The two populations were derived from crosses sharing the same female parent: (Z9 × Leda)–2, a breeding line developed at IVIA, where ‘Z9’ is IRRI’s (International Rice Research Institute) breeding line IR66155-2-1-1-2 and ‘Leda’ is a variety obtained in our IVIA Department. Populations are named following the male parent: J is (Z9 × Leda)–2 × JSendra (where ‘JSendra’ is a variety obtained in our Department); and population MS is (Z9 × Leda)–2 × (Marjal × Senia)–16B (where the male parent is also a breeding line from IVIA).

The F₁ from each cross was selfed to obtain the F_2. In 2004, during the regular growing season in Valencia, more than 1450 plants of each F₂ population were transplanted to concrete basins in rows, under a high plant density: 15 cm between rows and 10.5 cm between plants. A minimum of 1000 plants per population were transplanted to the field as well, but at a wider spacing of 50 cm between rows and 15 cm between plants. Within these populations, only 150 plants from each F₂ could be evaluated (as they were used for a QTL study as well): 75 in the field and 75 in basins. In both cases, the soil was a clay-loam, with typical cultural practices for the region, but the F₂ populations were transplanted to separate fields and basins (Fig. 1). Plants from variety ‘Senia’ (also from IVIA) were used as a check in the four combinations population/environment, to estimate the environmental variance of each combination, because this variety is one of the checks used in our breeding program, and an ancestor of both populations.

Each of the 150 F₂ plants per cross gave rise to an F₃ family by natural selfing. In 2005, 30 plants per line (family) of all the 300 F₃, plus the parents, were grown in the field, in rows, spaced at 15 cm × 50 cm. Every 25 lines, a row of ‘Senia’ was inserted. The seed of 6 random plants per F₃ line was collected; the selected F₄ were grown the following year in the same field conditions.

In each F₂ plant, and in 6 plants/line in F₃, F₄, checks and parents, the following traits were evaluated: days to maturity (MAT) as the number of days from seeding to 95% of mature grains; plant height (PH) to top of the panicle; panicle length (PL) and culm diameter (CD) at the second basal internode, measured in 5 culms/plant; average stem length (SL = PH – PL); growth habit (GH), visually scored with a scale of 1 (erect) to 6 (prostrated); panicle number (PN); total panicle weight (TW); average panicle weight (PW = TW/PN).

In the F₃, days to maturity and growth habit were scored only once per row, since families were homogeneous. Flag leaf length (FL, in 4 culms/plant), 100 grains’ weight and grain measurements were only evaluated in the F₂.

Grain measurements were determined in 30 dehusked (whole) grains per plant: length (GL), width (GW) and their ratio (GS) were scored with an image analyser associated to Inspector programme (from Matrox); percent of grains with a white core (WP) and its mean size (WS) were visually scored; WS was calculated with the following index: (3 × Nº of grains with big cores + 2 × Nº of grains with medium cores + Nº of grains with small cores) / Total Nº of white-cored grains.

Lodging resistance traits were measured only in F₃ and F₄: pulling resistance (PR) as is described in Torró et al. (2011); second basal internode length (IL) of two culms per plant (because shorter internodes there could increase lodging resistance).

Analyses of trait distributions and means, as well as ANOVA and correlation between traits, were performed in all populations, generations and genotype/environment combination, with the statistical packages: Statgraphics Plus 5.0® (Statistical Graphics corp. 1994-2000) and SAS 9.1 (SAS Inst. Inc., Cary, NC, USA). The ANOVAs made for comparison between environments and between populations had one factor (between groups) and the error term (within groups). Some traits were transformed to standardize their distribution: PN and PR were transformed by the logarithm, CD in F3 was transformed by the inverse (1/x), and WP by the arc sin of its square root (Bliss’ transformation, to compress the tails of this typically binomial distribution). As mentioned, MAT and GH in F₃ had only one score per line, and consequently narrow sense heritability was not estimated. For multivariate analyses (principal components analysis), only the traits not correlated (data not shown) were considered (for example, we did not include stem length, as it is highly correlated with plant height).

Heritabilities were calculated by analyses of variance. Using the notation of population genetics, Vp (phenotypic variance of a trait in a population) = Vg + Ve, where Vg is its genetic variance and Ve its environmental variance. The broad-sense heritability is the genetic proportion of its total variation, and is thus estimated as: H² = Vg/Vp.

In F_2s, only H² could be calculated: Vg = Vpl – Ve, where Vpl is the variance between population plants and Ve (environmental variance) is the variance between homozygote check plants. In traits with more than one measurement per plant, the analysis is one case of the general ANOVA (Table 1).

There is only one family in an F2: f = 1.

Ve = Ve1 + Ve2, where Ve1 is the variance of plant means (σ²pl) in a homozygous check. Ve2 is the variance within plants (σ²).

Vg = σ²pl – Ve1.

The error of H² was estimated as the significance of Vp/Ve (Kearsey & Pooney, 1996).

From the analyses of variance of F₃, both H² and narrow-sense heritability (h²) can be calculated:

Vg = Va + Vd, where Va is the additive genetic variance (the variation due to gene effects fixable in homozygosis) and Vd is the dominant genetic variance.

h² = Va/Vp.

Using Mather’s notation of self-compatible crosses, Va = (1/2)D and Vd=(1/4)H, where D represents the contribution of additive genetic effects and H represents dominance contributions.

The ANOVA is another case of Table 1, where the within families variance is: σ²pl = (1/4)D + (1/8)H + E1, being E1 the variance of the specific environment, that is the environmental variance between plants within families (estimated by σ²pl in the homozygote check [variance within rows], in which D = H = 0).

σ²B = (1/2)D + (1/16)H + E2, where E2 is the variance of the common environment, the environmental variance between families (estimated by σ²B in the homozygote check).

In F₃ traits with more than one measurement per plant, Table 1 applies fully. Here: σ²pl is the variance of plant means within families; σ² is the error term (E3); Ve = E1 + E2 + E3; σ²pl = (3/32)H + E1 – (1/2)E2 + (1/2)σ²B; σ²B, E1 and E2 are the same estimators as described for traits with one measurement per plant (Chandraratna, 1964).

Two selection processes were carried out independently: the complete F₂ populations underwent selection as part of our breeding programme (“global selection”), taking into account mainly PH, MAT, PN and TW, however PL was also measured. A 2% selection pressure was applied in basins (30 plants were selected in population MS, and 29 in population J), while in the field, the selection pressure was 1% (11 and 8 plants selected in MS and J, respectively). Due to lack of space, the F₃J families derived from the field could not be grown; as for the MS cross, we analysed 3 and 14 F₃ families derived from F₂ mother plants selected in the field and in basins, respectively. The derived F₃ lines were transplanted at wide spacing (30 cm × 30 cm) in basins for further selection as is the practice in our Department.

In the 300 unselected F₃ lines (derived from the 75 F₂ plants of each genotype/environment combination) grown in the field, traits related to lodging (PH, CD, PN, IL and PR) were selected (“Lodging selection”) using a selection pressure (proportion of selected plants: ρ) of 5% between families and of 1/6 within families. It resulted in 7-8 lines being selected per cross with the extreme values in each of the abovementioned traits; and, within the selected lines, seeds from the plant with the extreme value were grown as F₄ the following season. CD and PR were positively selected, PH was negatively selected, and for the PN and IL a bidirectional selection was used. Selection differential (S) was calculated as the difference between the mean of selected F₃ lines (M_selec) and the overall F₃ mean (M_F3); response to selection (R) was calculated as the difference between the mean of the F₄ lines derived from selected F₃ lines (M_F4) and M_F3, checking its significance with the Student t test. Since both generations were grown in the same field, but on consecutive years, values were corrected with the check ‘Senia’ for environmental effects; R is, therefore, only a rough estimate of response to selection.

Since 1991 we have selected F₂ in the field (wide spacing) and in basins (narrow spacing). As a result, nine varieties were commercially released until 2014. Four of these crosses had their F₂ selected in both environments: Three of their resulting varieties (‘Gavina’, ‘Cormorán’ and ‘Sivert’) came from an F₂ mother plant selected in basins, and the fourth (‘JSendra’) had its F₂ mother plant selected in the field. Four other varieties (‘Albufera’, ‘Argila’, ‘Antara’ and ‘Fleixa’) came from crosses whose F₂ were only selected in basins. One variety (‘Sarcet’) came from a cross whose F₂ was only selected in the field.

Prior to the analysis of the different genotype/environment combinations, a comparison of environments was performed by ANOVA of the corresponding checks. In Table 2, contrast significances shadowed in the E, Pf and Pb columns indicate that differences between checks are significant, showing environmental differences between basins and fields, between fields or between basins. The two basins used in the F₂ trial were homogeneous for most studied traits (except FL, MAT and some grain traits), while many differences were found between the two field plots where the F₂ were grown. On the contrary, the field plots used for the F₃ (adjacent, not separate like the F₂ field plots) proved to be more suitable for comparison of the populations’ performance, since only the GW of the checks varied significantly between them.

Table 2 shows that many traits expressed higher mean values in the F₂ plants grown in the field under wide spacing than in basins under narrow spacing, but in some cases differences were small. The most affected traits were those related to grain yield: panicle number and total grain weight showed a significant increase when cultivated in the field with respect to culture in basins (duplicating or even quadruplicating the values), especially in population J. The increase of panicle weight was not so marked. Culms were also significantly thicker in the field. An exception was flag leaf length: It was longer in basins than in the field. Plants showed a less open growth habit in the field.

In Table 2, contrast significances underlined in the E, Pf and Pb columns, indicate that differences between environments or between populations are significantly bigger than differences between checks, and so they can be attributed to genetic or G×E effects. Plant height and culm length showed G×E interaction: In the field, J plants were taller on average than in basins, while MS plants were shorter on average than in basins.

Plants were on average more productive and tillered more in F₂J than in F₂MS and F₃J. As a mean, the F₃ plants tillered less than F₂ plants grown in the field. Perhaps as a consequence, culm diameter was bigger, although part of this difference was due to the different way of measuring the trait. In both generations, MS plants had thicker tillers than J plants.

The trait means of both populations in the field differed more in the F₂ than in the F₃, since only PL, MAT, GH and CD showed significant differences between the F₃; the two traits measured only in F₃ (IL, PR) also proved to be different. F₃J had longer internodes and shorter panicles, matured later, had narrower culms, its plants were more open and presented higher pulling resistance than F₃MS.

In the F₂, broad sense heritabilities (H²) of plant height, panicle length, flag leaf length, 100-grain weight and mean grain-chalk size were higher in basins than in the field. But the H² of culm diameter, mean panicle weight and growth habit were higher in the field. Concerning total panicle weight, the only heritability (broad or narrow) different from zero was that of F₂J in the field, but it was not significant, according to the Ftest (Vp/Ve=1.68). On the other hand, broad sense heritabilities were higher in F₃ than in F₂, except for plant height, mean panicle weight and growth habit. Narrow sense heritabilities of many traits were low, i.e., additive variance had little or no weight in determining them. However, it was high in panicle length (in both populations), and in the plant height and mean panicle weight of F₃J (the latter only relatively high).

The highest broad sense heritabilities in the F₂ are those of elongation related traits, days to maturity, growth habit, grain dimensions and weight; pulling resistance and presence of white core are also noteworthy.

In the multivariate analysis of plant character means in both populations and both environments (Table 3), only the first three principal components (orthogonal —independent— linear combinations of characters) were informative. The first two principal components explained 80% of the variance of the variables’ correlation matrix. The first principal component includes mainly panicle number, total panicle weight, growth habit, culm diameter and flag leaf length. The second includes mainly panicle length, grain length and width, and days to 95% maturity. The third includes mainly plant height, one-hundred grain weight, chalk percentage and size. Figure 2 represents the four combinations population/environment against the first two principal components. Populations grown in basins appear in the negative part of the first principal component. When grown in the field, they appear in the positive side. This confirms the effect of the environment mentioned before: plants in basins tillered less on average, with more slender culms, lower grain production, a more open growth habit and longer flag leaves than in the field. In both environments, J population appears in the negative part of the second principal component, while MS population appears in the positive side. This shows the following genetic effect: Plants of MS population matured earlier than those of J, while their panicles and grains were longer.

In order to test the suitability of both environments for selection, we compared F₃ plants derived from F₂ individuals selected in basins or field. Table 4 shows the population means of traits evaluated in selected F₂ and F₃ plants within the breeding programme (except maturity, which was not accurately estimated in the F₂). The last column is a rough comparison between environments. Number of panicles was the only trait which presented significant differences between F₃ plants selected from different F₂ environments: Those derived from field selection tillered more than those derived from basins’ selection. The selection of the other traits was not influenced either by the environment or the different selection pressure used.

In both crosses, F₃ selection in the field (“lodging selection”) was effective in reducing plant height and in increasing culm diameter and internode length (Table 5).

In F₄J, panicle number and pulling resistance were also effectively increased. However, internode length and number of panicles did not show a significant response to negative selection.

Comparing the success of both types of selection in the F₂ of the nine varieties obtained in the last years, narrow-spaced selection seems quantitatively more effective than wide-spaced selection, however ‘JSendra’ (selected in the field, widely-spaced) is the most cultivated variety among the nine.

As expected, most of the traits showed higher expression (growth potential) in the field under wide spacing (except FL in both populations, and PH in F₂MS). The traits which showed higher heritability (H²) in the field were PW, GH, CD, GW and WP (in both populations), and TW in F₂J. In basins, the shadow avoidance response and the stronger competition under a higher planting density restricted plant yield (panicle number and weight, and consequently total weight) and stem thickness; plants also grew more open in basins. But the weaker environmental influence in basins than in field plots resulted in higher heritabilities for most traits (ten). This can be explained by the fact that basins are a more homogeneous environment, demonstrated by the lack of significant differences found when comparing checks (shadowed entries of the last column in Table 2).

This study detected genetic, environmental and G×E effects in some of the traits studied. Most of the genetic and environmental differences were confirmed by principal components analysis, whose first axis is correlated with traits with significant and consistent differences between basins and field (PN, TW, CD, GH and FL), and therefore differentiates field values from basin values. Tillering (PN), and consequently yield, is well known for being highly sensitive to environmental factors, including cultural practices such as planting density (Martínez, 2010). Broad sense heritabilities of these traits were low, in agreement with the strong environmental influence, and with previous studies (see Yamagata, 1997 for FL, or Xiong, 1992 for PN). An exception was growth habit in the F₂, but its H² turns to moderate-low in F3. The results concerning the F₂ populations are, however, approximate, since each combination population/environment consisted of 75 plants only.

The second component reflects the influence of genotype on traits such as earliness, panicle structure and grain shape (MAT, PL, GL and GW). Although most of the traits showed some significant difference between crosses in F₂ or F₃, the four traits mentioned above are the only ones where such differences were consistent in both environments and generations. Broad sense heritabilities of these traits were, accordingly, quite high in F₂. In fact, grain dimensions are easily fixed in early generations, and MAT is so from the F₃ (Jennings et al., 1979).

Some characters (PH, SL, and WS) displayed G×E interaction. This was due to the particular situation of the field used for the study of F₂MS: It was placed along a natural barrier which caused the rise of temperature and humidity levels (Fig. 1b,c). An environmental difference between both F₂ fields was therefore probably the cause of the apparent G×E interaction, and also the cause of the generally higher difference in trait means between the F₂ populations in the field than between the F₃ populations (except for CD and GH).

We can see true G×E interactions when comparing the behaviour of the checks vs that of the F₂ or F₃ populations: Although most significant differences between field plots and basins for both F₂ populations are maintained when comparing the checks, both F₂J and F₂MS had thicker culms and shorter flag leaves in the field than in the basins, while checks did not differ when comparing both environments. And on the contrary, the environment had a stronger influence on PL of ‘Senia’ checks than on the same trait in F₂J or F₂MS. Another trait which shows environmental plasticity in ‘Senia’ is the presence and length of awns (data not shown).

In traits such as PH (and SL), MAT and grain traits (weight, dimensions and white core presence), R ² values (coefficients of determination) are very low, both in genotypes’ and environments’ ANOVAs (data not shown) of the F₂, indicating that G×E, not included in the model, must be an important factor in determining variation. On the other hand, in yield related traits (PW, PN, TW), as well as GW and CD, the ANOVA across environments explains better their variation than the ANOVA across genotypes. Finally, PL and FL are traits whose variation is mainly determined by both “population” and “environment” factors.

As mentioned earlier, populations F₃J and F₃MS only differed significantly in a few traits. These differences can be assigned to genetic differences and the G×E component (underlined significances in Table 2); this is in agreement with the higher R ² obtained in the F₃ ANOVA, where the model (which leaves the environmental variance in the residue) explains quite well the observed variation. This is because their field plots were adjacent and highly comparable, as demonstrated by the checks’ contrasts of significance (Table 2).

Although we have noted before that the difference in selection pressure did not affect the majority of traits evaluated, we may draw attention that in the field it doubled that in basins. The reasons for the higher selection pressure applied on the F₂ selected in the field are two: 1) within our breeding program, we devoted at least two weeks to select four or five F₂s in basins, while field selection required only one day per F_2, which can affect the number of plants that have reached maturity that given day; 2) wide-spaced plants are easier to differentiate, because in the field plants were more productive and tillered more. Therefore, they can be selected more strictly.

Response to selection could not be determined for two reasons: because the only plants measured were the selected plants (not the whole F₂ and F₃ populations) and because generations where grown in different years and conditions.

From the results of the MS cross, of four characters, only panicle number showed significant differences between environments of origin in the F₂: the F₃ derived from field F₂ continued to tiller more than those derived from F₂ selected in basins, under the same F₃ plant density. This can be explained in this way: since plants tiller more under wide spacing, higher tillering genotypes were more effectively selected in the field, and some genetic differences in panicle number have been fixed; although they were not detected in the heritabilities calculated by ANOVA, probably because they are less reliable estimators than realized heritability (h²=R/S). Plant yield had the same tendency, but differences were not significant. It may be concluded that both environments (field and basins) can be used for selection in the F₂, although some different response might be expected in yield related traits; but we cannot draw conclusions out of only three lines selected in the field.

The objectives were: (1) to see the effect of a moderate reduction in height as tall plants are clearly prone to lodging, but not drastic enough to compromise yield; 2) to analyse the effect of reducing the length of the first internode on total plant height; 3) to determine if longer internodes can contribute to the resistance of the basal part of the rice plant (by improving its elasticity, and so its ability to recover), that was the reason for applying both positive and negative selection to IL to increase stem sturdiness by selecting thicker culms; 4) to see the effect of a change in panicle number; and 5) to check if the measure of the pulling resistance can serve as an estimator of susceptibility to lodging. It was not surprising that selection succeeded in reducing plant height, as this trait showed a high heritability. But the effective increase in stems diameter or internode length was unexpected, considering the observed low heritabilities, with no significant additive variance in one of the crosses. One possible explanation for the improvement of culm diameter lies in the presence of several QTLs displaying overdominance in the selected MS lines (QTL analysis was not performed in the J cross; Torró, 2010). As for IL, interestingly selection for obtaining plants with shorter internodes was useless in both crosses. This suggests that the observed reduction in plant height in both crosses after selection was achieved either by shortening of other internodes, or by reducing the number of nodes (which is different among the ancestors of both populations); but these traits were not evaluated in the segregating populations. The other two traits with low heritabilities that responded to selection —but only in the J cross—, were panicle number and pulling resistance. Similarly to IL, only positive selection to increase PN was effective; again, this might be due to the abundant positive dominance and overdominance observed in the QTLs for PN (Torró, 2010). Alternatively, the response to selection in traits with low observed heritability could be due, again, to the existence of additive variance not detected through analysis of variance of sibs, since realized heritability is more reliable. Finally, pulling strength showed correlation with lodging susceptibility (Torró et al., 2011), but in that study we concluded that the angle difference of recovery after bending a stem would be a more advisable criterion for selection of lodging resistance.

We may conclude that the two ways of F₂ cultivation described can be used for selection,_, although different responses might be expected in yield related traits.