Variation in seed longevity among diverse Indica rice varieties

Abstract Background and Aims Understanding variation in seed longevity, especially within closely related germplasm, will lead to better understanding of the molecular basis of this trait, which is particularly important for seed genebanks, but is also relevant to anyone handling seeds. We therefore set out to determine the relative seed longevity of diverse Indica rice accessions through storage experiments. Since antioxidants are purported to play a role in seed storability, the antioxidant activity and phenolic content of caryopses were determined. Methods Seeds of 299 Indica rice accessions harvested at 31, 38 and 45 d after heading (DAH) between March and May 2015 and differing in harvest moisture content (MC) were subsequently stored at 10.9 % MC and 45 °C. Samples were taken at regular intervals and sown for germination. Germination data were subjected to probit analysis and the resulting parameters that describe the loss of viability during storage were used for genome-wide association (GWA) analysis. Key Results The seed longevity parameters, Ki [initial viability in normal equivalent deviates (NED)], −σ−1 (σ is the time for viability to fall by 1 NED in experimental storage) and p50 [time for viability to fall to 50 % (0 NED)], varied considerably across the 299 Indica accessions. Seed longevity tended to increase as harvest MC decreased and to decrease as harvest MC increased. Eight major loci associated with seed longevity parameters were identified through GWA analysis. The favourable haplotypes on chromosomes 1, 3, 4, 9 and 11 enhanced p50 by ratios of 0.22–1.86. Conclusions This is the first study to describe the extent of variation in σ within a species’ variety group. A priori candidate genes selected based on rice genome annotation and gene network ontology databases suggested that the mechanisms conferring high seed longevity might be related to DNA repair and transcription, sugar metabolism, reactive oxygen species scavenging and embryonic/root development.


INTRODUCTION
Orthodox seeds can remain viable for many years if stored at low moisture content (MC) and low temperature (Bewley and Black, 1994). This 'storability' or 'longevity' is exploited as a means of ex situ conservation of biodiversity in seed-/ genebanks. However, the length of time that seeds can survive varies among species (Justice and Bass, 1978;Walters et al., 2005;Probert et al., 2009;Merritt et al., 2014). It also varies among seed lots within a species and among the individual seeds within a seed lot. The variation in seed longevity within a seed lot is usually described by fitting a normal distribution of seed deaths over time, the basis of the seed viability equations (Ellis and Roberts, 1980). Collecting fully matured seeds is critical to ensuring maximum longevity during storage (Hay and Smith, 2003), but how the seeds are dried, processed and held before storage and the length of time that these steps take also affect subsequent longevity (Probert et al., 2007;Whitehouse et al., 2015). Understanding why some seeds remain alive for a long time and others, under the same conditions, for a short time, has been a fundamental question in seed biology. As well as being important for predicting viability during storage, it is important in terms of managing seed vigour. Seed lots with high viability are likely to have high vigour; reduced vigour is an early symptom of seed ageing.
A number of seed constituents have been purported to play a role in conferring stability in storage, including sugars, proteins and antioxidants. High levels of raffinose family oligosaccharides, or high levels relative to the amount of sucrose, have been correlated with seed longevity (Bailly et al., 2001;ElSayed et al., 2014) and De Souza Vidigal et al. (2016) concluded that galactinol, an intermediate metabolite in the biosynthesis of raffinose oligosaccharides, can be used as a marker of longevity among different varieties of Brassicaceae or tomato. However, some authors concluded that proteins must contribute to the stabilizing effect of sugars, in particular proteins that are expressed late in seed development (Chatelain et al., 2012).
Repair processes that are initiated during the early stages of seed germination and which overcome the damage to macromolecules incurred during ageing have also been identified as key to 'longevity' (Waterworth et al., 2010(Waterworth et al., , 2015. Much of the damage that occurs during seed ageing is attributed to the generation and action of reactive oxygen species; hence, the ability to neutralize these reactive oxygen species through an effective antioxidant system is also important (Noctor and Foyer, 1998;Bailly, 2004;Sattler et al., 2004;Nguyen et al., 2015). Given these different hypotheses relating to seed longevity, and the plasticity of the trait, it is not surprising that a complex network of putative longevity-related genes has been reported (Righetti et al., 2015).
Other work relating genes and seed longevity has taken a quantitative genetics approach. Through genetic linkage mapping using inbred line or recombinant inbred line maize populations, a number of quantitative trait loci (QTLs) and candidate genes associated with seed longevity under natural (low MC and low temperature) and artificial ageing (high MC and high temperature) conditions have been identified (Revilla et al., 2009;Han et al., 2014). A genome-wide association (GWA) study for a diverse barley panel discovered over 100 genetic markers correlated with seed longevity (Nagel et al., 2015). Sano et al. (2017) conducted QTL analysis and RNA sequencing for Arabidopsis recombinant inbred lines and identified a negative correlation between brassinosteroids and seed longevity. Relatively few studies have reported candidate genes involved in rice (Oryza sativa) seed longevity. Most have been QTL studies using biparental mapping populations, leading to the identification of large chromosome regions rather than candidate genes. The marker interval of the most well-defined QTL spanned a region of 0-72 cM on chromosome 9 (Miura et al., 2002;Sasaki et al., 2005;Xue et al., 2008;Li et al., 2012). Sasaki et al. (2015) fine-mapped this region using near-isogenic lines and identified the candidate gene, TPP7, a trehalose-6-phosphate phosphatase known to be involved in anaerobic germination tolerance (Kretzschmar et al., 2015). However, these studies on rice seed longevity were limited to genes where the two parents (usually Japonica × Indica) differ. Different studies also used different ageing/storage conditions and measures of 'longevity' (Hay et al., 2019). Furthermore, they did not take into account variation in the timing of harvest maturity, which is known to have an effect on subsequent seed longevity . Compared with conventional biparental QTL mapping, GWA has several advantages: (1) it can identify a range of genes with allelic variation among a large number of diverse accessions; and (2) there is typically higher resolution of QTLs, allowing the direct identification of candidate genes without need for further fine mapping (Collard et al., 2008). Although a GWA panel could span different variety groups, haplotypes are often subpopulation-specific (Famoso et al., 2011). Consequently, restricting a panel to a single variety group increases the power of QTL detection for genes that are polymorphic only within that variety group (Al-Tamimi et al., 2016;McCouch et al., 2016).
The International Rice Genebank at the International Rice Research Institute conserves over 127 000 accessions of Oryza sativa, O. glaberrima and their wild relatives. With such a large number of accessions, routine viability monitoring involves thousands of germination tests each year, requiring the destructive use of seeds, and considerable time and money. Appropriate adjustment of monitoring intervals based on known variation in seed longevity across genotypes is one way in which genebank management can be improved. The aims of this study were to: (1) characterize the survival curves of seeds of diverse Indica rice accessions; and (2) identify loci/candidate genes associated with seed longevity in Indica rice through GWA analysis, taking into account the effects of seed maturity and harvest moisture content.

Plant materials
A diverse panel of 299 Indica rice accessions from the 3,000 Rice Genomes Project (2014) were used (Supplementary Data Tables S1 and S2; Supplementary Data File S1). Seed samples were taken out of genebank storage in December 2014. After breaking dormancy (50 °C for 7 d), seeds were germinated in seedling plots and 20-d-old seedlings were transplanted to the Zeigler Experiment Station, with 200 mm spacing of plants within and between rows; 28 kg ha −1 NPK (14-14-14) was applied during field preparation. Nitrogen (20 kg ha −1 ) was applied 30 and 50 d after transplanting. Insect, disease and weed control was consistent with standard International Rice Research Institute procedures. A total of 12 plants per sample (seed lot) were planted and harvested as bulk. The day when 80 % of plants had completed heading was considered the heading date. Seeds were harvested at 31, 38 and 45 d after heading (DAH) and threshed. Harvest MC was determined by weighing three replicate samples of ground seeds before and after 2 h of oven-drying at 130 °C followed by 1 h at room temperature (ISTA, 2018). Blown and sorted seed lots were dried at 15 °C, 15 % RH. For each seed lot, three samples of 100 dried seeds were weighed to determine 100-seed weight. The bulk of each seed lot was then sealed inside aluminium foil bags and stored at −20 °C.

Seed longevity phenotyping
A standard protocol was used for the seed storage experiments (Hay et al., , 2019Whitehouse et al., 2015), done in batches of ten accessions. Samples of 60 seeds for each accession × harvest maturity were equilibrated at 60 % RH, 25 °C in a climate test chamber (Model VC3 0034-M; Vötschtechnik, Germany) and then transferred to laminated aluminium foil bags, which were immediately heat-sealed and placed at 45 °C. Samples were taken at 7-d intervals from 0 to 63 d and tested for ability to germinate. For each sample, 30 seeds were placed on two layers of Whatman No. 1 filter paper with 7 mL of distilled water in each of two 90-mm-diameter Petri dishes. Seeds were germinated at 30 °C with 12 h light per day and scored daily up to 21 d after sowing (seeds sampled at 0 d storage) or at 5, 9, 14 and 21 d after sowing (all other samples). Two or three sets of three samples each were also used for monitoring MC before, during and/or at the end of storage.
Germination data for each seed lot were analysed by probit analysis in GenStat v. 18 (VSN International, Hemel Hempstead, UK), thereby fitting the Ellis and Roberts (1980) viability equation: where v is the viability in normal equivalent deviates (NED) after p days storage, K i is the initial viability (NED) and σ is the time it takes for viability to fall by 1 NED. Thus, −σ −1 is the slope of the transformed survival curve. The time for viability to fall to 50 %, p 50 , was also estimated. For seed lots for which germination initially increased or was maintained at ~100 %, only the data covering the period when viability was declining were included in the analysis.

Determination of antioxidant and phenol content
Antioxidants in the caryopses of the seed lot with greatest longevity (p 50 ) as determined from the seed storage experiment were screened following Shimada et al. (1992) and Dewanto et al. (2002). Ground and then freeze-dried caryopses (0.3 g) were extracted with 6 mL of 80 % MeOH at 40 °C for 3 h. To quantify free radical scavenging activity, the extract was reacted with 1,1-diphenyl-2-picrylhydrazyl (DPPH; Sigma Co. Singapore) at 25 °C for 30 min in darkness, then scavenging activity was determined by maximum absorption at 518 nm using a DU 800 UV/Vis spectrophotometer (Beckman Coulter, Germany). The other sample extract was mixed with 50 % Folin-Ciocalteu's phenol reagent, 2 N (Sigma Aldrich, USA) plus 2 % Na 2 CO 3 solution at 25 °C for 30 min. Total phenolic content was determined by maximum absorption at 725 nm.

Association analysis
Descriptive statistics of individual traits and correlations between seed longevity and related traits were analysed using STAR 2.0.1 (International Rice Research Institute, Philippines). Genome-wide association analysis was conducted in TASSEL 5.2.7 (Bradbury et al., 2007) using the 18 million single-nucleotide polymorphism (SNP) base set of the 3,000 Rice Genomes Project (Wang et al., 2018). After filtering SNPs for 20 % missing data and minor allele frequencies (<5 %), 988 k SNPs were retained for association analysis. A mixed linear model (MLM) with kinship matrix and general linear model (GLM) with principal components analysis (PC 5 ) were applied. Based on quantile-quantile plots for the GLM, there was considerable overestimation (false positives), which may mislead the candidate gene selection. When PC 5 was included in the GLM, the estimation of quantile-quantile plots was improved. Plot-peaks above the threshold (P < 9.99E−06) where the -log 10 P values were not linear in the quantile-quantile plot and detected in major loci of both MLM and GLM with PC 5 were considered to be associated with the traits.

Variation in heading date, seed harvest moisture content and 100-seed weight
Length of time from sowing to 80 % heading ranged between 67 and 100 d, with a mean of 82 d. Due to this variation in heading date, there was also variation in the dates of harvest (Fig. 1). For seeds harvested at 31 DAH, the first accessions were harvested on 19 March and the last accessions were harvested on 21 April. The last harvests of 45 DAH seeds were on 5 May. Seed harvest MC (% fresh weight) across all seed lots ranged between 11.57 and 25.74 %, but showed a decreasing trend with harvest date and maturity (Fig. 1A, B). Across accessions, the mean harvest MC of 31 and 38 DAH seeds was significantly greater than that of 38 and 45 DAH seeds, respectively (paired t-test, P < 0.001). The 100-seed weight ranged between 1.04 and 4.28 g (Fig. 1C, D; Table 1). Across accessions, the 100-seed weight of seeds harvested at 38 DAH (2.487 g) was significantly higher than that of seeds harvested at 31 DAH (2.470 g; P = 0.03) or 45 DAH (2.475 g; P = 0.016).

Seed longevity
The mean MC during seed storage (across all seed lots and sampling times) was 11.13 %; the range across batches was between 10.77 and 11.66 %. The seeds did not consistently lose or gain moisture during the seed storage experiment, or differ in MC within an accession depending on maturity (data not shown). The plots of ability to germinate following different periods of storage were typically sigmoid (Supplementary Data Fig. S1). When fitting eqn (1) to the seed storage data, it was possible to constrain the fitted survival curves for the different seed maturities to a common slope (−σ −1 ) without a significant increase in residual deviance (approximate F-test, P > 0.05) for 95 accessions; for the other accessions, parameters could not be constrained. For most accessions, there was an increase in K i and the gradient of the survival curve decreased (i.e. σ increased) between 31 and 38 DAH; hence, p 50 increased. For 69 accessions, 38 DAH was the optimum harvest time for maximum p 50 , but for 227 seed lots seed longevity (p 50 ) was greatest for seeds harvested at 45 DAH.
For seeds that were losing moisture between sequential harvests (i.e. between 31 and 38 DAH or between 38 and 45 DAH), there was a significant relationship between the decline in MC and the improvement in p 50 as a proportion of the p 50 of the less mature seed lot ( Fig. 2A). Conversely, if the seeds gained moisture between sequential harvests, the greater the amount of water absorbed the greater the potential decline in p 50 (Fig. 2B). The greatest proportional decline in p 50 was by 66 %, from 8.34 to 2.82 d between 38 and 45 DAH, for accession IRGC 125739.
Considering seeds harvested at 38 DAH, K i [hereafter expressed as K i(38DAH) ] ranged between 0.12 (IRGC 127432) and 5.84 NED (IRGC 127789) with a mean of 2.70 NED ( Fig. 3; Table 1). As the initial quality of most seed lots improved at 45 DAH, the range increased overall from 0.53 to 6.29 NED with a mean of 3.32 NED. The slope of the survival curves, −σ ( The free radical scavenging capacity of caryopsis extracts ranged from 30.30 to 91.50 % (Table 1). Total phenolic compounds (non-enzymatic antioxidants) varied between 18.60 and 2282.70 μg g −1 dry weight. There was significant correlation between antioxidant activity and phenolic content (P < 0.001; Fig. 4J), but antioxidant activity and phenolic content were not significantly correlated with seed longevity (p 50 ) for seeds harvested at 38 DAH or with the highest p 50 (Fig. 4F-I). The 100-seed weight was significantly correlated with p 50 for seeds harvested at 38 DAH (P < 0.05; Fig. 4A) but not for seeds with the highest p 50 (Fig. 4B), even though there was a significant correlation between the two sets of p 50 values (P < 0.001; Fig. 4E).

GWA analysis
Association analysis was conducted using filtered (MAF > 0.05) 988 k SNP markers. An MLM detected several loci that were associated with individual traits (Supplementary Data Fig.  S2), and a number of co-localized trait associations. For example, a haplotype containing marker 10706067855 at position 6 067 855 on chromosome 7 was strongly associated with antioxidant capacity (P = 2.82E−11) and total phenolic compounds (P = 4.98E−17). However, seed longevity loci did not overlap with any other loci detected for other traits in this study. Harvest MC was used as a covariate of seed longevity traits. For seeds harvested at 38 DAH (typical harvest time in the International Rice Genebank), a single major peak on chromosome 4 associated with p 50(38DAH) was consistently detected with both MLM and GLM analyses (Fig. 5). Two peaks associated with K i(38DAH) , on chromosomes 3 and 11, were detected, but P values were below the MLM threshold, likely due to underestimated genetic effects, as indicated by the quantile-quantile plot (Supplementary Data Fig. S2). For the slope [−σ (38DAH) −1 ], a consistent major peak on chromosome 3 was detected in both MLM and GLM. Further association analysis, to identify additive genetic effects in relation to continued on-plant seed maturation, was performed using the maximum p 50 observed for an accession (i.e. for seeds harvested at either 38 or 45 DAH; Fig. 6). Compared with the Manhattan plot of 38 DAH seed lots, four additional loci appeared on chromosomes 1, 3, 9 and 11. The allelic effects of identified loci were estimated through haplotype mapping (Fig.  7). The presence of favourable haplotypes (shaded green) at one or two loci (genotypes 6-16) was associated with moderate enhancements of seed longevity (p 50 ), by 22-103 % compared with genotype 17 having unfavourable haplotypes (shaded grey) at all loci. Accessions with favourable haplotypes at three loci (genotypes 3-5) showed 53-152 % increase in longevity, while accessions with four-locus combinations of favourable haplotypes (genotypes 1 and 2) showed enhancements of 153-186 %, suggesting additive effects of the haplotypes.

Candidate gene selection
Our GWA peaks tended to locate inside linkage disequilibrium blocks (data not shown). The size of each GWA peak was determined based on the physical distance between markers with subsequently increased and decreased P values between a threshold. Since we do not know which markers are causal SNPs, we first considered all genes located at GWAS peaks as potentially associated with the trait. Annotation using MSU7 (Kawahara et al., 2013) identified 66 genes located in the eight GWA peak regions (on chromosomes 1, 3, 4, 9 and 11 for p 50 ; on chromosomes 3 and 11 for K i ; and on chromosome 3 for slope) (Figs 5 and 6; Supplementary Data Table S3). Based on gene annotation plus ontology with networking genes (RiceNet v2, accessed on 2 November 2017; Lee et al., 2015), 36 genes were considered as potentially relevant to seed longevity. We also considered whether gene expression occurred in seed tissues (The Bio-Analytic Resource for Plant Biology Database). Using the SNP-Seek Database (Mansueto et al., 2017), we aligned allelic variations and focused on gene regions where significant P-value markers were located in exons and/or introns. In total, eight genes were identified as a priori candidate genes associated with seed longevity (Table 2).

DISCUSSION
Understanding the molecular basis of seed longevity and why seeds of some species have the potential to remain alive for decades or centuries under optimal storage conditions is an ongoing focus of seed biology. It is not only of direct relevance to those concerned with storing seeds for long periods of time, such as genebank managers, but also, since seed longevity is an inherent aspect of seed physiological quality, to anyone involved in producing or using seeds. Seeds with greater longevity will retain their planting value for longer periods, regardless of storage conditions. There is considerable variation among orthodox species in seed longevity (Priestley et al., 1985;Hong et al., 1996); thus, it has been possible to identify trends in relative seed longevity associated with, for example, the climate in the country of origin (Walters et al., 2005;Probert et al., 2009).
There is much less variation within a species or even a genus. Early studies assumed that the 'rate' of loss of viability, i.e. the slope of the survival curve (−σ −1 ) did not differ between seed lots of the same species stored at the same MC and temperature (Ellis and Roberts, 1980). Thus, it was suggested that only the initial quality of the seeds, through K i , determines the longevity of a seed lot of a particular species in storage. However, Whitehouse et al. (2018) have found that both K i and σ can vary among rice varieties. Variation in σ has also been reported for foxglove (Digitalis purpurea; Hay et al., 1997) and Trifolium ambiguum (Hay et al., 2010) seeds. Similarly, here we found that there was considerable variation in both K i , as expected (between 0.53 and 6.29 NED for seeds with maximum longevity), and σ (between 2 and 33.3 d) within 299 Indica rice varieties (Table 1). This is the first time variation in both seed longevity parameters have been explored so extensively through seed storage experiments (Hay et al., 2019). Given this variation, we have been able to conduct GWA analysis for the different longevity parameters (K i , −σ −1 and p 50 ).
Seed longevity is known to improve throughout seed development, as the seeds acquire desiccation tolerance and during the first part of the desiccation phase, when the seeds are still metabolically active but starting to dry to equilibrium with ambient conditions (Ellis and Pieta Filho, 1992;Angelovici et al., 2010;Hay et al., 2010;Chatelain et al., 2012;Whitehouse et al., 2015). This late 'accumulation' of seed longevity can make it difficult, especially for a shatter-resistant crop like rice, to know when to harvest seeds for maximum physiological quality. Previous studies (Kameswara Rao and Jackson, 1996a, b) have identified the optimum harvest time for rice accessions to be at around 35 d after anthesis (DAA, approximately equivalent to DAH − 1). More recently, Whitehouse et al. (2015) found that the harvest date and hence MC at harvest was more critical than chronological age. For the study reported here, we attempted to take into account the potential effect of seed maturity by harvesting each accession on three occasions and characterizing the seed survival curves for each seed lot. The mean (across all accessions) 100-seed weight changed significantly, but only slightly between 31 and 38 DAH (mean ± s.d. change +0.017 ± 0.103 g) and between 38 and 45 DAH (mean ± s.d. change −0.012 ± 0.097 g) (Fig. 1C, D), confirming that the seeds were harvested after mass maturity. Indeed, Kameswara Rao and Jackson (1996b) found that for rice accessions grown in the same location and season, mass maturity occurs between 18.5 and 21.6 DAA. Hence, we infer that the rice seeds in our study were in the desiccation phase of seed development. Of the 299 accessions included in the study, 250 declined in MC between 31 and 38 DAH, of which 139 continued to lose moisture between 38 and 45 DAH. However, harvest MC depends on ambient conditions and, as seen in Whitehouse et al. (2015), the general trend was for harvest MC to decline over the harvesting period (Fig. 1A). Thus, seeds that were harvested later were less likely to show continued decline in MC over successive harvests, because they had already dried by 31 DAH. Seed longevity was influenced by the harvest MC. For seeds that were drying on the plant, there was, on average, an 8.7 % improvement in longevity (p 50 ) for every 1 % reduction in MC ( Fig.  2A). Conversely, if seeds took up water between consecutive Correlation coefficients (r) were significant at *P < 0.05 or ***P < 0.001 (two-sided tests).
harvests, p 50 was likely to decline (Fig. 2B). A decline in seed longevity if harvest is delayed is not unexpected (Kameswara Rao and Jackson, 1996a, b;Hay et al., 2015;Nasehzadeh and Ellis, 2017). In the case of non-shattering wild species, it has been suggested that determining the moisture status of a sample of seeds can be used to determine whether the seeds are ready to harvest and/or the optimum way to dry the seeds (or not) after harvest  a relatively high temperature (45 °C; Whitehouse et al., 2018). Thus, high-temperature drying substitutes for on-plant desiccation, and even stimulates an enhanced response (Whitehouse et al., 2017), when ambient conditions are too humid to allow drying in situ. In the present study, all seeds were dried at 15 % RH/15 °C, conditions that conform with the Genebank Standards for seed drying (FAO, 2014). The results of the GWA analysis might have differed if the seeds had been initially dried at the higher temperature; this is now standard practice for regenerated accessions at the International Rice Genebank.  . Arrows indicate loci that appeared when the maximum p 50 for each accession was used. Nagel et al. (2015) found that antioxidants in barley seeds were highly associated with longevity. In our study, p 50 was not correlated with free radical scavenging capacity of antioxidants in the caryopsis or total phenolics (Fig. 4E-I). This could be partly due to our assay, which measured all antioxidants rather than specific antioxidants. Our recent study on correlation between seed longevity and vitamin E in the caryopses of 20 rice accessions found that among eight types of vitamin E, only γ-tocotrienol was significantly correlated with seed longevity (Lee et al., 2017). Another possibility is that antioxidants play a lesser role in the seed longevity of Indica rice. Seed longevity varies across subgroups of rice (Lee et al., 2017). This might be due to intraspecific variation in genomic structure. Hence, each subgroup might have different mechanism(s) conferring extension of seed longevity. Candidate gene network ontology identified through the current study also suggests that oxidation reduction may play an important role in seed longevity (Table  2). For example, LOC_Os12g25690 and LOC_Os12g25700, part of the network of candidate gene LOC_Os04g01280, are related to oxidation reduction. Therefore, individual antioxidants need to be tested further for association with seed longevity. Ideally, this would involve looking at embryo-specific antioxidant levels rather than whole seeds.
Based on the database of network prioritization server (RiceNet v2), various genes in networks that are connected to our candidate genes were identified. We prioritized direct neighbourhoods of candidate genes, which, through the combination of log likelihood scores (top 3), annotation and ontology, were considered relevant to seed longevity mechanisms (Table 2,  Supplementary Data Table S3). Sasaki et al. (2015) reported that 'TPP7' and/or its flanking gene on chromosome 9 conferred the difference in seed longevity between seeds of 'Milyang23' (an Indica variety, high longevity) and 'Akihikari' (a temperate Japonica variety, poor longevity). However, these genes were not indicated as important in the Indica panel used in this study. Indeed, a large deletion on 'TPP7' is present in many Indica accessions (Kretzschmar et al., 2015).
Association analysis was conducted for K i , −σ −1 and p 50 ; loci for each trait were identified in different locations ( Supplementary  Data Fig. S2). Ontology-mediated gene networks suggested a     common mechanism of seed longevity across longevity parameters: DNA-dependent transcription and repair of damaged DNA (Table 2). Waterworth et al. (2016) showed how sensor kinases control DNA repair and the germination of aged Arabidopsis seeds. A similar mechanism was reported for Medicago truncatula seeds (Verdier et al., 2013). In their results, unfolded protein binding and RNA processing genes were highly associated with seed longevity. LOC_Os04g01160 (SOR1) (Hanzawa et al., 2013) and LOC_Os11g01439 (RLCK57) (Vij et al., 2008), associated with p 50 , and LOC_Os03g06890, associated with slope, were annotated to root and embryonic development. LOC_ Os03g51050 (PTR8) (Léran et al., 2014), LOC_Os09g37250 and LOC_Os11g01439 (RLCK57) were 'additionally associated' with on-plant desiccation due to delayed harvest (Fig. 6). LOC_Os04g01280 and LOC_Os11g01439 (RLCK57) with the largest allelic effects on p 50 (51.8 and 49.0 %, respectively) are involved in sugar metabolism and reactive oxygen species scavenging (Bailly, 2004;ElSayed et al., 2014). Both mechanisms were previously demonstrated as determinants of seed longevity. LOC_Os04g01280 and LOC_Os04g01160 are known to control auxin induced root gravitropism and root architecture, such as primary root length and lateral root number (Shi et al., 2009). As the main advantage of GWAS using a diverse rice panel with high-density genotype data, we provide the first report of these genes as potentially being involved in seed longevity. Similarly, Righetti et al. (2015) revealed that pathogen-resistant genes indeed play an important role in seed longevity extension. From a genebank management perspective, testing Indica accessions for favourable haplotypes with respect to longevity through cost-efficient genotyping assays will make it possible to flag particular seed lots requiring less frequent cycles of seed multiplication. Conversely, Indica accessions without favourable haplotypes should be prioritized for more frequent viability monitoring to ensure that seed lots are above the viability standard. Seed longevity markers are also important from a breeding perspective. Rapid decline in seed viability leaves farmers vulnerable in terms of seed security, especially since the incidence of hot, humid weather is likely to increase under climate change. Viable seeds are critical to successful crop establishment and weed competitiveness during the early stages of crop growth, particularly under direct-seeded conditions, and thus improved seed longevity will be an increasingly important adaptive trait for farmers. In the context of direct seeding of rice, longevity should be recognized as part of the essential trait package. It could be of particular interest to hybrid programmes, where the value of the seeds is higher and poor germination becomes a liability. The loci detected in this study demonstrated sufficient improvement in longevity to merit further validation for potential use in breeding programmes for direct-seeded and hybrid varieties.
For this study, as well as characterizing the seed survival curve and taking into account the effect of seed maturity, the seed MC used was lower than in many other studies involving 'artificial ageing' or 'controlled deterioration' (Clerkx et al., 2004;Xue et al., 2008;Chatelain et al., 2012;Rehman Arif et al., 2012;Li et al., 2014;Hay et al., 2019). Some studies have additionally considered storage under 'natural' or 'conventional' storage conditions, at lower MC, and identified different QTLs contributing to differences in longevity in the different environments (Schwember and Bradford, 2010;Nagel et al., 2011). The target MC for the storage experiments here was 10.9 %, the predicted MC of desorbing rice seeds at 60 % RH. These conditions were used to ensure the results were comparable with other studies on cultivated and wild rice Whitehouse et al., 2015Whitehouse et al., , 2017Timple and Hay, 2018), and across diverse species (Probert et al., 2009;Mondoni et al., 2011;Merritt et al., 2014). Interestingly, this MC is less than that recommended for farmers and seed companies storing rice seeds (12-14 %). It is also within the linear part of the moisture desorption isotherm, where we expect there to be a linear relationship between seed longevity (log 10 σ) and MC (log 10 MC; Hay and Timple, 2016). In the case of accelerated ageing or controlled deterioration experiments that are carried out at much higher MC, particularly those above 85 % RH, different metabolic process might be expected to be occurring (Vertucci and Leopold, 1984), and hence identification of different 'seed longevity' loci is probable. Since we used a lower RH, we expect that seed lots (accessions) that lose viability more quickly in this storage environment would also lose viability more quickly in genebank storage. The accessions used are relatively new, derived from single-seed descent of original accessions, and hence there is not yet enough viability monitoring data to confirm that the identified genes may be influencing longevity in long-term storage. Nonetheless, based on the results of the GWA, we propose the use of the candidate markers to predict the relative longevity of Indica rice accessions to better manage the International Rice Genebank collection.

SUPPLEMENTARY DATA
Supplementary data are available online at https://academic. oup.com/aob and consist of the following. Table S1: origin of a diverse Indica rice panel used in the GWAS of seed longevity. Table S2: information on 299 Indica rice accessions used in the GWAS of seed longevity. Table S3: list of candidate genes associated with seed longevity traits of 299 Indica rice accessions. Figure S1: seed survival curves of 299 Indica rice accessions. Figure S2: GWA analysis on traits of 299 Indica rice accessions measured in this study. File S1: additional references for information on 299 Indica rice accessions.