Grain Yield Performance of Hybrid Rice in Multi-Environment Tests for Thirty-two Years in India

Background: Hybrid F 1 genotypes with higher yields or improvement in other traits of economic value due to heterosis as compared to inbred local check varieties (ILCV), are identied and released as hybrid commercial varieties. We analyzed the yield data of 2070 hybrid F 1 genotypes with ILCV evaluated over 32 years (1988 to 2019) in 2376 multi-environment experiments executed at 102 locations in the irrigated ecosystem across India. Results: The genetic gain or loss in yield of hybrid F 1 genotypes estimated over the test duration was non-signicant. Hybrid F 1 genotypes produced 10% more grains (728-2588 kg/ha) than ILCV in many experiments at several locations. Our analyses have established that grain yields of 7.0 to 7.9 t/ha, were harvested in hybrid F 1 genotypes with early, mid-early and medium maturity duration, and in those with medium slender grains at many locations in 362 experiments. A higher level of rice productivity per day (62 to 63 kg/ha) was recorded with the early maturing and mid-early maturing hybrid F 1 genotypes in these tests. The N requirement to produce 8 t/ha of hybrid rice grains was 15 kg N/t as compared with a minimum of 20 kg N/t used in China. Both the hybrids and inbreds in these experiments produced grain yields that were easily attained previously with high yielding ( ≥ 10 t/ha) commercial inbreds since 1968. Unless the attainable yields are reached in inbred checks with the proven appropriate crop production practices in an experiment, it is futile to estimate a genetic gain or loss for grain yields in new genotypes developed. Conclusions: Hybrid genotypes bred in India produced yields of 7 to 8 t/ha which matched with reports from China on hybrids and green super rice; these India-bred hybrids showed higher productivity per day and shorter maturity periods than super hybrids of China. Opportunities still exist to breed indica/japonica hybrids to obtain more heterotic early and mid-early maturing hybrids, and develop ecient agronomical practices to realize the potential advantages from hybrids. There is scope for breeders to limit test locations to represent specic target areas to avoid data loss. Focusing on removing obstacles in hybrid seed

In the dynamic hybrid rice-breeding program, hybrid genotypes with improvement in yields or in other traits of economic value, when compared with inbred local check variety (ILCV) are identi ed through phenotype-based selection and released for commercial cultivation. Each year hybrid F 1 genotypes (hybrid genotypes) with improvement in yields over ILCV are assessed by All India Coordinated Project (AICRIP) in nationally coordinated multi-environment tests (METs) in the irrigated ecosystem 5 . During the evaluation of hybrid genotypes, yield, grain quality and resistance to biotic and abiotic stresses are ranked. The grain yield of hybrid genotypes that are superior because they yield 10% more grains, in comparison with ILCV, is a decisive factor for primary selection in the METs. The mean grain yields of the top-three hybrid genotypes (T3HM), inbred local check variety (ILCVM) and hybrid F 1 breeding stock or experimental mean (EXPM) represent the hypothetical oating checks of Jensen 6 that adjust to yield gains, if any, annually. Many inbred genotypes developed at the breeding centers in the country have been assessed for their yield performance in the METs of AICRIP during 1974-1994.The analysis of three oating checks showed a signi cant and positive annual increase in yields of 1.2% or 52 kg/ha in inbred genotypes developed for the irrigated ecosystem 7 . ANOVA and regression analyses with and without environmental effects 8 by deducting the check or experimental mean from the mean yield of the top-three genotypes to estimate differences and gain in yield over time, revealed no evidence for either a genetic gain or loss in grain yields of inbred genotypes from 1974 till 1994 7 . The observed gain in yield has been attributed to improved crop-husbandry skills and farm infrastructure development at test locations over the years. The absence of any genetic gain for yields has also been demonstrated in inbred genotypes tested in 11 different rice ecosystems from 1995 to 2013 9 . Such an absence of genetic gain in yields was proved in similar analyses performed earlier on the grain yield of inbred rice genotypes developed by breeders and tested in worldwide experiments from 1976 to 1997 in the International Rice Testing Program (IRTP) -later termed as International Network for Genetic Evaluation of Rice (INGER) by the IRRI in different ecosystems 10,11 . The three oating checks in these international trials in the irrigated ecosystem showed non-signi cant annual grain yield increases of 16-18 kg/ha. The lack of progress in grain yield of inbred genotypes over the years was more or less similar in the national (AICRIP) and international (IRRI) testing programs 7,10,11 . We examined the realization of attributed yield advantage due to heterosis in the hybrid (F 1 ) genotypes developed using three-line system through an intensive national breeding program during the 32-year period from 1988 to 2019 using the same methods employed in earlier studies 7,9,11 .

Comparison of mean grain yields of hybrid genotypes in experiments
Data sets on grain yield assessment of hybrid F 1 genotypes along with inbred LCV in 2376 experiments executed between 1988 and 2019 were used for this study (Table 1). # Source data are provided in Additional le 1: Table S1-Sheet 1. # IHRT = Initial hybrid rice trial; IHRT-E with early maturity (no test in 1988, 1989, 1990, 1992, 1996 and 1999; IHRT-ME with mid-early maturity (no test in 1988, 1989 and 1990); IHRT-MED with medium maturity (no test in 1990 and 1992) and IHRT-MS with medium slender grains (no test in1988 to 2005) groups. AICRIP = All-India coordinated rice improvement project A comparison was done between the grain yields produced by hybrid genotypes in four Initial Hybrid Rice Trials (IHRT) with early (IHRT-E 110-120 days), mid-early (IHRT-ME 121-130 days), medium (IHRT-MED 131-140 days) maturity periods, and an exclusive group with medium slender grains (IHRT-MS 130±5 days) ( Table 2, Fig 1-2). The overall grain yields of T3HM in IHRT-E (110-120 days) were signi cantly higher (P ≤ 0.01) than those of the EXPM (by 761 kg/ha) or the ILCVM (by 1369 kg/ha). The EXPM was signi cantly (P ≥ 0.01) higher than that of ILCVM (by 608 kg/ha) ( Table 2). The linear regression models on the three oating checks over the years showed non-signi cant annual yield increases of 5 kg/ha in T3HM and 11 kg/ha in EXPM. The ILCVM however, showed a highly signi cant (P ≥ 0.01) annual decrease of 41 kg/ha in this maturity group trial (Fig. 1). The coe cients of determination for these three oating checks (R 2 , 0.008, 0.080 and 0.303, respectively) were very low and non-signi cant. In IHRT-ME (121-130 days), the overall grain yields of T3HM were signi cantly higher (P ≤ 0.01) than the EXPM (by 802 kg/ha) or the ILCVM (by 992 kg/ha) ( Table 2). The EXPM yield was also signi cantly higher than that of ILCVM (by 191 kg/ha). The linear regression models on the three oating checks over the years showed non-signi cant annual yield increases of 5 kg/ha in T3HM, and 16 kg/ha in ILCVM. The coe cients of determination values (R 2 , 0.035 and 0.176) were very low and non-signi cant. Strangely, there was a signi cant increase of 21 kg/ha in EXPM (R 2 , 0.540, R a 2 , 0.523, P ≤ 0.01, MSE, 171) in this mid-early maturity group trial (Fig. 1). In IHRT-MED (131-140 days), the overall grain yields of T3HM were signi cantly higher (P ≤ 0.01) than those of the EXPM by 742 kg/ha or ILCVM by 869 kg/ha ( Table 2). The EXPM was higher than ILCVM by 127 kg/ha but statistically on par with it. The linear regression models on the three oating checks over the years showed signi cant yearly yield increases of 29 kg/ha in T3HM (P = 0.05), 35 kg/ha in EXPM (P = 0.01)and 24 kg/ha in ILCVM (P = 0.05) in this maturity group trial (Fig.2). The coe cients of determination were signi cant for these three oating checks (R 2 = 0.378, 0.568 and 0.371, respectively). But the R a 2 was non-signi cant for ILCVM. In IHRT-MS (130±5 days), the overall grain yields of T3HM were signi cantly higher (P ≤ 0.01) than those of EXPM by 580 kg/ha or ILCVM by 972 kg/ha ( Table 2). The EXPM was also higher than ILCVM by 392 kg/ha. The linear regression models on the three oating checks over the years showed signi cant (P = 0.05) yearly decreases of 37 kg/ha in yields of T3HM, 29 kg/ha in EXPM and 46 kg/ha in ILCVM grain yields (R 2 , 0.605, 0.525 and 0.584, respectively) in this trial (Fig. 2) In each of the maturity group trials, we detected four major grain types in hybrid F 1 genotypes 5 : long bold (kernel length >6 mm, and length: breadth ratio (L/B) of <3.0); long slender (kernel length >6 mm, and L/B of >3.0); medium slender (kernel length <6 mm, and L/B of 2.5 -3.0); and short bold (kernel length <6 mm, and L/B of <2.5) (Additional le 4, Table S2). Hybrid F 1 genotypes with the consumer preferred medium slender grains (kernel length <6 mm, and L/B of 2.5 -3.0) were present in different maturity group trials that exactly matched in number with those in medium slender trial (181). Hybrid F 1 genotypes with short slender grains (kernel length <6 mm, and L/B of >3.0) have also been found in different maturity trials except in early maturing (<120 days) trial.
Performance of hybrids across the country Hybrid F 1 genotypes with different maturity periods and those with medium slender grains were assessed for grain yields in tests for only 1-4 years in 270 experiments executed between 1988 and 2019 (Table 3). These data sets were inadequate to control error degrees of freedom in analysis 12 . Mean grain yields in the remaining data set from 1947 experiments executed at different locations for 5-24 years, were used for analysis (Additional le 1, Table S1). Further scrutiny indicated that hybrid F 1 genotypes recorded lower mean grain yields than inbred checks over the years in 962 experiments and were well-distributed in IHRT-E (286), IHRT-ME (218), IHRT-MED (316) and IHRT-MS (142 experiments) ( Table 3). Only the data from 985 experiments showed higher yields of hybrid F 1 genotypes than inbred checks at several locations and in one or more trials.  and Voluntary -partially public funded. †DMRT computed based on unequal replications. Any two means followed by a common alphabet are not signi cantly different at P = 0.05. CV = Coe cient of variation Hybrid F 1 genotypes with mid-early maturity (121-130 days) produced higher grain yields than ILCV at 21 locations in 242 experiments (Table 5). In the analysis of variance, the F-value showed highly signi cant (P < 0.01) differences in the mean grain yields of locations compared with ILCV. The yields achieved in hybrid F 1 genotypes were within the hypothesized 10% extra grain yield over ILCV due to heterosis at three AICRIP, one private and one voluntary location, besides HCV. Thirteen locations were in four distinct statistically signi cant groups. Hybrid genotypes showed higher yields by 980-2130 kg/ha than ILCV, and by 641-1791 kg/ha than HCV. The ILCV registered only 5.7 t/ha in this trial. The hybrid genotypes showed mean grain yields ranging from 7.1 to 7.9 t/ha at nine (4 AICRIP and 5 private funded) locations in three statistically signi cant groups. The grain productivity per day varied from 57 to 63 kg/ha at these locations.  (Table 6). In the analysis of variance, the F-value showed highly signi cant (P < 0.01) differences in the mean grain yields of locations compared with ILCV. The yields achieved in hybrid F 1 genotypes were within the hypothesized 10% extra grain yield over ILCV due to heterosis at two (1 private funded and 1 voluntary) locations and HCV. Hybrid genotypes showed higher yields 656-2184 kg/ha than ILCV, and by 118-1646 kg/ha than HCV at 16 other (6 AICRIP funded, 9 private funded and 1 voluntary) locations. The ILCV registered only 5.7 t/ha in this trial. The hybrid genotypes showed mean grain yields ranging from 7.0 to 7.9 t/ha at 11 locations. The grain productivity per day varied from 52 to 58 kg/ha at these locations. #Source data are provided in Additional le 1: Table S1-Sheet 2. # ILCV -inbred local check; HCV -hybrid commercial variety (used only from 1996). *AICRIP (All-India coordinated rice improvement project) -public funded, Private -seed industry funded, and Voluntary -partially public funded. †DMRT computed based on unequal replications. Any two means followed by a common alphabet are not signi cantly different at P = 0.05. CV = Coe cient of variation Hybrid F 1 genotypes with medium slender grains (130±5 days) recorded higher mean yields than ILCV at 15 locations in 177 experiments (Table 7). In the analysis of variance, the F-value showed highly signi cant (P < 0.01) differences in the mean grain yields of locations compared with ILCV. The yields achieved in hybrid F 1 genotypes were within the hypothesized 10% extra grain yield over ILCV due to heterosis at four AICRIP locations. At 11 other (3 AICRIP, 5 private funded and 3 voluntary) locations, hybrid genotypes recorded higher yields by 828-2525 kg/ha than ILCV and by 93-1789 kg/ha than HCV. The ILCV registered only 5.0 t/ha in this trial. The hybrid genotypes showed mean grain yields ranging from 6.5 to 7.6 t/ha at seven locations in three statistically signi cant groups. The grain productivity per day varied from 49 to 57 kg/ha at these locations. #Source data are provided in Additional le 1: Table S1 Table 3). The mean grain yield varied in these ILCV also; it varied from 3.6 to 6.0 t/ha in early, from 3.8 to 7.4 t/ha in mid-early, and 4.1 to 7.9 t/ha in medium maturing ILCV, and from 4.5 to 5.0 t/ha in ILCV with medium slender grains. Nevertheless, several inbred varieties in many experiments produced high yields of 7 to 9 t/ha (Additional le 6, Table S4).

Discussion
Hybrid rice seeds are genetically packed to produce high yields due to heterosis. Seeds are also the most effective and inexpensive medium to transfer hybrid rice technology to farmers. Isolated evaluation reports on hybrid rice were biased and conclusions were based on insu cient data 13, 14 . We have used the largest data generated on hybrid rice by AICRIP in METs 5 for 32 years (Additional le 1: Table S1). Although the irrigated ecosystem is made homogenous by regular applications of water, environmental and yearly variations may in uence the yields recorded at test locations. Therefore, yield data sets were statistically corrected suitably to neutralize the year effect as well as the location effect before any analysis. Further, the three oating checks namely T3HM, EXPM and ILCVM were used to remove pitfalls in estimating absolute yield performance. When similar trend is observed in the oating checks, the actual genetic gain or loss over years is obtained in the re-analysis by deducting year-wise experimental or check mean yield from the mean yield of top-three genotypes (Fig 1-2).

Grain yields of hybrid genotypes
The top-three hybrid genotypes in IHRT trials produced signi cantly higher grain yields than that of the experimental mean or the inbred check variety mean yields over the years and across locations evaluated ( Table 2). Over 32 years (1988-2019), the grain yields of T3HM increased by 0.6 to 0.8 t/ha than that of the EXPM and by 0.9 to 1.4 t/ha than that of ILCVM. The EXPM of hybrid genotypes assessed was also higher by 0.1 to 0.6 t/ha than that of the inbred LCVM ( Table 2). The contention on yield advantage due to heterosis in hybrids was not apparently present in most of these hybrid F 1 genotypes. The linear regression models on the three oating checks over the years showed signi cant annual increases of 5-29 kg/ha in T3HM and 11-34 kg/ha in EXPM in early, mid-early and medium maturing hybrids. However, the mean grain yield of ILCV decreased in early maturing hybrids and those with medium slender grains by 41-46 kg/ha but increased in mid-early and medium maturing hybrids by 16-24 kg/ha. The consumers in India are well-known to prefer and accept short bold or long slender grains. The stress on market preferred grain type rather than on the level of heterosis of hybrid F 1 genotypes was the evident cause of such declining yields of oating checks in IHRT-MS trial over the years. The classical stability parameters often do not consider the absolute performance of genotype cohorts evaluated in experiments. A few statistical parameters were developed to combine both stability and performance 6,15,16 . The oating checks of Jensen 6 however, combine both stability and performance 17 . The trend observed was more or less similar across the three oating checks studied using AICRIP experimental data. When a linear model showed a signi cant increase or decrease in T3HM, the environmental effects were eliminated in that data year-wise by deducting EXPM or ILCVM from T3HM. The linear models developed after removing environmental effects became statistically non-signi cant. Evidence thus obtained clearly revealed lack of genetic gain or loss in hybrid genotypes developed over the 32 years. Our analyses have established that grain yields of 7.0 to 7.9 t/ha were harvested in hybrid F 1 genotypes with early, mid-early and medium maturity, and in those with medium slender grains at 20 locations and in 362 experiments performed during 1988 to 2019 (Tables 4-7).

Disarray in the assessment of hybrids
The assessment of hybrid genotype performance has generally been de ning rather than experimenting with them due to following reasons. During 1988-2019, 102 locations were used to test 2070 hybrid F 1 genotypes in 2376 experiments across rice growing irrigated areas in India. Due to ambiguity in nomenclature of some locations in the initial years, data from 159 experiments were removed before analyses. Grain yield assessments were done on hybrid genotypes for only 1-4 years in 270 experiments performed and were omitted as inadequate for analysis 12 . Boyles et al 18 had also excluded data from locations that were not represented for ve or more years from the long-term analysis of red wheat experiments. The remaining assessments were done for 5-24 years. In 962 experiments, hybrid F 1 genotypes had produced yields lower than that of inbred LCV in one or more trials. Thus, nearly two-thirds of all experiments performed had demonstrated the lack of hybrid vigor in hybrid F 1 genotypes evaluated. This is a colossal loss of efforts at 102 locations on eld evaluation, time and money used to conduct these experiments made with hybrid F 1 genotypes from 1988-2019 (Additional le 1, Table S1). Initiation of an additional singular trial for hybrids with medium slender grains from the year 2006 further demonstrated the speculative organization of testing in METs of AICRIP as early, mid-early and medium maturing hybrid F 1 genotypes with this grain type were already available in adequate quantities (Additional le 4: Table S2). The limited use of released 10 commercial hybrids from 2005 as checks (Table 8) further con rmed the di culty of producing and supplying hybrid seed in su cient quantity for tests. However, 54 inbred checks have been used with frequent and rather irrational replacements. Astoundingly, while high yields (~9 t/ha) were recorded in ILCV in some experiments, at most locations (Additional le 6, Table S4) but the ILCVM remained low (~6 t/ha). Apparently there was no attempt to select an appropriate inbred variety for use as check. Our analysis further adds to doubts raised by Muralidharan et al 7,9 that either the yield of inbred local check is underestimated or the yield of hybrid F 1 genotype is over estimated. The lack of required repetitions of experiments has made the bulk of resource and efforts on assessment of grain yields of expensive hybrid F 1 genotypes irrelevant. There was evidently no critical assessment of the data on hybrid rice experiments in AICRIP from time to time.
Critical appraisal of test locations for evaluation of hybrid rice For commercial release of a hybrid genotype, the criterion set is the ability to produce at least 10% higher grain yield than the inbred check due to heterosis 3 . During 1996-2019, rice hybrid F 1 genotypes were tested at 102 test locations in the irrigated ecosystem in four trials across India. The mean grain yields produced by hybrid F 1 genotypes were higher than the yields harvested in ILCV at 30 locations (13 AICRIP, 11 private and 6 voluntary) (Tables 4-7). The coe cient of variation (CV) is a standardized, dimensionless measure of dispersion relative to a data set's average. It enables the comparison of several data sets on genotypes 19 . It is also used as a measure to compare the robustness of different biological traits 20 . At locations where heterosis in yield was detected in hybrid F 1 genotypes, the CV remained low (10 to 11%) indicating robustness of analysis. Only 23 locations, hybrid F 1 genotypes recorded ≥10 per cent higher grain yields than that of ILCV. These locations

Legitimacy of hybrid trials
Hybrid F 1 genotypes, inbred checks and commercial hybrid varieties have been organized into four trials with IHRT-E (early 110-120 days), IHRT-ME (mid-early 121-130 days), IHRT-MED (medium maturity 131-140 days), and IHRT-MS (medium slender grains maturing in 130±5 days). Our analyses have shown the presence of medium slender grain types in the hybrid genotypes included in these trials in nearly equal numbers. Therefore, the IHRT-MS is redundant. The productivity per day was comparatively higher in early and mid-early maturing than medium maturing commercial hybrids or ILCV (Tables 4-6).
China has so far cultivated only commercially released hybrids that have a long maturity duration (>140 days).Yuan and Sun 22 had reported on the development of an early maturing hybrid Weiyou 64. The compatibility of early maturity and high yield in rice has been a subject of debate 23 . Using the early-maturing restorer line Ce64 derived from IR9761-19-1, Minghui77 24 , Zaohui89 25 and other early maturing hybrids were developed in China. Mapping of a dominant earliness gene, Early owering-completely dominant (Ef-cd) encoding a long noncoding RNA (lncRNA) 26 and its subsequent cloning 27 have revealed the genetics behind the possibility of exploring this trait to combine early maturity and high yield for enhancing productivity per day. High-yielding rice hybrids in China usually take 160 to even 180 days from sowing to harvest. The newly developed early maturing (125 days) hybrid variety G3-1S/P19 tested at two sites produced 1.57 kg/m 2 in central China's Hunan Province 28 . Several high yielding early maturing inbred commercial varieties have been developed in India 29 and elsewhere, unconscious of the earliness gene (Ef-cd), based on simple phenotype selection 30 . We have demonstrated that early (110-120 days) and mid-early maturing (121-130 days) hybrid F I genotypes produce high yields (7 to 8 t/ha) with increased productivity per day (61-63 kg/ha). As the maturity duration is increased, water requirement, the cost of cultivation and the associated risks also increased. A scrutiny of AICRIP data indicated that most hybrid F 1 genotypes matured between 115 and 135 days. Therefore, efforts may be more focused to breed early and mid-early maturing hybrids with high productivity per day.

Hybrid vigour and its exploitation
Hybrid genotypes had produced higher yields by 728-2588 kg/ha than the yields of inbred LCV (4922-5648 kg/ha) at only 30 locations tested in 5-24 experiments (Table 4-7). None of the hybrids evaluated for 32 years had shown a mean yield of ≥10 t/ha that has been long-established repeatedly with inbred LCV in METs of AICRIP 7,9,11,31 . Yet, the grain yields of 7.0 to 7.9 t/ha in hybrid genotypes attained in many experiments reveal the achievable higher yield owing to unhindered and better crop husbandry. This inference is con rmed by the lack of genetic gain for grain yields of hybrids in any of the trials examined in this study. The claim of heterosis can deceptively come from an under-estimation of inbred LCV, especially when inbred yields are as low as 5-6 t/ha as compared with their easily attainable level of ≥10 t/ha under e cient crop management. Heterosis is considered as an important trait to increase biomass and yield in hybrid F 1 genotypes than in the speci c parents used, but not over the yield of commercially cultivated high yielding inbred varieties. Rice-breeding using abundant and variable germplasm accessions for many decades has effected countless -and often unknown -changes in the genetic composition of inbred varieties. Breeding activity however, is solely dependent on the meager availability of male sterile lines to develop hybrid rice varieties. The genetic basis of heterosis is less understood 32 . The recent advancements in cloning the heterotic QTL GW3p6 and development of a near-isogenic rice line indicate the possibility of realizing high yield in inbred rice lines without even needing to develop hybrid rice 33 .

Unsettled issues
Breeding hybrids has demonstrated improved yield potential in many cross-pollinated crops such as sorghum, maize and cotton 34,35,36 . Hybrid breeding takes advantage of heterosis (hybrid vigour). It is a phenomenon where F 1 hybrids derived from crosses between genetically distinct inbred varieties exhibit superior phenotypic performance over their parents. Superior performance may specify an improved function of any biological process in the hybrid offspring 37 . Rice is a self-pollinated crop and commercial hybrid breeding mostly depends on a few cytoplasmic male sterile systems. It requires a lot of nancial resource to evaluate eld performance of all possible crosses among a large number of inbred lines 38 . In general, only a small proportion of crosses can be evaluated in the eld and many potential superior crosses may not get tested. The evidence generated in our study has con rmed the production of the stipulated 10% more grains by some hybrids than what was produced by inbreds in many experiments. The grain yields recorded in these experiments however, were well below the attainable yield levels of inbreds established ever since 1968. The genuineness of the claim of a yield advantage in any hybrid over inbred variety is untenable. The optimized canopy, architecture, dark green leaves, erect ag leaf, narrow leaf, dwarf plants with a plant height of ~one meter, panicle length of 25 cm, 180 grains per panicle, grain lling rate of 81%, 1000-grains weight of 26 g, and adaptability to a wide range of growing environments determine yield level of rice (data collected from 2013 ~ 2015 by China Rice Data Center, http://www.ricedata.cn/variety/varis/604222.htm) 39 . The architectural and physiological features associated with high yield were recently studied in two elite historical hybrid rice cultivars, i.e., YLY1 and LYP9 39 . The canopy photosynthesis was found to increase the proportion of biomass allocation to above ground tissues (1.5%), productive tillers (25%), photosynthate reserve in leaf sheath (5-11%) before grain lling and photosynthate translocation to grains. Yet, the yields of hybrids have remained below or at best on par with the yield records of inbred varieties (~10 t/ha).
A longer root, a larger number of tips, a better developed aerenchyma, a higher capacity for N uptake, and reduced NH4 + e ux from roots are associated with higher N-use e ciency and growth performance in hybrid super rice Yongyou 12 and Jiayou 6 40 . Lin et al 41 used a linkage map consisting of high-density SNPs, to identify heterosis-associated ve genes, which contributed to the high yield, with repeated occurrence of qSS7 and qHD8 in both hybrid populations. The two super hybrids Yongyou12 and Jiayou 6, however, produced grain yields of 11.0-11.8 t/ha compared to 9.1 t/ha in common (inbred) variety Xiushui 134 at 200 kg N/ha 40 .
The expression high yield of any variety is to be used in relation to the one that produces lower yield. However, whenever high yields of hybrids are reported, the yields of inbreds have been less than their demonstrated attainable yield. The question therefore, arises of whether the extra yield in the hybrid is due to heterosis? Is it only an assumption that the heterosis of hybrids developed leads to increased yields over inbreds? There is little evidence to indicate that heterosis of hybrids bred is due to induction of better traits in F 1 plants leading to enhanced yields. It is essential to prove that hybrids have certain improved traits compared with those of inbreds to produce more grain yields.
Opportunities to achieve desirable heterosis for yield Dominance 42 and overdominance 43,44 hypotheses were coined long ago to explain heterosis. The molecular mechanism of heterosis still remains mysterious despite research on heterosis for more than a century. Unfolding research ndings indicate the potential to improve heterosis by exploiting inter-speci c heterosis between African rice and Asian rice 45

Stagnating yield in inbred varieties in breeders' experiments
Absence of any genetic gain for grain yield has been demonstrated in successively generated inbred commercially released varieties compared to the early release of Indian semi-dwarf inbred varieties in 1968 by analyzing yield data from AICRIP 7,9 and from IRRI's international rice testing program 11,31 . Yields of major crops have stabilized or even stagnated in many regions of the world 31,55,56,57 . Any genetic gain of wheat, barley, rice and other cereal crops after 1967 is rightly attributed to modi cation of plant architecture, especially reduction in plant height, and the consequent non-lodging at high levels of nutrient application 7 and increased harvest index and grains per unit area 58 . Nevertheless, AICRIP's progress in conventional breeding has added numerous traits of value in rice especially the ability to withstand abiotic and biotic stress conditions leading to enhanced stability in yield performance and overall production in the country. Improvements in the quality of grains of inbred varieties have enabled huge rice exports. Increases in national rice production have con rmed the continued improvement in crop production skills at the farm level along with effective use of irrigation, water-use-e cient varieties and other factors 9 . Inbred LCV had recorded overall mean grain yields ranging from 4.8 to 5.5 t/ha (Table 2) in 2376 experiments across the country executed for 32 years along with hybrid F 1 genotypes. The mean grain yields produced by inbred LCV were low even at the best performing 20 locations (5.0 to 5.7 t/ha). Nonetheless, the inbred check had shown high yields ranging from 9.1 to 9.8 t/ha at several locations. Overwhelming evidence is available to indicate the stagnating grain yields of LCVs since the release of inbred variety Jaya that established 10 t/ha as the easily realizable yield in 1968. A decisive introspection is warranted concerning crop husbandry practices that lead to low grain yields in inbred varieties compared to earlier records of high yields reported in support of their release. Unless the attainable yields are reached in inbred checks with appropriate crop production practices in an experiment, it is futile to make any comparison with new genotypes to estimate a genetic gain for grain yields.

Dilemma in choosing inbred or hybrid variety
Many indica and japonica hybrids have been cultivated in addition to inbred varieties in China. SY63 is a rice hybrid derived from the female parent Zhenshan 97 A (ZS97A, a WA CMS line) and the male parent MH63 in China 59 . The indica/japonica hybrid has more productive tillers, larger sink size, increased accumulation of non-structural carbohydrate in the stems at heading time and its remobilization to grains, higher enzyme activity for sucrose-to-starch conversion process in the grains, and greater photosynthesis during the ripening period; the hybrid has greater root biomass, deeper soil distribution at heading time, and higher root oxidation activity during the ripening period. Due to enhanced agronomic and physiological traits, indica/japonica hybrid produces improved yields under low N input conditions 60 71 . Potential yield is a location-speci c attribute as it depends on the crop growth duration and local weather. However, the current farm yield in China varied from 5.2 to 8.8 t/ha 61 . Huanghuazhan is the most common inbred rice cultivar planted in central and south China and is widely grown in 4.5 million ha across seven major rice-producing provinces (http://www.ricedata.cn/variety/) with high and stable yield, good quality, and wide adaptability. Although Huanghuazhan is an inbred cultivar, the yield is comparable to that of hybrid cultivars 61,72 . Therefore, it is clear that to harvest grain yields beyond 10 t/ha in hybrid super rice even in a limited area at select locations, adoption of intensive crop management and application of high levels of fertilizers and other input will be required. Rice grain yields of 10 to 11.6 t/ha in inbred varieties were recorded in yield trials performed at many locations in India 7,31 and in international trials across rice growing areas in many countries of the world 11,31 . Muralidharan et al 9 further concluded that the potential yield is limited to 15-16 t/ha and attainable yield to 10-11 t/ha with choice inbred varieties in the best rice growing stress-free environment under intensive management. He et al 73 estimated hybrid rice to yield 7.5 t/ha when compared with 6.5 t/ha yields of inbreds in Jiangsu Province, China. From 1976 to 1995, hybrid rice varieties covering 15.7 million ha in China have yielded on an average 6.6 t/ha compared with 5 t/ha for conventional varieties 74 . Shanyou 63 (SY63), the most widely cultivated indica hybrid rice, had recorded 7325 kg/ha in various regional, provincial and national yield trials from 1982 to 1985. In spite of such yield reports, the increased grain production in China from 1984 to 2012 due to SY63 was estimated at an average yield increase of 300 kg/ha over check varieties 59 75 indicated that mean yields were 9.2 t/ha (4.6 to 12.4 t/ha) from 1809 japonica hybrids (156 days) compared to 8.7 t/ha (3.3 to 13.9 t/ha) from 296 japonica inbreds (153 days) 75 . The late maturing japonica (glutinous with low amylose content of 10-15% in grains or sticky rice) are grown in areas exposed to long days, and where grain lling continues for a longer period. Li et al 75 further reported that the mean yields were 8 t/ha (3.8 to 13.6 t/ha) from 4814 indica hybrids (132 days) compared to 6.7 t/ha (3.2 to 12.4 t/ha) from 767 indica inbreds (123 days). Our analysis of 2070 indica hybrid F 1 genotypes in trials of AICRIP performed in India from 1988-2019 also showed at 42 locations similar higher grain yields of 7.0 to 7.9 t/ha in early (110-120 days), mid-early (121-130 days) and medium (131-140 days) maturing hybrid F 1 genotypes, and 5.8 to 7.6 t/ha in those with medium slender grains (130±5 days) as compared to 5.0 to 5.7 t/ha in 583 indica inbreds (Tables 4-7). In all these yield trials of AICRIP, yields of inbred varieties were lower than their known attainable levels. None of the hybrids tested could match the mean grain yields of ≥10 t/ha recorded by indica inbred varieties such as Rasi, IR 36 and IR 50 (early 115-120 days), Jaya and IR 8 (medium 131-135 days), and Swarnadhan and Savitri (late maturing >140 days) in rice yield trials 7,9,11,31 . The inbred variety Pusa 44 is well known to produce grain yields of 7 to 8 t/ha 29 (Tables 4-7). The N requirement to produce 8 t/ha of hybrid rice has been estimated at a minimum of 160 kg N/ha or 20 kg N/t in China 68,69 compared to 120 kg N/ha or 15 kg N/t used in AICRIP experiments with hybrid genotypes across India. Elevating the expression or activity of NGR5 (nitrogen-mediated tiller growth response) in rice can possibly reduce nitrogen fertilizer use while increasing grain yields further 77 . Application of an optimum level of nitrogenous fertilizer can additionally reduce emission of nitrous oxide, a greenhouse gas that is 300 times more potent than carbon dioxide in affecting climate change 78 . Nitrogen use is ine ciently distributed spatially across global food systems 79 . India has a large population and a large yield gap in rice production (~ 6 t/ha) 9 . The funds saved by reducing hybrid rice trials and test locations may be diverted to support sustainable intensi cation of rice production and research on precision farming and on increasing nitrogen use e ciency to reduce pollution.
Incidence of diseases and pests were also documented on hybrid genotypes tested in AICRIP trials 80 . Genetic gains in terms of resistance to diseases and pests have been remarkable. Within a decade after the release of Jaya in 1968, rapid strides were made in the development of stable and widely adaptable varieties that were insulated with resistance against biotic and abiotic stresses 7,29 . The Indian varieties that produced stable high yields in the international tests, or those which possessed resistance to stresses or good quality traits found wide acceptance and claimed release in several countries around the world 31,81 . Genetic gain for resistance was amply proved when several AICRIP improved genotypes were shown to possess resistance 7,81,82 . Besides yield attributes, it is a simple task to incorporate genes possessing resistance to pathogens and insect pests in susceptible inbreds to achieve de nite genetic gains. Although restorer lines with resistance genes are used to generate hybrids, little information is available on their full expression in hybrids. It is apparently easier to manage resistance and prevent yield losses from diseases and pests in inbreds than in hybrids. The emerging molecular understanding appears to indicate a metabolic con ict between heterosis and defense mechanisms. Down-regulation of defense response genes in hybrid has been reported to lead to heterosis 83,84 . Yields in hybrid seed production of the three-line system are still in the range of 1.5 to 2.0 t/ha in Asia 85 . Hybrid rice seed production requires adoption of appropriate but non-farmer friendly cultural practices and skilled labor at premium wages to ensure a good seed set, which ultimately increases the cost of F 1 seeds.
Focus on increasing hybrid seed yield is needed. Apomixis, the asexual formation of seeds can be used by breeders to x heterosis in hybrid seeds and rapidly generate doubled haploid crop lines to produce more hybrid seeds at a low cost. Using CRISPR/Cas9 gene editing technology, synthetic apomixis was established, but it reduced the numbers of clonal seeds produced 86 . This uneconomical and scanty hybrid seed production forces the Government of India to provide a massive nancial support for training, production and distribution of hybrids seeds by public organizations and private seed business rms 87 . Yet, due to easy availability and much less price of seeds, farmers would undoubtedly prefer to grow high yielding inbred commercially released rice varieties to reduce the production cost. We have demonstrated that at the same level of cultural and agronomic management from 1988-2019, hybrid F 1 genotypes with different maturity periods produced yields that are comparable or less than inbred LCV in 1391 experiments; in 985 experiments hybrid F 1 genotypes produced yields that are ≥10% higher than yield of inbred varieties. Further, records on high yields of 10-13 t/ha in hybrid F 1 genotypes and 9-10 t/ha in inbred varieties at many experiments beyond doubt proves the need to use e cient agronomic management practices in rice production to acquire higher yield bene ts.

Conclusions
We analyzed the rice grain yield data of 2070 hybrid F 1 genotypes with inbred varieties evaluated over 32 years (1988 to 2019) in 2376 multi-environment experiments executed at 102 locations in the irrigated ecosystem across India. The genetic gain or loss in yield of hybrid F 1 genotypes estimated in tests over years was non-signi cant. Hybrid F 1 genotypes produced grain yields of 7.0 to 7.9 t/ha, which matched with the yields of hybrids and green super rice reports from China. India-bred hybrids showed higher productivity per day (62 to 63 kg/ha) and shorter maturity periods than super hybrids of China. The N requirement to produce 8 t/ha of hybrid rice grains was 15 kg N/t as compared with a minimum of 20 kg N/t used in China. Hybrids and inbreds produced grain yields that were easily attained with high yielding (≥10 t/ha) commercial inbreds since 1968, very similar to the yields reported from China. Only in less than 20% of test locations, hybrid F 1 genotypes produced 10% more grains than the yields of inbred varieties. Yet, high yields of 9-10 t/ha were recorded in yield trials with both inbred varieties and hybrids at many locations in India as reported from other rice growing areas in many countries of the world. Therefore, doubts arise on whether the yield of inbred local check is underestimated or the yield of hybrid F 1 genotype is overestimated. Further investigation is needed to nd the reasons for the lower yields of inbred at these locations using uniform crop husbandry. Unless the attainable yields (≥10 t/ha) are reached in inbred checks with the proven appropriate crop production practices in an experiment, it is futile to estimate a genetic gain or loss for grain yields in new genotypes developed. Opportunities still exist to breed more heterotic early and mid-early maturing hybrids, and develop e cient agronomical practices to realize the potential advantages from hybrids. The presence of de nite improved traits must be demonstrated in hybrid F 1 genotypes before making any comparison with inbred varieties to produce more grain yields. There is scope for breeders to limit test locations to represent speci c target areas to avoid data loss. Focusing on removing technical barriers in hybrid seed production is essential to exploit yield heterosis in hybrids, and to make hybrid rice technology pro table to farmers.

Methods
METs of hybrid genotypes and data sets on grain yield As many as 2070 hybrid F 1 genotypes in Initial Hybrid Rice Trials (IHRT) were evaluated in METs of AICIRP for 32 years 5 .
Hybrid F 1 genotypes were grouped depending on the maturity duration and grain type into four different trials: IHRT-E (early maturing -110-120 days), IHRT-ME (mid-early maturing -121-130 days), IHRT-MED (medium maturing -131-140 days) and IHRT-MS (medium slender grains maturing in 130±5 days). These experiments have led to the identi cation and release of 105 (36 public and 69 private seed industries bred) rice hybrids for commercial cultivation in India 88 . Data sets on grain yield assessment of hybrid F 1 genotypes along with inbred LCV in 2376 experiments executed between 1988 and 2019 were used for this study ( Table 1). The ILCV were nationally released commercial inbred varieties grown widely at each test location.
These experiments were performed in the irrigated elds at 102 locations in 23 states of India. Trials were conducted with different sets of hybrid F 1 genotypes each year belonging to early (IHRT E 110-120 days), mid-early (IHRT ME 121-130 days) and medium maturity (IHRT MED 131-140 days) durations. As Indian consumers predominantly prefer medium slender grain types, an additional trial (IHRT MS130±5 days) was introduced for this category in 2006 to identify such hybrid F 1 genotypes.
In some years, one or the other trial was abandoned apparently due to paucity of hybrid genotypes or their seeds in adequate quantity.
All AICRIP's experiments were conducted in randomized block design with three replications. Every test hybrid F 1 genotype was sown in nursery using 12-15 kg seeds/ha. In the irrigated ecosystem, 20-25-dayold seedlings were transplanted manually with a single seedling per hill in experimental plots, at a spacing of 20 cm between rows and 15 cm between hills. The experimental plot size varied with locations; but in most cases the minimum sub-plot size was 10 m 2 . Efforts were made at all locations to ensure crop growth of hybrid genotypes by adjusting the time of planting, and fertilizer application to suit the maturity period where a particular experiment was conducted. A basal dose of farm-yard manure at 10 t/ha was applied before ploughing the eld and fertilizers were applied at a uniform rate (115 kg N:60 kg P 2 O 5 :60 kg K 2 O/ha). The entire dose of phosphatic and potassic fertilizers together with one-half of the nitrogenous fertilizer were also applied as a basal dose in the last plough. The remaining nitrogenous fertilizer was applied in equal doses at maximum tillering and booting stages of crop growth. Appropriate recommended plant protection measures were used to control weeds, diseases and insects to prevent yield losses to the best possible extent. At the owering stage, a dose of nitrogen (5 kg N/ha) was applied. In all treatment plots, one border row was excluded and grains from the remaining plants were harvested and expressed as kg/ha at 14% moisture 7,9,11 .

Assessment of genetic gain in hybrid F 1 genotypes
The methodologies employed for the yield assessment of hybrid F 1 genotypes in the irrigated ecosystem to coincide with the rice cropping season and crop management were approximately uniform. Individual trial data was scrutinized and analyzed by AICRIP at the end of each year. From the results presented by AICRIP for each trial over a period of 32 years 5 , the mean annual yields of the top-three hybrid genotypes (T3HM) were derived across all test sites. Similarly, ILCV mean yield and experimental mean (EXPM) yield estimated year-wise over locations were also derived and saved for later analysis. The estimates of T3HM, EXPM and ILCV are presented in data set 1 (Additional le 1, Table S1, sheet 1). The grain yields produced at 102 locations by cohorts of hybrid F 1 genotypes HYC and ILCV in experiments performed under four IHRT trials between 1996 and 2019 are given in data set 2 (Additional le 1, Table S1, sheet 2). Details on geographical position, soil conditions etc on the locations used are presented in Additional le 1, Table S1, sheets 3-4. The commercially released rice hybrids were used at locations from 2006 to 2019 as hybrid check variety (HCV) depending on F 1 seed availability. Hybrid genotypes which consistently recorded more than 5% yield advantage over the HCV and 10% yield advantage over the inbred LCV were selected for commercial release. Details on the number of hybrid F 1 genotypes, locations and experiments involved in this investigation are summarized in Table 1.

Treatment of the yield data
All the trials on hybrids were made in the irrigated ecosystem and with the use of cohorts of hybrid F 1 genotypes including checks each year and adoption of a set of agronomical practices recommended by the AICRIP 5 . Yet, the characteristics of locations and varying climatic conditions may exert some in uence on the yields from year to year. To analyze the performance of yield or to identify genetic gain for yield in hybrid F 1 genotypes using available data from four different trials, it is crucial to remove the extraneous data or <noise> if any observed to the yearly variations in climatic and eld conditions in locations. The methodology used was to choose a reference year and then to correct the yields step by step, based on the values over two consecutive years in the data set.
Let Y mj be the yield of the highest j th year, j = 1,2,3,….n, and Ȳ be the grand mean.
Thus "year effect" is calculated relative to the year j+1, using the previously calculated "year effect" corrected yield value, and so on for all the years posterior to j, and identically for all the years anterior to j. In the data set 1, the year variable for T3HM, EXPM and ILCV (Additional le 1, Table S1, sheet 1) is corrected with the calculated respective grand means, Ȳm t3h , Ȳm exp and Ȳm ilcv , for these three oating checks. For the data set 2 (Additional le 1, Table S1, sheet 2), grand mean yield of locations over years, Ȳm ly was calculated to make corrections. This process was followed for each of the four IHRT trials before any data analysis. Further, grain yield performance of HCV and ILCV in experiments was also tabulated 5 . Before analysing this data, individually the Ȳm hcv and Ȳm ilcv were calculated for each hybrid rice trial, duration-wise to make corrections.
Analysis of variance of the mean grain yields derived over experiments (data set 1, Additional le 1, Table S1, sheet 1) at each of the locations was computed for each hybrid rice trial, duration-wise in the irrigated ecosystem 12,89 . Initially, the variances of mean grain yields in the T3HM, EXPM and ILCV were analyzed using the F-test. Wherever variance was homogenous, the data sets were pooled trial-wise. Mean comparisons were made for equality by t-test on the basis of analysis of variance. If variance was not homogenous, weighted means were derived. The mean grain yields of the top-three hybrid genotypes, inbred checks and hybrid breeding stock (experimental mean)that represent the hypothetical oating checks of Jensen 6 adjust to yield gains, if any, annually. Instead of considering the top-most entry alone, the mean performance of the top-3 ranking hybrid genotypes provides a better measure of changes if any, in the genetic potential of successively developed new hybrid genotypes. For each test year in different hybrid trials, the mean grain yields of the three oating checks, T3HM, EXPM and ILCV, were calculated across locations. The linear regressions were performed with mean grain yields recorded by the three oating checks over the years, to estimate the change in yields due to genetic and environmental causes. Residual plots were evaluated for occurrence of visual pattern and coe cients of determination were calculated. In whichever trial the mean grain yields showed a statistically signi cant increase or decrease, genetic gain or loss was tested. To this end, in each test year under different trials, the yield of EXPM or ILCV was deducted from the grain yield of T3HM to eliminate the environmental effect, and regression analyses were repeated to nd genetic gain or loss, if any, for grain yield. The criteria used to identify the best-t models were the signi cance of model parameters (Student's t-test), coe cients of determination (R 2 and R a 2 that adjusts for the number of explanatory terms in model relative to the number of data points), and the lowest root means square of standard error (RMSE), while meeting the assumptions of normality, independence and homogenous variance in regression analysis 90,91,92 .
To assess the performance of hybrid genotypes at locations, the number of experiments performed to test hybrid genotypes in different maturity groups in AICRIP experiments performed between 1996 and 2019 were tabulated ( Table 1). The mean grain yields of cohorts of hybrid F 1 genotypes, HCV and inbred LCV recorded in the experiments at each test location were culled and assembled in each trial for each location year-wise. ANOVA was performed 12

Availability of data and materials
All data generated or analysed in this study are included in this published article and its additional les.
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.