Impact of rice GENERAL REGULATORY FACTOR14h (GF14h) on low-temperature seed germination and its application to breeding

Direct seeding is employed to circumvent the labor-intensive process of rice (Oryza sativa) transplantation, but this approach requires varieties with vigorous low-temperature germination (LTG) when sown in cold climates. To investigate the genetic basis of LTG, we identified the quantitative trait locus (QTL) qLTG11 from rice variety Arroz da Terra, which shows rapid seed germination at lower temperatures, using QTL-seq. We delineated the candidate region to a 52-kb interval containing GENERAL REGULATORY FACTOR14h (GF14h) gene, which is expressed during seed germination. The Arroz da Terra GF14h allele encodes functional GF14h, whereas Japanese rice variety Hitomebore harbors a 4-bp deletion in the coding region. Knocking out functional GF14h in a near-isogenic line (NIL) carrying the Arroz da Terra allele decreased LTG, whereas overexpressing functional GF14h in Hitomebore increased LTG, indicating that GF14h is the causal gene behind qLTG11. Analysis of numerous Japanese rice accessions revealed that the functional GF14h allele was lost from popular varieties during modern breeding. We generated a NIL in the Hitomebore background carrying a 172-kb genomic fragment from Arroz da Terra including GF14h. The NIL showed superior LTG compared to Hitomebore, with otherwise comparable agronomic traits. The functional GF14h allele from Arroz da Terra represents a valuable resource for direct seeding in cold regions.


Introduction
Low-temperature seed germination (LTG) is a pivotal agronomic trait in rice (Oryza sativa).As rice originated from tropical and subtropical regions, it is highly susceptible to low-temperature conditions compared to other cereal crops such as wheat (Triticum aestivum) and barley (Hordeum vulgare) (Okuno, 2004).Nevertheless, rice is produced in temperate and high-altitude regions, where it frequently experiences temperatures below 20°C.In Japan, rice is abundantly cultivated in relatively cold areas such as Tohoku and Hokkaido.In recent years, there has been an increasing demand to shift from conventional transplantation-based rice cultivation to direct seeding to reduce labor and costs.However, direct seeding raises the risk of exposure to low temperatures during seed germination (Iwata et al., 2010).Therefore, to expand the use of direct seeding, it is crucial to breed rice cultivars with enhanced LTG.
OsSAP16 presumably acts as a regulator of LTG.14-3-3 proteins are regulatory proteins that are widely conserved in eukaryotes.
QTL pyramiding has been proposed as a breeding concept (Ashikari & Matsuoka, 2006) for bringing together several QTLs (or genes) related to agronomically important traits in the genetic background of locally adapted elite cultivars.In practice, it is essential to generate pre-breeding materials for QTL pyramiding, i.e., near-isogenic lines (NILs) that harbor one or a few genomic segments introgressed from the donor parent into the genome of the recipient parent through a combination of continuous backcrossing and selfing via marker-assisted selection (Zhang et al., 2022).In this study, we determined that GF14h is responsible for an LTG-related QTL in Portuguese rice variety Arroz da Terra.We generated a NIL in the background of rice cultivar Hitomebore, which is adapted for growth in northern Japan, by replacing its GF14h genomic fragment with that from Arroz da Terra and tested its LTG performance.

Evaluation of germination rate
Seeds for each line were harvested 45 days after heading, air-dried at 30°C for two days, and stored at 4°C until use.The seeds were air-dried at 50°C for seven days in the dark to break dormancy.For germination tests, 50 seeds per replicate were incubated in a Petri dish filled with distilled water in the dark at 15°C (low-temperature conditions) or 25°C (optimal temperature conditions).The germination rate was calculated as the total number of germinated seeds at each time point divided by the number of seeds tested.Seeds were considered to have germinated when the white coleoptile was visible.
To evaluate resistance to pre-harvest sprouting, panicles were harvested from NIL-GF14h Arroz and Hitomebore 30 days after heading.The panicles were incubated in dark, wet conditions (by covering them with filter paper moistened with water) at 28°C for 10 days, and seed germination was scored.

QTL-seq analysis
LTG data for the 200 RILs derived from a cross between Iwatekko and Arroz da Terra (Fig. S1) were analyzed (Takagi et al., 2013).The top 20 RILs showing high-LTG and the bottom 20 RILs showing low-LTG phenotypes were selected to assemble the two bulk samples with contrasting LTG phenotypes.All seedlings with high or low LTG were pooled, and DNA was extracted from each bulk as previously described (Takagi et al., 2013).The genomic DNA of the two bulks was used to generate DNA-seq libraries and sequenced on a GAⅡx sequencer (Illumina, CA, USA).QTL-seq was performed to identify QTLs related to LTG (Takagi et al., 2013;Sugihara et al., 2022).

Map-based cloning of qLTG11
To narrow down the qLTG11 region, genotyping was performed using a cross population of qLTG11-NIL (BC2F3) backcrossed to Hitomebore.The germination rate under lowtemperature conditions (15°C) was measured to characterize LTG activity.Highresolution fine mapping with ten markers (markers B−K) between 23.466 Mb and 23.609 Mb on chromosome 11 identified six informative recombinants in the target region.
Primers used for mapping are listed in Table S1.

De novo assembly of the Hitomebore and Arroz da Terra genomes
To reconstruct the qLTG11 regions in Hitomebore and Arroz da Terra, de novo assembly was performed for each genome using Nanopore long reads and Illumina short reads according to a published method (Sugihara et al., 2023).To extract high-molecularweight genomic DNA from leaf tissue for Nanopore sequencing, a NucleoBond highmolecular-weight DNA kit (MACHEREY-NAGEL, Germany) was used.Following DNA extraction, low-molecular-weight DNA was eliminated using a Short Read Eliminator Kit XL (Circulomics, MD, USA).Library preparation was then performed using a Ligation Sequencing Kit (SQK-LSK-109; Oxford Nanopore Technologies [ONT], United Kingdom) according to the manufacturer's instructions, and sequencing was performed using MinION (ONT, UK) for Arroz da Terra.For Hitomebore, Nanopore long reads sequenced by Sugihara et al. (2023) were used.Base-calling of the Nanopore long reads was performed using Guppy 4.4.2(ONT, UK).Sequences derived from the lambda phage genome were removed from the raw reads with NanoLyse v1.1.0(De Coster et al., 2018).The first 50 bp of each read were then removed, as were reads with an average read quality score below 7 and reads shorter than 3,000 bases, using NanoFilt v2.7.1 (De Coster et al., 2018).The clean Nanopore long reads were assembled using NECAT v0.0.1 (Chen et al., 2021), setting the genome size to 380 Mb.To improve the accuracy of assembly, Racon v1.4.20 (Vaser et al., 2017) was used twice for error correction using Nanopore reads, and Medaka v1.4.1 (https://github.com/nanoporetech/medaka)was subsequently used to correct mis-assembly.Two rounds of consensus correction were then performed using bwa-mem v0.7.17 (Li & Durbin, 2009) and HyPo v1.0.3 (Kundu et al., 2019) with the Illumina short reads.Redundant contigs were removed using purgehaplotigs v1.1.1 (Roach et al., 2018)

Plant transformation
To generate GF14h knockout mutants, two single guide RNAs (sgRNAs) targeting exon 4 or exon 5 of GF14h were designed using the web-based service CRISPRdirect (crispr.dbcls.jp)(Naito et al., 2015) and cloned individually into the pZH::OsU6gRNA::MMCas9 vector (Mikami et al., 2015).The resulting vectors were introduced into Agrobacterium (Agrobacterium tumefaciens) strain EHA105 for transformation into qLTG11-NIL plants (Toki et al., 2006).The target sites in the positive transformants were sequenced by Sanger sequencing to detect mutations.To obtain overexpression constructs, the full-length coding sequence of functional GF14h was amplified from total RNA extracted from qLTG11-NIL and cloned into the plant binary vector pCAMBIA1300 under the control of the cauliflower mosaic virus (CaMV) 35S promoter.The overexpression plasmid was introduced into Agrobacterium strain EHA105 for transformation of rice variety Hitomebore (Toki et al., 2006).All primers used are listed in Table S1.

Expression analysis
Total RNA was extracted from germinating seeds using an RNA-suisui S kit (Rizo).Total RNA was treated with RNase-free DNase I (Nippon Gene).The resulting samples were reverse transcribed into first-strand cDNA using a PrimeScript RT Reagent Kit (Takara Bio).Quantitative PCR (qPCR) was conducted using a QuantStudio 3 system (Thermo Fisher Scientific) with Luna Universal qPCR Master Mix (New England Biolabs).The cycling parameters were 1 min at 95°C, followed by 40 cycles of amplification (95°C for 15 sec and 60°C for 30 sec).The Actin gene (Os03g0718100) served as an internal control, and the Delta CT method was used to calculate the relative expression levels.The primer sets are listed in Table S1.

RNA-seq
Total RNA was extracted from Hitomebore and qLTG11-NIL seeds at 0, 1, 2, and 3 days after the onset of seed hydration under low (15°C) or optimum (25°C) temperature conditions using an RNA-suisui S kit (Rizo, Ibaraki, Japan).Sequencing libraries were prepared using an NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs Japan, Tokyo, Japan) following the manufacturer's protocol.The libraries were sequenced in paired-end mode using an Illumina HiSeq X instrument (Illumina, CA, USA).The raw reads have been deposited in the DNA Databank of Japan (BioProject accession No. PRJDB17450).For quality control, reads shorter than 50 bases and those with an average read quality below 20 were discarded using Trimmomatic v0.36 (Bolger et al., 2014), and poly(A) sequences were trimmed using PRINSEQ++ v1.2 (Cantu et al., 2019).The resulting clean reads were aligned to the de novo assembled Hitomebore and Arroz da Terra genomes with HISAT2 v2.1 (Kim et al., 2019).BAM files were sorted and indexed with SAMtools v1.10 (Li et al., 2009), and aligned reads were assembled into transcripts with StringTie (Pertea et al., 2015) by combining bam files for each variety.

Haplotype network analysis
Sequencing datasets were obtained for 503 rice accessions.Of these, 379 were FASTQ files downloaded from the DNA Data Bank of Japan Sequence Read Archive (DRA) (Yabe et al., 2018;Tanaka et al., 2020;Tanaka et al., 2021;Shimono et al., 2023) and 124 were sequenced in this study (Table S2).Details about DNA extraction, wholegenome sequencing techniques, and construction of the genotype datasets in VCF format are provided in a previous report (Shimono et al., 2023).This study specifically focused on the coding region of GF14h.Genotype information related to the coding region of GF14h was extracted from the VCF dataset.In addition, the k-mer analysis program (https://github.com/taitoh1970/kmer)(Itoh et al., 2020) was used with Illumina short reads to detect the 4-bp deletion with high sensitivity.Genotype information for the presence of the 4-bp deletion was added to the VCF file, and 81 samples with heterozygous genotypes were discarded.A haplotype network was then constructed using the median-joining network algorithm (Bandelt et al., 1999) implemented in Popart v1.7 (Leigh et al., 2015).

GF14h Arroz
The grain yields of Hitomebore and NIL-GF14h Arroz were investigated in experimental paddy fields in 2023.Field experiments were conducted at the Iwate Agricultural Research Center (39°35'N, 141°11'E) in Kitakami, Iwate, Japan.A fertilization regime of N:P2O5:K2O = 6:6:6 g m −2 was applied as a basal dressing, and N:K2O = 2:2 g m -2 was applied as a top dressing.Seeds were sown in a seedling nursery box on 21 April, and seedlings were transplanted to the paddy field on 18 May.To evaluate agronomic traits, the seedlings were transplanted at a rate of one plant per hill, with a planting density of 22.2 hills m −2 .Culm length, panicle length, panicle number, grain number per plant, and grain weight per plant were measured at maturity.To evaluate yield performance, seedlings were transplanted with three plants per hill at a planting density of 16.7 hills m −2 .The 0.9 × 5.0 m experimental plots in the paddy fields were arranged in a randomized complete block design with three replicates.At maturity, 50 hills were harvested from each plot to measure brown rice yield.The hulls were removed using a rice huller (Model 25MC, Ohya Tanzo Factory Co., Ltd., Japan), and the hulled grains were screened with a grain sorter (1.9-mm sieve size).Brown rice yields were adjusted to 15% moisture content and converted to weight per hectare.

Evaluation of QTLs associated with low-temperature germination using the Portuguese rice variety Arroz da Terra
The Portuguese rice variety Arroz da Terra germinates more vigorously under lowtemperature conditions than the Japanese cultivars Iwatekko and Hitomebore (Fig. 1a).
To identify the genes responsible for this difference, we searched for QTLs involved in the high LTG of Arroz da Terra.We previously generated a set of 200 RILs at the F7 generation derived from a cross between Arroz da Terra and Iwatekko (Fig. S1) (Takagi et al., 2013).We phenotyped all RILs for LTG and selected the 20 RILs with the highest LTG and the 20 RILs with the lowest LTG.We assembled two pools of seedlings with low or high LTG and extracted their genomic DNA for whole-genome sequencing on the Illumina platform (Fig. S1) (Takagi et al., 2013).We mapped the resulting sequencing reads to the Nipponbare rice reference genome (IRGSP-1.0)and performed QTL-seq analysis using our new high-performance pipeline (Sugihara et al., 2022).Based on the ΔSNP index, we identified three QTLs related to LTG on chromosome 3 (qLTG3-1 and qLTG3-2) and chromosome 11 (qLTG11) (Fig. 1b), which is consistent with the results of a previous study (Takagi et al., 2013).
The qLTG3-1 region contained the gene Os03g0103300, which was reported to be involved in LTG in a study using rice cultivar Italica Livorno, which has high LTG, and Hayamasari, which has low LTG (Fujino et al., 2008).An examination of its coding sequences in Arroz da Terra, as well as Iwatekko and Hitomebore, revealed that they were identical to those found in Italica Livorno and Hayamasari, respectively (Fig. S4).While Italica Livorno harbored a functional haplotype for this gene, Hayamasari carried a lossof-function haplotype due to a 71-bp deletion (Fig. S4) (Fujino et al., 2008).Therefore, we propose that the causal gene for the QTL LTG3-1 is Os03g0103300.
To evaluate the contribution of the two other QTLs to LTG, we generated NILs harboring a segment from the Arroz da Terra genome for each QTL (approximately 5 Mb) in the Hitomebore background (Figs. 1c,S2a,S3a).We detected no clear effect on LTG, as qLTG3-2-NIL and Hitomebore showed similar seed germination rates at 15ºC (Fig. S3b).By contrast, the germinability of qLTG11-NIL was more vigorous than that of Hitomebore at 15ºC (Fig. 1d), indicating that qLTG11 enhances LTG.qLTG11-NIL seeds also germinated more rapidly than Hitomebore seeds under normal conditions (25ºC), although with a smaller difference between the two genotypes than at low temperature (Fig. S5).We therefore focused our analysis on qLTG11.

Identification of GF14h as the candidate gene for qLTG11
To delineate the qLTG11 region, we carried out map-based cloning using a segregating population derived from a cross between BC2F3 line qLTG11-NIL and Japanese elite cultivar Hitomebore (Fig. S2a).For mapping, we conducted germination tests at 15°C.We narrowed down the genomic region containing the QTL to a 52-kb segment (from 23.512 bp to 23.564 Mb) on chromosome 11 based on the Nipponbare reference genome (IRGSP-1.0)(Fig. 2a).This interval contains two annotated genes based on the Nipponbare genome sequence (Fig. 2b).We compared the genomic sequence of Hitomebore and Arroz da Terra across the candidate region using de novo genome assembly obtained from Nanopore long reads.The cultivars Hitomebore and Nipponbare had an identical genomic sequence over the entire candidate region (Fig. S6a).By contrast, the genome sequence from Arroz da Terra was substantially different from that of Nipponbare, with the equivalent candidate region spanning approximately 94 kb (Figs 2b,    S6b).
As the causal gene behind the variation in LTG is likely expressed in seeds, we performed transcriptome deep sequencing (RNA-seq) during seed germination in Hitomebore and qLTG11-NIL.Within the candidate region, the gene Os11g0609600, corresponding to the 14-3-3 gene GF14h, was expressed in both Hitomebore and qLTG11-NIL.The GF14h gene structure and haplotypes in Arroz da Terra and Hitomebore are shown in Fig. 2c.We detected a 4-bp deletion in the GF14h coding region in Hitomebore, causing a frameshift mutation predicted to introduce a premature stop codon (Figs.2c, S7).These results suggest that Hitomebore carries a loss-of-function allele of GF14h.To assess the role of GF14h in LTG, we examined the expression pattern of the putative functional GF14h (GF14h Arroz ) allele during seed germination at low temperature (15°C) using qLTG11-NIL.RT-qPCR analysis of GF14h expression levels showed that they were comparable in the embryo and endosperm at 1 and 3 days after the onset of seed imbibition (Fig. 2d).At the beginning of germination, when a white coleoptile was visible (5 and 7 days after seed imbibition), GF14h expression levels rose in the endosperm, but not in the embryo (Fig. 2d).By nine days of imbibition, when most seeds had germinated, GF14h expression in the endosperm returned to basal levels (Fig. 2d).These results support the notion that GF14h plays a role in seed germination at low temperature.

GF14h plays a vital role in LTG
To investigate the contribution of GF14h to LTG, we knocked out the functional GF14h copy present in qLTG11-NIL by clustered regularly interspersed short palindromic repeat (CRISPR)/CRISPR-associated nuclease 9 (Cas9)-mediated gene editing and evaluated LTG.Specifically, we introduced two single guide RNA (sgRNA) constructs targeting the exons of GF14h individually into qLTG11-NIL by Agrobacterium-mediated transformation.We chose to knock out GF14h in the qLTG11-NIL background rather than Arroz da Terra to evaluate the specific contribution of GF14h to LTG without the influence of qLTG3-1, which would be present in the Arroz da Terra background.To accurately evaluate the phenotypes of the edited plants, we selected heterozygous plants in the T0 generation and isolated homozygous mutant lines and their unedited homozygous siblings in the T1 generation.We obtained four knockout lines (gf14h-1, gf14h-2, gf14h-3, and gf14h-4) and their wild-type sibling (WT Arroz ) (Fig. S8).We detected significant drops in LTG rates in all four knockout lines compared to WT Arroz (Fig. 3a,b).Furthermore, we generated transgenic lines in the Hitomebore background overexpressing the functional GF14h allele from Arroz da Terra under the control of the CaMV 35S promoter.In these overexpression lines, GF14h expression increased approximately 1,000-fold compared to the wild-type sibling (WT Hitomebore ) (Fig. S9).
Importantly, the overexpression lines showed higher LTG than WT Hitomebore when tested at 15ºC (Fig. 3c,d).Taken together, these data indicate that GF14h is a key gene involved in LTG.

Loss-of-function alleles GF14h and qLTG3-1 increased in frequency during rice breeding in Japan
We reconstructed the GF14h haplotype network using genotype data obtained from whole-genome resequencing of 492 O. sativa accessions from various collections, including the World Rice Core Collection (Tanaka et al., 2020), the Rice Core Collection of Japanese Landraces (Tanaka et al., 2021), and a set of Japanese landraces and modern varieties (Shimono et al., 2023), in addition to 11 wild rice (O.rufipogon) accessions (Zhao et al., 2018) (Table S2).We distinguished 10 haplotypes for the GF14h coding region based on 11 polymorphic sites comprising one frameshift mutation caused by a 4bp deletion, four nonsynonymous single nucleotide polymorphisms (SNPs), and six synonymous SNPs (Table S3).The conversion of the functional allele Hap2 to its nonfunctional allele Hap1 required only a single step: a 4-bp deletion (Fig. S10).
We analyzed the haplotype frequencies of GF14h and qLTG3-1 in Japanese landraces and cultivars, which we grouped according to their time of release.More than half of all Japanese landraces carried functional alleles of both GF14h and LTG3-1 (Fig. 4a).The next most common allele combination was a nonfunctional GF14h allele with a functional LTG3-1 allele (Fig. 4a).The percentage of lines with loss-of-function alleles at both GF14h and LTG3-1 has increased since the beginning of crossbreeding in Japan in the early 20th century, with more than 80% of varieties released after 2001 carrying loss-of-function alleles for both genes (Fig. 4a,b).Looking at each gene separately in landraces, only a few lines carried a loss-of-function allele for LTG3-1, whereas roughly half of all lines already harbored a loss-of-function allele for GF14h (Fig. 4b).Modern breeding thus appears to have increased the proportion of loss-of-function alleles for these two genes, with a substantial increase in LTG3-1, reaching almost 90% among lines bred after 2001 (Fig. 4b).

The Arroz da Terra GF14h allele could be valuable for rice breeding
To assess how useful the above findings might be to practical breeding programs, we developed new breeding materials.QTL pyramiding, a strategy for introducing multiple QTLs for desired traits into a single genetic background, is a key strategy employed in current breeding.An essential step in QTL pyramiding is the generation of NILs containing the desired QTLs.Therefore, we developed a NIL, termed NIL-GF14h Arroz , using the elite cultivar Hitomebore as the genetic background into which we introgressed a 172-kb region from the Arroz da Terra genome containing GF14h (Fig. S2).This NIL showed a higher seed germination rate under low-temperature conditions compared to Hitomebore (Fig. 5a,b).In addition, NIL-GF14h Arroz was slightly more susceptible to preharvest sprouting than Hitomebore (Fig. S11).Importantly, we observed no substantial differences in five agronomic traits (culm length, panicle length, panicle number, grain number, and grain weight) between NIL-GF14h Arroz and Hitomebore (Fig. 5c-g).Brown rice yield was slightly lower in NIL-GF14h Arroz compared to Hitomebore, but a sufficient yield was guaranteed (Fig. 5h,i).These results suggest that NIL-GF14h Arroz could be a valuable parental line for breeding via QTL pyramiding.

Discussion
Here, we demonstrated that the functional GF14h allele present in Portuguese rice variety Arroz da Terra plays a pivotal role in supporting seed germinability under lowtemperature conditions.Although the regulation of seed germination by GF14h was recently documented, its activity under low-temperature conditions remained unclear (Sun et al., 2022;Yoshida et al., 2022).While many genomic regions associated with LTG have been detected through QTL mapping and GWAS, only a few studies have identified the causal genes (Miura et al., 2001;Fujino et al., 2004;Han et al., 2006;Jiang et al., 2006;Fujino et al., 2008;Nguyen et al., 2012;Li et al., 2013;Fujino et al., 2015;Pan et al., 2015;Satoh et al., 2015;Jiang et al., 2017;Shakiba et al., 2017;Wang et al., 2018;Jiang et al., 2020;Shim et al., 2020;Yang et al., 2020;Pan et al., 2021;Mao et al., 2022).Indeed, LTG is a quantitative trait involving the cumulative effects of multiple genes and their epistatic relationships, making it difficult to assess the specific effect of a single genomic region on LTG.To eliminate the influence of other chromosomal regions on LTG, we first generated qLTG11-NIL containing only one of three QTLs detected in the Arroz da Terra background for genetic analysis.The analysis of qLTG11-NIL revealed that GF14h participates in LTG.Significantly, the NIL harboring the functional GF14h allele from Arroz da Terra in the Hitomebore background provides valuable pre-breeding materials for QTL pyramiding.These findings provide a genetic understanding of lowtemperature germinability as well as new resources for rice breeding.

Influence of GF14h on low-temperature germination
In this study, we identified GF14h as being implicated in LTG.In a previous study, a genetic complementation assay with a functional GF14h allele introduced into the rice cultivar Nipponbare background increased the germination rate at 30°C, but only to a limited extent at 15°C (Yoshida et al., 2022).This result is not consistent with our finding that introducing functional GF14h into Hitomebore resulted in a significant improvement in germination at low temperature.Perhaps this discrepancy is due to differences in the rice varieties used in the germination assays.Notably, the haplotypes of qLTG3-1, a major QTL behind LTG (Fujino et al., 2008), are different between Nipponbare and Hitomebore: whereas Nipponbare, which was released in 1961, harbors the functional allele of qLTG3-1, Hitomebore carries a loss-of-function allele with a deletion of 71 bp (Fig. S4) (Hori et al., 2010).Moreover, the cultivars Koshihikari and Hayamasari, which carry the same loss-of-function qLTG3-1 allele as the Hitomebore variety, were reported to exhibit lower germination rates at low temperatures than Nipponbare (Hori et al., 2010).
LTG tests using chromosome segment substitution lines derived from a cross between Koshihikari and Nipponbare indicated that qLTG3-1 contributes to the difference in LTG between the two varieties (Hori et al., 2010).Based on these observations, it is likely that Nipponbare has a better LTG ability than Hitomebore, which may have masked the effect of functional GF14h on LTG in the previous study (Yoshida et al., 2022).Our study confirmed the involvement of GF14h in LTG through map-based cloning and analysis of knockout and overexpression lines.It is also worth mentioning that our experiments were performed in the qLTG11-NIL background, which allowed us to isolate the contribution of GF14h to LTG without any influence from qLTG3-1 or other genes in the Arroz da Terra background.In summary, we provided multiple lines of evidence that GF14h contributes to LTG.
The expression pattern of functional GF14h during seed germination was previously unclear.While GF14h has been shown to be expressed in the aleurone layer surrounding the embryo (Yoshida et al., 2022), it is also highly expressed in the endosperm (Sun et al., 2022).In the current study, we showed that GF14h was expressed throughout the seeds, but with a transient induction in expression in the endosperm at roughly the time of initiation of seed germination.GF14h was reported to regulate seed germination by interacting with the abscisic acid and gibberellin signaling pathways at optimal temperatures (Sun et al., 2022;Yoshida et al., 2022).We therefore suggest that GF14h controls LTG by interacting with various phytohormone signaling pathways.

Low-temperature germinability in rice was lost due to selection in modern Japanese breeding
In this study, we performed haplotype network analysis of GF14h using many Japanese rice varieties.We identified ten distinct haplotypes based on 11 polymorphic sites in the GF14h coding region.Of these, Hap4, encompassing the aus, indica, tropical japonica, temperate japonica, and O. rufipogon accessions, was defined at the center of the haplotype network.Furthermore, we determined that a 4-bp deletion converted the functional haplotype Hap2, which was derived from Hap4, into the nonfunctional haplotype Hap1.This finding is consistent with the relationship between Hap6 and Hap1 (which we defined as Hap2 and Hap1, respectively) observed by Sun et al. (2022).These results suggest that the nonfunctional allele represented by Hap1 was introduced into temperate japonica varieties from tropical japonica varieties carrying Hap2 and then spread to Japanese cultivars.
We studied the haplotype frequencies of GF14h and qLTG3-1 in various Japanese landraces and cultivars, considering their time of release from breeding programs into the field.More than half of the Japanese landraces analyzed carried both functional GF14h and qLTG3-1 alleles.However, the frequency of varieties carrying loss-of-function alleles for both GF14h and qLTG3-1 has increased since crossbreeding began in the early 20th century.This trend has continued to the present, perhaps as a result of artificial selection to improve resistance to pre-harvest sprouting, with more than 80% of all varieties bred since 2001 carrying these loss-of-function alleles.
Our study provides a historical perspective on allelic shifts in Japanese rice breeding while highlighting the influence of modern breeding on genetic diversity.Further research is needed to elucidate the potential effects of higher frequencies of loss-of-function alleles on the overall phenotypic characteristics and ecological adaptability of Japanese rice varieties.In addition, as mentioned in previous reports (Sun et al., 2022;Yoshida et al., 2022), we believe that the reintroduction of functional alleles should be considered in order to develop cultivars suitable for labor-saving cultivation techniques such as direct seeding.

Application to direct seeding for rice cultivation
Cultivation stability under direct seeding conditions is important for managing rice production costs and reducing labor.However, since rice is sensitive to low temperatures, improving seed germination and seedling establishment at low temperatures is a desirable breeding trait in high-latitude rice production areas such as Japan.In the current study, we developed a potentially useful NIL (NIL-GF14h Arroz ) by introducing the functional GF14h allele into the Hitomebore background.Overexpressing this functional GF14h allele was previously shown to improve anaerobic germination and tolerance to seedling establishment under anaerobic conditions in laboratory experiments (Sun et al., 2022).However, whether NIL-GF14h Arroz exhibits strong seedling vigor at low temperatures in rice fields remains to be determined.We previously identified a QTL associated with seedling vigor, qPHS3-2 (QTL for plant height of seedling 3-2) (Abe et al., 2012).
qPHS3-2 most likely corresponds to the gibberellin biosynthesis gene GA20 oxidase1 (OsGA20ox1), a paralog of Semi Dwarf1 (sd-1, corresponding to OsGA20ox2) (Abe et al., 2012).Therefore, the pyramiding of qPHS3-2 in NIL-GF14h Arroz by marker-assisted selection represents a promising approach for further improving seedling vigor in NIL- In recent years, "early-winter direct seeding" has been experimentally tested as a new system of direct seeding for rice production in Japan (Shimono, 2020).In this system, seeds are directly sown in the early winter of the previous year instead of the spring.The sown seeds thus overwinter in snow-covered soil and germinate the following spring.The major advantage of this approach is that it can significantly decrease the amount of labor required by farmers during the busy spring season.However, it is challenging to overwinter the seeds of modern rice varieties in the soil and achieve good seedling establishment (Shimono et al., 2012;Oikawa et al., 2019;Shimono, 2020).We expect that reintroducing beneficial alleles like GF14h Arroz that were lost during modern breeding into future rice varieties will enable the implementation of new cultivation practices and increase productivity.A total of 350 Japanese varieties were examined, including the World Rice Core Collection (Tanaka et al., 2020), the Rice Core Collection of Japanese Landraces (Tanaka et al., 2021), and the collection of Japanese core cultivars (Shimono et al., 2023).The allele type at each gene was determined to be functional or nonfunctional by the k-mer method using Illumina short reads for each variety.

Fig 1 .
Fig 1. Effects of quantitative trait loci (QTLs) on low-temperature seed germinability.(a) Representative photographs showing the germination of seeds from the Iwatekko, Hitomebore, and Arroz da Terra varieties 10 days after the onset of seed imbibition at 15°C.Scale bar, 1 cm.(b) Map positions of QTLs for low-temperature germination, as determined by QTL-seq.The Δ (SNP-index) values (red lines) were plotted for chromosomes 3 and 11, with statistical confidence intervals under the null hypothesis of no QTL (green, P < 0.05; orange, P < 0.01).(c) Diagram showing the genotype of qLTG11-NIL.qLTG11-NIL harbors the Arroz da Terra allele at qLTG11 on chromosome 11.Light blue indicates genomic fragments from Hitomebore; red indicates genomic fragments from Arroz da Terra; dark blue indicates heterozygous regions.(d) Germination time courses of Hitomebore, qLTG11-NIL, and Arroz da Terra at 15°C.Values are means ± standard deviation (SD) from biologically independent samples (Hitomebore and NIL, n = 10; Arroz da Terra, n = 5).

Fig 2 .
Fig 2. Positional cloning of qLTG11.(a)Fine mapping of qLTG11 to a 52-kb region between markers E and J.The chromosomal positions are based on the Nipponbare reference genome (Os-Nipponbare-Reference-IRGSP-1.0).Germination percentage was determined at 9 days of incubation at 15°C.Red and blue rectangles indicate chromosomal segments homozygous for Arroz da Terra or Hitomebore, respectively.Different lowercase letters indicate significant differences (n = 3 biologically independent samples, P < 0.001, Tukey's HSD test).(b) Genomic structure of the candidate genomic region in Arroz da Terra and Hitomebore.Os11g0609600 (shown in red), encoding GF14h, is expressed in germinating seeds.(c) Diagram of the GF14h gene structure and sequence polymorphisms between Arroz da Terra and Hitomebore.The chromosomal positions are based on the Nipponbare reference genome.The coding region of GF14h in Hitomebore is identical to that in Nipponbare.The 4-bp deletion in Hitomebore causes a frameshift and the introduction of a premature stop codon.(d) Relative GF14h expression levels in germinating seeds of qLTG11-NIL.In the boxplots, the box edges represent the upper and lower quantiles, the horizontal line in the middle of the box represents the median value, whiskers represent the lowest quantile to the top quantile, and the black squares show the mean.Five biological replicates were measured independently.Different lowercase letters indicate significant differences based on Tukey's HSD test (P < 0.05).OsActin1 (Os03g0718100) was used for normalization.

Fig 3 .
Fig 3. Effect of GF14h mutation and overexpression on low-temperature germination.(a) Representative photographs showing seed germination in wild-type harboring Arroztype GF14h (WT Arroz ) and CRISPR/Cas9 knockout lines (gf14-1) at 8 days after the onset of seed imbibition.Scale bar, 1 cm.(b) Seed germination rate of WT Arroz and its CRISPR/Cas9 knockout lines at 7 days of seed imbibition at 15°C.The two target constructs (Fig. S8) were introduced into the qLTG11-NIL line.Data are means ± standard error (SE, n = 3).Different lowercase letters indicate significant differences based on Tukey's HSD test (P < 0.01).(c) Representative photographs showing seed germination of wild-type (WT Hitomebore ) and OsGF14h Arroz overexpression lines (OsGF14h Arroz -Ox #2) at 7 days of seed imbibition at 15ºC.Scale bar, 1 cm.(d) Seed germination rate of WT Hitomebore and GF14h Arroz overexpression lines in the Hitomebore background at 7 days of seed imbibition at 15°C.Data are means ± SE (n = 3).Different lowercase letters indicate significant differences based on Tukey's HSD test (P < 0.05).

Fig 4 .
Fig 4. Gradual selection of loss-of-function alleles in GF14h and qLTG3-1 during rice breeding in Japan.A total of 350 Japanese varieties were examined, including the World Rice Core Collection(Tanaka et al., 2020), the Rice Core Collection of Japanese Landraces(Tanaka et al., 2021), and the collection of Japanese core cultivars(Shimono et al., 2023).The allele type at each gene was determined to be functional or nonfunctional by the k-mer method using Illumina short reads for each variety.(a) Proportion of allele type combinations at GF14h and qLTG3-1 sorted by breeding year.(b) Proportion of nonfunctional allele types at GF14h and qLTG3-1 sorted by breeding year.

Fig 5 .
Fig 4. Gradual selection of loss-of-function alleles in GF14h and qLTG3-1 during rice breeding in Japan.A total of 350 Japanese varieties were examined, including the World Rice Core Collection(Tanaka et al., 2020), the Rice Core Collection of Japanese Landraces(Tanaka et al., 2021), and the collection of Japanese core cultivars(Shimono et al., 2023).The allele type at each gene was determined to be functional or nonfunctional by the k-mer method using Illumina short reads for each variety.(a) Proportion of allele type combinations at GF14h and qLTG3-1 sorted by breeding year.(b) Proportion of nonfunctional allele types at GF14h and qLTG3-1 sorted by breeding year.