Introduction

Rice (Oryza sativa L.) is the most important cereal crop, providing sustenance to most people worldwide. Its yield has almost reached a plateau in most of the rice-growing areas of the world. There is limited scope for expanding rice cultivation areas in rapidly developing countries like India and China. It is becoming challenging to increase the yield potential of rice by developing new genotypes through increasing intrinsic yield potential, as evidenced by the low genetic gain during the last two decades1. In several parts of India, China, and Bangladesh, rice, also called wet rice, is cultivated mainly in the rainy season. One of the main reasons for the low rice production in this region is the low intensity of light during the wet season, which results in low productivity of the crop2,3,4. Rice experiences significant yield loss due to low light intensity caused by seasonal overcast clouds and global dimming5,6,7. India's eastern and northeastern hill regions contribute to 48% of the total rice production from 58% of the area under rice cultivation8. In the wet season, light intensity is around 500 µmol/m2/s compared to 1400 µmol/m2/s in the summer (Boro) season. 400 µmol/m2/s PPFD reduces 30–50%net photosynthesis, ultimately reducing yield by 34–55%2. This reduced light intensity also hinders grain filling and quality through reduced amylose, starch content (6–37%), grain protein content and increased chalky grain numbers9,10.

The yield is a complex trait influenced by source and sink strength and translocation efficiency between them. Panicles per plant, grains per panicle, and seed weight are the major sink strengths directly controlling rice yield. Rice is a widely researched crop, with numerous studies on its yield and component traits conducted under various conditions, including biotic and abiotic stresses11,12 Although a wide genetic variation is observed among rice genotypes for shade tolerance ability, its genetic or physiological mechanism is not unveiled except in a few with inconclusive transcriptome analysis8,9. Many low-light-tolerant genotypes, including Swarnaprabha, Rudra, and Purnendu, have been reported earlier13, but the tolerance mechanisms vary depending on the genotype. Furthermore, no universal shade tolerance mechanism is observed; genotype-specific plasticity significantly reduces yield loss in low light intensity13. However, low-light intensity grown rice showed reduced panicle number compared to other yield-attributing parameters in almost all the genotypes. No studies have been conducted to identify the genomic region controlling rice yield loss under reduced light intensity in rice-growing countries, including India. Improving the productivity of wet season rice is hindered by a poor understanding of the genetics of traits that alleviate yield and panicle number loss in low light intensity. Some tolerant genotypes maintain almost the same number of tiller and panicle in open and shade, while a few maintain equal grains per panicle and seed weight2. It is also unclear whether the QTLs previously identified in ambient light are involved in yield enhancement in low light. Genetic analysis of low-light tolerant genotypes, including identifying QTLs and genomic regions controlling yield loss in low-light conditions, can aid in the development of high-yielding, low-light tolerant varieties of rice, which can improve yield potentiality in the wet season by reducing the yield gap between the wet and summer rice.

Therefore, to identify QTLs related to low light (LL) tolerance in Swarnaprabha (SP), the study was conducted where a RIL population resulting from a cross between SP and IR64 (LL susceptible genotype) was used as a mapping population. The mapping used yield components like panicle number, yield per plant, test weight, panicle weight, seed per panicle for four seasons and genotyping data of a 7 K SNP array. The further association analysis of a panel of one hundred seventeen genotypes was used to validate the QTLs identified in the RIL population.

Results

Parental differences: Swarnaprabha vs IR64

The performance of Swarnaprabha (SP) and IR64 were compared both in the open field (Ambient light, AL) and under the white net (35% light cut, LL) for four consecutive wet seasons (Kharif): 2019, 2020, 2021, and 2022 (Supplementary Table 2). The Mean grain yield of four years in Swarnaprabha (539 g/m2) was higher than IR64 (450.8 g/m2). The thousand seed weight (TW) of Swarnaprabha was higher in all four years and is the main reason for a marginally higher yield than IR64. In ambient light, the average number of panicles (PN) for the variety SP was 12.6, and for IR64 it was 11.1. In low light, SP had a 19% reduction in panicle number and a remaining PN of 10.2, while IR64, with a 35.7% reduction, had only 7.1 panicles. One hundred thirty RILs derived from SP x IR64 showed a wide range of variations with continuous frequency distribution for all the yield-attributing parameters (Supplementary Table 3). The mean values for all the traits in all four seasons are significantly reduced in LL compared to the AL condition (Table 1).

Table 1 Descriptive statistics of the major yield attributing parameters in 130 RILs and the percentage of reduction under low light compared to ambient light for each trait.
Full size table

All five traits showed several extreme values, either larger or smaller than those of the parents, i.e., bi-directional transgressive segregation was observed for all the traits. Analysis of variance confirmed the significant variations among the RILs for all the traits (Supplementary Table 4). The interaction between the RILs and the years has the predominant contribution to the total variation of PN and GY on ambient conditions, where the highest contributions are by the PN and GY under LL. The effect of RILs and the RILs x years interaction contributed almost equally to the total variations in PW and GPP.

QTL mapping

2268 SNPs were considered good data based on the parameters described in the methodology, and 927 SNPs showed polymorphism between two parents, and more than fifty minor alleles were observed for all. So, finally, 927 SNPs were considered for linkage map construction. The polymorphic SNPs were evenly distributed throughout the genome (Supplementary Fig. 1). (Chromosome 1 contains 124 SNPs; chromosome 2 contains 71 SNPs; chromosome 3 contains 65 SNPs; chromosome 4 contains 105 SNPs; chromosome 5 contains 81 SNPs; chromosome 6 contains 75 SNPs; chromosome 7 contains 82 SNPs; chromosome 8 contains 81 SNPs; chromosome 9 contains 36 SNPs; chromosome 10 contains 57 SNPs; chromosome 11contains 99 SNPs and on chromosome 12 contains 51 SNPs). This linkage map covers the 2024 cM linkage distance of the total rice genome with a marker density of 1 SNP per 2.2 cM of the rice genome. QTL mapping was carried out using the year-wise data separately, along with the mean data of all seasons. For each season, mapping was performed separately for AL and LL conditions. The linkage map construction was followed by QTL analysis using Inclusive Composite Interval Mapping (ICIM software, IciMapping, Version4.0) with mapping parameters, i.e., step (1 c M) 1.0 and PIN 0.00100. A minimum of LOD 3 was considered for being a putative QTL (Supplementary Fig. 2). Forty four putative QTLs for all the yield-attributing parameters were identified in ambient and low light (Table 2). Exclusively, a major QTL on chromosome 1 was identified in low light conditions for PN and GY (qPNLL1.1 and qGYLL1.1,), where Swarnaprabha contributes favourable alleles. The locus was between 35825579 bp and 37692801 bp with 3 SNPs, 1163456, SNP-1.37415410. and 1212517 inside the QTL. When the four-year mean value was considered, qPNLL1.1 explained 47% of the total phenotypic variance with LOD 19, where the additive effect was 1.1. As expected, the same locus also contains yield-enhancing QTL, qGYLL1.1, which explains 38.8% PVE at LOD 14.3. Another region on chromosome 6 between the marker interval of 5950897-5951985 and physical interval of 4641044-4680281 bp was identified for GY under LL (qGYLL6.1) with an average LOD 6.4 and PVE% of 10.3 in three consecutive seasons as well as from the mean data. The study also identified two TW-improving QTLs in lowlight located on chromosomes 1 and 7, where favourable alleles from both the parents SP and IR 64.

Table 2 The trait-specific QTLs, their locations LOD, phenotypic variation explained (PVE), and additivity (Add) are given for season-wise (WET 2019, 2020, 2021, 2022) and the mean data.
Full size table

Validation of QTLs in an association panel

One hundred seventeen genotypes' yield and attributing traits reduced significantly in low light compared to AL (Supplementary Table 5). In the LOSS curve, the half decay distance of this association panel at the arbitrary nominal level of r2 = 0.10 was found to be 403032 (Supplementary Fig. 3). The identified QTNs in at least two models are shown in Table 3. SNP 1163456 on chromosome 1 was associated with low light-responsive PN and GY in almost all four models (Table 3). In two models, another QTN (SNP 5951985), was identified for GY and PN on chromosome 6.

Table 3 The GWAS-based QTNs identified in at least two models in low light (LL) and ambient light (AL), their chromosomal positions (Chr), minor allele frequency (MAF), and the additive effect.
Full size table

The regression analysis using the allelic variants of SNP 1163456 on chromosome 1 showed a significant deviation from zero for the lowlight-responsive GY but not in ambient light (Fig. 1). On the other hand, when PN was considered for the regression analysis, both lowlight ambient light responsive regression curves deviated significantly from zero but in contrasting directions.

Fig. 1
figure 1

The regression curve between allelic variants of the SNP-1163456 and grain yield (GY) and panicle number (PN) in low and ambient low light.

Full size image

Predicting shade-responsive genes inside newly identified QTLs

A stable QTL on chromosome 1, improving panicle and yield in lowlight (qPNLL1.1 and qGYLL1.1), was considered for further genome analysis. The marker interval between 1163456 and 1212517 was between the physical distance from 35825579 to 37692801nt. Exploring the Rice Genome Annotation Project database, ten transcripts were selected based on their expression in leaf, culm or shoot apical meristem region from RNA-Seq FPKM Expression Values, (Supplementary Table 6).

Relative transcript abundance of the selected genes in seven-day-old flag leaves (Table 4) and shoot apical meristem (Table 5) of the AL and LL-grown plants was compared through qRT-PCR. The analysis was first done in parental pairs, i.e., Swarnaprabha and IR64. In flag leaves, the relative expression of MOC2, PsbS1 and qSH1 genes were downregulated in shade in both the parents. On the other hand, in the apical meristem, DUF668-1, AUX1, and qSH1 genes showed upregulation in the shade for both parents. However, MOC2 was upregulated in shade-grown Swarnaprabha compared to ambient, but the repression was observed in IR64. The transcript availability of DGP1 was higher in shade-grown IR64, but MOC2 transcript availability in apical meristem was higher in Swarnaprabha compared to IR64. Three RILs with SP-carrying QTL showed upregulation in the shade compared to their open-grown, whereas downregulation was observed in RILs carrying IR64 alleles (Fig. 2).

Table 4 Transcript abundance of selected genes in flag leaves of ambient light (AL) and lowlight (LL)-grown Swarnaprabha (SP) and IR64.
Full size table
Table 5 Transcript abundance in the apical meristem of selected genes from the yield and panicle improving QTL of chromosome 1.
Full size table
Fig. 2
figure 2

The relative transcript abundance of the MOC2 gene in open and shade-grown apical meristem in IR64 and Swarnaprabha (SP) and three RILs, R57, R119 and R124. p-value of the t-test between open and shade-grown plants is given above.

Full size image

A nucleotide sequence comparison of the DGP1 gene between SP and IR64 identified an SNP in 856th nt on the 3rd exon, which has the potential to convert an amino acid Glu to Val. MONOCULM 2 (MOC 2, LOC_Os01g64660) carries two nucleotide polymorphisms between SP and IR64. There are two nucleotide indel at 1613 nt at the 7th intronic region of the gene and a single base pair polymorphism (C/A) at 2243 bp at the 10th intronic region (Table 6). There are also three polymorphisms in the upstream regulatory element of MOC2. A 2 bp in/del is observed in SP at 777 bp upstream of the gene. There is one SNP at − 735 bp upstream (A/T) and another at − 730 bp upstream (A/G) (Table 6).

Table 6 Types and position of the nucleotide polymorphism between SP and IR64 and its effect on amino acid (AA) modifications.
Full size table

Discussion

Rice is usually grown in the rainy season across the globe. In several countries, prolonged overcast clouds affect rice production during the growing season. In Eastern Indian states, the mean yield of summer rice is almost double that of wet rice (kharif)14. One of the reasons for a huge yield gap is the low light intensity during the main rice-growing season. Although lowlight responsive rice genotypes were identified, their analysis is yet to be deciphered2,13. Thus, the study aims to identify the robust QTLs and predicted genes from the QTL so that a breeder can use them in future rice breeding programs. Earlier reports identified that lowlight-grown rice reduces panicle numbers drastically2. So, based on the earlier observation, a contrasting photoperiod-insensitive parental pair, Swarnaprabha and IR64, is considered. The differences in yield and its attributing parameters between the two parents were minimal when grown under ambient light (AL). However, when grown in lowlight, GY (SP vs IR64:: 25% vs 48%) and PN (SP vs IR64:: 19% vs 36%) reduction were significantly less in SP compared to IR64. Thus, the RIL population (F9) developed from Swarnaprabha x IR64 was ideal for mapping the yield-enhancing traits in low light (35% light cut). In this study, the performance of rice genotypes in low light was assessed by covering the field with a white shade net (35% light cut), thus preventing the absorption of any specific colour from the natural sunshine. Also, shade-net covering was made by keeping a 1.5mt gap between the field and the net, which confirms the insignificant differences in the relative humidity and temperature between open and shade-net-grown plants. The QTL mapping was conducted to minimize the environmental effect by taking the four-year’s data for yield and its component characters in both AL and LL. The SNP orders perfectly match the order deduced from their physical position, which implies that the RIL population used in this study is unbiased and ideal for mapping any traits using the linkage map. The highest number of polymorphic SNPs was observed in chromosome 1, which supports the observation of most saturated markers in chromosome 1 in the meta-analysis11.

Forty four putative QTLs were identified for the GY, PN and other yield components in both AL and LL conditions. Among them, the QTLs on chromosome 1 (qPN_LL1.1 and qGY_LL1.1) were identified between 35.82 to 37.69 Mbp for the trait of GY and PN in lowlight only, which means they are not found in AL. Additionally, GWAS by five different models also identified a QTN (1163456 on chromosome 1), associated with the grain yield and panicle numbers, which is exactly one of the flanking markers of the previously identified grain yield and panicle improving QTL (qPN_LL1.1 and qGY_LL1.1) under low light. As the Swarnaprabha alleles contribute favourable phenotypes followed by the confirmation of the same in a panel of rice genotypes by GWAS, the QTL with a high additive effect may be targeted for its yield and panicle loss alleviation in shade-grown rice. One previous report confirms that two closely located QTLs in the same region controlling PN in a cross between IR64 and O. rufipogon where favourable alleles come from IR6415 in ambient light. Although no panicle-improving QTL in ambient light was identified, the allelic variation-based regression curve confirmed a panicle-improving QTL contributed by the IR 64-type allele in an association panel. Hittalmani et al.16 reported approximately two closely located thousand seed weight controlling QTLs, one at 174 cM and another at 197 cM. In this study, one seed weight-controlling QTL (qTW_AL1.1) was identified in a similar location in wet-2020, where a favourable allele comes from SP. Here, a PN-controlling QTL located on chromosome 1, (qPN_AL1.1) at 56 cM, the region also carries a consensus grain yield and yield components QTL11. Only one QTL on chromosome 2 has been reported to control PN in rice by several earlier reports11; in this study, the same region identified qPN_AL2.1 in 2019 wet. Thus, the identified lowlight responsive PN and GY-improving QTL on chromosome 1 are newly reported QTLs. The study also identified panicle-number improving two QTLs on chromosome 1 in ambient light but far away from the qPN_LL1.1. A recent report confirms that the same region carries tiller number-improving QTLs in ambient light17.

Another QTL on chromosome 6, qGY_LL6.1, from 4.6 to 4.7 Mbp, was identified to control panicle and grain yield under LL conditions. It was also stable across the seasons, controlling 8–12% of total PVE with 4–6 LOD. Reference18 reported a panicle improving QTL on chromosome 6 near the lowlight responsive panicle and yield improving QTL, qGY_LL6.1, identified in this study. Associated QTN, SNP 5,951,985, was identified for panicle and grain yield on chromosome 6, which is also the right marker of previously identified QTL qGY_LL6.1 using SPxIR64 derived RIL population in this study. Thus, association mapping validates that the identified lowlight responsive QTLs are not only Swarnaprabha or IR64 specific. So, the identified PN and GY improving QTL on chromosome 1 and chromosome 6 can be recommended to introduce other high-yielding backgrounds to reduce the yield gap between wet and summer rice and to develop new varieties targeting the lowlight-prone areas. An association panel of one hundred seventeen genotypes further validated the identified QTLs. ANOVA analysis deciphered the significant contribution of light intensity to the total variation.

Although six genes are differentially expressed in low light-grown flag leaves, there are no significant differences between SP and IR64. However, eight genes expressed differentially between ambient and lowlight-grown apical meristem, and only DGP1 and MOC2 showed significant differences between SP and IR64. DGP1 (LOC_Os01g62060) has previously been reported to have a role in photosynthesis regulation19. However, as RQ was higher in IR64 than SP, it was not considered for the probable reason, as the yield-enhancing favourable gene is coming from Swarnaprabha. However, sequencing of the DGP1 gene showed an SNP (ASP/TIR), which can convert the Glutamate to Valine in the polypeptides. It can be targeted for allele-specific marker development to elucidate its role in controlling panicle number and photosynthesis in low light. As the Swarnaprabha is more efficient in tiller development under shade, MOC2 (LOC_Os01g64660), within the identified QTL might be a yield and panicle-enhancing gene for the purpose. MOC2, also a Fructose 1–6 bisphosphatase gene, has previously been reported to have a role in controlling tiller outgrowth20 and has shown upregulation in SP under LL inside the shoot meristem in this study. The earlier reports confirmed that the gene does not inhibit auxiliary bur formation. Rather, its absence promotes outgrowth deficiency. Higher expression and shade responsiveness in Swarnaprabha might have a role in preventing tiller number reduction in the shade. Nucleotide polymorphism in coding and 5’-UTR between SP and IR 64 gives scope for its deeper analysis in mitigating lowlight stress, particularly tiller development. Although MOC1 is previously reported as a tillering gene in rice, located on chromosome 621, no QTLs were identified in any year in a similar location in this study, probably due to a lack of difference between the two parents.

Thus, the study identified two QTLs in chromosomes 1 and 6 as responsible for alleviating yield loss in shade-grown rice. It also predicted MOC2 as a gene for their shade-responsive transcript upregulation in the tolerant and a few favourable RILs. Nucleotide polymorphism between Swarnaprabha and IR64 will help transfer the allele more easily and accurately in another rice background. However, further fine mapping of the QTL may lead to identifying the candidate gene, followed by allele mining in the vast rice repository and developing allele-specific markers. Therefore, due to the significant additive effect and validation across various genotypes, the low-light-responsive yield-improving QTL on chromosome 1 can be directly utilized by introducing the trait in other high-yielding genotypes through marker-assisted selection easily.

Methods

Field-phenotyping

A mapping population of 130 biparental RILs developed from a cross between Swarnaprabha x IR64 was used to evaluate the yield and its attributing parameters in open and shaded fields. Both are released varieties in India. The RILs were developed from F2 through single-seed descent. The F7-derived bulked seeds of each RIL were selfed for two additional generations, and they have been maintained at the Crop Research Unit of the university experimental farm. A panel of one hundred seventeen genotypes comprised of Indica landraces, released varieties, and aromatic and aus subtypes were evaluated similarly as described in RILs. One hundred thirty RILs were grown in both open (Ambient light) and shade (low light) with 3 replications following RBD design, considering a single line of 4 m as one replication. Twenty-six plants were grown per line with 20 cm between rows and 15 cm plant-to-plant following the recommended package of agronomic practices in four consecutive seasons, i.e. Kharif 2019 (S1), Kharif 2020 (S2), Kharif 2021 (S3), and Kharif 2022 (S4). Similarly, one hundred seventeen genotypes were evaluated in Kharif-2023 for association analysis. As described earlier, a white shade net (30% light cut) was used in the field to reduce the light intensity2,13. Temperature and photosynthetically active radiation (PAR) were recorded daily at 9 h, 13 h, and 16 h during transplanting to maturity in both open and shade (Supplementary Table 1). No significant temperature differences between open and shade-grown fields were noticed. The mean of ten plants in each row was considered for recording effective tiller number or panicle number (PN), yield per plant (PY), hundred seed weight (HSW), panicle weight (PW), and filled spikelet per panicle (SPP) for each replication. ANOVA analysis for partitioning of genotypic and light intensity across the seasons for yield and other parameters was performed by GraphPad Prism 10.2.3. The significant difference between different parameters of plants grown in open-field and shade was estimated using a paired t-test (non-parametric). Descriptive statistics for RILs and genotypes were also analysed using GraphPad Prism 10.2.3.

Genotyping

DNA was extracted from young leaves of one hundred thirty RILs, their parents and one hundred seventeen genotypes. For genotyping, Illumina's custom-designed 7 K iSelect array was used. 7 k array, also known as C7AIR, consisted of 7098 SNPs that represented an improved version of Cornell_6K_Array_Infinium_Rice (C6AIR) with additional content from the Rice Haplotype map project22,23. One hundred seventeen genotypes were genotyped with the same 7 K SNP chip as in the RIL population. After calling the data, 927 polymorphic SNPs between the parents were used for mapping. SNPs with a call rate of less than 90% and minor allele frequency of less than 40% were removed from the dataset (Supplementary Tables 7 and 8). The polymorphic SNPs were used for linkage mapping.

Construction of linkage map and QTL analysis

Integrated QTL software, IciMapping, Version 4.0, available from http://www.isbreeding.net24 was used for linkage map preparation and QTL mapping. Based on the genotypic data of 130 RILs, the linkage map was prepared with the following parameters i.e. grouping with anchor only, ordering with input order and rippling by DIS (sum of adjacent distances) where genetic distances (cM) were calculated based on Kosambi mapping function which was further used for preparing the input file (BIP file) for QTL mapping. Before QTL mapping the test of normality of distribution at p >  = 0.05 was performed following Kolmogorov–Smirnov test for goodness of fit. Mapping of QTLs was performed through Inclusive Composite Interval Mapping (ICIM) using the same software with mapping parameters i.e. step (cM) 1.0 and PIN 0.00100. For epistatic QTL analysis, ICIM-EPI function was used in the same software.

Validation of QTL through association analysis

In 2023-kharif, yield and its attributing parameters were measured further in a panel of one hundred seventeen genotypes comprising local landraces, released varieties, aus, and aromatic rice. All the genotypes were collected from our University's Crop Research Unit field gene bank. SNPs with a call rate of less than 95% were removed from the dataset, and 2928 good-quality SNPs were used to create a diversity tree using the neighbour-joining method, giving rise to 5 separate sub-groups. Linkage disequilibrium (LD) among the polymorphic markers was created by Tassel 5 with a sliding window size of 50 markers. To determine the decay of LD, obtained r2 values were graphed against physical and genetic distances, and a LOESS curve was created. A previously used approach25 was used to estimate LD decay distance and evaluated at the commonly accepted r2 threshold of 0.2, as described earlier26. Genome-Wide Association Study (GWAS) was carried out in R studio using five different models, i.e., 'FarmCPU'27 and 'BLINK'28 in the GAPIT package, mrMLM29 and FastmrMLM30 in the mrMLM package, and 3VmrMLM31 in the IIIVmrMLM package.

Relative transcript quantification

The leaves and meristematic tissues were collected from 8 to 10 days of flag leaves and 45 days of meristematic tissues of field-grown rice plants. Three independent plants were marked and considered as one replication. Collected tissues were washed properly with phosphate buffer saline and immediately placed in an RNA-protector solution. Good-quality RNA was extracted from the collected sample leaves following the RNeasy plant mini kit as described earlier32, and RNase-free DNase treatment was conducted to avoid DNA contamination. Extracted RNA was immediately converted to cDNA using a high-capacity reverse transcriptase (Applied biosystem kit). qPCR was performed using cDNA as a template using gene-specific primer and tubulin, and relative transcript quantification was estimated following ∆∆CT, as described earlier33.