Introduction

In 2021, oil palm (Elaeis guineensis Jacq.) was the main contributor from the agriculture sector to Malaysia’s Gross Domestic Product (GDP) at 35.2%, followed by other agriculture (29.3%), livestock (16.7%), fisheries (11.3%), forestry and logging (5.2%), as well as rubber (2.3%) (DOSM 2022). In the same year, the area of oil palm plantations in Malaysia reached 5.7 million hectares, an increase of 54,000 hectares from 1960 (MPOB 2022), indicating the remarkable growth of the industry. Moreover, global demand for vegetable oils is expected to increase to 240 million tonnes by 2050 (Barcelos et al. 2015). As the highest productivity oil crop among cultivated oil crops, the oil palm industry worldwide strives to meet this increasing demand.

The dura fruit form from the Deli breeding stock was the first commercial oil palm planting material in Malaysia in 1917. Discovery of the monogenic shell thickness by Beirnaert and Vanderweyen (1941) showed the importance of the tenera fruit form of the dura × pisifera (D × P) hybrid. Compared to the dura, the tenera contains a higher proportion of oil-bearing mesocarp, which results in a higher oil yield. Thus, the Malaysian Department of Agriculture (DOA) began production of the country’s first D × P planting material in 1953, and the oil palm industry responded by switching to D × P planting materials a few years later (Kushairi et al. 1999). Switching from the dura to the tenera (D × P) planting materials improved yields by up to 30% in the country (Kushairi 2009).

Since the Algemene Vereniging van Rubberplanters ter Oostkut van Sumatra (AVROS) pisifera has shown good general combining ability (GCA) with the Deli dura, the Deli dura × AVROS pisifera (D × P) has been the common commercial planting material in Malaysia and worldwide (Kushairi et al. 1999). The Deli populations in Malaysia are descended from four dura seedlings planted in Bogor Botanical Garden, Indonesia in 1848, while AVROS pisifera originated from the “Djongo” palm found in Eala Botanical Garden, Zaire. Although, the performance of Deli dura × AVROS pisifera families has been extensively studied to identify superior families in order to increase oil productivity (Noh et al. 2010), the narrow genetic base of the existing Deli and AVROS populations would be a setback for yield improvement, as both populations are descended from a limited number of palms. This drawback led to a series of expeditions conducted in Africa and Latin America to search for E. guineensis and E. oleifera genetic materials, respectively (Rajanaidu 1994; Rajanaidu et al. 2017; Rajanaidu and Rao 1988). These germplasm collections were then field-planted, evaluated, and utilised in several ways for crop improvement. One of them is to develop elite D × P planting materials through the introgression of elite palms from the germplasm collections into current breeding material (Rajanaidu et al. 2017).

Development of new high-yielding planting materials is much needed as the national yield performance has stagnated at 3.7 t ha−1 for a few decades (Parveez 2019). Due to land scarcity, increasing plantation areas is not an option to boost yields. Therefore, the most effective and sustainable way to increase oil yield is by utilising planting materials from high-yielding genetic bases. Generally, development of new and improved planting materials relies greatly on populations of dura and pisifera. In Malaysia, the modified recurrent selection (MRS) is the most widely practised scheme by oil palm breeders, involving inter-crossing between selected parents from both dura and pisifera populations with the aim to exploit heterosis (Kushairi and Mohd Din 2020). Selected families derived from the inter-crosses would be planted to improve bunch and oil yields. Hence, the objective of this study was to select the best families from a set of 38 dura × pisifera crosses for the development of new high-yielding planting materials.

Materials and methods

Materials

A total of 38 D × P families were derived from inter-crosses between 31 Deli dura palms and 11 pisifera palms (four AVROS, four MPOB-Nigeria and three MPOB-Nigeria × United Plantations) (Table 1). The Deli dura palms were from several sub-populations, such as Banting, Highlands Estate, Ulu Remis, Johor Labis, Highlands Estate × Elmina, Ulu Remis × Highlands Estate, and Ulu Remis × Elmina. The AVROS pisifera male parents were from the fifth cycle of MPOB’s tenera/pisifera breeding populations, which is the descendant of BM 119 from Oil Palm Research Station (OPRS), Banting. The MPOB-Nigeria pisifera palms were from the second generation of Nigeria germplasm in MPOB, while the MPOB-Nigeria × United Plantations pisifera palms were derived from a collaborative crossing programme with the industry known as BK 20: United Plantations (UP) Nigeria Crossing Programme. The 38 D × P families were planted in Trial 0.491 at MPOB Research Station Hulu Paka, Terengganu, in 2007 in a randomised complete block design (RCBD), in three replications, with 16 palms per family per replicate at a planting density of 148 palms ha−1. Annual rainfall ranged from 2,545 to 6,176 mm year−1, with a mean rainfall of 3,948 mm year−1 from 2007 to 2018.

Table 1 Dura × pisifera families in Trial 0.491, MPOB Research Station Hulu Paka
Full size table

Data collection

Data collection from the 38 D × P families was carried out for vegetative measurements, bunch yield, and bunch quality components. Bunch yield was recorded for each palm at two harvest rounds per month starting at 36 months after field planting by recording the bunch weight (BWT) and bunch number (BNO). Fresh fruit bunch (FFB) yield, BNO and average bunch weight (ABW) were calculated based on the following formulae;

$${\text{FFB yield (kg palm}}^{ - 1} {\text{ year}}^{ - 1} {)}\;{ = }\;\sum\nolimits_{{{\text{i}} = 1}}^{{\text{n}}} {{\text{BWTi}}}$$
$${\text{BNO}}\,{\text{(no}}{.}\,{\text{palm}}^{{ - 1}} \,{\text{year}}^{{ - 1}} ) = \sum\nolimits_{{{\text{i}} = 1}}^{{\text{n}}} {{\text{BNOi}}}$$
$${\text{ABW}}\,{\text{(kg)}}\,{ = }\,{\text{FFB/BNO}}$$

where n is the number of harvest rounds, and i is the number of palm.

The bunch yield record between January and December was summarised for each year, and the average over four consecutive recording years (2014–2017) was used for data analysis. In addition, two to five bunches from each palm were sampled from 2011 to 2018 to determine the bunch quality components using the bunch analysis method (Blaak et al. 1963; Rao et al. 1983). The bunch quality components include bunch weight (BWT), mean fruit weight (MFW), mean nut weight (MNW), parthenocarpic to bunch (P/B), mesocarp to fruit (M/F), kernel to fruit (K/F), shell to fruit (S/F), oil to dry mesocarp (O/DM), oil to wet mesocarp (O/WM), fruit to bunch (F/B), oil to bunch (O/B), kernel to bunch (K/B), moisture content (MC), oil to fibre (O/F), oil yield (OY), kernel yield (KY), total economic product (TEP), and total oil (TOT). In 2015, one round of vegetative measurements using the non-destructive method by Corley and Breure (1981) was carried out for each palm. The vegetative characters measured were frond production (FP), petiole cross-section (PCS), rachis length (RL), leaflet length (LL), leaflet width (LW), leaflet number (LN), trunk height (HT), leaf area (LA), leaf area index (LAI), and trunk diameter (DIA). In terms of height, the oil palm stem is not visibly apparent in the first two years of growth and is not considered in the annual trunk height increment (HTi) estimation (Breure and Powell 1988). Therefore, the annual HTi in this study is the height of the palm at year eight after field planting, divided by six. Physiological traits such as vegetative dry matter (VDM), bunch dry matter (BDM), total dry matter (TDM), and bunch index (BI) were derived using bunch yield and vegetative measurements (Table 2).

Table 2 Mean squares for bunch yield and its components in 38 D × P families
Full size table

Statistical analysis

All data were analysed on an individual-palm basis, where simple statistics such as mean and coefficient of variation (CV) were determined for each trait. The analysis of variance (ANOVA) for all traits was performed using the Statistical Analysis Software (SAS 9.4) programme, where the general linear model (PROC GLM) was used due to the unequal number of palms from the families studied. The family mean comparison was conducted using Fisher’s least significant difference (LSD) at a 5% level of probability. Broad-sense heritability (h2B) of all traits in families was estimated using variance components from ANOVA as follows;

$${\text{Broad - sense heritability }}\left( {{\text{h}}^{{2}}_{{\text{B}}} } \right) = {2}(\sigma^{2}_{{\text{g}}} /\sigma^{{2}}_{{\text{g}}} + \, \sigma^{{2}}_{{{\text{gr}}}} + \, \sigma^{{2}}_{{\text{e}}} )$$

where σ2g = family variance, σ2gr = interaction between family and replication variance and σ2e = within palms variance. The sum of σ2g, σ2gr and σ2e is phenotypic variance (σ2p).

The minimum variance method (Ward 1963) was used to cluster the families into groups, which was analysed using SAS 9.4.

Results and discussion

Bunch yield and its components

A previous study by Noh et al. (2010) revealed limited genetic variability among 40 Deli dura × AVROS pisifera progenies planted in MPOB Research Station Keratong, Pahang, in terms of FFB yield and BNO, which could be a hindrance to future selection and breeding. However, 38 D × P families in this study and 34 D × P families studied by Arolu et al. (2017) proved otherwise, where analysis of variance (ANOVA) showed highly significant differences for bunch yield and its components. A significant difference was also detected for the interaction effect between family and replicate (g × r), suggesting inconsistencies in bunch yield and its components across the three replicates in this study.

The performance of 38 D × P families in terms of bunch yield and its components is presented in Table 3. Family PK 4044 from the Deli Highlands Estate × (MPOB-Nigeria × UP) cross produced the highest FFB yield at an average of 214.4 kg palm−1 year−1. However, no difference was observed in the FFB yields from families PK 4144, PK 4159, PK 4189, ECP HP 496, ECP HP 500, ECP HP 502, ECP HP 504, ECP HP 519, ECP HP 529, and ECP HP 449 based on Fisher’s LSD. The majority of these families were from the Deli Ulu Remis × MPOB-Nigeria cross. In contrast, family PK 4039 from the Deli Banting × (MPOB-Nigeria × UP) cross, was ranked the lowest in terms of FFB yield, at 149.4 kg palm−1 year−1. Fisher’s LSD indicated no significant difference between family PK 4039 and four other families, namely ECP HP 415, ECP HP 617, ECP HP 626 and PK 4169. The results also showed that families with the highest (PK 4044) and lowest (PK 4039) FFB yields had comparable ABW means at 14 kg. However, a higher BNO for family PK 4044 (a mean of 15.3 bunches palm−1 year−1) than for family PK 4039 (a mean of 10.6 bunches palm−1 year−1) may have contributed to the differences in their FFB yields. Moreover, FFB yield is one of the requirements listed in the Malaysian Standard of Oil Palm Seeds for Commercial Planting (MS 157) by the Department of Standards Malaysia. Based on the current standard, MS 157:2017, the tenera in the progeny test should meet the minimum requirement of FFB yield at 170 kg palm−1 year−1 (DOSM 2017). In this study, the majority of the families, i.e. 33 of 38 families representing 87%, had higher FFB yields than the value listed in the MS 157:2017 standard, demonstrating the effectiveness of the parental selection.

Table 3 Family means for bunch yield and its components from the 38 D × P families
Full size table

In terms of the number of fruit bunches produced per family, the highest BNO was recorded in family ECP HP 496 from the Deli Ulu Remis × MPOB-Nigeria cross at 20.2 bunches palm−1 year−1, though it had the lowest ABW of 10.3 kg. These values were significantly different from the corresponding mean values for BNO and ABW of other families through Fisher’s LSD analysis. Conversely, family ECP HP 550 from the Deli Banting × AVROS cross had the lowest BNO but highest ABW at 8.2 bunches palm−1 year−1 and 22.2 kg, respectively. These observations suggested a negative relationship between BNO and ABW. The negative correlation between BNO and ABW in oil palm is widely known (Tanya et al. 2013). As oil yield (OY) is an important trait for productivity improvement, relationships between OY and yield components have been studied. The BNO and ABW exhibited by the MPOB-Senegal germplasm were moderately and positively correlated to OY production, while FFB yield was highly and positively correlated to OY (Myint et al. 2019). Four families derived from the MPOB-Nigeria and MPOB-Nigeria × United Plantations male parents, namely PK 4044, ECP HP 502, PK 4144, and PK 4189 were the top FFB yielders, with comparable ABW (14.3 to 16.5 kg) and higher BNO (13.0 to 15.3 bunches palm−1 year−1) than the trial means. The results thus suggest that moderate ABW with high BNO traits should be considered for high FFB yield production. The findings also agreed with Arolu et al. (2016), who suggested that Nigeria pisifera male parents may have contributed to higher FFB yield in their 34 Deli dura × Nigeria pisifera families.

Bunch quality components

There were highly significant differences (P < 0.01) among the families for all bunch quality component traits (Table 4). As bunch quality components consist of major economic traits such as OY, TEP, and TOT, significant genetic variation detected among families would assist in the selection programme for improvements of these traits. The g × r interaction was also highly significant for the majority traits such as BWT, P/B, K/F, O/DM, O/B, O/F, OY, TEP, and TOT, while significant for MFW and K/B. This suggests inconsistencies in the bunch quality component traits across the replicates. Mhanhmad et al. (2011) found that F/B, K/B, and O/B were significantly higher in the dry season (January to April 2008) compared to the wet season (July to October 2007) in Pathio district, Chumphon province, Thailand. Nevertheless, non-significant differences in the g × r interaction were observed for MNW, M/F, S/F, O/WM, F/B, MC, and KY, implying consistencies in the performance of families for these traits across replicates.

Table 4 Mean squares for bunch quality components in 38 D × P families
Full size table

The performance of bunch quality components for 38 D × P families is presented in Table 5. Bunch weight (BWT) ranged from 9.4 kg to 24.7 kg, with a trial mean of 13.2 kg. Family ECP HP 550 from the Deli Banting × AVROS cross with the highest BWT, or 86% higher than the trial mean, was significantly different from all the other families. In contrast, family ECP HP 496 from the Deli Ulu Remis × MPOB-Nigeria cross with the lowest BWT was not significantly different from families PK 4169, PK 4181, ECP HP 500, ECP HP 504, PK 4176, PK 4039, and ECP HP 502. Meanwhile, four families, namely ECP HP 630, ECP HP 425, ECP HP 414, and ECP HP 450, were found to have the highest MFW of 15.2 g, 14.3 g, 14.0 g, and 13.8 g, respectively. Their BWT means were not significantly different based on Fisher’s LSD. On the other hand, eight families (PK 4160, PK 4044, PK 4169, PK 4144, ECP HP 502, PK 4161, PK4159, and PK 4189) had the lowest MFW and were not significantly different from each other. Families ECP HP 617 (Deli Banting × AVROS) and ECP HP 519 (Deli Ulu Remis × MPOB-Nigeria) had contrasting results in terms of MNW, M/F, and K/B. Family ECP HP 617 had the highest MNW (3.3 g) and K/B (5.9%), while having the lowest M/F (73.5%). On the contrary, due to its lowest MNW (1.1 g) and K/B (2.3%), family ECP HP 519 had the highest M/F at 90.0%, or 12% higher than the trial mean, with significant variation from the other families. Some 17 families representing 45%, had a higher M/F than the trial mean and were mostly from the Deli Ulu Remis × MPOB-Nigeria cross. Shell to fruit (S/F) ranged from 6.2% to 17.3%, with a trial mean of 12.8%. Family PK 4144 from the Deli (H. Est. × Elmina) × (MPOB-Nigeria × UP) had the highest S/F and was not significantly different from families PK 4189, PK 4039, PK 4161, ECP HP 617, and PK 4181. The lowest S/F, which was 48% lower than the trial mean, was recorded by family ECP HP 519, with significant variation from all the other families. Kushairi et al. (2003) suggested that high M/F and low S/F determine a high O/B, whereby an increase in mesocarp content would reduce the shell content without changing the kernel size. Family ECP HP 519 had a mean O/B of 30.9%, the third highest after families ECP HP 500 and ECP HP 531. These three families were from the Deli Ulu Remis × MPOB-Nigeria cross and were not significantly different for O/B. Kernel to fruit (K/F) ranged from 3.70% by family ECP HP 500 to 10.6% by family ECP HP 617, where both families also showed contrasting results for O/B at 31.2% and 21.6%, respectively.

Table 5 Family means for bunch quality components from the 38 D × P families
Full size table

Meanwhile, P/B varied from 2.3% for family ECP HP 635 (Deli Banting × MPOB-Nigeria) to 10.8% for family ECP HP 450 (Deli Ulu Remis × AVROS). Family ECP HP 635 also had the lowest F/B at 59.1%. However, its lowest P/B and F/B means were not significantly different from families PK 4032, PK 4189, PK 4160, PK 4144, and PK 4044, derived from the MPOB-Nigeria × UP male parents. Family ECP HP 626 (Deli Banting × MPOB-Nigeria) had the lowest O/DM (74.0%) and O/WM (41.4%), which were 6% and 19% lower than the trial means, respectively, with significant variation from the other families based on Fisher’s LSD. On the contrary, the highest O/DM was found for family ECP HP 519 at 81.1%, while family ECP HP 500 had the highest O/WM of 55.0%. Moisture content (MC) varied from 31.1% to 44.3%, with a trial mean of 35.6%. Family ECP HP 626 with the highest MC (or 24% higher than the trial mean) showed no significant difference from the other families. The high MC reflected a negative influence on the oil-related traits such as O/F (291.6%), OY (35.4 kg palm−1 year−1), TEP (39.4 kg palm−1 year−1), and TOT (38.7 kg palm−1 year−1). Meanwhile, family ECP HP 500 with the lowest MC had the highest OY (70.0 kg palm−1 year−1), TEP (73.1 kg palm−1 year−1), and TOT (72.6 kg palm−1 year−1). Oil yield (OY) is a derived trait, of which O/B and FFB yield are its main components. Thus, the high OY of family ECP HP 500 was associated with the highest mean of O/B. Based on Fisher’s LSD, OY, TEP, and TOT means of family ECP HP 500 did not differ significantly from ECP HP 519, with the highest M/F and the lowest S/F. This suggests that the selection of palms for high oil production should also consider the M/F and S/F traits as selection criteria. KY varied from 4.8 kg palm−1 year−1 to 12.4 kg palm−1 year−1, with a trial mean of 8.2 kg palm−1 year−1. Family ECP HP 449 from the Deli Ulu Remis × AVROS had the highest KY and was not significantly different from families PK 4152, ECP HP 550, and ECP HP 452. These progenies, with KY at least 38% higher than the trial mean are potential candidate for multi-location evaluation in efforts to develop varieties with kernel oil.

Family ECP HP 626 displayed the worst performance in bunch quality components, having the lowest means of O/DM, O/WM, O/F, OY, TEP, and TOT. However, all 38 D × P families in this study had OY exceeding 35 kg palm−1 year−1 or 5 t ha−1 year−1, which is above the 2021 average national OY of 3.1 t ha−1 (MPOB 2022). This suggests the potential use of these families in this study to improve the performance of national oil production. In the meantime, factors underlying the gap between potential and actual results are important to understand. According to Woittiez et al. (2017), yield-reducing factors such as unsuitable cropland as well as pests and disease infections have an impact on actual yields. Besides that, factors underlying bunch production would require further studies, especially regarding sex determination and the failure of bunches to form. In the current Malaysian Standard, MS 157:2017, the minimum requirements for bunch quality components for the tenera in progeny tests are an O/B of 25%, K/B of 3%, and OY of 42.5 kg palm−1 year−1 (DOSM 2017). A total of 18 families representing 47%, namely ECP HP 423, ECP HP 425, ECP HP 428, ECP HP 450, ECP HP 456, ECP HP 466, ECP HP 593, ECP HP 618, ECP HP 496, ECP HP 504, ECP HP 529, ECP HP 531, ECP HP 633, PK 4032, PK 4044, PK 4159, PK 4176, and PK 4181, fulfilled these minimum requirements, including FFB yields of more than 170 kg palm−1 year−1. Half of these families were from the Deli Ulu Remis female parents, which could be selected for future breeding programmes.

Vegetative and physiological traits

ANOVA results showed highly significant differences between families for all vegetative and physiological traits (Table 6), demonstrating the presence of variability in these traits. In addition, the ANOVA also revealed that the performance in the vegetative and physiological traits was inconsistent across the replicates. This was supported by the significant g × r interaction effect obtained. Frond production (FP) ranged from 21.7 fronds palm−1 year−1 to 29.1 fronds palm−1 year−1, with a trial mean of 24.8 fronds palm−1 year−1 (Table 7). Family ECP HP 502 from the Deli Ulu Remis × MPOB-Nigeria cross had the highest means of FP, or 17% higher than the trial mean. It was significantly different from all the other families. Typically, palms with a high number of fronds would potentially produce a higher FFB yield as bunches are produced at the frond axils (Noh et al. 2010). This study has also proven this assertion where family ECP HP 502 was the second highest FFB yielder after family PK 4044 (Table 3) with an FP of 26.9 fronds palm−1 year−1. As such, FP should be considered in the selection of high-yielding planting materials. Family ECP HP 437 from the Deli Johor Labis × AVROS cross had the highest PCS (35.8 cm2), while family ECP HP 496 had the lowest PCS (23.8 cm2). This is advantageous because palms with narrow PCS allow for easier frond cutting during FFB harvesting. On the other hand, short RL is a selected trait for high-density planting at 180–200 palms ha−1 (Barcelos et al. 2015), compared to 136–160 palms ha−1 in standard commercial plantings. For breeding for compactness, palms with an RL shorter than 5 m are desirable (Norziha et al. 2020). In this study, PK 4144 was the only family with an RL less than 5 m. It was however, not significantly different from three other families from the Deli Ulu Remis × AVROS cross, namely ECP HP 450 (5.1 m), ECP HP 449 (5.1 m), and ECP HP 452 (5.0 m). Family PK 4169 from the Deli Banting × (MPOB-Nigeria × UP) cross had the highest RL at 5.9 m, with no significant difference from families ECP HP 633 (5.8 m) and ECP HP 529 (5.8 m). Trunk height (HT) varied from 1.7 m (PK 4039) to 2.6 m (ECP HP 500), with a trial mean of 2.2 m. HT of less than 1.8 m (or HTi lower than 0.3 m year−1) is also a selection criterion for compactness besides RL (Norziha et al. 2020). Furthermore, low HT is a critical trait of interest for oil palm breeders as it increases the ease of FFB harvesting and the economic lifespan (Marhalil et al. 2014). In this study, families PK 4039, ECP HP 496, and ECP HP 504 had the lowest HT (or an average HTi of 0.3 m year−1), with no significant variation among them. Compared to current planting materials with HTi between 0.40 m year−1 and 0.75 m year−1 (Kushairi et al. 1999; Kushairi and Mohd Din 2020), these two families were approximately 28% shorter and can be considered in breeding for shorter palms.

Table 6 Mean squares for vegetative and physiological traits in 38 D × P families
Full size table
Table 7 Family means for vegetative and physiological traits from the 38 D × P families
Full size table

In terms of LL and LW, they ranged from 80.8 cm (ECP HP 519) to 94.0 cm (ECP HP 437) and 4.7 cm (ECP HP 496) to 5.8 cm (ECP HP 635), respectively. Family ECP HP 496 not only exhibited the lowest LW but also recorded the smallest LA, measuring 7.3 m2, and the lowest LAI at 4.3. Breure (2010) stated that yields are likely to decrease when the LAI is higher than 6, due to competition among palms. All families in this study had LAI of below 6 except for ECP HP 617, but this was not significantly different from the LAI values of ECP HP 437 (6.0), ECP HP 626 (5.9), ECP HP 635 (5.9), PK 4169 (5.8), PK 4159 (5.7), ECP HP 529 (5.7), and PK 4044 (5.7) based on Fisher’s LSD. Three of them, namely ECP HP 617, ECP HP 626, and PK 4169, from Banting female parents, were among the five lowest FFB yielders, which is in line with observations by Breure (2010). However, other traits may contribute to better FFB yield, such as high PCS and LL for ECP HP 437, high LW for ECP HP 635, and high LN for ECP HP 529. The leaflet number (LN) varied from 163.1 (PK 4144) to 179.9 (ECP HP 529), with a trial mean of 170.6. Trunk diameter (DIA), ranging from 0.6 m to 0.8 m. Family ECP HP 502, with the highest DIA, had the highest VDM and TDM at 17.5 t ha−1 year−1 and 33.9 t ha−1 year−1, respectively.

For physiological traits, the bunch index (BI) is one of the selection criteria for pisifera male and dura female parents (Breure 1986). Increasing BI, or the proportion of BDM to TDM, is one of the strategies to boost oil production (Hardon et al. 1972; Breure and Corley 1983). In this study, family ECP HP 449 had the highest BI (0.6) and BDM (17.6 t ha−1 year−1), with no significant difference from families ECP HP 496, ECP HP 456, PK 4161, ECP HP 529, ECP HP 504, PK 4189, and PK 4159. In contrast, family ECP HP 415 from the Deli Banting × AVROS cross had the lowest BI and BDM means, measuring 0.4 and 12.2 t ha−1 year−1, respectively. These values did not differ significantly from those of family ECP HP 633 from the Deli Banting × MPOB-Nigeria cross. Strong positive correlations between BI and FFB yield have been reported by Junaidah et al. (2004) and Fadila et al. (2016). Thus, BI should not be neglected in the selection of high-yielding materials.

Heritability estimates for D × P families

Statistically, the heritability estimate is used to describe the percentage of phenotypic variation that can be attributed to genetic variation. It indicates the reliability of the phenotype as an indicator of genotype, with possible values between 0% (all environmental variation) and 100% (all genetic variation). Heritability estimates can be categorised as low (less than 30%), moderate (30% to 60%) or high (more than 60%) (Johnson et al. 1955). Variance components and broad-sense heritability (h2B) estimates for bunch yield and its components, bunch quality components, as well as vegetative and physiological traits were calculated for the 38 D × P families (Table 8). Broad-sense heritability (h2B) estimates were low for FFB yield (15.8%) but high for BNO (80.9%) and ABW (66.5%). These results are expected as the h2B estimate for FFB yield is usually low but high for both BNO and ABW (Corley and Tinker 2016). High h2B estimates for FFB yield, BNO, and ABW ranging from 88% to 94% were observed among 42 E. oleifera × E. guineensis progenies in Brazil (Gomes Junior et al. 2016). This was attributed to the high genetic variability among the evaluated progenies, mainly from male and female parents from natural populations in America. Thus, introduction of new materials should always be a priority to increase genetic variability in developing new varieties with selected traits, especially high yield.

Table 8 Variance components and heritability estimates for bunch yield and its components, bunch quality components, vegetative and physiological traits
Full size table

The variance for family, family by replication, within palms, and phenotype are presented in Table 8. The results showed that family variance (σ2g) for FFB was lower than the family × replication variance (σ2gr), suggesting a higher influence of local environmental factors on FFB than the family genotype. A few studies reported that the magnitude and dimensions of environmental effects are different for each genotype (Okoye et al. 2009; Rafii et al. 2001). Therefore, studying the interaction of genotypes with the environment is essential to evaluate specific genotypes, especially in terms of FFB. Meanwhile, σ2g for BNO and ABW were higher than σ2gr, indicating the superior effect of genotype in the expression of both traits. In terms of bunch quality components, σ2g estimates were higher than σ2gr for all traits. Therefore, most traits can serve as selection criteria because family genotypes contribute more variation to bunch quality components than environmental factors. On the contrary, most vegetative and physiological traits, such as RL, LN, LA, LAI, BDM, and TDM, had lower σ2g estimates than the σ2gr. Stronger environmental influence on vegetative and physiological traits, compared to bunch yield and its components, as well as bunch quality components, suggests the need for additional studies on the adaptability and stability of the test genotypes in multi-environment trials.

The h2B estimates were high for the majority of the bunch quality components, such as MFW (100%), MNW (100%), O/B (93.2%), K/F (91.9%), S/F (85.2%), M/F (83.4%), O/WM (78.8%), MC (73.9%), O/DM (73.5%), K/B (73.3%), O/F (72.0%) and OY (62.7%). Moderate h2B estimates were detected for the rest of the traits, namely TOT (55.2%), KY (55.0%), TEP (53.8%), BWT (50.9%), F/B (50.3%), and P/B (31.5%). Analysis of 24 dura × pisifera progenies from 10 genetic origins showed that the h2B estimates for most bunch quality components were moderate, ranging from 30% to 60% (Swaray et al. 2020). Besides the fact that the aforementioned study involved dura and pisifera parents from more varied backgrounds compared to this study, the difference in the h2B estimates could also be due to the different study environment, one of the determining factors for h2B estimates (Acquaah 2020). Most of the vegetative and physiological traits, in contrast, had low h2B estimates, ranging from 0% (DIA and BI) to 26.1% (RL). This suggest that most vegetative characters are influenced more by the environment. These results were also contrary to observations by Breure and Corley (1983), where heritability was generally high for vegetative characters. In our study, a high h2B estimate was only detected for FP (61.3%), while h2B estimates were medium for VDM (35.2%), HT (33.3%), PCS (33.1%), and LL (32.5%).

Principal component analysis

Principal component analysis (PCA) is a useful method for variable-reduction when there are many variables. The variables can be reduced to a few principal components that represent the majority of the variance in the observed variables. For specific crucial and practically significant principal components, maintaining an eigenvalue greater than one is recommended as a general guideline (Ekezie 2013; Iezonni and Pritts 1991). In this study, seven components (PC1-PC7) showed an eigenvalue greater than one, accounting for 89.0% of the total variation (Table 9). Studies on some MPOB oil palm germplasm reported between four and six principal components with eigenvalues greater than one, which explained a total variation of over 85% (Li-Hammed et al. 2016; Norziha et al. 2019; Suzana et al. 2016; Wan Nor Salmiah et al. 2022).

Table 9 Variables correlation loading matrix, eigenvalues, variance and cumulative variance of the seven principal components of 38 D × P families
Full size table

The PC1, with an eigenvalue of 12.3, has the highest variance of 35.2%. It exhibited a strong and positive correlation with bunch quality component traits such as OY (0.263), TOT (0.252), TEP (0.250), O/WM (0.244), and O/B (0.237). Simultaneously, it demonstrated the highest negative relationship with K/F (−0.238) and MC (−0.239). Variables with significant positive and negative effects contribute significantly to the diversity, especially those on PC1 (Li-Hammed et al. 2016). Thus, these bunch quality component traits can be considered for selection in breeding programmes. Meanwhile, other components which had eigenvalues from 1.4 to 5.6 and variance between 3.9% and 16.0%, were primarily associated with vegetative and physiological characteristics. The PC2 was associated mainly with VDM (0.351), PCS (0.337), and HT (0.300). Both PC3 and PC5 were characterised by LA (−0.309 and 0.352) and LAI (−0.308 and 0.352), with the addition of LN (−0.321) for PC3 and LL (0.364) for PC5. On the other hand, PC4 and PC6 exhibited high correlations with BDM (0.339 and 0.312), along with FFB (0.330) and TDM (0.387) for PC4. Additionally, PC6 showed correlations with BI (0.317) and S/F (−0.345). The last component, PC7, was primarily associated with LL (0.487), DIA (0.413), and LW (−0.368).

Through multidimensional preference analysis, a PCA biplot was generated to examine the inter-relationship between the D × P families and variables, as well as to determine which variables contributed the most to the families. Traits in the right quadrant, such as O/B, M/F, OY, TEP, and TOT, were positively correlated with PC1, while traits that were negatively correlated with PC1, namely K/B, MNW, K/F, MC, and KY, were located in the left quadrant (Fig. 1). Vegetative and physiological traits such as PCS, LW, LN, VDM, and RL contributed positively to PC2. Contrastingly, BI and S/F correlated negatively with PC2. Positive associations between variables are indicated by their proximity to one another. Thus, TEP and TOT depend positively on OY. Other positive associations were detected between K/B and MNW, as well as K/F and MC. On the contrary, a strong negative correlation can be observed between ABW and BNO, among others. There was a lack of correlation between VDM and BDM as the angle between these two variable vectors is 90 degrees or orthogonal. Families ECP HP 500 and ECP HP 519 were distinctly located in quadrant I, with O/B being the variable that contributed most to these families. Family ECP HP 496 was located furthest in quadrant IV of the plot due to having the highest BNO as well as the lowest means of leaf-related traits such as LW, LA, and LAI. It is suggested that these three families (ECP HP 500, ECP HP 519, and ECP HP 496) are unique and can be selected for the development of new high-yielding planting materials.

Fig. 1
figure 1

Biplot based on 35 traits of 38 D × P families on the first and second principal component axes

Full size image

Cluster analysis

Cluster analysis is a method that uses dendrograms to organise genotypes into groups, which shows how different genotypes might be distinguished. This method has been used to identify unique populations of oil palm germplasm for conservation and utilisation in breeding programmes, such as MPOB-Nigeria (Li-Hammed et al. 2016), MPOB-Sierra Leone (Suzana et al. 2016), MPOB-Guinea (Norziha et al. 2019), and MPOB-Cameroon (Wan Nor Salmiah et al. 2022). The results of those studies also revealed that the grouping of the populations did not associate with their geographical origins. Cluster analysis was also used to evaluate D × P progenies for parental selection in developing new high-yielding planting material (Arolu et al. 2017).

All 38 D × P families were grouped into three main groups (clusters) with several sub-clusters (Fig. 2). Generally, the grouping of families did not associate with their backgrounds or origins, as crosses from different backgrounds were grouped in the same cluster. Cluster I comprised 13 families, most of which were crosses between Deli (Banting, Johor Labis and Ulu Remis × Elmina) dura and AVROS pisifera. The second cluster comprised the lowest number of families (nine), with most of them from the Deli Ulu Remis × MPOB-Nigeria cross. On the other hand, Cluster III comprised the highest number of families (16), with families from the Deli dura × (MPOB-Nigeria × UP) pisifera cross as the majority.

Fig. 2
figure 2

Dendrogram of the 38 D × P families based on 35 traits evaluated with root mean square distance between observations (R2) = 1.01206

Full size image

Cluster I consisted of two sub-clusters: I-A and I-B, generally exhibiting large bunch sizes (ABW and BWT), high MC, and wide petioles (PCS) (Table 10). Moreover, sub-cluster I-B, which comprised crosses from all three different pisifera male parents (AVROS, MPOB-Nigeria, and MPOB-Nigeria × UP) had the highest values of kernel and leaf-related traits such as K/F (8.6%), K/B (5.1%), LL (90.3 cm), LN (173.4 leaflets), LA (9.6 m2), and LAI (5.7). However, families within both sub-clusters in Cluster I produced low bunch yield with poor bunch quality, as demonstrated by their low FFB yield (176.3–177.1 kg palm−1 year−1), BNO (10.3–10.4 bunches palm−1 year−1), O/WM (46.2–49.9%), OY (42.5–47.9 kg palm−1 year−1), TEP (48.2–53.1 kg palm−1 year−1), and TOT (47.2–52.3 kg palm−1 year−1). The results indicated that families from different genetic backgrounds but exhibiting similar performance for certain traits were clustered together. This was similar to the observations by Arolu et al. (2017), where 34 Deli dura × Nigeria pisifera progenies were clustered according to the performance of certain characteristics. Conversely, 25 D × P full-sib progenies derived from crosses between Deli duras and four different pisifera male parents (AVROS, Dumpy AVROS, La Me, and Yangambi) were grouped into various clusters according to their pisifera source rather than the morphological performance of the progenies (Junaidah et al. 2011).

Table 10 Performance means for each sub-cluster
Full size table

Cluster II consisted of families producing more fronds (FP), longer rachis (RL), broader trunk (DIA), and higher dry matter (BDM and TDM) compared to the other two clusters. Both sub-clusters II-A and II-B also demonstrated the best performance in bunch yield and bunch quality components. They had the highest means of FFB yield (202.6–206.1 kg palm−1 year−1), BNO (14.4–16.0 bunches palm−1 year−1), M/F (81.5–84.9%), O/WM (52.7–53.8%), OY (57.0–62.8 kg palm−1 year−1), TEP (61.1–66.5 kg palm−1 year−1), and TOT (60.4–65.8 kg palm−1 year−1). Furthermore, families within sub-cluster II-A excelled in oil-related traits such as O/DM (80.0%), O/B (30.1%), and O/F (405.7%). Based on the pedigree of the families, sub-cluster II-A consisted of families from Ulu Remis × MPOB-Nigeria crosses. Outstanding bunch yield and bunch quality performance of D × P progenies from the Ulu Remis dura has also been reported by Swaray et al. (2020). Among 24 D × P progenies derived from 10 genetic origins, Ulu Remis × Yangambi produced the highest FFB yield, while Ulu Remis × AVROS exhibited higher values of bunch quality components compared to the trial mean.

In this study, families from the Ulu Remis × AVROS cross in sub-cluster III-A produced the highest MNW (2.7 g), F/B (66.5%), and KY (9.8 kg palm−1 year−1), while having the lowest values for some vegetative traits such as RL (5.3 m), LW (5.1 cm), LN (167.6 leaflets), and DIA (0.7 m). Interestingly, sub-clusters III-A and III-B consisting of families from the Deli dura × (MPOB-Nigeria × UP) pisifera cross, had contrasting fruit sizes. On average, the highest MFW was recorded for families within sub-cluster III-A at 13.2 g, while the lowest was recorded in families within sub-cluster III-B at 8.0 g. Besides that, sub-cluster III-B displayed the lowest M/F (77.9%), PCS (25.6 cm2), and HT (2.1 m), but produced the highest S/F (15.08%).

Conclusion

The 38 D × P families examined in this study demonstrated a high level of genetic variability for all traits. High heritability estimates were observed for OY, BNO and FP, which all relates to high FFB yield. Therefore, these traits should be prioritised in selecting high-yielding palms. Four families, namely PK 4044, ECP HP 496, ECP HP 500, and ECP HP 502, were identified to have high FFB yields (205.5 to 214.4 kg palm−1 year−1), which could be attributed to their high BNO (14.6 to 20.2 bunches palm−1 year−1) and moderate ABW (10.3 to 14.7 kg). Families ECP HP 496 and ECP HP 500 produced the highest OY at an average of more than 9 t ha−1 year−1, outperforming the national average of 3.7 t ha−1. Family ECP HP 496 also had an advantage in terms of low HT, which was 38% shorter than current recommended planting material in Malaysia, a characteristic that will help to ease FFB harvesting and extend its economic lifespan. Parental palms of these families will be utilised for breeding of high-yielding planting materials. Additionally, superior individual palms from these families can be selected as ortets for clonal propagation of high-yielding commercial clonal planting material. However, since the present study was carried out in only one location, it is important that the identified high-yielding families undergo multiple location testing to verify their suitability under different environmental conditions prior to being recommended for commercial planting. In the interim, using the high-yielding planting material identified in the present in conjunction with good agricultural practices will help increase the national oil yield and contribute to the advancement of the oil palm industry.