25)). The NERD1 protein is predicted to be an integral membrane 236 protein with a signal peptide and one transmembrane domain ( Fig 5A). The majority of 237 the protein is predicted to be extracellular, with the transmembrane domain located near The nerd1 ovule number and fertility phenotypes suggest that NERD1 should be 246 expressed in developing flowers. We used a NERD1pro::gNERD1-GUS fusion to 247 examine NERD1 expression throughout Arabidopsis development. In 248 NERD1pro::gNERD1-GUS inflorescences, GUS activity was detected throughout flower 249 development, including inflorescences, developing and mature anthers, and in the 250 stigma, ovules, and carpel walls of mature pistils (Fig 6). NERD1-GUS activity was 251 present in the carpel margin meristem (CMM) in stage 9 flowers, where ovule initiation 252 occurs (Fig 6C and E). NERD1 reporter expression in the CMM during pistil 253 development is consistent with a role for NERD1 in ovule initiation. During seedling 254 development, the NERD1-GUS reporter was detected in shoot and root apical 255 meristems (SAM and RAM) and in the vasculature (Fig 6I). 256 257

Overexpression of NERD1 increases plant productivity 258
We examined NERD1 transcript levels in the low ovule number accession Altai-5 259 compared to Col-0. NERD1 transcript accumulation was reduced in Altai-5 buds as 260 compared to Col-0 but similar in Altai-5 and Col-0 leaves (Fig 7). We hypothesized that 261 the low ovule number in Altai-5 may be linked to reduced NERD1 expression in 262 developing flowers. In order to determine whether increasing NERD1 expression is 263 sufficient to increase ovule number, we transformed Col-0 and the low ovule number 264 accession Altai-5, with a NERD1-GFP fusion construct driven by the constitutively 265 expressed Cauliflower Mosaic Virus 35S promoter (35S::NERD1-GFP). Overexpression 266 of NERD1 had no effect on ovule number in the Col-0 background, but significantly 267 increased ovule number in the Altai-5 background (Fig 8A-C), indicating that the NERD1 268 effect on ovule number is background-dependent. The 35S::NERD1-GFP plants 269 displayed an even more striking phenotype when overall plant architecture was 270 examined (Fig 8A). In both the Altai-5 and Col-0 backgrounds, NERD1 overexpression 271 led to increased branching ( Fig 8D) and shortened internode lengths between flowers, 272 leading to an overall increase in flower number in the overexpression plants compared 273 to untransformed controls (Fig 8E-H). Thus, NERD1 overexpression leads to increased 274 biomass and reproductive capacity, with up to a 2.5-fold increase in total flower number 275 over the lifespan of the plant. While all independent transformants displayed increased 276 branching and flower number, some of the 35S::NERD1 plants were male sterile (Fig  277   S7). This male sterility correlated with NERD1 expression levels and plants with higher 278 NERD1 transcript levels had more severe male sterility (Fig S7). The sterility effect was 279 more severe in Col-0 than in Altai-5 ( Fig S7). The lower endogenous NERD1 280 expression in Altai-5 inflorescences might explain the lower sensitivity of Altai-5 to 281 NERD1 overexpression with respect to male fertility, demonstrating background-282 dependent sensitivity to NERD1 levels for both ovule number and male sterility. to the Col-0 control, indicating that root development is also sensitive to NERD1 289 expression levels ( Fig S8). 290 291 Three SNPs in NERD1 showed significant correlation with ovule number in our GWAS 292 (Fig 9A-B). Two of them are synonymous SNPs that are not predicted to change the 293 amino acid sequence, but the third is a non-synonymous SNP (C to A change in 294 comparison to Col-0 reference) causing a Serine to Tyrosine change at amino acid 230 295 of NERD1 (Fig 9A-B). This non-synonymous SNP was present in 10 out of the 16 296 lowest ovule number accessions and not present in the 16 highest ovule number 297 accessions ( Fig 9A). Across all of the accessions in our GWAS panel, the "A" allele at 298 this position was significantly associated with lower ovule numbers ( Fig 9C). However, 299 some accessions with the "C" allele of NERD1 have low ovule numbers, including the 300 Altai-5 accession (Fig 9A), underscoring that multiple mechanisms influence ovule 301 number in Arabidopsis.

GWAS reveals new ovule number-associated loci in Arabidopsis 313
A quantitative trait locus (QTL) mapping study utilizing variation between Ler and Cvi 314 identified QTLs residing on chromosomes 1, 2 (near the ERECTA gene), and two QTL 315 on chromosome 5(23). No follow-up study has identified the genes underlying these 316 QTLs. Our population displayed a much larger range of ovule numbers per flower (39-317 84) than that seen in Cvi vs Ler (66 and 56, respectively). We identified 9 significant 318 associations in our GWAS, but only one potentially overlaps with a known ovule number 319 QTL (see chromosome 5 in Fig S2). Of the genes with an identified effect on ovule 320 number, only BIN2 overlapped with a GWAS peak, indicating that our study has 321 revealed at least seven novel regulators of ovule number in Arabidopsis. We 322 molecularly identified two new genes that control ovule number that were linked to two 323 GWAS peaks on chromosome 3. This expands that number of known ovule number 324 determinants and confirms the value of GWAS as a forward genetics tool. The male fertility defect in nerd1 plants is more severe than the female defect. 389 Homozygous nerd1 anthers are empty, indicating that pollen development is not 390 initiated or aborted very early. Transmission efficiency tests using pollen from 391 heterozygous nerd1 plants crossed to wild-type females revealed that nerd1 also has 392 gametophytic effects on pollen function. The sporophytic effects could be related to 393 early stages of anther development. In particular, specification of the tapetum is critical 394 for pollen development (reviewed in (42)). In our 35S::NERD1 experiment, transformants 395 that accumulated the most NERD1 transcripts were male sterile. Together these results 396 indicate that pollen development is sensitive to NERD1 levels, i.e. either too much or 397 too little NERD1 is detrimental to pollen development. Future experiments should focus 398 on determining the stage of anther and/or pollen development that is affected in nerd1 399 mutants and the specific cell types that express NERD1 in developing anthers. in (46)). Gibberellins also repress shoot branching in Arabidopsis, maize, and rice, with 420 GA-deficient mutants displaying increased branching compared to wild-type controls(47-421 50). In contrast, cytokinins and brassinosteroids are both positive regulators of 422 branching(49-53). The mechanism through which these plant hormones work together 423 to regulate branching is unknown, but auxin has been proposed to control cytokinin and 424 strigolactone biosynthesis(54, 55), while the brassinosteroid signaling regulator BES1 425 may inhibit strigolactone signaling to promote branching(56). The promotion of 426 branching in 35S::NERD1 plants could be related to regulation of one or more of these 427 hormonal pathways. Like NERD1, upregulation of both cytokinin and brassinosteroid 428 signaling pathways have been shown to positively regulate branching and ovule 429 number (15,53,56,57), suggesting that NERD1 may be intimately connected to these 430 pathways. Future research is needed to explore the intersection between NERD1 and 431 hormonal pathways. 432 433

NERD1-induced increases in ovule number are background-dependent 434
Overexpression of NERD1 in the Col-0 and Altai-5 backgrounds led to increased 435 branching and flower number, but ovule number was only increased in the Altai-5 436 background, suggesting a background-dependence on the ovule number trait. In 437 Arabidopsis, natural accessions were shown to respond differently in their ability to 438 buffer GA perturbations caused by overexpressing GA20 oxidase 1, which encodes a 439 rate-limiting enzyme for GA biosynthesis(58). Genetic background dependence has 440 been shown to be a wide-spread phenomenon in C. elegans, with approximately 20% of 441

Genome-Wide Association Study 482
GWAS was performed using GWAPP, which is a GWAS web application for Genome-483 Wide Association Mapping in Arabidopsis (http://gwas.gmi.oeaw.ac.at) (22). In our 484 study, 148 accessions had single nucleotide polymorphisms (SNPs) data available on 485 the 1001 Full-sequence dataset; TAIR 9. Logarithmic transformation was applied to 486 make the results more reliable for parametric tests. A simple linear regression (LM) was 487 used to generate the Manhattan plot by using GWAPP(22). SNPs with P values ≤ 1 × 488 10 -6 were further considered as candidate loci linked to alleles that regulate ovule 489 number (a horizontal dashed line in Fig. 2 shows the 5% FDR threshold -log10p 490 value=6.2, which was computed by the Benjamini-Hochberg-Yekutieli method). SNPs 491 with <15 minor allele count (MAC) were not considered to help control false positive 492 rates. 10 genes flanking the highest SNP for each locus were tabulated as candidate 493 genes for each significant association. 494 495

Cloning and Generation of Transgenic Lines 496
For complementation and overexpression experiments, Gateway Technology was used 497 to make all the constructs. Genomic DNA fragments corresponding to the coding 498 regions of candidate genes were amplified from either beginning of the promoter 499 (defined by the end of the upstream gene) or the start codon to the end of the CDS 500 (without stop codon) by PCR with primers that had attB1 and attB2 sites from Col-0 501 genomic DNA (see supplementary table 3

for primer sequences). For amplifying 502
At3g60660 and At3g51050, PHUSION High-Fidelity Polymerase (NEB, M0535S) was 503 used. All the PCR products were put into entry vector pDONR207 by BP reactions and 504 then were recombined into destination vector pMDC83 (GFP) by LR reaction (61). 505 Native promoter constructs were amplified from ~2KB promoter region to the end 506 (without stop codon) by PCR with primers that had attB1 and attB2 sites from Col-0 507 genomic DNA. The PCR products were put into entry vector pDONR207 by BP 508 reactions and then were recombined into destination vector pMDC32 (tdTomato) and 509 The cladogram tree was generated in MEGA7, which nucleotide distance and neighbor-545 join tree file were calculated by PHYlogeny Inference Package (PHYLIP, version 3.696). 546 The phylogenetic tree of NERD1 was inferred using neighbor-joining method in 547          SNPs are identified based on comparison to the Col-0 reference genome. The red boxes highlight low-ovule number associated SNPs (the turquoise bar indicates two SNPs in adjacent nucleotides, a non-synonymous C-A SNP and a synonymous C-G SNP). Only genomes that were available on the SALK 1,001 genomes browser (http://signal.salk.edu/atg1001/3.0/gebrowser.php) were considered. (B) One of the ovule number-associated SNPs in NERD1 causes a serine-tyrosine substitution in low ovule number accessions. (C) Average ovule number in NERD1 "A" and "C" allelecontaining accessions. The "A" allele is significantly associated with lower ovule numbers. "***" indicates statistical significance at p value<0.001 determined by Student's t-test.