Heterosis is found to depend on environment and genetic background
We identified 4,625,141 SNPs in the parent panel (N = 267), after excluding SNPs with minor allele frequency (MAF) less than 5% and missing rate larger than 50%. According the neighbor-joining tree of 267 parental lines based on the above SNPs and the posterior validation errors in different number of run K in admixture, all the 267 lines could be classified into japonica and indica subspecies (Additional file 1: Figure S1). Thus, our obtained 418 hybrids include two kinds of intra-subspecific combinations, i.e., japonica×Nippponbare (J×Nip) and indica×9311 (I×9311), and two kinds of inter-subspecific combinations, i.e., japonica×9311 (J×9311) and indica×Nipponbare (I×Nip).
We phenotyped the 418 hybrids and their 267 parents in 2013 at Changsha (CS) (28° 13′ N, 112° 58′ E, a long-day environment) and Sanya (SY) (18° 10′ N, 109° 28′ E, a short-day environment) of China. Six yield-related traits were investigated, including spikelet number per panicle (SPP) and its two component traits (both primary and secondary branch numbers per panicle (PBP and SBP)), 1000-grain weight (KGW), panicle number per plant (PNP), and grain weight per plant (GWP) (Additional file 2: Table S2).
Obviously, both the phenotype of inbred parent and hybrid were influenced by the environmental condition; however, the influence of environment on hybrid was generally stronger than that of inbred parents (Fig. 1a). All six traits exhibit the same phenomenon, except for KGW of indica hybrids and PNP of japonica hybrids in Changsha and Sanya, respectively (Additional file 1: Figure S3c-d). To further investigate the environmental effect on yield traits of inbred parents and hybrids, we first evaluated the correlations of phenotypes between the two environments. The result showed that the correlation of parental phenotype between the two environments is generally much higher than that of hybrid except for PNP and GWP trait, which show a larger proportion of residual errors than the other four traits (Additional file 1: Figure S4a), implying that hybrids are more variable than inbred parents across environments. We then performed the two-way analysis of variance (ANOVA) including environment as a factor. The results indicated that the proportion of environment effect on hybrids was generally much higher than that on the corresponding inbreds, especially for SPP and its related traits (Fig. 1b and Additional file 2: Table S3). These observations also indicated that hybrids are more sensitive to the environment than their parents. Considering that there were only two environments in the current study, there may be deviations; hence, in order to ascertain whether this observation is a common phenomenon, we further analyzed the published data of 2808 maize hybrids and their parents across five environments [37], 265 maize hybrid and their parents across 4 environments [38], and 1800 wheat hybrid and their parents across 11 environments [39], the results showed that the effect of environment on hybrids was generally stronger than that of the inbred parents (Additional file 1: Figure S5). Taken these observations together, we concluded that hybrids are more sensitive to the environment than their inbred parents is a predominant phenomenon across species.
Examining the strength of heterosis in terms of different traits, combinations, and environments, the result indicated that heterosis, especially the better-parent heterosis, are just the potentiality rather than the inevitable result of hybridization, despite which is predominant over the cases (Additional file 1: Figure S6). It was clear that not all combinations showed hybrid vigor. On average, 10.65% of intra-subspecific hybrids and 10.29% of inter-subspecific hybrids even displayed hybrid weakness (Additional file 1: Figure S6a-h). The degree of middle-parent heterosis (dHmp) from all combination types, in terms of SPP-related trait and GWP, ranged from 7.87 to 70.13% with an average of 35.7% in Changsha, this was apparently higher than that in Sanya (−8.53 to 53.30% with an average of 8.21%) (Fig. 1c, d). However, for PNP and KGW, the dHmp of all combinations generally appeared to be weaker difference between the two locations (Fig. 1c, d). Particularly, the proportion of positive overdominant (POD) heterosis of all traits across the combinations in Changsha (averagely 63.19%) was much higher than that in Sanya (averagely 32.40%). And the decrease of POD from long-day to short-day environments was distinctly represented by spikelet number-related trait (PBP, SBP, and SPP), compared to the other traits (Additional file 1: Figure S6a-h). Compared to the intra-subspecific combinations across two environments, the proportion of POD heterosis of inter-subspecific combinations for all traits was apparently higher than that of intra-subspecific combinations in Changsha, except for the PNP trait with consistent heterosis across the two environments (Fig. 1c and Additional file 1: Figure S6a-d). In contrast, for SPP-related traits and GWP, the proportion of POD heterosis of inter-subspecific combinations was even lower than that of intra-subspecific combinations in Sanya (Fig. 1d; Additional file 1: Figure S6e-h). These lines of evidence indicated that the degree of heterosis is apparently dependent on environments, traits, and combinations. Thus, it is of significance to uncover the genetic basis and mechanism underlying inbreds, hybrids, and especially heterosis so as to highlight the opportunity to produce strong heterosis and elite hybrids.
Genome-wide identification of QTLs affecting yield traits of rice hybrids
We carried out genome-wide association studies (GWAS), using 120 sets of genetic and phenotypic data. The phenotypic data consists of three types of datum panels evaluated for six yield traits (PBP, SBP, SPP, KGW, PNP, and GWP) under two environments (Changsha and Sanya). The three types of datum panels include 20 sets of data for each of the traits, i.e., (1) four sets from parents (both japonica and indica in two environments), (2) eight sets from F1 (four types of combinations in two environments), and (3) eight sets of calculated middle-parent heterosis value (Hmp) (four types of combinations in two environments) (see “Materials and methods”). The Manhattan plot of the results of association analysis is shown in Additional file 1: Figure S7 to S18.
Totally, we identified 635 and 624 QTLs in Changsha and Sanya, respectively, from the parental datum panel (P_QTL), 828 and 895 QTLs from the F1 datum panel (F1_QTL), and 636 and 818 QTLs from the Hmp datum panel (Hmp_QTL) (Additional file 2: Table S4). When comparing the two environments, P_QTLs appeared apparently to be more environment-stable (proportion of shared QTLs, 38.4% on average), than F1_QTLs (9.8% on average) and Hmp_QTLs (6.6% on average), regarding the traits related to grain number (PBP, SBP, and SPP) and grain size (KGW). As for PNP, the situation is combination-dependent. The three panels of QTLs related to grain weight per plant (GWP) were rather environment-specific (Additional file 1: Figure S19).
Comparing the shared QTLs from the three panels (P_QTL, F1_QTL, and Hmp_QTL), we found that the genetic architecture affecting hybrids synchronizes more with that impacting heterosis (24.09±21%), compared to that affecting inbred parent (12.28±10%), but there were some exceptions for some combinations, environments, and traits (Additional file 1: Figure S20). The situation with more colocalized F1_QTL and P_QTL than colocalized F1_QTL and Hmp_QTL was more in Sanya than Changsha, more for spikelet number than PNP and GWP, more for 9311 combinations than Nipponbare ones. These results implied that the improvement of hybrids should concern both heterosis and the genetic background of the inbred lines, but their respective contribution varied depending on the combinations, environments, and traits.
Nonadditive-preferred QTLs, which are more variable than additive-preferred ones, are the main contributors to heterosis
As expected according to the quantitative genetic theory, we can detect the QTLs showing significant additive effect in the parental datum panel, nonadditive effect in the Hmp datum panel, and both additive and nonadditive effects in the F1 datum panel. The above results indicated that one QTL can be detected based on different types of genetic effects (additive or nonadditive) in different types of panels by GWAS. Therefore, in order to estimate the relative degree of additive and nonadditive effect of each QTL and throw light on the understanding of the genetic basis underlying heterosis, we estimated the additive effect (a) and dominance effect (d) of each QTL on six yield traits using the parental, F1 and Hmp datum panels (see “Materials and methods” for detail). A QTL is referred as overdominance preferred if the absolute ratio of dominant effect to additive effects (|d/a|, degree of dominance) is no less than 1.5, and dominance preferred if 0.5≤|d/a|<1.5 (including partial dominance), and additive preferred if |d/a|<0.5 (see “Materials and methods” for detail). We called those QTLs being dominance preferred or overdominance preferred as nonadditive QTLs.
Among the 44 scenarios (five traits of four types of combinations under the two environments, plus GWP of two types of intra-subspecific combinations under the two environments), both F1_QTLs and Hmp_QTLs showed apparently more nonadditive effects than P_QTLs (Additional file 1: Figure S21 and Additional file 2: Table S5), except for primary branch number per panicle of J×9311 in Sanya. Particularly, the majority of F1_QTLs and Hmp_QTLs displayed overdominant effects (69.27% and 77.71%, respectively), with only a small portion of QTL represented additive effect (10.66% and 7.55%, respectively). Conversely, the majority of P_QTLs demonstrated additive (44.16%) and dominant (37.08%) effects, and only a small proportion (18.74%) showed overdominance. Consistent with the observation that SPP-related trait (PBP, SBP, and SPP) did not exhibit obvious heterosis in Sanya (Fig. 1d), fewer overdominant F1_QTLs and Hmp_QTLs were identified in Sanya than that in Changsha. On average, 75.7% of the F1_QTLs identified in Changsha expressed as overdominant for the SPP-related traits, while the proportion significantly reduced to 42.6% in Sanya (Additional file 1: Figure S21a-c). Comparing the two subspecies, we found that the reduction is more remarkable in japonica hybrids (from 71.0% in Changsha to 22.3% in Sanya) than that in indica hybrids (from 80.4% in Changsha to 62.8% in Sanya) (Additional file 1: Figure S21a-c). These observations indicated that the effects of environment on SPP-related trait in japonica hybrid was stronger than that in indica hybrid. When regarding the trait of KGW and PNP, the proportion of overdominance identified in F1_QTLs and Hmp_QTLs did not show such consistent changes between subspecies and between environments, and varying by combinations (Additional file 1: Figure S21d-e).
To compare the response of different genetic components to the environment and their genetic background, comparative analysis from the levels of environmental and genetic background were conducted. First, we examined the environmental stability among the QTLs of additive, dominant, and overdominant ones, the result showed that a larger proportion of additive QTLs showed environment-stable than nonadditive QTLs, regarding most types of combinations for all traits except for PNP. Meanwhile, a higher proportion of dominant QTLs showed environment-stable than overdominant QTLs for most of the combinations and traits (Additional file 1: Figure S22). These results indicated that the higher magnitude of the dominant effect of a QTL, the stronger its environmental sensitivity. The distinctively larger proportion of unstable factors including overdominant and dominant QTLs identified in hybrids or heterosis than that in inbreds, consistent with the fact that the response of hybrid to the environment was generally stronger than that of inbreds (Fig. 1b and Additional file 1: Figure S4 and S5).
Second, in order to compare the stability of the genetic effect for each QTLs with additive, dominant, and overdominant under different genetic background (different combinations), we estimated the phenotypic coefficient of variation of the same genotype under different genetic backgrounds of individuals in the tested population, the result showed that homozygous genotypes of QTLs with nonadditive effects exhibited higher variability, as compared to those with additive effects (Additional file 1: Figure S23). Examining the variance of QTLs identified in an immortalized F2 population, in which the frequency of each genotype was more balanced, a similar phenomenon was observed (Additional file 1: Figure S24) [31]. Thus, these observations implied that the higher degree of dominance exhibited stronger background response and variability for most of the QTLs.
Nonadditive expressed genes are highly dependent on the expression of their upstream transcription factors
As mentioned in introduction and above, the heterosis, dominant and overdominant phenomenon at the phenotype level are often resulted from the integrated effects of multi-factors at various intermediate and fundamental levels (such as different genes, QTLs, gene expression, and physiological traits), thus it is challenging to investigate the molecular mechanism of heterosis at the phenotypic level. Transcription is such an intermediate step for a gene to perform its functions in development of complex phenotypes. Therefore, it is informative to explore gene expression patterns between parents and their F1 hybrid, in order to understand molecular mechanisms underlying heterosis. Here we investigated transcriptome profile of young panicles from the hybrid LYP9 and its two parents PA64S and 9311. As a whole, 8248 genes showed differential expressions between the two parents and their F1 hybrid in at least one of three tissues (1 mm, 2 mm, 3 mm young panicles). Expression patterns can be classified as additive preferred (A) (13%), dominant preferred (D) (39%), and overdominant preferred (OD) (48%) in at least one of three tissues (Fig. 2a, Additional file 2: Table S6). When compared the stability of genes with additive and nonadditive expression, we found that the dominant and overdominant expression showed dramatically more variability across tissues than the additive expression (Fig. 2b); in another words, nonadditive or heterosis of expression is more tissue-specific and may be more background-dependent. This is consistent with the results mentioned above, where hybrids are more variable than the inbred and nonadditive QTLs are more variable than additive QTLs.
We investigated the possible direct relations between nonadditive effect and genetic background by analyzing the correlation of expression levels between the target genes and their direct background, i.e., the transcription factors. The results indicated that the expression of genes with dominant and overdominant effects represented apparently stronger dependency on their upstream transcription factors than those genes with additive effects (Fig. 2c). We also analyzed the transcriptomes of three Arabidopsis thaliana combinations and observed the same phenomenon (Additional file 1: Figure S25) [12]. Thus, these results indicated an important phenomenon that expression of genes with nonadditive effects are more sensitive to the dosage changes of their upstream genetic backgrounds.
One core molecular mechanism of dominance and overdominance—homo-insufficiency under insufficient background (HoIIB)
Does the genetic background dependency of nonadditive effects represent an essential molecular mechanism underlying heterosis? It is well known that no factor is absolutely independent in the biology system and that ligand-receptor binding, including the binding of transcription factor to a target gene, is obviously the most common dependent relationship between molecules, where the ligand and receptor can be the genetic background of one another and their binding reaction is described by the Hill equation [41]. In order to investigate the possible internal relationship between genetic background and the occurrence of dominance and overdominance, simulated genetic effects of one polymorphic site of one receptor were compared among the diploid parents and their F1, according to the Hill equation with different ligand concentrations as the background. We here considered the following three major regulation scenarios with the assumption that the ligand concentration is consistent among parents and their F1 for simplicity (see Additional file 3: Simulation1 for details).
Scenario 1: Null allele vs one functional allele of one polymorphic site under one genetic background, that is, one of two alleles of one polymorphic site of the receptor is loss-function and the other allele can be bound by one ligand as the background of the receptor (Additional file 1: Figure S26).
For the positive regulation, when the activator as the background is insufficient (smaller X/K) for the functional allele of the receptor, the receptor will express as positive (partial-) dominance. In contrast, when the background is sufficient (larger X/K), the receptor will express as additive effect (Fig. 3a). Apparently, it is the insufficient ligand background that can only activate partial function of two homo-alleles in parents, but relatively full function of one allele in the F1, which results in the positive (partial-) dominance. For the negative regulation, the performance is similar, but the receptor expresses as negative (partial-) dominance, when the insufficient ligand background can only suppress partial function of two homo-alleles in parents, instead relatively full function of one allele in the F1 (Additional file 1: Figure S27). It is common between positive and negative regulations that the reaction is dramatically more sensitive to the ligand (activator or repressor) concentration change under insufficient ligand background, where the (partial-) dominance is observed easily. It should be noted that there is no overdominance for this scenario, if no synergistic effect were involved (when n is equal to 1).
Scenario 2: Two alleles of one polymorphic site under two independent backgrounds, that is, two alleles of one polymorphic site of the receptor, can be bound by two respective and independent ligands as the backgrounds of the receptor (Additional file 1: Figure S28).
For the positive regulation, as expected in Scenario 1, the receptor easily appears positive dominance when the activator background for the allele with the larger maximum function of the receptor is insufficient (smaller X/K). As different from Scenario 1, we can also observe positive overdominance under Scenario 2, when the receptor in F1 can cumulate the effect from the (partial-) dominant allele with a larger function and that from the other allele with a smaller function. When both backgrounds of two alleles are sufficient (higher X/K), the receptor in both parents and F1 can express the full function as two alleles and one allele, respectively, as a result the receptor expresses as additive (Fig. 3b and Additional file 1: Figure S29). The performances of negative regulation are similar, but the receptor expresses as negative (partial-) dominance or overdominance under insufficient background (Additional file 1: Figure S30). It is common between positive and negative regulations that the reaction is dramatically more sensitive to the ligand (activator or repressor) concentration change under insufficient ligand background (smaller X/K), where the nonadditive effect is easy to be observed.
Scenario 3: Two alleles of one polymorphic site with shared background, that is, two alleles of one polymorphic site of the receptor can be bound by the same ligand as the background of the receptor (Additional file 1: Figure S31). But these two alleles may have different affinities (K) to the ligand and show different maximum functions (μ). Thus, we considered two situations: (1) One allele has higher affinity and shows a larger maximum function, and the other has lower affinity and shows a smaller maximum function (abbreviated as HALF/LASF) (Fig. 3c and Additional file 1: Figure S32). (2) One allele has higher affinity, but shows smaller a maximum function, and the other has lower affinity, but shows a larger maximum function (abbreviated as HASF/LALF) (Fig. 3d and Additional file 1: Figure S33). Before considering the above two situations, we found from the simulation that there is only additive effect if the ligand randomly and equally binds to two alleles (see Additional file 3: Simulaiton1). In spite of positive or negative regulations (Additional file 1: Figure S34 and S35), the performance is similar to scenario 2 that the reaction tends to appear nonadditive under insufficient ligand background, especially for the allele with a larger maximum function. The insufficient ligand background renders the reaction dramatically more sensitive to the ligand (activator or repressor) concentration, compared to the sufficient ligand background. But we can only observe the nonadditive effect, when the background is dramatically insufficient under HALF/LASF (Additional file 1: Figure S33). In addition, the degree of nonadditive effect is apparent weaker in the HALF/LASF situation, compared to the HASF/LALF, because in the latter situation the background in F1 can be reallocated to the allele with LALF from the allele with HASF when the background for the latter has been saturated (Additional file 1: Figure S32). Taken together, we suppose that overdominance results from the cumulation or compensation between the (partial-) dominance of the allele with a larger function and the effect of the other allele with a smaller function.
According to the above simulations, we put forward one model that explains a core molecular mechanism underlying the nonadditive effects and heterosis: homo-insufficiency under insufficient background (HoIIB) (Fig. 3e). As indicated by HoIIB, it is the genetic background insufficient to maximize the function of two homo-alleles in parents, but relatively or even completely sufficient to maximize the function of one allele in F1, thus resulting in the insufficient function of two homo-alleles in parents, but the relatively or completely sufficient function of one allele in F1, that renders the target locus nonadditive in effect, as contributing to heterosis. And there were three main features of this theoretic model, according to the simulation. First, the background insufficiency for the allele with a larger function is the driving force for nonadditive effects. What we see dominance and heterosis is not the consequence of a stronger heterozygous, but the consequence of the weakened parent with homo-allele of larger function. In other word, the observable function of two homo-alleles is lower than their maximum function due to insufficient background. Second, if there is no synergy (n = 1), the overdominance can only be found when both alleles are functional, which result from the cumulation or complementation between the (partial-) dominance of the allele with a larger function and the effect of the other allele with a smaller function (Additional file 1: Figure S33). Third, we observed one general phenomenon in the three scenarios mentioned above, that is, the reaction is dramatically more sensitive to the ligand (activator or repressor) concentration under insufficient ligand background, where the nonadditive effect is easy to be observed.
The HoIIB model was supported by different levels of evidence
It is intriguing that in the observed experiments we have found extensive evidence that can represent the three features of the HoIIB model mentioned above. First, we observed the homo-insufficiency of the allele with a large function and the cumulation or complementation from the allele with a smaller function at various levels including transcription, QTL, and traits. Using transcriptome profile from the 1, 2, and 3 mm young panicles of 9311, PA64S, and their hybrids (LYP9), we investigated expression levels in the two parents for those genes with additive, positive dominant, and positive overdominant effects, respectively. The homo-insufficient expression was substantially observed in the higher parent for genes with dominant and overdominant transcription, compared to those with additive transcription (Fig. 4a and Additional file 1: Figure S36). Meanwhile, the homozygous genotypes in lower parent showed increased expression for the positive overdominance in most cases. Then we compared the QTL with different types of genetic effects that were identified by our GWAS of the three main yield components (SPP, KGW, and PNP). Apparently, the parents of genotypes with lager effects of the dominant and overdominant QTLs represented decreased phenotype, compared to those of the additive QTLs (Fig. 4b and Additional file 1: Figure S37 to S39). We also compared the QTLs that were identified by the 278 immortal F2 lines from the crosses between randomly selected RILs derived from Minghui 63 and Zhenshan 97 [31]. The four yield traits showed apparent HoIIB phenomenon for the dominant and overdominant QTLs, that is, the genotype with higher effect for dominant and overdominant QTLs represented decreased effect, compared to the additive QTLs (Additional file 1: Figure S40). We further investigated the distribution of the degrees of middle-parent heterosis for the five yield traits (SPP, PBP, SBP, PNP, and KGW) among the MCC combinations evaluated under the two environments. The stronger heterosis tended to be found among the combinations whose higher parents show decreased phenotypes (Fig. 4c and Additional file 1: Figure S41). Secondary, as predicted in HoIIB model, overdominance is more likely to occur when both alleles are functional. Examining the occurrence of overdominance in the transcriptomes of rice hybrid 9311×PA64S and Arabidopsis hybrid Col×C24, the results showed that overdominance was more frequently observed when both parental alleles were functional than that when only one parental allele was functional (Fig. 4d and Additional file 1: Figure S42). Thirdly, the HoIIB model implied that the expression or the observable function of those genes with stronger heterosis are subject to more serious homo-insufficiency background and thus will show a stronger response to the change of background, compared to those with weaker heterosis. This explains previous observation that the coefficient of variation of the QTLs identified in the MCC hybrid or by the immortalized F2 mapping panel showed that both homozygous and heterozygous genotypes of QTLs with (over-) dominant effects exhibited higher variability [31], compared to those with additive effects (Additional file 1: Figure S23 and S24). At the expression level, the instability of the genes with (over-) dominant expressions was reflected by their higher variance of expression across the 3 tissues, compared to the genes with additive effects (Additional file 1: Figure S43). Phenotypically, the combinations with higher degree of dominance also showed higher variability for most of the traits (Additional file 1: Figure S44).
The HoIIB model was experimentally validated in yeast
To validate the HoIIB model, we designed an experiment to see whether we can manipulate the performance of heterosis of one gene by changing its background sufficiency within a living organism. In order to reduce the experimental complexity as much as possible, we used the transcription level as the performance indicator (phenotype) of the target gene and the transcription factor as its background, and carried out the experiment in the simple diploid organism, yeast. We screened the reported transcription factors and its target genes in yeast according to the following criteria: (1) the promoter region being bound by a transcription factor has been clearly validated; (2) there is strong and simple regulatory relationship between the transcription factor and its target gene. After investigating the co-expression of six pairs of genes (WAR1 vs PDR12, VHR1 vs VHT1, VHR1 vs BIO5, AZF1 vs CLN3, AFT1 vs FIT3 and FZF1 vs SSU1), we found that SSU1 showed a strong co-expression with its transcription factor FZF1 in strain BY4743 of Saccharomyces cerevisiae (R2 = 0.88, see Additional file 2: Table S7 and Table S8). So we selected FZF1 and its target gene SSU1. According to the reported binding features between two genes [42], we knocked out the FZF1 recognition motif 5′-CGTATCGTATAAGGCAACAATAG-3′ in SSU1 promoter region, then constructed the heterozygous (SSU1/ssu1) and homozygous (ssu1/ssu1) knockout strain of SSU1 in BY4743 (Additional file 1: Figure S45a-c). The ssu1/ssu1 genotype showed apparently decreased expression compared to wild genotype of SSU1 (SSU1/SSU1), indicating the effective mutation. Apparently, the SSU1/ssu1 genotype showed obvious nonadditive expression (d/a = 1.899) in the system comprising genotypes SSU1/SSU1, SSU1/ssu1 and ssu1/ssu1 under normal FZF1 background (Fig. 4e and Additional file 2: Table S9), implying that FZF1 supply the insufficient background to SSU1 in BY4743 according to our HoIIB model. We may expect that we can decrease the dominance degree of SSU1 if we can regulate up the expression of its background FZF1. In the strains with native or overexpressed FZF1, we investigated the transcription level (representing the phenotype) of genotypes SSU1/SSU1, SSU1/ssu1 and ssu1/ssu1, and the transcription level of FZF1 (representing the background sufficiency). We really observed dramatically decreased dominance degree of SSU1 along with the increasing of background, i.e., expression level of FZF1, and SSU1 even nearly transited into additive expression when the expression of FZF1 upregulated more than 10 folds (Fig. 4e, f). The results can be confirmed by a repeat experiment (Additional file 1: Figure S45d-e and Additional file 2: Table S10).
To validate our HoIIB model at a more complex phenotype level, we tried to investigate the effect of SSU1 on growth rate of yeast. Given that SSU1 may regulate the growth rate of yeast in the sulfur environment [43], we first examined the relationship between the transcription level of SSU1 and yeast maximum growth rate in the liquid medium. It showed that the transcription level of SSU1 was linearly correlated with the maximum growth rate of yeast, but when the transcription of SSU1 was upregulated higher than 10 folds, the growth rate decreased to some extent. This indicates that high expression may not be beneficial to growth, and the relationship between them is complicated [44] (Additional file 1: Figure S46a-f). We thus selected the events with SSU1 expression upregulated less than 10 folds to validate the HoIIB model. The HoIIB phenomenon was observed in all conditions including normal, 2 mM and 4 mM K2S2O4-treated SD-Ura medium (Additional file 1: Figure S46g-h and Additional file 2: Table S11). Thus, our designed experiments using both transcription and growth rate as the phenotype indicators indicated that the dominance degree of downstream genes can be manipulated by changing the level of background sufficiency.
The systematic HoIIB phenomenon related to rice yield heterosis
The model and the results mentioned above revealed that insufficient background contributing to the homo-insufficiency is not only the limiting factor for (over-) dominant loci to reach their maximum function, but also the one that causes the instability of the target genes. Therefore, identification of (over-) dominant loci will provide us with a start point or hint to discover the key limiting factors along the genome, or gene regulatory network that impacts such important traits as yield, and thus guide the improvement of hybrids.
In order to investigate the possible systematic HoIIB factors impacting rice yield heterosis, we firstly compared the MCC QTLs identified from different combinations and environments (Additional file 2: Table S12), followed by gene set enrichment analysis using the candidate genes repeatedly identified by GWAS (Additional file 1: Figure S47). Results showed that the Nipponbare combinations have apparently more colocalized nonadditive QTLs than did the 9311 combinations, consistent with the fact that Nipponbare is less productive than 9311 and suggesting that Nipponbare may represent a more constrained background and thus easily result in nonadditive effect in its F1 hybrids compared to 9311. Regarding different subspecific combinations, negative overdominant QTLs were identified more frequently in indica combinations for traits related to SPP and PNP but in japonica ones for KGW; however, negative dominant and positive nonadditive QTLs tended to be detected in japonica combinations for all traits. Regarding different environments, the colocalized nonadditive QTLs tended to be detected in Sanya compared to Changsha (Additional file 2: Table S12). These results indicated that the HoIIB appeared to be taxa- and environment-systematic to some degree, but mainly determined by two specific parents in the combination investigated. Secondly, the GO enrichment indicated that those genes within additive QTLs seldom show enrichment, but those genes within nonadditive (dominant and overdominant) QTLs are frequently involved in many kinds of catalytic activities and binding functions (Additional file 1: Figure S48 and Additional file 2: Table S13). Compared to those genes with nonadditive performance in nonlethal deletion yeast strains grown in five different media [43], we also found that they enriched in the GO terms of catalytic activity (Additional file 1: Figure S49). The enrichment in catalytic activity for nonadditive genes may be explained by the reports that most enzymes in organism usually operate at an unsaturated substrate concentration [44], i.e., at the lower level of substrates, which may result in the insufficient background of these enzymes and thus their nonadditive performance. Further checking those genes encoding rate-limiting enzymes (RLE) showed that the proportion of RLE genes in nonadditive QTLs was generally higher than that in additive QTLs (Additional file 1: Figure S50), consistent with the fact that most of the RLEs usually contain the distinctly larger Kcat values compared to the available concentration of substrate [45]. These results suggested that the background/substrate of RLE may be the kind of important limiting factors that confer the systematically enriched catalytic activity in the pathway of these RLE genes. Thus, identifying and improving these limiting factors may provide the chance to make breakthroughs in future breeding of both inbreds and hybrids.