Rice (Oryza sativa L.), a global staple food now grown on all inhabited continents, was domesticated from its wild progenitor, O. rufipogon Griff., in tropical and subtropical regions of Asia (Oka, 1988). After domestication, the expansions of rice landraces into the present-day range required a diverse array of adaptations to local environments, which included changes in daylight sensitivity, expanded thermal tolerance (for excess cold and heat), adaptations to water availability (drought and waterlogging), and resistance to biotic stresses (Garris et al., 2005; Glaszmann, 1987). Although a large amount of genomic data has been available for wild and cultivated rice varieties, and genetic characterizations of important agronomic traits were obtained in the past two decades (Gutaker et al., 2020; Huang et al., 2011, 2012; Wang et al., 2016), a complete landscape of genomic variations underlying regional adaptations remains elusive.

We selected 185 wild rice (O. rufipogon) and 743 cultivated rice varieties, which represent 33 major rice-growing regions worldwide (Figure 1a). Of them, all 185 wild rice varieties and 371 cultivated rice varieties (203 japonica and 168 indica varieties) were newly sequenced (see methods and material on https://github.com/vya-caas/zheng). The sequencing depth for these varieties was >5, which is for the first time to sequence such a large amount of wild rice varieties at this sequencing depth. Through a combination of principal component analysis (PCA), we confirmed five cultivated rice subgroups, including aus, indica (ind), tropical japonica (trj), temperate japonica (tej), and aromatic (aro) (Figure 1b). In particular, the aus and ind varieties, cultivated primarily in tropical and subtropical regions of Asia, were derived from the wild rice clade designated as OR-I. The trj and tej varieties, both originated from the OR-J clade of wild rice, were cultivated primarily in tropical, high-elevation regions of Southeast Asia (trj) and colder regions of Northeast Asia (tej).

To understand the selection on critical traits and the recent rapid speciation, three robust indicators of selective sweeps were examined to identify top outliers: (i) maximum differentiation from wild rice; (ii) maximum negative residuals of diversity/differentiation metrices based on the Hudson–Kreitman–Aguade test; and (iii) top scoring windows from a site frequency spectrum composite likelihood ratio test based on the length of haplotype. Using high-stringency cut-offs, considerable overlaps in outliers were observed among these methods. We identified 131 genomic regions with strong selection signals. Within individual rice subgroups, the numbers of targeted regions specific to tej, trj, ind, and aus were 20, 11, 20, and 12, respectively, including 791, 1477, 2647, and 2786 genes, respectively. Some of them were strong candidates for selection imposed by human preferences and local environmental conditions.

To identify loci associated with local adaptation, we collected temperature and day-length information in May and August of the last 68 years across the sampling sites in the Figure 1a and performed a genotype-environment association analysis based on a Bayesian analysis based on allele frequency data (Coop et al., 2010). As shown in Manhattan plots (Figure 1c,d), 128 significant association signals/loci (defined by –log₁₀^p > 3) were identified. Among these significant association signals/loci, 25 were located in genic regions (exonic or intronic), including eight for day length (OsLFL1, RNC3, OsDof2, APG, RFT1, Hd1, Ghd7, and ROC4; Figure 1c) and one for temperature sensitivity (COLD1; Figure 1c). The remaining 103 loci were located in non-genic regions, more than half of which (59, or 57.28%) coincided with top selective sweep candidate regions detected in the four cultivated rice subgroups, including 32 specifics to tej, 16 to trj, 8 to aus, and 3 to ind. These were strong candidates as targets of selection for local adaptation.

Based on the adaptation-associated results, the top five significant sites were found on chromosome (Chr.) 8 (Figure 1d). We only identified a non-synonymous SNP (C/A) with a significant association peak (P = 1.31 × 10⁻²²). This SNP was located in the coding region of LOC_Os08g36000 on Chr. 8 (+22 689 418 bp, C > A; +40 aa, Val > Phe) (Figure 1e). The encoded protein contained an F-box and two LRR domains. Many genes in this family respond to biotic and abiotic stresses, including temperature and hormones (Yan et al., 2010). The ‘C’ haplotype (Hap^tej) was almost fixed in the tej subgroup, with a frequency of 0.985, whereas the ‘A’ haplotype (Hap^other) predominated in other subgroups included aus, ind, trj, and wild rice varieties, with a frequency of 0.961 (Figure 1f,h). To address the function of the Hap^tej allele, we performed cold treatments on 14 cultivated rice varieties with the Hap^tej allele and 19 cultivated rice varieties with the Hap^other allele in a growth chamber, and the results showed that under cold stress (4 °C for 48 h), the survival rate was higher in rice varieties with the Hap^tej allele (93.25%) than those with the Hap^other allele (10.12%; P = 1.66 × 10⁻¹³, two-tailed t-test) (Figure 1g,h,k). We therefore named the gene CHILLING-TOLERANCE DIVERGENCE F-box (COLDF).

To further characterize the function of COLDF, we obtained TILLING and CRISPR mutants of the gene in the ZH11 background (O. sativa ssp. japonica var. Zhonghua 11) (Figure 1i,j). The TILLING mutant carried a G to A mutation in the coding region (+22 688 780 bp, G > A; +256 aa, Ala > Thr), and the knockout mutant carried a 22-bp insertion in the coding region, designated as coldf-1 and coldf-2, respectively. Fourteen-day-old seedlings of the homozygous coldf-1 and coldf-2 mutants were subjected to cold stress at 4 °C for 48 h, and then returned to normal growth conditions at 30 °C for recovery. Results showed that these coldf seedlings were susceptible to cold stress, with a survival rate of 37% and 12% in coldf-1 and coldf-2, respectively, in comparison with a survival rate of 82% in ZH11 (Figure 1i,j; P = 1.07 × 10⁻¹⁰, two-tailed t-test for coldf-1 and P = 1.94 × 10⁻¹², two-tailed t-test for coldf-2). These results suggest that the Hap^tej of COLDF is responsible for enhanced cold tolerance in the tej cultivars. Our findings provide insight into the mechanistic basis of rice domestication and improvement by identification of adaptation-associated loci, which would help in mitigating these effects by facilitating rapid development of improved varieties that are better able to tolerate the stresses accompanying the changing climate.

水稻（Oryza sativa L.）是一种全球性的主食，现在种植在所有有人居住的大陆上，它是从其野生祖先O.驯化而来的。 rufipogon Griff.，在亚洲的热带和亚热带地区（Oka，1988 年）。驯化后，水稻地方品种扩展到今天的范围需要对当地环境进行各种适应，包括日光敏感性的变化、扩大的耐热性（对过冷和过热）、对水资源可用性的适应（干旱和内涝） )，以及对生物胁迫的抗性 (Garris et al ., 2005 ; Glaszmann, 1987）。尽管野生和栽培水稻品种的大量基因组数据是可用的，并且在过去的二十年中获得了重要农艺性状的遗传表征（Gutaker et al ., 2020 ; Huang et al ., 2011 , 2012 ; Wang et al., 2011, 2012） al ., 2016 )，区域适应的基因组变异的完整景观仍然难以捉摸。

我们选择了 185 个野生稻 ( O. rufipogon ) 和 743 个栽培稻品种，它们代表了全球 33 个主要水稻种植区（图 1a）。其中，185个野生稻品种和371个栽培稻品种（粳稻203个，籼稻168个）全部新测序（方法和资料见https://github.com/vya-caas/zheng）。这些品种的测序深度>5，这是第一次在该测序深度对如此大量的野生稻品种进行测序。通过主成分分析（PCA）的结合，我们确定了五个栽培稻亚群，包括澳大利亚、籼稻（ind）、热带粳稻( trj )、温带粳( tej ) 和芳香( aro ) (图 1b)。特别是，主要在亚洲热带和亚热带地区种植的aus和ind品种来自被指定为 OR-I 的野生稻进化枝。trj和tej品种均起源于野生稻的OR-J 进化枝，主要种植在东南亚的热带高海拔地区 ( trj ) 和东北亚较冷的地区 ( tej )。

为了了解对关键性状的选择和最近的快速物种形成，检查了三个稳健的选择性扫描指标以确定顶级异常值：（i）与野生稻的最大分化；(ii) 基于 Hudson-Kreitman-Aguade 检验的多样性/分化度量的最大负残差；(iii) 基于单倍型长度的站点频谱复合似然比测试的最高评分窗口。使用高严格性截止值，在这些方法中观察到大量异常值重叠。我们确定了 131 个具有强选择信号的基因组区域。在单个水稻亚组中，特定于tej、trj、ind和aus的目标区域的数量分别为 20、11、20 和 12 个，分别包括 791、1477、2647 和 2786 个基因。其中一些是人类偏好和当地环境条件强加的选择的有力候选者。

为了识别与局部适应相关的基因座，我们在图 1a 中的采样点收集了过去 68 年 5 月和 8 月的温度和日长信息，并基于基于等位基因频率的贝叶斯分析进行了基因型-环境关联分析数据（Coop等人，2010 年）。如曼哈顿图（图 1c，d）所示，识别出 128 个重要的关联信号/位点（由 –log ₁₀ ^p  > 3 定义）。在这些重要的关联信号/基因座中，25 个位于基因区域（外显子或内含子），包括 8 个日长（OsLFL1、RNC3、OsDof2、APG、RFT1、Hd1、Ghd7和ROC4；图 1c) 和一个用于温度敏感性 ( COLD1；图 1c)。其余 103 个位点位于非基因区域，其中一半以上（59 个，或 57.28%）与四个栽培水稻亚组中检测到的顶级选择性扫描候选区域一致，包括 32 个tej特异性、16 个trj特异性、8到aus和 3 到ind。这些是作为本地适应选择目标的有力候选者。

根据与适应相关的结果，在染色体（Chr.）8 上发现了前五个重要位点（图 1d）。我们仅鉴定了具有显着关联峰的非同义SNP (C/A) ( P  = 1.31 × 10 ^-22 )。该 SNP 位于Chr上LOC_Os08g36000的编码区。8（+22 689 418 bp，C > A；+40 aa，Val > Phe）（图 1e）。编码的蛋白质包含一个 F-box 和两个 LRR 结构域。该家族中的许多基因对生物和非生物胁迫作出反应，包括温度和激素（Yan et al ., 2010）。'C' 单倍型（Hap ^tej）几乎固定在tej亚组中，频率为 0.985，而'A' 单倍型（Hap^other ) 在其他亚组中占主导地位的包括aus、ind、trj和野生稻品种，频率为 0.961（图 1f，h）。为了解决 Hap ^tej等位基因的功能，我们在生长室中对 14 个具有 Hap ^tej等位基因的栽培稻品种和 19 个具有 Hap ^other等位基因的栽培稻品种进行冷处理，结果表明在冷胁迫下（4° C 48 h），具有Hap ^tej等位基因的水稻品种的成活率（93.25%）高于具有Hap ^other等位基因的水稻品种（10.12%；P  = 1.66 × 10 ^-13，双尾t-测试）（图1g，h，k）。因此，我们将基因命名为CHILLING-TOLERANCE DIVERGENCE F-box ( COLDF )。

为了进一步表征COLDF的功能，我们在 ZH11 背景（O. sativa ssp. japonica  var . Zhonghua 11）中获得了该基因的 TILLING 和 CRISPR 突变体（图 1i，j）。TILLING 突变体在编码区携带 G 到 A 突变（+22 688 780 bp，G > A；+256 aa，Ala > Thr），敲除突变体在编码区携带 22 bp 插入，命名为分别为coldf-1和coldf-2。将纯合coldf-1和coldf-2突变体的14日龄幼苗在4℃下冷胁迫48小时，然后在30℃下恢复正常生长条件进行恢复。结果表明，这些colf幼苗易受冷胁迫，coldf-1和coldf-2的成活率分别为37％和12％，而ZH11的成活率为82％（图1i，j; P  = 1.07× 10 ^{-10 ，} coldf-1的双尾t检验和P  = 1.94 × 10 ^{-12 ，} coldf-2的双尾t检验）。这些结果表明，COLDF的 Hap ^tej负责增强tej的耐寒性。品种。我们的研究结果通过识别与适应相关的基因座来深入了解水稻驯化和改良的机制基础，这将有助于通过促进能够更好地耐受气候变化带来的压力的改良品种的快速开发来减轻这些影响。