Oil palm selection based on kernel content

A. Alvarado, F. Sterling, C. Montoya 

Abstract

ASD de Costa Rica's breeding strategy has aimed to increase the ratio of mesocarp to fruit and of oil to mesocarp, which has resulted in a decrease in kernel content. Due to the genetic variability available, and to the high heritability of this characteristic, it is possible to revert the process and to produce materials with a high kernel content. Hence, four series of experiments, planted in the Southern Pacific region of Costa Rica between 1977 and 1991, with progenies of different genetic origins, were evaluated with the purpose of selecting those that would allow for a rapid increase in kernel content per bunch.

The components of the phenotypic variance were separated and combining ability values were estimated in order to determine the contribution of each parent to total kernel production. This trait was used as a dependent variable in a multiple regression model to determine the most important independent variables.

Kernel production per hectare was affected by general as well as specific combining ability. In the regression model, the following variables explained the behavior of two selection indexes used as dependent variables (one economic and the other regarding oil and kernel production): bunch total weight, trunk height, leaf length, kernel to fruit and oil plus kernel to bunch. When this model was applied to phenotypical data, in order to select the progenies with the highest kernel content, it was associated with higher production of fresh fruit, better vegetative growth and better oil and kernel production.

Kernel content was higher in Tanzania and La Mé genotypes. A high kernel content per bunch can be obtained with the conventional material Deli x AVROS, which guarantees a high fruit yield but with a modest kernel gain. Another way of reaching this objective is by using crosses that have been less exploited commercially, but that could rapidly increase kernel production, such as Tanzania x La Mé or Deli/Tanzania x La Mé.

Introduction

The objective of plant breeding in oil palm is to create genotypes with the maximum potential for oil and kernel production per unit of area. Breure & Bos (1992) and Breure & Verdooren (1995) described the strategy for selecting superior materials, which involves three steps: 1) phenotypic selection of superior palms; 2) selection based on general combining ability (GCA) by means of progeny tests; and 3) the evaluation of the best families, carrying out crossings between them to exploit GCA and specific combining ability (SCA).

Of both components of combining ability, GCA (which refers to the additive effect that parents contribute to total genetic variance) and SCA (that is the contribution of the interaction of both progenitors on the behavior of their descendants), the GCA generally makes the greatest contribution to total genetic variance, which means that the average behavior of the progenitors allows for the predicting of the behavior of the descendants (Falconer 1960; Sterling et al. 1994).

Another way of predicting the behavior of the progenies is through the estimation of heritability, which measures the easiness of a characteristic, phenotypically evaluated in the parents, to be transmitted to the descendance (Falconer 1960). The higher the heritability, the easier this character will be inherited, thus, the phenotypic selection on that variable is generally very effective. Kernel to fruit and kernel to bunch are variables that generally show high heritability values, from 0.5 to 0.9 (Breure and Bos 1992; Sterling et al. 1994).

In the Palm Research Program (PIPA) of ASD de Costa Rica, the breeding strategy has aimed to increase the percentage of mesocarp in the fruit and of oil in the mesocarp, causing a decrease in the kernel to fruit ratio. If there is enough genetic variability, and heritability is high, it is possible to revert the process and to produce materials with a high kernel content whenever the oil palm industry requests, for example, more lauric acid.

This paper presents the results of four series of experiments, planted with progenies from different genetic origins, the purpose of which was to select, within their respective parents, those that would allow for a rapid increase in kernel content per bunch.

Materials and methods

Progeny test trials were planted in the Central and Southern Pacific areas of Costa Rica in the localities of Quepos and Coto, respectively, in 1977. Those experiments include Deli x AVROS crosses, whose parental lines were imported from experimental stations in Asia (Banting, Chemara, Mardi and Dami) and other populations that include Ekona, Calabar, La Mé, and Tanzania male lines (Table 1). 

The region of Quepos is characterized by a mean annual rainfall of 3441 mm, a dry season from December to April, a temperature that ranges from 22.6 to 31.10°C, and a daily average of 5.8 hours of sunshine. In Coto, mean annual rainfall is about 4069 mm, the dry season lasts from December to March, temperature ranges from 21.6 to 32.30°C and the daily average of sunshine is close to 5.5 hours. 

Calculation of the genetic components of the variance

In each series of experiments a model was defined in order to separate the components of the total variance; then, estimates of the contribution of each progenitor, of their interaction and of the environmental effect were obtained. For this purpose, a procedure similar to the one described by Sterling et al. (1994) and Breure & Verdooren (1995) was used. Models are fixed because materials are not random samples of the world population but rather come from the ASD collection in Costa Rica. In order to obtain the variance components, observed mean squares were equated with their expected values, thus:

CM_e = s² 

CM_D = s²  + c₁s²p   +  c₂s²d 

CM_P = s² +  c₃s²p

where:

CM = error mean square (e), dura mean square (D) or pisifera mean square (p)

s² =dura general variance (d) or pisifera general variance (p.)

c₁, c₂ and c₃ = coefficients originated in the type of partial dialelic design used 

The degree of connection between crosses allowed using "minimum squares" to estimate variance components. Mean square errors for dura and pisifera were estimated using the procedures PROC GLM, and its option RANDOM/Q, or directly through PROC VARCOMP, of the SAS package. Then, variances for each source were obtained from the above-mentioned equations.

SAS provides the mean squares:

Dependent variable: FFB (fresh fruit bunch production)

Thus, 

1) 98.71 = s² 

2) 262.04 = s² + 0.8165 s²p + 1.7225 s²d 

3) 194.62 = s² + 2.4615 s²p

Substituting 1) in 3) gives s²p = 38.96. This value is used in 2) to get s²d = 76.35. The difference between this error (98.71) and the environmental and unknown effects (48.30) is the interaction: 98.7 - 48.3 = 50.4

The percentage represented by each source of variance with regard to the total variance is used to obtain an estimate of general and specific combining abilities. GCA is the sum of the percentages of the variances due to dura and to pisifera and SCA is the percentage of the variance due to interaction, as the following example shows:

Therefore, the percentage of the variance due to general combining ability equals 53.9 (35.7 + 18.2) and SCA equals 23.6.

Determination of a regression model for the selection of progenies

Subsequently, a predicting model was determined, following the stepwise multiple regression statistical procedure used by Sterling et al. (1994), and the variables related to total kernel production were identified. GCA and SCA values estimate the contribution of each progenitor to the value of the variable for each progeny. Then, real values of the progenies were multiplied by the GCA. For example, if the real yield (FFB) of a cross is 700 and its GCA value is 0.539, the predicted value (FFBp) = 700 * 0.539 = 377.3. This is the net general contribution of the progenitors, while the SCA and the environmental part are obtained by subtraction (700.0-377.3). 
These values were used as independent variables in multiple regression models, where two indexes were used as dependent variables: kernel selection index (ind₁) and oil plus kernel production per hectare (ind₂). Ind₁ is a weighted index that sums up, in a single algorithm, the economic value of oil and kernel production according to age, trunk height, leaf length and the ratio of kernel to oil per bunch:

ind₁ = (oil value + kernel value + height value + leaf length value)* (% kernel to bunch/% oil per bunch)

where,

palm oil value = (0.6* kg of palm oil per year*12*age)/height

kernel oil value = (0.6*0.5*kg of kernel per palm per year*12*age*100)/height

value of trunk height = (-2.56*height)/100

value of leaf length = ((oil value + kernel value)*(10000/P*(leaf length/100)²))/143

Age is measured in years after planting and trunk height and leaf length in centimeters

ind₂ = t of palm oil per hectare per year + t of kernel per hectare per year 

Weight of oil per hectare per year (t) = fruit yield * % oil per bunch*143/1000

Weight of kernel per hectare per year (t) = fruit yield * % kernel to bunch *143/1000

The models were analyzed with SAS procedures "COLLINOINT" and "PRINCOMP" to discard variables with multicollinearity. Mallows coefficients were used to determine the number of variables in the model, and only those with probabilities lower than 5% were included.

In the next step, the best model was applied to the GCA part of the real values of the progenies and an estimate of each of the indexes was obtained. Those estimates were used to classify the progenies. A selection intensity of 85% (Z = 1.04) was used. (Sterling et al. 1994).

Results and discussion

Separation of variance components

Table 2 shows estimated values for GCA and SCA for the variables related to oil and kernel production in the four series of experiments. The most important component of the total variance was GCA (mother plus father) in the variables fresh fruit bunch production (FFB) and oil production per hectare per year (O/HA/Y). In the first variable, GCA accounts for 54% to 79% of the total variation, and in the second one from 66% to 96%. 

For the variable kernel per hectare (K/ha) of series 3, SCA had a greater weight in the model than GCA (66.9% of SCA and 28.0% of GCA), while GCA was the most important component in series 2 and 4 (86.8% y 88.8%). These differences are probably due to the fact that most of the series 3 crosses came from only one parent, as opposed to the other two cases where genetic variability was higher.

When GCA was splited, it was found that the mother's contribution explained most of the variation in kernel content. For example, the GCA value of the variable k/ha in series 1 was 82.8%, with 59.5% corresponding to the mother and 23.3% to the father; in other words, 59.5% of the phenotypic variation was explained by the maternal effect. This implies that selection of Deli dura palms with high kernel contents will allow for a substantial improvement of this characteristic. 

Table 2 also shows the maximum progress, relative to the mean, associated with each of the GCA components (Breure & Bos 1992). It can be deduced, from series 2 and 4, that the exploitation of GCA will allow for greater progress in kernel content since the values of maximum progress due to GCA, associated with the traits kernel to fruit, kernel to bunch and kernel per hectare, are higher than those of SCA. For example, for the variable kernel per hectare, of series 2, the maximum progress due to GCA was 83.0 while for SC it was only equal to 17.4. Of both components of GCA, the female line had the greatest contribution. This means that, with appropriate selection of genotypes and the exploitation of GCA, it should be possible to advance quickly in the improvement of this variable. 

Regression models 

When ind₁ was used as a dependent variable, the best model for each of the four series included some common variables: bunch total weight, kernel to fruit (except in series 1 where the variable was kernel to bunch), trunk height and leaf length. However, this last variable was excluded in series 4 and percentage of fertile fruits was included instead. The coefficient of determination (R²) for the models ranged from 77.6% to 97.2% (Table 3).

With ind₂, the best-fitting model for the first three series included the following independent variables: bunch production, leaf length and percentage of oil plus kernel to bunch. In series 4 the variables included were total bunch weight, percentage of oil to bunch and percentage of fertile fruits. The coefficient of determination ranged from 73.6% to 97.5%. For example, in series 4 kernel index (ind₁) is explained by the trait percentage of kernel to fruit (70%), by trunk height (12%) and by bunch production (9%); while kernel production per hectare (ind₂ ) was determined 83% by bunch production, 5% by percentage of oil to bunch and 4% by percentage of fertile fruits (Table 3). 

Progeny selection 

Once the regression models were defined, progeny selection was done using both selection indexes. Phenotypical means for variables related to bunch and oil production, vegetative growth and bunch composition were compared to determine the efficiency of the estimates (Table 4). 

There were some differences between the indexes. Ind₁ allowed for the determination of progress in fresh fruit production, vegetative growth, oil production and kernel production. In series 4, for example, values for selected crosses versus non-selected ones of the traits total bunch weight, trunk height and kernel per hectare were 129.8 versus 119.9 kg, 92.8 versus 106.2 cm and 1.1 versus 0.7 t, respectively.

Ind₂ allowed for the establishment of differences in fruit, oil and kernel production. Also in series 4, selected crosses showed average production of 145.9 kg of fruit per palm per year and 1.0 t of kernel/ha versus 114.2 kg and 0.8 t, respectively, in non-selected crosses.

Due to the fact that proportions of mesocarp and kernel to fruit are complementary, it is reasonable to expect that an increase in kernel content in the bunch would coincide with a decrease in mesocarp content in the fruit and in oil to mesocarp. The gain in the content of kernel to bunch in selected crosses was considerable and was proportionally higher than the reduction in oil content, especially when ind₁ was used .

For the four series, the mean value of kernel to bunch was 5.1% in selected crosses and 4.3% in the whole population when ind₁ was used; and 4.5% versus 4.3% when ind₂ was used. In oil content to bunch selected crosses showed an average of 27.8% versus 29.1% of the total population; and, when the second criteria for selection was used, oil content to bunch was slightly higher in selected crosses (29.4% versus 29.1%).

In summary, with the use of the kernel selection index (ind₁) better efficiency in the selection of crosses was achieved, since not only did fruit yield increase, but also vegetative growth and kernel content to bunch improved, although there was a slight reduction in oil content. With the use of ind₂ the kernel gain was smaller but without the reduction in oil content. 

Genetic variability

Table 5 and table 6 show the percentages of oil and kernel to bunch observed in tenera palms of several origins in the progeny test planted in 1991. The average oil content to bunch was similar in most maternal origins (28.1% to 29.0%), with only Tanzania material showing a lower percentage (25.4%). Among paternal origins, Calabar and Ekona materials showed the highest oil content values (28.9% to 29.2%) and La Mé the lowest one (26.0%). 

Kernel content had an inverse behavior to that of oil, with the highest kernel values being seen in the materials Tanzania and La Mé, which are the ones with the lowest oil contents (7.1% and 6.5% respectively).

With the genetic variability that is available in maternal origins, it is possible to obtain progenies with high kernel content. For example, the combination of Tanzania and La Mé origins would produce the highest values of kernel content to bunch (8.6%). Due to this variability, the maximum expected genetic progress for total production of oil and for kernel production per hectare is superior to that which would be obtained with other variables. For example, the maximum progress associated with GCA + SCA for the variable kernel/hectare in series 4 was 100%, while in mesocarp to fruit was 15% (Table 2).

Maximum progress (Table 2) is an indication of the ease with which the phenotypical value of a variable in a progeny can be increased through the proper exploitation of the genetic potential of its progenitors. In spite of the restricted genetic origin of the Deli dura y AVROS populations, the results of the series 1, 2 and 3 indicate that there is still potential to improve kernel to bunch content. Families and individual palms of these origins, which show maximum kernel contents, should be used for this purpose.

However, a rapid increase in kernel content would also be possible through the use of other genetic origins, such as those evaluated in series 4. For example, the introgression of the Tanzania origin within the Deli dura line (materials Deli/Tanzania) would allow for the combination of the high bunch production potential of the Deli material with the high kernel production of the Tanzania origin. Also, in order to appropriately exploit the GCA, palms with good production attributes within other masculine lines, such as La Mé and IR1039, could be selected.

In summary, there are two ways to obtain a high kernel to bunch content: through the conventional Deli x AVROS material, with a modest kernel increase; or through the use of crosses that are less exploited commercially, but that could considerably increase kernel production, such as Tanzania x La Mé or Deli/Tanzania x La Mé. 

References

BREURE , C.J.; BOS, I. 1992. Development of elite families in oil palm (Elaeis guineensis Jacq.) Euphytica. 64:99-112.

BREURE , C.J.; VERDOOREN, L.R. 1995. Guidelines for testing and selecting parent palms in oil palm. Practical aspects and statistical methods. ASD Oil Palm Papers. 9:1-68.

FALCONER, D.S. 1960. Introduction to quantitative genetics. Oliver and Boyd, London. 365 p.
HARTLEY, C.W.S. 1977. The Oil Palm. Longman Group, U.K. 806 p.

OKWUAGWU, C.O. 1985. The genetic base of the NIFOR oil palm breeding programme. In Proc. Int. Workshop on Oil Palm Germplasm and Utilization. ISOPB, pp. 228-239.

RICHARDSON, D.L.; CHAVES, C. 1986. Oil palm germplasm of Tanzania origin. Turrialba. 36:493-498.

ROSENQUIST. E.A. 1985. The genetic base of oil palm breeding populations. Proc. Int. Workshop on Oil Palm Germplasm and Utilization. ISOPB, pp. 27-56.

STERLING, F. 1998. Materiales de siembra de palma aceitera. Reporte interno. Informe bianual 1996-1997, Mejoramiento Genético. Programa de Investigación en Palma Aceitera, ASD de Costa Rica, pp 78-103.

STERLING, F.; ALVARADO, A.; MONTOYA, C.; RICHARDSON, D.L. 1994. Testing and selecting parental oil palm trees using progeny testing procedures. In Proc. Inter. Seminar on the Oil Palm Planting Material. ISOPB, Barranquilla, Colombia.