Quantitative Genetics Identifies Cryptic Genetic Variation Involved in the Paternal Regulation of Seed Development

Embryonic development requires a correct balancing of maternal and paternal genetic information. This balance is mediated by genomic imprinting, an epigenetic mechanism that leads to parent-of-origin-dependent gene expression. The parental conflict (or kinship) theory proposes that imprinting can evolve due to a conflict between maternal and paternal alleles over resource allocation during seed development. One assumption of this theory is that paternal alleles can regulate seed growth; however, paternal effects on seed size are often very low or non-existent. We demonstrate that there is a pool of cryptic genetic variation in the paternal control of Arabidopsis thaliana seed development. Such cryptic variation can be exposed in seeds that maternally inherit a medea mutation, suggesting that MEA acts as a maternal buffer of paternal effects. Genetic mapping using recombinant inbred lines, and a novel method for the mapping of parent-of-origin effects using whole-genome sequencing of segregant bulks, indicate that there are at least six loci with small, paternal effects on seed development. Together, our analyses reveal the existence of a pool of hidden genetic variation on the paternal control of seed development that is likely shaped by parental conflict.


INTRODUCTION 69
Post-fertilisation development is a complex process that involves dynamic 70 interactions between maternally and paternally derived genomes. A correct balancing 71 of parental genomes is essential for embryonic development, and disruptions of this 72 balance (e.g. by crossing individuals with different ploidies) often lead to embryo 73 inviability [1][2][3][4][5][6]. Genomic imprinting, an epigenetic mechanism that leads to 74 differential expression of alleles in a parent-of-origin-dependent manner, is 75 responsible for many parental asymmetries during embryo and seed development in 76 mammals and flowering plants [7,8]. 77 Transcriptome profiling of developing seeds has revealed the existence of hundreds of 78 candidate imprinted genes in the embryo and/or endosperm, a biparental nourishing 79 tissue that derives from a second fertilisation event (reviewed in [9][10][11]). However, 80 the functional role of genomic imprinting is still a matter of considerable theoretical 81 debate [12]. The parental conflict (or kinship) theory of genomic imprinting proposes 82 that imprinting can evolve as the manifestation of a conflict of interests between 83 maternal and paternal alleles over resource allocation during embryogenesis or seed 84 development [13][14][15]. This conflict arises due to the asymmetric genetic relatedness 85 between maternal and paternal alleles in polyandrous (multiple paternity) species, 86 where maternal alleles are more likely to be shared between siblings than paternal 87 alleles. 88 The parental conflict theory is supported not only by mutant phenotypes in mice [16- (referred to as Arabidopsis hereafter). Only the maternal MEA allele is expressed 92 (before fertilization in the embryo sac that contains the female gametes and later in RESULTS 114 mea seeds can be paternally rescued 115 When mea ovules are pollinated with wild-type pollen from the Landsberg erecta 116 accession (hereafter referred to as Ler), seeds undergo excessive cell proliferation and 117 abort before completing embryogenesis [21]. However, mea ovules pollinated with 118 pollen from other Arabidopsis accessions (such as Cvi-0 or C24) can give rise to 119 viable plump mea seeds (Fig. 1A, S1 Figure). To dissect the relative paternal and 120 maternal contributions to mea seed rescue, we introgressed mea-2 (originally in the  Since MEA is an imprinted gene (only its maternal allele is expressed) a potential 145 explanation for mea seed rescue could be an activation of the paternal wild-type MEA 146 allele. However, this hypothesis cannot easily be tested using allele-specific 147 expression assays, because maternal mea mutations already induce low levels of 148 paternal MEA expression [20,[31][32][33]. To determine genetically if the paternal MEA 149 allele is required for mea seed rescue, we examined the F2 progeny of crosses 150 between mea-2 and different Arabidopsis accessions. If a paternal MEA allele was 151 required for the rescue, we would not expect to recover viable homozygous mea/mea 152 seeds. However, we recovered 9-20% of viable mea/mea plants in the F2 progeny of 153 crosses with the accessions C24, Hs-0, and Lomm1-1 (S1 Table). This result clearly 154 indicates that a paternal MEA allele is not required for mea seed rescue in these 155 crosses. In the crosses with Cvi-0 we only recovered 3% of homozygous mea/mea 156 seeds; when these different F2 homozygous Cvi-0 mea/mea individuals were self-crossed heterozygous fis2/FIS2 and fie/FIE plants with pollen from Ler, Cvi-0, and 166 C24 (Fig. 1C). While Cvi-0 could rescue fis2 seeds, there was no significant seed 167 rescue using C24 pollen. fie seeds could not be rescued by pollen of either Cvi-0 or 168 C24. These results indicate that the mea seed paternal rescue does not simply occur at 169 the FIS-PRC2 level; rather it supports the hypotheses that the different FIS-PRC2 170 subunits play distinct roles [34] and that MEA participates in multiple protein 171 complexes during seed development [32]. To determine the extent of species-wide variation on MEA-dependent parental 176 interactions, we pollinated 164 Arabidopsis accessions with mea-2 to generate F1s 177 that were allowed to self-fertilize to examine seed viability rates in the F2 generation.

178
Each of the F2 populations segregates (1) MEA and mea, and (2) chromosomes from 179 Ler and the respective parental accessions: therefore, we expected to obtain 50% 180 viable seeds from accessions that do not modify the penetrance of mea (such as Ler), 181 and up to 75% viable seeds from accessions with a strong paternal rescue effect 182 (assuming no epistatic effects). Accordingly, we observed 52% plump seeds in the 183 control Ler crosses, while in the Cvi-0 and C24 crosses there were 65% and 68% 184 plump seeds, respectively ( Fig. 2A-C). Roughly half of the accessions tested showed 185 between 55% and 70% plump seeds, suggesting that alleles that modify the 186 penetrance of mea seed abortion are widespread among natural Arabidopsis  conditions (Cvi-0, C24, Se-0, Ts-1 and Co) (S3 Figure). We did not find a correlation 208 between mea rescue and the size of self-fertilized or outcrossed Arabidopsis seeds 209 [30,36]. 210 We performed a genome-wide association study (GWAS) to identify regions in the 211 genome whose species-level variation is linked to mea rescue. However, we were 212 unable to detect clear statistically significant associations (Fig. 2D), likely due to the 213 weak power of GWAS to detect polygenic traits with low effect sizes [37]. 214 Nevertheless, some of the most highly associated SNPs were in the vicinity of the 217 QTL analyses identify six loci that contribute to the rescue of mea seeds 218 We crossed homozygous mea/mea plants (generated using an inducible MEA-219 glucocorticoid receptor system) with pollen from 80 Cvi-0/Ler recombinant inbred 220 lines (RILs) for which a detailed genetic map is available [38]. The percentage of 221 plump seeds that originated from these crosses followed a continuous distribution 222 (Fig. 3A), indicating that the rescue of mea seeds is a polygenic trait. The broad-sense 223 heritability H 2 (the percentage of total phenotypic variance that can be explained by 224 genetic factors) is 85%, indicating that mea seed rescue is under strong genetic 225 control. We used maximum likelihood standard interval mapping to identify regions 226 that are significantly associated with mea seed rescue. As expected from the 227 continuous phenotype distribution, we identified multiple QTL peaks on several proportion (5-11%) of the overall phenotypic variation ( Figure 3D and Table 1).

235
Nevertheless, the effect of multiple QTLs increases exponentially: every additional 236 Cvi-0 QTL increases the rescued seed rate by roughly 50% (e.g. pollen donors with 237 two, three or four Cvi-0 QTLs generate on average 18%, 28% or 42% plump seeds, 238 respectively) (Fig. 3E).   isogenic Cvi-0 fragments. While Cvi-0 pollen gave rise to 85% mea plump seeds, 264 almost all the NILs showed no significant differences from Ler (3% viable seeds) 265 (Fig. 4A). The three NILs that clearly showed an effect (13-23% viable seeds) 266 actually contain multiple Cvi-0 fragments that overlap two or three of the identified 267 QTLs (Fig. 4B). Together, the RIL and NIL analyses suggest the existence of at least 268 six loci in Cvi-0 that contribute to the rescue of mea seeds. We also scored seed abortion in the F2 progeny of a cross between mea-2 and C24 282 (genotyped at 14 markers throughout the genome). Despite the low statistical power 283 caused by the segregation of MEA in this population, we found evidence for one QTL 284 at the bottom of chromosome 1 that could explain 15% of the observed phenotypic 285 variation (S4 Table). Thus, even this analysis with limited power identified one of the 286 loci on chromosome 1 that was mapped using RILs and NILs. To independently validate the results of the QTL analyses, we developed a novel 291 method for mapping parent-of-origin effects using whole-genome sequencing. The 292 strategy is to create an F2 population that contains one set of chromosomes from one 293 parent but inherits two segregating sets from the other parent. These two sets should 294 have opposing effects in pre-or post-fertilisation fitness or viability, so that they will 295 not be equally transmitted. DNA is then extracted from pools of viable F2 seedlings, 296 and whole-genome sequencing is used to identify genomic regions that exhibit biased 297 transmission of the two segregating paternal (or maternal) genotypes.

298
In this case, we took advantage of the differential survival of mea seeds depending on 299 the inheritance of Cvi-0 against Ler paternal alleles. First, we generated F1 hybrid 300 plants by reciprocally crossing Ler and Cvi-0 plants (Fig. 5A). The Ler/Cvi-0 hybrids 301 were then used to pollinate (1) Ler plants and (2)  to be close to 25%; in the 'mea pool', regions that are associated with mea rescue will 315 be enriched in Cvi-0 reads (up to 50%).  Figure 5C, Table 2). These peaks 341 were reproducible between the three biological replicates (S4 Figure). Each of the 342 peaks identified by Bulk-Seq is located in the vicinity of the QTLs identified by the 343 RIL-QTL analysis ( Table 1); some of the peaks (particularly b, d, and g) are also 344 close to SNPs that were identified by the GWAS analysis as associated (although non-345 significantly) with mea rescue (Fig. 2D). Taken together, the Bulk-Seq analysis 346 provides strong support to the existence and predicted location of the multiple Cvi-0 347 alleles that underlie the rescue of mea seeds.  growth.

363
We hypothesize that, in a maternal Ler background, the (potential) paternal growth Cvi-0 and C24 are not as strong and allow mea seeds to complete development.

369
Interestingly, paternal effects on mea seed development are, in turn, dependent on the 370 maternal genetic background: we showed that C24 paternal alleles can strongly rescue 371 mea seeds in a maternal Ler or C24 background, but this rescue is much weaker in a 372 Cvi-0 maternal background (Fig 1B). This indicates that there are multiple reciprocal 373 interactions between maternal and paternal alleles in the regulation seed growth.  overcome inter-specific hybridizations [72][73][74][75].

Plant material and growth conditions 445
The   replicates were generated, each using different parental individuals and at different 533 days. In total, around 10,000 and 4,000 F2 seeds were generated from mea/mea and 534 Ler mothers, respectively. Seeds were surface sterilised for 10 minutes using 1% 535 sodium hypochlorite, washed extensively with water, and sown in MS medium. After    Table). The 584 same quality filtering steps were performed in each of the three replicates. To 585 calculate the relative proportion of Cvi-0 and Ler reads across the genome, we first 586 summed the read counts of groups of 50 neighbouring SNPs across the genome using 587 a rolling window. We then calculated the proportion of Cvi-0 and Ler reads, and the 588 relative enrichment of Cvi-0 reads in the mea pool relative to the WT pool:

878
The 12 individuals are homozygotes from the F2 Cvi-0 x mea-2 population described in S1 879