Genetic and environmental manipulation of Arabidopsis hybridization barriers uncovers antagonistic functions in endosperm cellularization

Speciation involves reproductive isolation, which can occur by hybridization barriers acting in the endosperm of the developing seed. The nuclear endosperm is a nutrient sink, accumulating sugars from surrounding tissues, and undergoes coordinated cellularization, switching to serve as a nutrient source for the developing embryo. Tight regulation of cellularization is therefore vital for seed and embryonic development. Here we show that hybrid seeds from crosses between Arabidopsis thaliana as maternal contributor and A. arenosa or A. lyrata as pollen donors result in an endosperm based post-zygotic hybridization barrier that gives rise to a reduced seed germination rate. Hybrid seeds display opposite endosperm cellularization phenotypes, with late cellularization in crosses with A. arenosa and early cellularization in crosses with A. lyrata. Stage specific endosperm reporters display temporally ectopic expression in developing hybrid endosperm, in accordance with the early and late cellularization phenotypes, confirming a disturbance of the source-sink endosperm phase change. We demonstrate that the hybrid barrier is under the influence of abiotic factors, and show that a temperature gradient leads to diametrically opposed cellularization phenotype responses in hybrid endosperm with A. arenosa or A. lyrata as pollen donors. Furthermore, different A. thaliana accession genotypes also enhance or diminish seed viability in the two hybrid cross-types, emphasizing that both genetic and environmental cues control the hybridization barrier. We have identified an A. thaliana MADS-BOX type I family single locus that is required for diametrically opposed cellularization phenotype responses in hybrid endosperm. Loss of AGAMOUS-LIKE 35 significantly affects the germination rate of hybrid seeds in opposite directions when transmitted through the A. thaliana endosperm, and is suggested to be a locus that promotes cellularization as part of an endosperm based mechanism involved in post-zygotic hybrid barriers. The role of temperature in hybrid speciation and the identification of distinct loci in control of hybrid failure have great potential to aid the introduction of advantageous traits in breeding research and to support models to predict hybrid admixture in a changing global climate.

Speciation involves reproductive isolation, which can occur by hybridization barriers acting in the endosperm of the developing seed. The nuclear endosperm is a nutrient sink, accumulating sugars from surrounding tissues, and undergoes coordinated cellularization, switching to serve as a nutrient source for the developing embryo. Tight regulation of cellularization is therefore vital for seed and embryonic development.
Here we show that hybrid seeds from crosses between Arabidopsis thaliana as maternal contributor and A. arenosa or A. lyrata as pollen donors result in an endosperm based post-zygotic hybridization barrier that gives rise to a reduced seed germination rate. Hybrid seeds display opposite endosperm cellularization phenotypes, with late cellularization in crosses with A. arenosa and early cellularization in crosses with A. lyrata. Stage specific endosperm reporters display temporally ectopic expression in developing hybrid endosperm, in accordance with the early and late cellularization phenotypes, confirming a disturbance of the source-sink endosperm phase change. We demonstrate that the hybrid barrier is under the influence of abiotic factors, and show that a temperature gradient leads to diametrically opposed cellularization phenotype responses in hybrid endosperm with A. arenosa or A. lyrata as pollen donors. Furthermore, different A. thaliana accession genotypes also enhance or diminish seed viability in the two hybrid cross-types, emphasizing that both genetic and environmental cues control the hybridization barrier. We have identified an A. thaliana MADS-BOX type I family single locus that is required for diametrically opposed cellularization phenotype responses in hybrid endosperm. Loss of AGAMOUS-LIKE 35 significantly affects the germination rate of hybrid seeds in opposite directions when transmitted through the A. thaliana endosperm, and is suggested to be a locus that promotes cellularization as part of an endosperm based mechanism involved in post-zygotic hybrid barriers. The role of temperature in hybrid speciation and the identification of distinct loci in control of hybrid failure have great potential to aid the introduction of advantageous traits in breeding research and to support models to predict hybrid admixture in a changing global climate.
The endosperm is a triploid tissue that requires tight control of genome dosage (2:1 maternal:paternal ratio). Cellularization of the endosperm marks a transition in seed development, as up to this point, the endosperm functions as a nutrient sink. At this developmental time point, the endosperm concurrently switches from nutrient sink to source for the developing embryo (Lafon-Placette and Köhler, 2016), and manipulating the timing of endosperm cellularization through interploidy crosses arrests embryo development (Scott et al., 1998;Hehenberger et al., 2012). Similarly, hybridization between plant species was shown to result in embryo arrest due to endosperm cellularization failure (Haig and Westoby, 1988;Haig and Westoby, 1991;Comai et al., 2000;Bushell et al., 2003). Manipulating the ploidy of parents in interspecies crosses has also shown to improve the success of hybridization, demonstrating a requirement for genome balance in the endosperm (Comai et al., 2000;Bushell et al., 2003;Lafon-Placette et al., 2017).
The strength of the endosperm-based hybridization barrier can be influenced by accession specific genetic variation (Burkart-Waco et al., 2012;Burkart-Waco et al., 2013;Burkart-Waco et al., 2015;Bjerkan et al., 2020). In crosses between diploid A. thaliana to diploid A. arenosa, the choice of accessions in both species significantly acts to repress or enhance the endosperm barrier (Bjerkan et al., 2020;Burkart-Waco et al., 2012). The current knowledge on the effect of temperature in early seed development is limited (Paul et al., 2020), but an effect of temperature on hybrid seed development in reciprocal crosses of wheat and barley has previously been reported (Molnaŕ-Lańg and Sutka, 1994). The sensitivity of endosperm cellularization to heat stress during early endosperm development has been demonstrated in rice (Folsom et al., 2014) and type I MADS-box transcription factors (TFs) are deregulated during moderate heat stress (Chen et al., 2016). We have previously reported a temperature effect on endosperm based post-zygotic hybrid lethality of diploid species in the genus Arabidopsis (Bjerkan et al., 2020). This temperature effect was also shown using different accessions of both A. arenosa and A. thaliana, demonstrating a combinatorial effect of accessions and temperature (Bjerkan et al., 2020;Burkart-Waco et al., 2012).
Genomic imprinting is an epigenetic phenomenon, which infers parent-of-origin allele specific expression of maternally or paternally inherited alleles (Hornslien et al., 2019;Batista and Kohler, 2020). As proper endosperm development depends on a correct ratio of parental genomes, it is suggested that differences in genomic imprinting programs may be responsible for the evolution of sexual incompatibility in crosses between divergent individuals (Haig and Westoby, 1988;Haig and Westoby, 1991;Bushell et al., 2003;Schatlowski and Kohler, 2012). Alternatively, or additionally, epigenetic remodeling upon hybridization due to combination of diverged maternal and paternal siRNAs may lead to comprehensive failure of genomic imprinting and ectopic expression of transposons and imprinted genes (Martienssen, 2010;Ng et al., 2012). Recent evidence supports this emerging role of imprinted genes, and in Arabidopsis interspecies hybrids, paternally expressed genes (PEGs) shifted to be maternally expressed genes (MEGs) (Josefsson et al., 2006). Importantly, both PEGs and MEGs have been shown to erect hybridization barriers, and mutational loss of these genes has been reported to bypass hybridization barriers in interspecies crosses (Walia et al., 2009;Wolff et al., 2015). A major part of the MEGs and PEGs encodes proteins that activate pathways in the endosperm consistent with the prominent role of cellularization in seed survival.
Proper endosperm development in Arabidopsis is reliant on the FERTILIZATION INDEPENDENT SEED-Polycomb Repressive Complex 2 (FIS-PRC2) (Grossniklaus et al., 1998;Kiyosue et al., 1999;Luo et al., 1999;Köhler et al., 2003). FIS-PRC2 is important in endosperm development indirectly through the genes it regulates, which include several type I MADS-box TFs (Zhang et al., 2018). Deregulation of type I MADS-box TFs in interspecies crosses has been postulated to induce the endosperm-based hybridization barrier, but unfortunately most of these TFs have no clear function because of extensive genetic redundancy (De Bodt et al., 2003;Parrěnicováet al., 2003;Walia et al., 2009;Bemer et al., 2010). One exception is AGAMOUS-LIKE 62 (AGL62), which is found to suppress cellularization of the endosperm in Arabidopsis (Kang et al., 2008). The clear function of AGL62 is further emphasized through its interaction with the FIS-PRC2 complex (Hehenberger et al., 2012), with mutation of FIE, MEA, FIS2 and MSI1 resulting in an ectopic proliferation of nuclear endosperm (Grossniklaus et al., 1998;Köhler et al., 2003;Guitton et al., 2004), whereas the agl62 mutant results in precocious cellularization (Kang et al., 2008). AGL62 mutation has also been shown to alleviate the hybridization barrier in the A. thaliana × A. arenosa cross, resulting in a higher germination rate (Walia et al., 2009;Bjerkan et al., 2020).
Here we report that hybrid seeds from crosses between A. thaliana mothers with A. arenosa or A. lyrata pollen donors result in diametrically opposed endosperm phenotypes, both giving rise to reduced seed germination rate, albeit caused by late cellularization in crosses with A. arenosa and early cellularization in crosses with A. lyrata. We demonstrate that the hybrid barriers are under the influence of abiotic factors, and show that a temperature gradient leads to opposed cellularization phenotype and seed viability in hybrid endosperm with A. arenosa or A. lyrata as pollen donors. In addition, A. thaliana accession genotypes also influence seed viability in the two hybrid cross-types in opposite directions. Using stage specific endosperm reporters, we demonstrate that the source-sink endosperm phase change is delayed or precocious in seeds of the two hybrids. Our data suggests an A. thaliana type I MADS-BOX family locus to act as a promoter of endosperm cellularization, affecting the germination rates of A. arenosa or A. lyrata hybrid seeds in opposite directions.

Crosses, temperature-and germination assays
A. thaliana plants were emasculated 2 days before pollination. Crossed plants were placed at experimental growth temperature until silique maturity/harvesting. For each cross combination, 4-8 different individual plants were used as pollinators and 15-85 siliques (biological replicates) were harvested individually (Supplementary Datasheet S2). After short-term storage at 4°C seeds from induvidual siliques were surface-sterilized ON using chlorine gas (Lindsey et al., 2017). All seeds from the harvested individual siliques were counted and planted on individual 0.5 MS growth-medium containing petri-dishes and scored for germination by counting protrusions through the seed coat after 10 days at 22°C growth conditions. On day 20, germinated seedlings were checked for A. thaliana accidental self-pollination (formation of floral shoots without vernalization). These rare events occurred at a frequency less than 1% and self-plants were removed from the analyses if present.

Statistics
R-studio (version 2023.03.1 + 446) (R Core Team, 2023) was used for data analyses. Plots were generated using the ggplot2 (Wickham, 2016) and dplyr (Wickham et al., 2023) packages. For statistical analyses the car package (Fox and Weisberg, 2019) and ggpubr package (Kassambara, 2023) were employed. To assess the homogeneity of variance for the germination assays we conducted Levene's test (Levene, 1960). If the null hypothesis was rejected, the Welch's t-test (Welch, 1947) was used for statistical analyses. In all other cases statistical analyses were performed using Wilcoxon rank sum test (Mann and Whitney, 1947).

Microscopy
Feulgen stained seeds were harvested at 6 days after pollination (DAP), stained using Schiff's reagent (Sigma-Aldrich S5133), fixed and embedded in LR White (London Resin) (Braselton et al., 1996). Imaging was performed using an Andor DragonFly spinning disc confocal microscope with a Zyla4.2 sCMOS 2048x2048 camera attachment and excitation 488 nm/emission 500 to 600 nm. Seeds were scored for embryo and endosperm developmental stage and the number of endosperm nuclei. Endosperm nuclei counts were assigned an endosperm division value (EDV), which estimates the number of divisions to reach the corresponding number of endosperm nuclei (Ungru et al., 2008). Mean EDV was calculated using the formula: 2 x = mean number of nuclei, where x is the EDV (x = LOG(mean number of nuclei)/LOG (2)).
Crosses with EE-GFP and TE1-GFP markers were imaged using the Andor DragonFly as described. Whole-mount imaging of seeds was performed using an Axioplan2 imaging microscope after 24 h/ 4°C incubation in 8:2:1 (w/v/v) chloral hydrate:water:glycerol (Grini et al., 2002). Mature dry seeds were imaged on a 1.5 x 2.5 cm grid using a Leica Z16apoA microscope connected to a Nikon D90 camera. Seed size and circularity were measured by converting images to black and white and then using the ImageJ "Threshold" and "Analyze particles" functions (https://imagej.nih.gov/ij/).

Antagonistic effects on endosperm barriers in hybrid seeds from A. lyrata and A. arenosa crossed to A. thaliana
In order to compare the success of interspecific hybrids of A. thaliana crossed with A. lyrata or A. arenosa, we first observed the seed-set. Hybrid crosses with A. lyrata fathers had significantly reduced seed set per silique compared to crosses with A. arenosa fathers or A. thaliana (Col-0) self crosses (Supplementary Figures  S1A, B). The lower seed-set correlates with failure of pollen tube burst after entering the female gametophyte (Supplementary Figure S1C), a pre-zygotic barrier previously described (Escobar-Restrepo et al., 2007). Interestingly, pollen tube burst failure can also be observed in crosses between A. thaliana and A. arenosa, but to a lower degree, correlating with the observed seed-set frequency (Supplementary Figures S1A, C). Differences in seed set between the two hybrid crosses thus have a gametophytic pre-zygotic base, and therefore do not affect the postzygotic hybridization barrier in these crosses.
A. thaliana self seeds and A. thaliana × A. arenosa or A. thaliana × A. lyrata hybrid seeds were comparable in size at 6 DAP ( Figure 1A), but most A. thaliana × A. lyrata hybrid embryos were at an earlier developmental stage ( Figure 1B). Large variation in embryonic stages including developmental arrest was observed in 12 DAP whole-mount chloral hydrate cleared hybrid seeds (Supplementary Figure S2A). A time series of Feulgen stained A. thaliana × A. lyrata hybrid seeds further identified variation in endosperm cellularization, starting at 3 DAP and resulting in developmental arrest at the globular embryo stage (Supplementary Figure S2B). The frequency of A. thaliana × A. lyrata early cellularization was higher than for A. thaliana self, and in strong contrast to the late cellularization in hybrid seeds from A. thaliana crossed with A. arenosa [ Figure 1B; (Josefsson et al., 2006;Bjerkan et al., 2020)]. In order to compare the observed cellularization phenotype with nuclear proliferation in the syncytial endosperm, we investigated the number of nuclei in hybrid endosperm. The endosperm nuclei-number was significantly lower in A. thaliana × A. lyrata compared to A. thaliana × A. arenosa hybrid seeds, suggesting a reduction in endosperm proliferation rate ( Figure 1C). Nonetheless, germination rates of hybrid seeds from crosses with A. lyrata were significantly higher than crosses with A. arenosa ( Figure 1D; Supplementary Datasheet S2), indicating that the endosperm hybrid barrier is influenced in a diametrically opposed manner, both in terms of the endosperm cellularization phenotype and the effective output measured as the ability of hybrid seeds to germinate.
3.2 Endosperm phase change and cellularization in hybrids are not synchronized with embryo development Compared to A. thaliana, viable seeds resulting from crosses between A. thaliana mothers and A. arenosa or A. lyrata fathers appear to develop along a slower embryo and endosperm developmental path. This suggests that as long as endosperm cellularization and embryo development are synchronized, seed viability is not impacted. In A. lyrata × A. arenosa hybrid seeds with unsynchronized endosperm and embryo development, the embryo can be rescued in vitro, indicating that the barrier is caused by lack of nutrient support to the growing embryo (Lafon-Placette and Köhler, 2016).
To investigate synchronization of embryo and endosperm in A. thaliana (Col-0) hybrid seeds with A. arenosa or A. lyrata fathers, we used genetic markers of endosperm development that are expressed before and after endosperm cellularization in A. thaliana (van Ekelenburg et al., 2023). In A. thaliana the EE-GFP genetic marker is expressed after fertilization and up to endosperm cellularization at 6-7 DAP (van Ekelenburg et al., 2023) and not expressed after cellularization (Figures 2, S3). In A. thaliana × A. arenosa hybrid seeds, expression of the EE-GFP marker was observed for the full duration of the time series (15 DAP) and no visible downregulation could be observed (Figures 2, S3). This expression pattern supports the observation that A. thaliana × A. arenosa hybrid seeds fail to initiate or have delayed endosperm cellularization (Figure 1) as described previously (Josefsson et al., 2006;Bjerkan et al., 2020). In contrast, when crossed to A. lyrata, expression of the EE-GFP marker decreased from 5 DAP and only a low frequency of seeds expressed the marker at cellularization around 9 DAP (Figures 2, S3). Taking the developmental delay in the A. thaliana × A. lyrata hybrid into account, the EE-GFP marker is prematurely terminated ( Figure 2B), in accordance with the early cellularization phenotype (Figure 1). In A. thaliana self seeds, downregulation of the EE-GFP marker coincides with endosperm cellularization (van Ekelenburg et al., 2023). In hybrid seeds from A. arenosa or A. lyrata fathers, continued expression or premature EE-GFP downregulation, respectively, ( Figure 2B) coincides with diametrically opposed cellularization phenotypes both indicating that the endosperm phase change and cellularization is not synchronized with embryo development, leading to embryo and seed failure.
indicating premature endosperm phase change initiation and supporting the observed precocious endosperm cellularization phenotype (Figure 1).
In A. thaliana seeds the decrease of EE-GFP expression and increase of TE1-GFP expression is strictly coordinated, with limited overlap. This is contrasted by the EE-GFP and TE1-GFP expression in seeds from hybrid crosses (Figure 4). Since the marker transgenes are expressed from the maternal A. thaliana genomes, the only difference in these crosses is the paternal contribution. For A. thaliana × A. arenosa an overlap in expression of the markers was observed from 9 DAP, caused by the prolonged expression of the EE-GFP marker (Figure 4). In A. thaliana × A. lyrata overlapping expression was observed from 6 to 10 DAP (Figure 4), showing a shift in TE1-GFP expression towards earlier developmental stages, although the A. thaliana × A. lyrata hybrid seeds develop slower compared to both A. thaliana selfed and A. thaliana × A. arenosa seeds (Figure 1).

Temperature alters the hybrid barrier strength in diametrically opposed directions
Lowering of temperature from 22°C to 18°C ameliorates the germination efficiency of the hybrid seeds from A. thaliana mothers crossed to A. arenosa fathers (Bjerkan et al., 2020). In order to investigate if the phenotypically contrasting hybrid barrier observed in seeds from A. lyrata fathers was affected by temperature in a corresponding way, we performed crosses between A. thaliana (Col-0) mothers and A. lyrata or A. arenosa fathers at 4°C temperature windows ranging from 14°C to 26°C. Interestingly, the germination rate of A. thaliana × A. arenosa hybrid seeds was significantly enhanced by progressive lowering of the temperature, and contrasted by the germination rate of A. thaliana × A. lyrata hybrid seeds that was significantly enhanced by progressive increase of the temperature ( Figure 5; Supplementary Datasheet S2). In the temperature window, the effects on the hybrid barriers followed a close to linear, but opposed reaction norm. We applied more extreme temperatures previously reported to be within the normal, and not stress inducing, growth-range of A. thaliana , but a further enhancement could not be obtained (Supplementary Figure S5). lyrata are crossed to A. thaliana mothers, we next investigated if different A. thaliana accessions influence the hybrid barrier in similar or opposite manner. We performed hybrid crosses with A. arenosa and A. lyrata fathers to the diploid A. thaliana accessions Col-0, C24 and Ws-2. Additionally, the tetraploid accession Wa-1 was used as a control, as tetraploid A. thaliana crossed to diploid A. arenosa has been shown to increase hybrid seed survival (Josefsson et al., 2006).

Accessions of A. thaliana influence
Compared to the Col-0 accession crosses, the C24 accession enhanced seed survival significantly when crossed to A. arenosa, contrasted by the A. lyrata hybrid where the germination rate declined highly significantly compared to the Col-0 cross ( Figure 6; p ≤ 0.0001). For Ws-2, the germination rate was severely reduced in the A. arenosa hybrid cross, contrasted by moderately high (though lower than for Col-0) germination rate in the A. lyrata hybrid cross ( Figure 6, Supplementary Datasheet S2).
In the tetraploid A. thaliana Wa-1 to diploid A. arenosa hybrid cross, the seed germination rate was enhanced, as previously reported (Josefsson et al., 2006). In contrast, in the Wa-1 to A. lyrata hybrid cross, the germination rate was significantly decreased compared to the Col-0 cross ( Figure 6). Notably, the effect on hybrid seed viability was higher with the diploid C24 accession than with the tetraploid Wa-1 accession.
Crosses performed in parallel at 18°C and 22°C demonstrated the same trends at both temperatures (Supplementary Figure S6, Supplementary Datasheet S2). Although large differences in the barrier strength was observed, as measured by germination and seed viability, no obvious correlation could be found between seed survival and seed size and circularity (Supplementary Figure S7).

A. thaliana accession effects are not readily explained by endosperm cellularization phenotype
To investigate if the endosperm phenotype reflects the influence of accessions on hybrid seed viability, we inspected Feulgen stained 6 DAP hybrid seeds by confocal microscopy. We scored the number of endosperm nuclei in A. thaliana accessions and accession hybrids with A. arenosa and A. lyrata at 18°C and 22°C. The endosperm division value [EDV; (Ungru et al., 2008)] was generally higher when A. arenosa was involved (Figure 7; Supplementary Datasheet S3). No obvious correlation between the number of endosperm nuclei and hybrid seed viability was found (Supplementary Figure S8), however a significant correlation between endosperm proliferation rates and growth temperature could be observed in all crosses except A. thaliana C24 x A. lyrata. The latter hybrid cross did indeed exhibit very low germination rates (Figure 6), however similar low germination rates were found in A. thaliana Ws-2 x A. arenosa hybrid seeds but here accompanied by a high endosperm proliferation rate (Supplementary Figure S8, Supplementary Datasheet S3).
Seed phenotypes were scored for defined stages of embryo and endosperm development. A. thaliana accession self crosses at 22°C displayed embryo stages in the late heart to walking stick stage, with C24 exhibiting the fastest embryonic development. Endosperm cellularization was partly complete, though fully completed in most A. thaliana C24 self-seeds (Figure 7; Supplementary Datasheet S3). In A. thaliana accession x A. arenosa hybrid seeds, embryo development ranged from globular to transition stages. The endosperm was mainly syncytial or had initiated cellularization in the micropylar endosperm. In the C24 cross almost half of the seeds exhibited advanced cellularization stages and also complete cellularization. In this cross, higher germination rate was correlated with temporally correct timing of endosperm cellularization (Figures 6, 7).
In A. thaliana accession x A. lyrata hybrid seeds, embryonic stages ranged from globular to heart, where the C24 accession displayed a majority of heart stages, and Ws-2 a majority of globular stages. In the Col-0 and C24 accession hybrids, near uniform complete endosperm cellularization was observed, contrasted by early peripheral cellularization in Ws-2 (Figure 7). In the case of the latter hybrid cross, embryo development and endosperm cellularization appeared to be synchronized, leading to higher seed viability ( Figure 6). However, large differences in seed viability between C24 (low) and Col-0 (high) hybrid crosses ( Figure 6) were not reflected by the endosperm cellularization phenotype as both crosses had mostly fully cellularized endosperm and appeared to be in a similar embryonic stage (Figure 7). A. thaliana accession hybrid crosses at 18°C exhibited a similar pattern (Supplementary Figure S9). Major significant differences in seed viability between C24 and Ws-2 (low vs medium-high; Supplementary Figure S6) were not reflected by endosperm cellularization as both accession hybrids exhibited mostly micropylar endosperm (Supplementary Figure S9). We conclude that the effect of using different accessions in the hybrid crosses can not readily be explained by a direct effect on the endosperm cellularization phenotype alone and that a more complex interaction between different genotypes occur.

Mutation of the MADS-box transcription factor AGL35 influences A. arenosa and A. lyrata hybrid barriers antagonistically
Deregulation of type I MADS-box TFs has been correlated with endosperm-based hybridization barriers (Walia et al., 2009), and FIGURE 4 Overlapping expression of early and late endosperm markers in hybrid seeds. Percentage of seeds expressing proAT5G09370>>H2A-GFP (EE-GFP) and proAT4G00220>>H2A-GFP (TE1-GFP) at 1-18 days after pollination (DAP) in A. thaliana (Col-0) self-crosses and from crossing A. arenosa (A.a.) or A. lyrata (A.l.) as pollen donor to A. thaliana (Col-0). EE-GFP expression in the Col-0 × A. arenosa hybrid seeds were not documented after 15 DAP. many of these TFs are epigenetically regulated by the so-called FIS-PRC2 and the histone methyltransferase MEDEA (MEA) (Zhang et al., 2018). Mutation of MEA results in ectopic proliferation of endosperm nuclei and delayed cellularization (Grossniklaus et al., 1998;Köhler et al., 2003;Guitton et al., 2004) and we therefore investigated if endosperm overproliferation in A. thaliana mea mutant mothers crossed to A. arenosa or A. lyrata enhances or alleviates the hybrid barriers, respectively.
Heterozygous self-crossed A. thaliana mea mutants resulted in a reduced germination rate of 60% meaning that 80% of seeds carrying the mutant maternal allele failed to germinate due to delayed endosperm cellularization (Supplementary Figure S10). Crossing A. thaliana mea to A. arenosa or A. lyrata resulted in a significant decrease in seed survival compared to Col-0 crosses (Supplementary Figure S10 We studied the effect of single candidate genes regulated by FIS-PRC2 (Zhang et al., 2018). The mutant agl35-1 in the Col-0 background was previously shown to strengthen the barrier when A. thaliana was crossed to A. arenosa (Bjerkan et al., 2020) and AGL35 was upregulated in the same hybrid cross (Walia et al., 2009). AGL40 is similarly expressed in the endosperm (Zhang et al., 2018), upregulated in hybrids (Walia et al., 2009) and mutant seeds have reduced seed size (Kirkbride et al., 2019). To investigate if mutation of single candidate genes could produce opposed effects on the hybrid barrier when crossing A. thaliana mothers to A. lyrata or A. arenosa, as observed when changing temperature or A. thaliana accession (Figures 5, 6), we performed A. lyrata and A. arenosa crosses to A. thaliana agl35-1 and agl40-1 and scored seed germination.
Interestingly, in crosses where agl35-1 was crossed to A. arenosa or A. lyrata a highly significant decrease or increase in germination rate was observed, respectively, compared to wild type Col-0 crosses ( Figure 8A). Single mutation of AGL35 affected the hybrid barrier strength in diametrically opposed directions, as in crosses to A. lyrata the germination rate was significantly enhanced, in contrast to A. thaliana x A. arenosa crosses where the germination rate was significantly reduced ( Figure 8A). Mutation of AGL40 crossed to A. arenosa did not significantly affect germination rate, but in crosses to A. lyrata germination rate was significantly reduced. Col-0 crossed to A. lyrata displayed reduced seed size (Figures S2, S7) due to early endosperm cellularization (Figure 1), and thus the mutation of AGL40 may increase the frequency of early cellularization. To test if cryptic genetic variation in the agl35-1 mutant line could account for the observed phenotype, heterozygous agl35-1 was introgressed twice to Col-0 and segregating progeny of selfed heterozygotes were crossed to A. arenosa. The segregating agl35-1 A. thaliana mothers were genotyped for the agl35-1 insert and germination rate of hybrid seeds was scored ( Figure 8B). Importantly, the germination rate of segregating wildtype plants (agl35+/+) was not significantly different from wildtype (Col-0) when crossed to A. arenosa. Both homozygous (agl35-1 -/-) and heterozygous agl35-1 +/-plants crossed to A. arenosa displayed significantly lower germination rate than Col-0 crossed to A. arenosa, indicating that the increased strength of the hybrid barrier was caused by mutation of AGL35. Furthermore, the observation that the strength of the hybrid barrier can be caused by heterozygous (agl35-1 +/-) plants crossed to A. arenosa indicates that the observed phenotype is caused by genetic interaction occurring in the fertilization products, the embryo or the endosperm.
individuals in a hybrid cross can partly bypass the endosperm-based post-zygotic barrier without any change in ploidy (Walia et al., 2009;Wolff et al., 2015), the role of maternal and paternal genomes cannot be generalized.
Here we demonstrate that interspecies hybrid seeds from crossing A. lyrata or A. arenosa as the paternal parent to A. thaliana mothers show antagonistic endosperm cellularization phenotypes, with late cellularization in crosses with A. arenosa Effect of temperature, accession and hybridization on endosperm development. Confocal images showing endosperm cellularization of 6 DAP Feulgenstained seeds from crossing A. arenosa (A.a.) or A. lyrata (A.l.) as pollen donor to A. thaliana (accession Col-0 or C24 or Ws-2) at 22°C. Scale bar = 50 mm. Mean endosperm division value (EDV) is shown within each image, n EDV = 10 seeds. Beneath each image quantification of the described embryo and endosperm stages is shown as bar charts: Col-0, n = 36; Col-0 × A.a., n = 23; Col-0 × A.l., n = 23; C24, n = 32; C24 × A.a., n = 22; C24 × A.l., n = 31; WS-2, n = 26; WS-2 × A.a., n = 51; WS-2 × A.l., n = 37. Embryo stages (EM): 4c, 4-cell; Oc, Octant; Gl, Globular; Tr, Transition; He, Heart; To, Torpedo; WS, Walking stick; Endosperm cellularization stages (ES): Sy, Syncytial endosperm; Mi, Micropylar endosperm cellularization; Pe, Peripheral endosperm cellularization; PC, Partially complete endosperm cellularization; Co, Complete endosperm cellularization. and early cellularization in crosses with A. lyrata. In both cases, cellularization failure results in an endosperm-based hybrid barrier and reduced viability of germinating seeds. This compares to previous studies where timing of endosperm cellularization is influenced by the paternal species in reciprocal A. arenosa and A. lyrata interspecies crosses and in crosses within the genus Capsella (Rebernig et al., 2015;Lafon-Placette et al., 2017). Intriguingly, we find that a temperature gradient leads to diametrically opposed cellularization phenotype responses in hybrid endosperm with A. arenosa or A. lyrata as pollen donors. In addition, A. thaliana accession genotypes also influence hybrid seed viability in opposite directions. To this end, we demonstrate that single gene mutation in A. thaliana MADS-box TF AGL35 independently can affect the germination rates of A. arenosa or A. lyrata hybrid seeds in opposite directions.
connected to endosperm cellularization is disturbed in hybrid seeds. In the Col-0 × A. arenosa hybrid seeds that fail to cellularize, the EE-GFP marker continues to be expressed throughout endosperm development, indicating phase change failure. In the Col-0 × A. lyrata hybrid seeds, characterized by early cellularization, the developmental time point of TE1-GFP expression indicates occurrence of a premature phase change. These findings support that not only the timing of endosperm cellularization is affected in these developing hybrid seeds, but also the developmental timing of the genetic network associated with endosperm phase change and maturation occurring at cellularization. In accordance with the incomplete endosperm hybridization barrier, prolonged expression of EE-GFP in Col-0 × A. arenosa seeds and precocious expression of TE1-GFP in Col-0 × A. lyrata seeds were not observed in all individual seeds. This suggests that gene regulation associated with the endosperm phase change within each hybrid seed varies and potentially is affected by genetic or epigenetic variation that modulates threshold levels for gene activation or repression.
In our system the sole difference between the two hybrids is the paternal parent, indicating a trans-acting mechanism, where differential expression from A. lyrata and A. arenosa genomes regulates the genetic markers expressed from the A. thaliana genome. Supporting this hypothesis, paternal transmission of mutants in NUCLEAR RNA POLYMERASE D1 (NRPD1) can bypass the cellularization phenotype in paternal excess inter-ploidy crosses (Erdmann et al., 2017;Martinez et al., 2018). NRPD1 is a main component in the the RNA-directed DNA methylation (RdDM) pathway, resulting in small RNA directed gene regulation by de novo DNA methylation (Law et al., 2013;Kirkbride et al., 2019) and could be a potential trans-acting regulatory mechanism (Erdmann et al., 2017). Future experiments to identify transcriptional differences from the parental genomes in hybrid seeds may point at key genes and mechanisms responsible for ectopic timing of the endosperm developmental phase change.

Temperature and accession affect the viability of hybrid seeds
We found that by increasing the temperature from 14°C to 26°C, Col-0 × A. arenosa seeds display significantly decreased germination rates (more than 30%). The same temperature range has an opposite effect in Col-0 × A. lyrata seeds resulting in increased germination rates (more than 50%). This demonstrates a temperature dependent genetic mechanism that acts antagonistically when A. thaliana is crossed to A. arenosa or A. lyrata and produces diametrically opposed cellularization phenotype responses in the hybrid endosperm.
Interestingly, in Brassica oleracea, temperature affects abscisic acid (ABA) levels specifically in the endosperm and cooler temperatures obstruct the breakdown of ABA in the desiccating endosperm (Chen et al., 2021). This is consistent with a recent report demonstrating that A. thaliana inter-ploidy uncellularized endosperm induced by paternal excess is correlated with increased ABA levels, suggesting that endosperm cellularization is connected to dehydration responses in the developing embryo (Xu et al., 2023). ABA catabolism in response to temperature may therefore be a potent mechanism to explain the temperature influence on the hybrid barrier when A. thaliana is crossed to A. arenosa. In a similar manner, we further speculate that precocious cellularization in crosses with A. lyrata may be associated with a similar mechanism that triggers ABA breakdown, but this needs further investigation.
Notably, the effect of using different A. thaliana accessions in the hybrid crosses is larger than the temperature effect (close to 70% difference), and even larger than the interploidy effect. While the Col-0 and Ws-2 A. thaliana accessions resulted in a generally higher germination rate when hybridized with A. lyrata compared to A. arenosa, the C24 accession had the opposite effect. The way the accessions affected hybrid seed viability in opposite directions may point at a similar mechanism as observed in the temperature experiment. However, our results do not readily explain the observed germination rates by cellularization phenotype alone, and further investigations are required to resolve these observations.
Crossing tetraploid A. thaliana Wa-1 to diploid A. arenosa increases hybrid seed survival (Josefsson et al., 2006). In addition, ploidy affects the strength of the hybrid barrier in crosses between A. arenosa and A. lyrata, where higher ploidy in A. lyrata increases the hybrid seed survival rate, while higher ploidy in A. arenosa causes total seed lethality (Lafon-Placette et al., 2017). Our data show that a similar effect is found using the diploid accession C24, suggesting that C24 may have a higher effective ploidy and endosperm balance number [EBN; (Johnston and Hanneman, 1982)] compared to Col-0 and Ws-2. This corresponds well with the hypothesis that A. lyrata has a lower EBN compared to A. arenosa (Lafon-Placette and Köhler, 2016), explaining why crosses with C24 or Wa-1 decrease seed viability in the A. lyrata hybrid. However, it does not explain why A. lyrata crosses with the diploid C24 is more detrimental than crosses with the tetraploid Wa-1, suggesting that accession genotypes, in addition to ploidy, has an effect on the endosperm-based hybridization barrier.

AGL35 influences endosperm cellularization in hybrid seeds
AGAMOUS-LIKE (AGL) type I MADS-box TFs are highly expressed in the seed, specifically during endosperm cellularization (Bemer et al., 2010;Zhang et al., 2018;Bjerkan et al., 2020). Their importance in the endosperm-based hybridization barrier has been suggested by several studies (Josefsson et al., 2006;Walia et al., 2009;Bjerkan et al., 2020) and it has been hypothesized that timing of endosperm cellularization requires a stoichiometric balance between members of different MADS-box protein complexes (Batista et al., 2019). In this study we demonstrate that mutation in A. thaliana AGL35 has a highly significant and opposite effect on the hybrid barrier phenotype when crossed to A. lyrata and A. arenosa, respectively. AGL35 is bi-allelicly expressed in the chalazal endosperm (Bemer et al., 2010;Bjerkan et al., 2020) and upregulated in crosses between A. thaliana and A. arenosa compared with compatible crosses (Walia et al., 2009). Our results indicate that AGL35 is involved in the transition from syncytial to a cellularized endosperm, and may function as a promoter of cellularization, as mutant crosses to A. arenosa result in lower seed survival, while mutant crosses to A. lyrata result in increased survival compared to Col-0 crosses. Interestingly, a massive-multiplexed yeast two-hybrid study identified interaction between AGL62 and AGL35 (Trigg et al., 2017). These AGL TFs have seemingly antagonistic functions as AGL62 is a suppressor of endosperm cellularization (Kang et al., 2008). Paternal excess interploidy crosses cause increased AGL62 expression, correlated with endosperm cellularization failure (Erilova et al., 2009). AGL62 is also a direct target of the FIS PRC2 complex (Hehenberger et al., 2012) whereas we could see no direct effect on the A. arenosa or A. lyrata hybrid with A. thaliana by mutation of FIS PRC2. The antagonistic effects of single gene mutation of AGL35 is intriguing, and we speculate that expression differences between A. arenosa and A. lyrata in the hybrid endosperm may account for our observations but future investigation of this interaction and the role of AGL35 in regulation of endosperm-based hybridization barriers is required.

Conclusions
The findings in this study introduce a rigorous model system for the dissection of the influence of abiotic and genetic parameters in hybrid admixture, and have a large potential to support breeding and climate research. Further examination and usage of these approaches could help pinpoint genes, networks or gene dosage balances that are involved in overcoming the endosperm-based hybridization barrier. Species previously thought to be unable to hybridize due to postzygotic seed lethality may be able to do so given favorable conditions, and a similar effect could also apply to interploidy hybrids.
Currently, it is not known if the temperature effect on hybridization success is mediated by the same genetic network that is operated by changes in ploidy or genetic variation. Phenotypically, the temperature effect restores defects in timing of cellularization, but it is not known if the trigger is upstream or downstream of the causative genetic network. Elucidation of the genetic, epigenetic and mechanistic basis for this cross talk between the genic and environmental factors is therefore essential for our understanding of the plasticity of endosperm-based hybridization barriers.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Files. Further inquiries can be directed to the corresponding author/s.

Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.