Marker-assisted identification of Rf3 and Rf4 fertility restorer lines in rice (Oryza Sativa L.) and their evaluation for osmotic stress tolerance at the seedling stage and grain yield under rainfed drought

Introduction

Rice (Oryza sativa L.) hybrids yield 15 to 20% more than the best inbred varieties grown under comparable management, and this advantage has made hybrid rice a central element of yield intensification programmes in South and East Asia (Virmani, 1994) [1] . In India, adoption has lagged behind expectation. Seed cost, inconsistent heterosis across seasons and the weak performance of many hybrids outside irrigated ecologies have all limited farmer uptake (Spielman et al., 2013) [2] . Almost all commercial hybrids in the country are three-line hybrids built on the wild abortive cytoplasmic male sterility (WA-CMS) system, in which a male sterile A line is crossed with a restorer or R line carrying nuclear genes that suppress the sterility. The R line thus contributes half of the hybrid genome and much of its adaptation, and a hybrid rarely tolerates stress better than the restorer from which it was made. Fertility restoration in the WA-CMS system is governed mainly by two dominant loci, Rf3 on chromosome 1 and Rf4 on chromosome 10, with Rf4 usually exerting the larger effect (Yao et al., 1997) [3] . Both loci were first placed on linkage maps with RFLP and RAPD markers (Zhang et al., 1997) [4] . Rf4 was later cloned and shown to encode a mitochondria-targeted pentatricopeptide repeat protein that lowers the abundance of transcripts of WA352, the mitochondrial chimeric gene responsible for sterility (Tang et al., 2014) [5] . The WA352 protein interacts with a nuclear-encoded mitochondrial protein and triggers premature degeneration of the tapetum. Restoration is in effect the outcome of nuclear control over a specific mitochondrial transcript, and the dosage and background of the restorer alleles determine how completely pollen fertility is recovered in the F 1 (Chen et al., 2014) [6] . Conventional identification of restorers requires test crosses with a CMS line and scoring of pollen and spikelet fertility in the resulting hybrids over at least one additional season. Marker-assisted screening shortens this cycle considerably. Sattari et al. (2007) [7] developed a two-gene marker system that detected Rf3 and Rf4 alleles with high accuracy across diverse germplasm, and Sheeba et al. (2009) [8] validated several SSR markers linked to these loci in Indian breeding material. Fine mapping later narrowed both intervals and produced tightly linked markers that predicted restorer status with considerably fewer false positives (Suresh et al., 2012) [9] . Screening large sets of elite lines with such markers before test crossing now allows breeders to concentrate field resources on candidates with a high probability of carrying functional restorer alleles. Drought is the most damaging abiotic constraint on rice in rainfed lowland and upland ecologies, which together account for a large share of rice area in eastern and central India. Water deficit around flowering is especially harmful because it impairs pollen viability, anther dehiscence and panicle exsertion, and the resulting spikelet sterility cannot be compensated later in the season (Pantuwan et al., 2002) [10] . Grain yield under stress has proved to be a more dependable selection criterion than most secondary traits, provided stress is applied uniformly and heritability under stress remains adequate (Venuprasad et al., 2007) [11] . Over the past decade, direct selection for yield under managed drought, supported by large-effect quantitative trait loci, has produced several drought-tolerant inbred varieties (Kumar et al., 2014) [12] . Comparable effort has rarely been directed at restorer lines, and most WA-CMS restorers in current use were developed and tested only under irrigation. Seedling-stage screening with polyethylene glycol (PEG) offers a rapid way to reduce a large set of lines to a manageable number before field testing. PEG of high molecular weight does not enter root cells and lowers the water potential of the nutrient solution in a predictable way (Michel et al., 1973) [13] . Seedling root and shoot growth under PEG-induced osmotic stress has discriminated tolerant from susceptible rice varieties in several studies (Swapna et al., 2017) [14] . Whether seedling tolerance predicts reproductive-stage performance is less certain, because drought resistance at flowering depends on processes such as spikelet fertility that are not expressed in young seedlings (Fukai et al., 1995) [15] . A combined approach that uses markers to confirm restorer genotype, seedling assays to discard clearly susceptible lines and field trials to confirm yield under stress has seldom been reported for restorer germplasm. The present study addressed this gap in a set of 60 rice lines. Its objectives were (i) to classify the lines for Rf3 and Rf4 using linked molecular markers, (ii) to evaluate the Rf4-positive lines for seedling growth under PEG-induced osmotic stress, (iii) to assess the best of these lines for yield and yield components under irrigated and rainfed conditions, and (iv) to identify restorer candidates combining markerconfirmed restorer alleles with stable yield under water deficit.

Materials and Methods

Plant material and experimental site

Sixty rice lines, coded ML-01 to ML-60, formed the base population. Seed of all lines was multiplied in a single season before the experiments to minimise differences in seed age and vigour. The marker assay and the seedling experiment were carried out in the laboratory and controlled-environment facilities of B.N.P.G. College, Rath, Hamirpur, Uttar Pradesh. The field experiment was conducted at the research farm of B.N.P.G. College, Rath, Hamirpur, Uttar Pradesh, during 2024-2025. Weather conditions during the crop season are presented in Figure -1 .

Genotyping for Rf 3 and Rf4

Genomic DNA was extracted from young leaves of each line by the CTAB method (Murray et al., 1980) [16] , and its quality and concentration were checked on agarose gel and by spectrophotometry. Each line was genotyped with markers linked to Rf3 on chromosome 1 and Rf4 on chromosome 10, namely RMS-SF21-5 for Rf 3 and RMS-PPR9-1 for Rf 4 , which have been validated for restorer identification in Indian germplasm (Sheeba et al., 2009) [8] and refined by fine mapping (Suresh et al., 2012) [9] . A known restorer and a maintainer line were included in every run as positive and negative controls. A line was scored P (present) when it amplified the allele of the restorer control and A (absent) when it amplified the allele of the maintainer control. Lines were then grouped into four marker classes, namely Rf3 only, Rf4 only, Rf3 + Rf4 and neither. The mapped effect of Rf4 on fertility restoration is larger than that of Rf3 in the WA-CMS system (Yao et al., 1997) [3] , and lines carrying the Rf4 allele were given priority for phenotyping. All 28 lines positive for Rf4, whether alone or together with Rf3, were advanced to the seedling experiment.

Seedling screening under PEG-induced osmotic stress

Surface-sterilised seeds were germinated on moist filter paper, and uniform seedlings were transferred to Yoshida nutrient solution (Yoshida et al., 1976) [17] . After acclimatisation, seedlings were exposed to three osmotic treatments, namely control (nutrient solution alone), 15% (w/v) PEG-6000 and 25% (w/v) PEG-6000. The osmotic potential of each PEG solution was derived from concentration and solution temperature following Michel et al. (1973) [13] . Solutions were renewed at regular intervals to maintain concentration and pH. The experiment followed a completely randomised design with three replications for each of the 84 genotype × treatment combinations. Visual injury under 25% PEG was scored on the 1 to 9 scale of the Standard Evaluation System for rice, where 1 indicates no visible symptoms and 9 indicates nearly all leaves dried (IRRI, 2013) [18] . Root length (RL, cm) and shoot length (SL, cm) were measured on the sampled seedlings. Root fresh weight (RFW) and shoot fresh weight (SFW) were recorded immediately after separation and blotting, and root dry weight (RDW) and shoot dry weight (SDW) after oven drying to constant weight. All weights are expressed in g per seedling.

Field evaluation under irrigated and rainfed conditions

Twenty of the 28 lines, spanning the full range of seedling response from highly tolerant to susceptible, were evaluated in two adjacent blocks. The irrigated block was kept flooded from transplanting to physiological maturity. In the rainfed block, irrigation was withheld before the reproductive stage and the crop depended on rainfall until maturity, following the approach to managed reproductive-stage stress used in drought breeding of rice (Venuprasad et al., 2007) [11] . Each environment had three replications. The crop was raised following the recommended package of practices for the region, with uniform fertiliser application and plant protection in both blocks. Observations were recorded on randomly selected competitive plants in each plot. Days to 50% flowering (DFF) was noted on a plot basis. Plant height (PH, cm), number of effective tillers per plant (ET), panicle length (PL, cm), filled grains per panicle (FG), unfilled grains per panicle (UFG), spikelet fertility (SF, %), grain length (GL, mm) and grain yield per plant (GY, g plant -1 ) were recorded following the Standard Evaluation System (IRRI, 2013). Spikelet fertility was calculated as the ratio of filled grains to total spikelets per panicle, expressed as a percentage. Yield reduction (YR) under rainfed conditions was calculated for each line as

where GY is the mean grain yield of a line in the respective environment. The same expression, applied to seedling trait means, gave the percentage reduction under each PEG level relative to the control.

Statistical analysis

Replicate observations were analysed as two-factor experiments, genotype × stress level for the seedling traits and genotype × environment for the field traits, following Gomez et al. (1984) [19] , with both factors treated as fixed. Standard error of the mean, 𝑆𝐸 ( 𝑚 ) ± = √ 𝑀𝑆𝐸 𝑟

, and critical difference at P = 0.05, CD = t 0.05 × √(2MSE/r), were derived from the pooled error mean square, with r = 3. The coefficient of variation was calculated as the ratio of the error standard deviation to the grand mean of the trait. Pearson correlation coefficients were computed among seedling traits at 25% PEG and among field traits under rainfed conditions using genotype means. Principal component analysis on standardised genotype means was used to examine how variation was distributed across traits (Jolliffe et al., 2016) [20] . Lines were grouped by k-means clustering on standardised means with k = 3 (Hartigan et al., 1979) [21] . A standardised multi-trait selection index, scaled from 0 to 1 with higher scores denoting closer proximity to the ideotype, was computed for each line from seedling traits at 25% PEG and from field traits under rainfed conditions. Agreement between seedling and field indices was tested by Spearman rank correlation. Analyses were performed in Python 3.11 with the statsmodels (version 0.15) and SciPy (version 1.17) libraries.

Results

Marker classification of the 60 lines

The two markers separated the lines into four classes of unequal size. Fourteen lines (23.3%) carried the restorer allele at Rf3 alone and 18 (30.0%) at Rf4 alone. Ten lines (16.7%) were positive at both loci, and the remaining 18 (30.0%) amplified the maintainer allele at both. Taken together, 28 lines (46.7%) carried the Rf4 restorer allele and were advanced to phenotyping, while the 14 lines positive only for Rf3 and the 18 double-negative lines were set aside.

Seedling growth under PEG-induced osmotic stress

Osmotic stress depressed every seedling trait, and the depression deepened with PEG concentration. Averaged over the 28 lines, 15% PEG reduced the six traits by 12.9% (RL) to 15.9% (SDW) relative to the control. At 25% PEG the reductions rose to between 32.6% (RL) and 35.7% (SFW), roughly two and a half times the loss at the lower level. Root length responded least in relative terms at both concentrations, whereas the fresh and dry weights of the shoot responded most. Analysis of variance showed highly significant effects of genotype, stress level and their interaction for all six traits (P < 0.001). Stress level contributed by far the largest variance ratio (F = 2504.85 to 3145.82), followed by genotype (F = 56.83 to 415.03). The genotype × stress interaction, though smaller (F = 3.24 to 14.96), was significant throughout and was strongest for RDW. Lines thus changed rank as stress intensified, and a ranking made under control conditions would not have identified the most tolerant lines. Error coefficients of variation stayed between 3.38 and 3.96%, which indicates good control of experimental error in the hydroponic system. Table 1 presents the means of the 28 lines at 25% PEG, the level used for selection. ML-02 recorded the highest RL, SL, SFW, RDW and SDW, and ML-13 the highest RFW. ML-32, ML-05, ML-57 and ML-59 followed closely, and the differences among these lines were often within the critical difference for individual traits. At the opposite end, ML-52 gave the lowest values for five of the six traits, and ML-45, ML-22, ML-53 and ML-47 were similarly poor. The spread between best and poorest lines was wide. SDW at 25% PEG ranged from 0.067 to 0.163 g, a 2.4-fold difference, and RDW from 0.0156 to 0.0422 g, a 2.7-fold difference. Values are means of three replications ± standard error of the replicate observations. RL, root length; SL, shoot length; RFW, root fresh weight; SFW, shoot fresh weight; RDW, root dry weight; SDW, shoot dry weight; weights are per seedling. SE (m)±, CD (p = 0.05) and CV (%) are derived from the pooled error mean square of the genotype × stress analysis (error df = 168) and apply to comparisons of genotype means within a stress level. Effects of genotype, stress and genotype × stress were significant at P < 0.001 for all traits. Relative reduction at 25% PEG also varied considerably among lines (Figure -2). Loss of root length ranged from 22.8% in ML-15 to 41.7% in ML-49, and loss of shoot dry weight from 24.1% in ML-05 to 47.6% in ML-45. A line with vigorous growth under control conditions did not necessarily retain it under stress. ML-49, for example, had among the longest roots in the control but lost more root length than any other line at 25% PEG. Visual injury scores at 25% PEG agreed well with the growth measurements. Two lines, ML-02 and ML-32, scored 1 (highly tolerant) and seven scored 3 (tolerant), namely ML-05, ML-08, ML-09, ML-13, ML-51, ML-57 and ML-59. Eleven lines were moderately tolerant (score 5), six susceptible (score 7) and two, ML-22 and ML-52, highly susceptible (score 9).

Field performance under irrigated and rainfed conditions

Water deficit in the rainfed block lowered the mean of every field trait except days to flowering and raised the number of unfilled grains.

Relative to the irrigated block, mean grain yield of the 20 lines fell by 11.4%, from 25.86 to 22.91 g plant -1

. The largest relative changes were a 31.5% increase in unfilled grains and a 10.2% loss of filled grains per panicle. Spikelet fertility fell by 6.1 percentage points and effective tillers by 7.0%, while plant height, panicle length and grain length declined by only 1.2 to 3.6%. Mean days to 50% flowering was almost identical in the two environments.

The environment effect was significant for eight of the nine traits (P < 0.001), the exception being DFF (F = 3.35, P = 0.071). Genotype and genotype × environment effects were significant for every trait, including DFF (P < 0.001). The interaction was strongest for spikelet fertility (F = 38.93), filled grains (F = 30.63) and grain yield (F = 25.85), which are precisely the traits that decide yield under stress. Lines thus differed not only in yield potential but also in the degree to which that potential was lost under water deficit. Field coefficients of variation were low, from 0.60% for DFF to 4.71% for UFG. Under rainfed conditions (Table 2), ML-57 (28.38 g plant -1 ), ML-13 (27.96 g plant -1 ), ML-32 and ML-05 (both 27.72 g plant -1

) and ML-02 (27.08 g plant -1 ) gave the highest grain yields, and these five lines did not differ from one another by more than the critical difference in most yield components. They combined the largest number of filled grains per panicle (132 to 146), spikelet fertility above 84% and ten or more effective tillers per plant. ML-12, ML-20, ML-60, ML-28 and ML-33 formed the poorest group, with grain yield between 14.75 and 17.21 g plant -1 , spikelet fertility below 68% and more than 23 unfilled grains per panicle. These low yielders also flowered later under stress. ML-28 and ML-60 reached 50% flowering in about 97 days, compared with 91 to 95 days for most of the high yielders. are derived from the pooled error mean square of the genotype × environment analysis (error df = 80). Genotype and genotype × environment effects were significant at P < 0.001 for all traits; the environment effect was significant at P < 0.001 for all traits except DFF (P = 0.071). Yield reduction under rainfed conditions differed widely among lines (Table 3). ML-59 yielded slightly more under rainfed than under irrigated conditions (-1.24%), and ML-10, ML-49 and ML-57 lost less than 2%. Most of the high-yielding lines lost between 3.5 and 8.2%.

At the other extreme, ML-12, ML-20, ML-28, ML-60 and ML-33 lost 26.0 to 32.0% of their irrigated yield. The loss of spikelet fertility followed the same pattern. ML-51, ML-32, ML-59, ML-09 and ML-02 lost no more than 0.34 percentage points of spikelet fertility, and the first four recorded marginally higher fertility under rainfed conditions, whereas ML-60, ML-20 and ML-28 lost 13 to 14 points. Low yield reduction did not by itself identify the best restorer candidates. ML-10 and ML-49 were stable but had modest irrigated yields (25.33 and 24.02 g plant -1

), so their rainfed yield remained below that of lines such as ML-13 and ML-05, which lost around 6% from a higher starting point. Rainfed and irrigated grain yield were closely correlated across the 20 lines (r = 0.92, P < 0.001), which shows that yield potential contributed strongly to rainfed performance in this material. VIS, visual injury score on the 1 to 9 scale of IRRI ( 2013), where 1 is highly tolerant and 9 highly susceptible. Seedling and field index, standardised multi-trait selection index computed from seedling traits at 25% PEG and from field traits under rainfed conditions respectively, with higher values indicating closer proximity to the ideotype. GY, grain yield per plant, mean ± SE (n = 3). Yield reduction = (GY irrigated -GY rainfed)/GY irrigated × 100; a negative value indicates higher yield under rainfed conditions. SE (m)±, CD and CV apply to grain yield (genotype × environment, error df = 80).

Associations among traits and agreement between seedling and field screening

Under rainfed conditions, grain yield correlated most closely with spikelet fertility (r = 0.98), followed by filled grains per panicle, effective tillers and grain length (r = 0.96 each), panicle length (r = 0.95) and plant height (r = 0.94). Unfilled grains correlated negatively with yield (r = -0.95). DFF showed a moderate negative association with all yield-related traits (r = -0.39 to -0.48), so later-flowering lines tended to yield less under stress. The first principal component explained 87.0% of the variation among rainfed line means and the second 8.6%, so that two components captured 95.6% of the total (Figure -3). The seedling and field selection indices were strongly concordant across the 20 lines (Spearman ρ = 0.97, P < 0.001). Seedling traits at 25% PEG predicted rainfed grain yield well, with correlations of 0.97 for SDW, 0.95 for RDW and 0.94 for RL (all P < 0.001, n = 20). Root length at 25% PEG was also negatively associated with yield reduction (r = -0.89, P < 0.001). Five of the seven lines with the highest seedling index, namely ML-02, ML-32, ML-05, ML-57 and ML-13, were also among the six best lines in the field index, the sixth being ML-51. The marker classes were not evenly distributed along the tolerance gradient. Of the nine lines that scored 1 or 3 for visual injury, seven carried both Rf3 and Rf4, and only two (ML-05 and ML-57) carried Rf4 alone. Of the 18 lines positive for Rf4 alone, 16 scored 5 or higher. The association was not absolute, since the double-positive line ML-47 scored 7 and had one of the lowest seedling indices (0.111).

Discussion

Frequency and meaning of restorer marker classes

Nearly half the lines (46.7%) carried the Rf4 restorer allele and one in six carried both loci. Such frequencies are typical of breeding material with indica ancestry, in which restorer alleles are common because many elite indica lines trace back to restorer parents used in early hybrid programmes (Virmani, 1994) [1] . The prominence of Rf4 in the present set agrees with the view that this locus is the principal restorer of WA-CMS and is under positive selection wherever restorers have been bred deliberately (Yao et al., 1997) [3] . A marker score, it must be stressed, is a prediction of restorer status and not proof of it. Markers linked to a locus rather than lying within the causal gene can be separated from the functional allele by recombination, and the predictive accuracy of earlier SSR markers varied across genetic backgrounds for exactly this reason (Sheeba et al., 2009) [8] . Gene-based markers developed after the cloning of Rf4 reduce this risk considerably (Tang et al., 2014) [5] , but even they cannot account for modifier loci and background effects on pollen fertility. The decision to advance only Rf4-positive lines was a practical one and carries a cost. Lines positive for Rf3 alone may still restore fertility partially, and in some backgrounds Rf3 contributes substantially to seed set in the hybrid. Sattari et al. (2007) [7] found that lines carrying restorer alleles at both loci gave the most complete and consistent restoration, which supports giving double-positive lines priority but does not exclude single-locus restorers altogether. Test crossing the selected lines with one or more WA-CMS lines and scoring pollen and spikelet fertility in the F 1 is the necessary next step before any of them is described as a restorer in the strict sense.

Why seedling growth declined under osmotic stress

PEG-6000 lowers the water potential of the root medium without being taken up, so the seedling must lower its own tissue water potential to continue absorbing water. The first process to suffer is cell expansion, which depends on turgor and responds to water deficits smaller than those needed to reduce photosynthesis or stomatal conductance (Hsiao, 1973) [22] . This explains why length and fresh weight fell sharply even at 15% PEG. At 25% PEG the reduction in dry weight was of similar magnitude to that in length, which suggests that carbon assimilation was restricted as well as expansion, a pattern commonly reported once osmotic stress becomes moderate to severe (Farooq et al., 2009) [23] . Root length was the least affected trait at both PEG levels. Roots of many species sustain elongation at water potentials that halt shoot growth altogether, through osmotic adjustment in the growing zone and through changes in cell-wall extensibility that allow expansion at reduced turgor (Sharp et al., 2004) [24] . The smaller relative loss in root length than in shoot biomass seen here is consistent with this preferential maintenance of root growth. Lines that kept their roots growing under 25% PEG, such as ML-02, ML-32 and ML-59, would in principle be better placed to exploit deeper soil moisture in the field, a relationship that gained genetic support when the DRO1 allele for steeper root angle was shown to raise yield under drought (Uga et al., 2013) [25] . The significant genotype × stress interaction for all seedling traits has a direct bearing on selection. ML-49 had one of the longest roots under control conditions but suffered the largest loss of root length at 25% PEG, whereas ML-15 had shorter roots initially but lost the least. Seedling vigour under favourable conditions is thus a poor guide to tolerance, and screening must be done at a stress level strong enough to reveal the interaction. The discrimination achieved with 25% PEG in this study is in line with the report of Swapna et al. (2017) [14] , who found that PEG-induced osmotic stress separated rice varieties into tolerance groups that corresponded with their known field reaction. The very high intercorrelation of seedling traits, with 96.3% of variation on the first principal component, needs cautious interpretation. Length and biomass of the same organ are allometrically linked, and part of the agreement among traits reflects this dependence rather than independent physiological evidence of tolerance.

Yield formation under rainfed conditions

The mean yield loss of 11.4% under rainfed conditions was modest compared with the losses usually reported when water is withheld over flowering in managed drought nurseries, where yield frequently falls by half or more (Venuprasad et al., 2007) [11] . The stress in the present trial was thus moderate in its effect on yield. Stress severity changes which lines appear tolerant, and the ranking of rice lines under drought can shift considerably between mild and severe stress (Kumar et al., 2014) [12] , so the present classification applies to stress of similar intensity. The traits most affected were those formed at and after flowering. Unfilled grains rose by almost a third while grain length and panicle length hardly changed. Panicle length and grain dimensions are largely fixed before heading, whereas the proportion of spikelets that set grain is decided in a narrow window around anthesis. Water deficit during meiosis and anthesis reduces pollen viability, impairs anther dehiscence and lowers the number of pollen grains landing on the stigma (Saini et al., 1999) [26] . Genotypic differences in the sensitivity of anther dehiscence to water deficit have been demonstrated directly in rice (Liu et al., 2006) [27] , and such differences could explain why lines such as ML-32, ML-51 and ML-59 kept their spikelet fertility intact while ML-20, ML-28 and ML-60 lost 13 to 14 percentage points. The correlation of 0.98 between spikelet fertility and grain yield under rainfed conditions confirms that spikelet fertility, rather than tiller number or grain size, was the main limit on yield under stress in this material. Days to 50% flowering did not differ between environments on average, yet the genotype × environment interaction for DFF was significant. Some lines flowered earlier and others later under rainfed conditions, so the mean concealed opposite responses. Delay in flowering under drought is a well-established sign of susceptibility in rice, because stressed plants slow panicle development and the delay is associated with lower grain yield under stress (Pantuwan et al., 2002) [10] . The negative correlation between DFF and yield under rainfed conditions, and the late flowering of ML-28 and ML-60, agree with this interpretation. Earlier-flowering lines may also have partly escaped the most severe period of stress, and this escape component cannot be separated from true tolerance in a single-season trial.

Yield potential, stability and the choice of selection criteria

Rainfed yield correlated closely with irrigated yield (r = 0.92), so under the stress level of this trial most of the ranking under rainfed conditions reflected differences in yield potential. This relationship typically weakens as stress becomes severe, and in upland and rainfed lowland trials with intense drought the correlation between yield under stress and yield without stress can fall close to zero (Bernier et al., 2008) [28] . Under mild to moderate stress, as here, selecting lines with high potential remains an efficient strategy, provided the lines do not lose that potential disproportionately. Yield reduction alone would have favoured lines such as ML-10 and ML-49, which were stable largely because their irrigated yield was already lower. Indices based solely on relative yield loss are known to reward low-yielding but stable genotypes, a limitation recognised since the drought susceptibility index was first proposed for wheat (Fischer et al., 1978) [29] . The combined consideration of rainfed yield, yield reduction and a multi-trait index avoided this bias and identified lines that were both productive and stable. ML-57 illustrates the ideal case, combining the highest rainfed yield with a reduction of only 1.63%. ML-59 yielded slightly more under rainfed conditions than under irrigation. With three replications and a difference smaller than the critical difference, this is best read as full stability rather than a genuine positive response to stress.

Agreement between seedling and field screening

The close agreement between seedling and field indices (ρ = 0.97) and the high correlation of seedling dry weight at 25% PEG with rainfed grain yield were stronger than expected. Reviews of drought breeding in rice have repeatedly cautioned that seedling traits measured in artificial media often fail to predict yield under reproductivestage stress, because the processes limiting yield in the field operate at flowering and depend on soil water extraction by the whole root system (Fukai et al., 1995) [15] . Kamoshita et al. (2008) [30] reached a similar conclusion after reviewing a large body of phenotypic and genetic studies. Several features of the present study may explain the difference. The stress in the field was moderate and yield potential dominated rainfed yield, so any trait reflecting general vigour would be expected to correlate with it. The 20 field lines were themselves selected on seedling performance, and although this restriction of range would normally weaken a correlation, the lines that were advanced still spanned the full distance from highly tolerant to susceptible. Finally, the indices at both stages were built from sets of strongly intercorrelated traits, which tends to inflate agreement between them. The concordance is encouraging, but it should be confirmed in unselected material and under more severe stress before seedling PEG screening is used to replace field evaluation rather than to precede it.

Restorer loci and drought tolerance

Lines carrying both Rf3 and Rf4 were over-represented among the tolerant group. Seven of the nine lines with injury scores of 1 or 3 were double positive, while 16 of the 18 lines with Rf4 alone scored 5 or higher. No mechanism links fertility restorer genes to drought tolerance. Rf4 encodes a pentatricopeptide repeat protein acting on a single mitochondrial transcript in the anther (Chen et al., 2014) [6] , and the Rf3 region carries no gene known to influence water relations. The association is more plausibly explained by shared ancestry. Lines carrying both loci are likely to descend from a small number of elite restorer parents that were also selected for vigour and broad adaptation, and pedigree analysis of the test lines would allow this explanation to be tested directly. ML-47, a double-positive line with poor seedling tolerance, shows that restorer markers cannot be used as surrogates for drought tolerance. Their value lies in confirming restorer status, while tolerance must still be phenotyped. The fact that three of the five best lines in both screens carried both restorer loci is nevertheless of practical value. Lines with restorer alleles at both loci restore WA-CMS more completely and more consistently across environments than single-locus restorers (Sattari et al., 2007) [7] . If ML-02, ML-13 and ML-32 are confirmed as effective restorers in test crosses, they would combine robust fertility restoration with stable yield under water deficit, which is the combination required of male parents for hybrids intended for rainfed ecologies. The Rf4-only lines ML-57 and ML-05 performed as well in the field and deserve equal attention, since a single dominant Rf4 allele is often sufficient for restoration in indica backgrounds.

Limitations of the study

The field evaluation was conducted in a single season at one location. Genotype × season interaction is large for yield under drought, and the ranking of lines may change in a year with an earlier or more severe dry spell (Kumar et al., 2007) [31] . Restorer status rests on linked markers, and test cross data confirming fertility restoration are not yet available. Grain yield was recorded on a per-plant basis, which does not capture competition effects and plot-level yield. Physiological traits that would explain the mechanism of tolerance, such as leaf water potential, canopy temperature, relative water content and root distribution in the soil profile, were not measured. PEG-induced stress in nutrient solution also differs from soil drying in the absence of mechanical impedance and in the speed at which stress develops (Price et al., 2002) [32] . These gaps define the programme of work needed before the candidate lines are released into hybrid breeding.

Conclusion

Marker screening for Rf3 and Rf4 reduced 60 rice lines to 28 Rf4-positive restorer candidates, and a two-stage evaluation separated these candidates clearly for drought response. Osmotic stress at 25% PEG cut seedling growth by about one third on average, with a significant genotype × stress interaction that allowed tolerant and susceptible lines to be distinguished. Under rainfed conditions grain yield fell by 11.4% on average, and the loss was governed mainly by spikelet fertility. Seedling performance under 25% PEG ranked lines in close agreement with their rainfed yield. ML-02, ML-13 and ML-32, which carry restorer alleles at both loci, and the Rf4-positive lines ML-57 and ML-05 combined high rainfed yield with small yield reduction and strong seedling tolerance. These five lines are recommended as male parents for hybrids targeted at rainfed ecologies, subject to confirmation of fertility restoration through test crosses with WA-CMS lines. Future work should test the hybrids and their parents over several seasons and sites under quantified reproductive-stage stress, measure the physiological basis of spikelet fertility maintenance, and use gene-based markers to confirm the restorer alleles.

References

1. Virmani SS. Heterosis and hybrid rice breeding. Springer-Verlag; 1994.
2. Spielman DJ, Kolady DE, Ward PS. The prospects for hybrid rice in India. Food Security. 2013;5(5):651-665.
3. Yao FY, Xu CG, Yu SB, Li JX, Gao YJ, Li XH, Zhang Q. Mapping and genetic analysis of two fertility restorer loci in the wild-abortive cytoplasmic male sterility system of rice (Oryza sativa L.). Euphytica. 1997;98(3):183-187.
4. Zhang G, Lu Y, Bharaj TS, Virmani SS, Huang N. Mapping of the Rf-3 nuclear fertility-restoring gene for WA cytoplasmic male sterility in rice using RAPD and RFLP markers. Theoretical and Applied Genetics. 1997;94(1):27-33.
5. Tang H, Luo D, Zhou D, Zhang Q, Tian D, Zheng X, Chen L, Liu YG. The rice restorer Rf4 for wild-abortive cytoplasmic male sterility encodes a mitochondrial-localized PPR protein that functions in reduction of WA352 transcripts. Molecular Plant. 2014;7(9):1497-1500.
6. Chen L, Liu YG. Male sterility and fertility restoration in crops. Annual Review of Plant Biology. 2014;65:579-606.
7. Sattari M, Kathiresan A, Gregorio GB, Hernandez JE, Nas TM, Virmani SS. Development and use of a two-gene marker-aided selection system for fertility restorer genes in rice. Euphytica. 2007;153(1-2):35-42.
8. Sheeba NK, Viraktamath BC, Sivaramakrishnan S, Gangashetti MG, Khera P, Sundaram RM. Validation of molecular markers linked to fertility restorer gene(s) for WA-CMS lines of rice. Euphytica. 2009;167(2):217-227.
9. Suresh PB, Srikanth B, Kishore VH, Rao IS, Vemireddy LR, Dharika N, Sundaram RM, Ramesha MS, Rao KRSS, Viraktamath BC, Neeraja CN. Fine mapping of Rf3 and Rf4 fertility restorer loci of WA-CMS of rice (Oryza sativa L.) and validation of the developed marker system for identification of restorer lines. Euphytica. 2012;187(3):421-435.
10. Pantuwan G, Fukai S, Cooper M, Rajatasereekul S, O'Toole JC. Yield response of rice (Oryza sativa L.) genotypes to drought under rainfed lowlands. Field Crops Research. 2002;73(2-3):169180.
11. Kumar A, Dixit S, Ram T, Yadaw RB, Mishra KK, Mandal NP. Breeding high-yielding drought-tolerant rice: Genetic variations and conventional and molecular approaches. Journal of Experimental Botany. 2014;65(21):6265-6278.
12. International Journal of Advanced Biochemistry Research https://www.biochemjournal.com 26174707, 2026, 10(10), Downloaded from https://www.doi.org/10.33545/26174693.2026.v10.i10Sc.10278 by International Journal of Advanced Biochemistry Research on [02/10/2026]. See the Terms and Conditions (https://www.biochemjournal.com) for rules of use; OA articles are governed by the applicable Creative Commons License
13. Michel BE, Kaufmann MR. The osmotic potential of polyethylene glycol 6000. Plant Physiology. 1973;51(5):914-916.
14. Fukai S, Cooper M. Development of drought-resistant cultivars using physio-morphological traits in rice. Field Crops Research. 1995;40(2):67-86.
15. Gomez KA, Gomez AA. Statistical procedures for agricultural research (2nd ed.). John Wiley and Sons; 1984.
16. Jolliffe IT, Cadima J. Principal component analysis: A review and recent developments. Philosophical Transactions of the Royal Society A. 2016;374(2065):Article 20150202.
17. Hartigan JA, Wong MA. Algorithm AS 136: A k-means clustering algorithm. Journal of the Royal Statistical Society. Series C (Applied Statistics). 1979;28(1):100-108.
18. Hsiao TC. Plant responses to water stress. Annual Review of Plant Physiology. 1973;24:519-570.
19. Farooq M, Wahid A, Kobayashi N, Fujita D, Basra SMA. Plant drought stress: Effects, mechanisms and management. Agronomy for Sustainable Development. 2009;29(1):185-212.
20. Sharp RE, Poroyko V, Hejlek LG, Spollen WG, Springer GK, Bohnert HJ, Nguyen HT. Root growth maintenance during water deficits: Physiology to functional genomics. Journal of Experimental Botany. 2004;55(407):2343-2351.
21. Uga Y, Sugimoto K, Ogawa S, Rane J, Ishitani M, Hara N, Kitomi Y, Inukai Y, Ono K, Kanno N, Inoue H, Takehisa H, Motoyama R, Nagamura Y, Wu J, Matsumoto T, Takai T, Okuno K, Yano M. Control of root system architecture by DEEPER ROOTING 1 increases rice yield under drought conditions. Nature Genetics. 2013;45(9):1097-1102.
22. Liu JX, Liao DQ, Oane R, Estenor L, Yang XE, Li ZC, Bennett J. Genetic variation in the sensitivity of anther dehiscence to drought stress in rice. Field Crops Research. 2006;97(1):87-100.
23. Bernier J, Atlin GN, Serraj R, Kumar A, Spaner D. Breeding upland rice for drought resistance. Journal of the Science of Food and Agriculture. 2008;88(6):927-939.
24. Kumar R, Venuprasad R, Atlin GN. Genetic analysis of rainfed lowland rice drought tolerance under naturally-occurring stress in eastern India: Heritability and QTL effects. Field Crops Research. 2007;103(1):42-52.
25. Price AH, Cairns JE, Horton P, Jones HG, Griffiths H. Linking drought-resistance mechanisms to drought avoidance in upland rice using a QTL approach: Progress and new opportunities to integrate stomatal and mesophyll responses. Journal of Experimental Botany. 2002;53(371):989-1004. . 2026;SP-10(10):201-210. doi: https://www.doi.org/10. 33545/26174693.2026.v10.i10Sc.10278