The mean values for seven quantitative characters of six generations of the three crosses 863B x P 7-7 (1), 843B x P 7-4 (2), and  81B x ICMP 451 (3)  in two environments, Agriculture Research Farm of Banaras Hindu University under irrigated conditions (E₁) and Rajiv Gandhi South Campus under rainfed condition (E₂) are presented in Table 1. In general, the performance of the three crosses of various characters over different generations was better in Varanasi (E₁₎ than in Mirzapur (E₂)_. The poor performance in Mirzapur (E₂₎ was mainly due to poor fertility and low water holding capacity of the soil. All these characters exhibited reduced expression under water stress; however, this was less for such characters as plant height and ear length. These observations confirm earlier findings of Govindaraj et al. (2010); Van Oosterom et al. (2006); Singh and Sagar (1989 & 2001); Soomro et al. (2008). The F₁s of all the crosses performed better than both the parents and mid-parents in both the environments for all the characters, except for days to maturity in crosses 843B x P 7-4 and 81B x ICMP 451 in Varanasi. This indicated prevalence of heterobeltosis, which could arise due to true over-dominance or dispersion of completely or incompletely dominant genes. Negative heterosis for days to maturity indicated dominance for earliness. Higher heterosis in case of pearl millet, an allogamous species, is expected. Virk (1986) has noted positive heterosis for quantitative characters including grain yield (-56.62 to 424.16%), but negative heterosis for days to flowering. The better performance of F₁ than both the parents for grain yield and other attributes even under moisture stress shows that hybrids will withstand moisture stress. The F₂ means, lesser than the F₁ in both the environments for all the characters and crosses except for ear length for cross 843B x P 7-4 in Varanasi indicated high amount of inbreeding depression. Smaller F₂ mean than F₁ could be due to elimination of dominance effects, as in all the crosses the dominant exceeded the additive component and may be responsible for bringing reduction in F₂ mean. The F₂ mean exceeding all other generations for ear length could be due to transgression and fixable epistatic effects. Both the backcrosses exceeded the respective recurrent parents in most of the cases, indicating the prevalence of allelic and non-allelic interactions for genetic control of important traits in pearl millet. Govindaraj et al. (2010); Gupta and Phul (1981); Girgla et al. (1985); Singh and Sagar (2001) also reported similar results.
Table (1):
Mean performance of six generations of seven quantitative traits in two environments.

To study the genetics and gene interactions, the data was subjected to three parameter model of Cavalli (1952) and later six parameter model of Jinks and Jones (1958). The estimates of six parameter model along with their significance have been presented in Table 2. The analysis revealed the inadequacy of additive-dominance model for all the seven traits studied. Some of the traits such as ear weight and grain yield per plant for cross 843B x P 7-4 indicated significant X² values. This analysis lead to the conclusion that variability present in the material could not be ascribed to only additive and dominance effects of the gene effects, albeit the epistatic gene effects also played an important role in genetic control of the characters studied. The present study also revealed that only additive effect was not sufficient and significant to explain the variability for all the traits. Further, significant dominance effects were important for only ear weight per plant for cross 81B x ICMP 451 in Varanasi environment. In general preponderance of dominant variation was observed for all the traits in both the environments and all the three crosses. Similar observations were made in Pearl Millet by Virk (1986); Singh and Sagar (1989); Govindaraj et al. (2010). The comparative analysis of simple vs. higher order of gene integrations indicated that while additive effect were less effective than the dominant effects when compared with their corresponding digenic or higher order of integrations. This indicated that the expression of the additive effects was affected at the cost of their digenic interactions. In few cases additive and additive x additive effects was important such as fodder yield in case cross 843B x P 7-4; and 81B x ICMP 451 in Varanasi environment. This suggests that fixable nature of variation could be exploited by simple selection for these specific characters and crosses. Additive x dominance type was the other lone significant digenic integrations for number of effective tillers per plant for cross 863B x P 7-7 in both the environments. The dominance x dominance type of digenic interactions was high and significant in a number of cases. This type of epistasis advocate for searching desirable specific cross combinations. In general additive x additive and additive x dominance integrations were observed for grain yield in cross 863B x P 7-7 in Varanasi environment, while three and four parameter combinations of simple and digenic interactions were found to be significant in a number of cases, but all the five parameters exhibited significance in case of ear weight for the cross 843B x P 7-4 in Varanasi environment. Days to maturity for cross 863B x P 7-7 and 843B x P 7-4 in Varanasi environment, ear length for cross 863B x P 7-7 in Varanasi environment, plant height for crosses 843B x P 7-4 in Mirzapu environment exhibited siginficance for four parameter. Even the six parameter model failed to detect the significance of any of five genetic parameters for tillers per plant for crosses 843B x P 7-4 and 81B x ICMP 451 in Varanasi Environment and Mirzapur environment, respectively. While the three parameter model was found to be inadequate in both the environments. The discrepancy for the significance of parameters studied in 3 and 6 parameters model may be due to non-orthognaility nature of the estimates in the former model. The present study indicates that dominance and epistatic effects were important in explaining the variability in the present material. The present finding supports the earlier findings of Singh and Sagar (2001) and Singh et al (1990a) but contradicts the findings of Grafius (1959). The Grafius observed that non-allelic interactions prevail for yield but not for its components. The digenic and trigenic epistasis and linked digenic integrations have been observed by Gupta and Phul (1981) in Peral millet.  The study revealed that characters such as plant height, ear length, ear weight, days to maturity, and dry fodder yield are complex traits as controlled by number of parameters significant in the 3 crosses in two environments.  As reported earlier grain yield and tiller number are also complex traits, the significance of only one or two parameters in present materials are probably due to non availability of enough variability for these traits in present genetic materials. The study indicated that dominance (h) and dominance x dominance (l) interactions accounted for high proportions of total genetic variability and even increased under moisture stress conditions in most of the cases. This indicated that hybrids will perform better even under moisture stress conditions. This is in conformity with earlier findings such as, Singh and Sagar (2001); Govindaraj et al. (2010). It is suggested that cyclic breeding particularly reciprocal recurrent selections should be practised to improve pearl millet rather than going only for simple selections methods.
Table (2):
Gene effect of six-parameter model for seven yield and component traits of Pearl Millet in two environments.

* Significant at 5 per cent level
** Significant at 1 per cent level
D Duplicate
C Complementary epistasis

The estimates of components of genetics variance viz., additive (D), dominance (H), environmental (E) are given in Table 3. The dominance variance exceeded the additive variance indicating thereby preponderance of dominance gene action for all the characters of all the cross in both the environments except for tillers per plant for crosses 863B x P 7-7 in Mirzapur environment and cross 843B x P 7-4 in Varanasi environment, for ear length for cross 843B x P 7-4 in Mirzapur environment, days to maturity for cross 843B x P 7-4 in Varanasi environment and ear weight for cross 863B x P 7-7 in Mirzapur environment. This concedes the nature of heterobeltiosis observed from the generations mean analysis. The magnitude of dominance gene effects was comparably higher than additive gene effect in 3 as well as 6 parameter model. The preponderance of dominant gene effects in pearl millet has been observed by Singh and Sagar (1989); Virk (1986); Sheoran et al. (2000); Govindaraj et al. (2010). In general importance of dominance effect was observed in the present materials except in some cases like fodder yield in cross 843B x P 7-4, grain yield in cross 81B x ICMP 451, where additive and additive x additive effects were important and this indicated that variation in these crosses were fixable nature and simple selection can improve these characters of particular crosses
Table (3):
Estimates of components of genetic variance for 7 characters in pearl millet.

The heritability values broad and narrow sense along with their genetic advance have been presented in Table 4. In general, both the type of heritability was low in Varanasi than in Mirzapur environment. Broad sense heritability estimates were generally high for all the characters for all the three crosses in both the environments. Low narrow sense heritability estimates support the prevalence of epistatic effects as also noted from the generations means. The high estimates of heritability in Mirzapur (rainfed) environment are encouraging and offer greater scope of selection in that environment, but further studies are necessary in this direction. The high estimates under water stress may be due to either narrowing down of the variance under stress, as is usually observed and has been reported by Johnson and Frey (1967); Moeljopaviro and Ikehashi (1981); Govindaraj et al. (2010);  Larik et al. (1997 & 2000); Singh and Sagar (1989 & 2001); Soomro et al. (2008). Estimates of expected genetic advance expressed as percentage of mean were low to medium for all the characters including grain yield. But very high gain was obtained for number of effective tillers (191.36%) for cross 863B x P7-7 and high estimates for gain yield (55.24%) for cross 843B x P7-4 in Mirzapur environment. It suggested greater scope for improvement of these characters.

Table (4):
Estimates of heritability and genetic variance, for quantitative characters in pearl millet.

In general the study suggest that the sufficient  variability  exist in the present pearl millet breeding materials with  preponderance of dominant simple and higher order of gene action effects (dominance x dominance  and dominance x additive) with high heritability of important traits. This suggest cyclic breeding with reciprocal recurrent selections should be adopted to improve pearl millet rather than going only for simple selections methods.