Morphological Variability Identification of Harumanis Mango (Mangifera indica L.) Harvested from Different Location and Tree Age

Harumanis is one of the main signatures of Perlis with regards to its delightful taste, pleasant aroma and expensive price. Harumanis authenticity and productivity had become the remarks among the farmers, entrepreneurs, consumers and plant breeders due to the existence of morphological characteristics variation among the fruits and high production cost. Assessment of Harumanis morphological characteristics of natural population and different tree ages may represent a possible source of important characteristics for development and breeding purposes of Harumanis. The aim of this study is to evaluate the morphological variation of Harumanis collected from different location in Perlis and tree age. A total of 150 Harumanis fruits from 50 trees with three different stages of development (young, middle-aged and old) were characterised using 11 traits; 10 quantitative and one qualitative morphological trait. The ANOVA analyses in combination with Dunn’s pairwise and Kruskal-Wallis multiple comparison test able to point out the existence of environmental factor and age influence towards the significant different of identified morphological traits except for Total Soluble Solid (TSS) and pulp percentage. Five clusters of 50 Harumanis accessions reflect a grouping pattern which not according to neither geographical region nor age. The result of Principal Component Analysis (PCA) using the first two principal components (PCs) provided a good approximation of the data explaining 84.09% of the total variance which majorly contributed by parameters of weight, fruit dimensional characteristics, peel percentage and hue angle, h. Preliminary screening of important morphological characteristics which contribute to the phenotypic diversity of Harumanis is successfully achieved. The findings can be employed by the plant breeders and farmers for the establishment of standard grading of Harumanis and advancement of breeding crop of Harumanis in future.

Abstract: Harumanis is one of the main signatures of Perlis with regards to its delightful taste, pleasant aroma and expensive price. Harumanis authenticity and productivity had become the remarks among the farmers, entrepreneurs, consumers and plant breeders due to the existence of morphological characteristics variation among the fruits and high production cost. Assessment of Harumanis morphological characteristics of natural population and different tree ages may represent a possible source of important characteristics for development and breeding purposes of Harumanis. The aim of this study is to evaluate the morphological variation of Harumanis collected from different location in Perlis and tree age. A total of 150 Harumanis fruits from 50 trees with three different stages of development (young, middle-aged and old) were characterised using 11 traits; 10 quantitative and one qualitative morphological trait. The ANOVA analyses in combination with Dunn's pairwise and Kruskal-Wallis multiple comparison test able to point out the existence of environmental factor and age influence towards the significant different of identified morphological traits except for Total Soluble Solid (TSS) and pulp percentage. Five clusters of 50 Harumanis accessions reflect a grouping pattern which not according to neither geographical region nor age. The result of Principal Component Analysis (PCA) using the first two principal components (PCs) provided a good approximation of the data explaining 84.09% of the total variance which majorly contributed by parameters of weight, fruit dimensional characteristics, peel percentage and hue angle, h. Preliminary screening of important morphological characteristics which contribute to the phenotypic diversity of Harumanis is successfully achieved. The findings can be employed by the plant breeders

INTRODUCTION
Harumanis is a fruit belongs to Mangifera genus which also known as MA 128 (Jabatan Pertanian Malaysia 1995) and has been registered on 28 May 1971 by the Department of Agriculture, Malaysia (Aziah 2006). This variety is known for its pleasant taste, texture and fragrance among consumers and is popular both in the local and international markets (Jha et al. 2013). These sensory parameters vary with individual mango and person involved in testing. There is little or no information available on the instrumental estimation of these sensory attributes. The present study, therefore, was conducted to correlate sensory and instrumental textural attributes to explore the possibility of predicting them for seven major cultivars of mango influenced by harvesting dates and ripening period. The fruit shape of Harumanis is oblong with prominent beak, while the skin shading is green and usually will turn to yellowish green when ripe. The fruit size varies, ranging from 300 to 650 grams, has a 16° to 17° Brix value and the flesh colour is usually orange and sweet (Sani et al. 2018).
Despite its appealing qualities, Harumanis is facing a problem which related to the production inconsistency due to phenotype variation which caused the flesh colour and sizes of the fruits to be varied although they are harvested from the same cultivation area. This situation is worsening as some other mangoes like Sala and Tong Dam look almost similar and leads to confusion about clear identification of this exceptional mango. The current grading system of Harumanis is via visual morphological inspection and the technique is also used for various mangoes including Sala, Chokanan and Maha (Chee 2009;Jusoh 2017). This has been a major drawback among consumers since they can misjudge other mangoes as Harumanis and buy them at expensive price (Bakhtiar 2011). Eventually, such concerns will give a negative consequences to the markets, especially in term of revenue and prominence of Perlis as the main producer of Harumanis in Malaysia (Jusoh 2017).
Morphological marker has been classified as one of the factors that play a major role in plant breeding and can be applied to pre-determine the agronomic traits of interest. Identification of variation and traits of interest, followed by its incorporation into germplasm are the influential components of any crop advancement for plant breeding programme. Such variation can be obtained from either crossing two distinctive parental genotypes or choosing existing variations from the highly available germplasm from natural populations (Rajeev et al. 2009). So far, very few reports have been published with regards to the evaluation of genetic and phenotypic diversity of Harumanis in this country with limited number of morphological traits being studied (Ariffin et al. 2015;Rahman et al. 2018;Sani et al. 2018;Yusuf et al. 2018a;2018b). The systematic study on morphological variation of Harumanis fruit samples collected from various location and tree age in Perlis has never been reported previously.
The presence of intra-varietal among the fruits is usually caused by various factors. It is probably due to phenotypic variation caused by spontaneously arisen genetic polymorphisms from their relatives and ancestors that are maintained by nature, adaptation to different natural environment and human preferences (Kliebenstein 2009). Furthermore, the development factor such as type of rootstock and scion may also cause the variation (Vieira et al. 2007). Therefore, having abundant choices of parents from different location for population development mapping and extending the mapping with the fruits' quantitative and qualitative characteristics by using morphological marker will be an effective method for selection of superior progenies. The parent crops with traits of interest such as higher yield, improved nutritional quality and appropriate colour or fragrance preferred by consumers can be singled out and be used as the preferred crop for breeding purposes.
Other than environmental effect, tree age also affects the nutritive properties, storage and physiology of harvested fruits (Meena & Asrey 2018). The tree age factor affects the pomological parameters of the olives and the physiochemical characteristics of the virgin olive oil (Bouchaala et al. 2014). Young tree of orange was reported to produce low quality of fruits compared to the older trees (Hearn 1993). The tree age also reported to affect the fruit yields and physical characteristics of pummelo (Nakorn & Chalumpak 2016) and apple fruits (Arshad et al. 2014). Ozeker (2000) reported that a 20-years-old "Marsh seedless" tree produced heavier grapefruits with higher juice content, soluble solids and acidity compared to fruits from a 34 years old tree.
Overall assessment of the important and highly potential mango traits needs to be analysed to be further utilised for Harumanis crop breeding improvement. Thus, this study is the first attempt for the assessment of Harumanis variability based on the morphological traits from different location in Perlis and tree age using morphological methods by performing statistical analyses by ANOVA and Kruskal-Wallis test in combination with Dunn's pairwise multiple comparison test, clustering analysis by Ward's method and squared Euclidean distances with the combination of PCA. Further identification of morphological key descriptors will be helpful in assisting the selection of desirable individuals with high commercial morphological values.

Study Area
A total of five different orchards around Perlis were selected for the study. The orchard selection was based on several criteria such as dispersion of locations which covering north, south and east of the state, different tree ages which categorised as young (5 to 10 years old), middle-aged (15 to 20 years old) and old (33 to 35 years old) (Khalid et al. 2017;Meena & Asrey 2018) and farmers who are certified with MyGAP (Malaysian Good Agricultural Practices). The location of each of the sampled trees was recorded using a handheld global positioning system (GPS) along with location information of the surveyed trees (Table 1). A random sample of 50 Harumanis mango trees were selected as stated in Table 1 for characterisation purpose.

Plant Materials
Fruit sampling and morphological characterisation were conducted during Harumanis harvesting season from March till June 2017. Triplicate of mangoes per accessions were hand-harvested as they reached maturity (13 weeks after flowering) and further undergo a post-harvest process prior analysis (Khidir & El Tayeb 2014).

Morphological Descriptors and Statistical Data Analysis
The morphological diversity observation within the studied population was determined by a total number of 10 quantitative and one qualitative scored morphological trait as indicated. The 10 quantitative traits analysed are fruit weight, length, width, thickness, peel firmness, flesh firmness, total soluble solid (TSS), peel percentage (% (w/w) of total fruit), pulp percentage (% (w/w) of total fruit) and stone percentage (% (w/w) of total fruit) while hue angle is the only qualitative trait. TSS was obtained by measuring the top, middle and bottom part of the pulp juice of the equatorial region of fruit using an Atago Digital Refractometer. The TSS was measured at different positions of the fruits (top, middle and bottom) as sugar levels often vary within the fruit, being higher at the stem-end and lower at the calyx-end. Thus, the value of the TSS is recorded as a mean from each position's reading in each sampled fruit (Jha et al. 2010). For flesh colour, the internal colour reading was measured using a portable colorimeter   The colour coordinate was able to show the variation between the basic colours. Minimum, maximum, mean and coefficient of variation (CV) were computed for each trait by using Excel 2013. The CV was determined as the variability indicator. The CV percentage was determined as the variability index (Markovski & Velkoska-Markovska 2015). The ANOVA, clustering, dendrogram and PCA were performed using SPSS and IBM 19.0 and Minitab 16. Comparison of traits heterogeneity among the accessions across location and tree ages (Meena & Asrey 2018) were performed by ANOVA followed by Tukey post hoc test for traits across categories (tree age: length, thickness and peel firmness and location: weight, length, width, thickness and peel firmness). Kruskal-Wallis test followed by Dunn-Bonferroni pairwise comparisons test were preferred for the remaining traits across both categories since the distributions are not symmetric (Bing et al. 2018). Where k is the number of groups, n i is the sample size of sample I, N is the total number of all the observations across k groups, r ij is the rank (among all the observations) of observation j from group i.
These tests were performed to test the hypotheses of the distribution of morphological traits mean (H 0 : µ U = µ E ) against the two-tailed alternative (H 1 : µ U ≠ µ E ) for ANOVA if normality is assumed or if normality is not assumed median (H 0 : median U = median E against H 1 : median U ≠ median E ) for Kruskal-Wallis test if the distribution of the accessions are not same across both categories of age and place depending on the result of a Shapiro-Wilk normality test (Dunn 2013).
Correlations between traits were computed using Spearman's correlation coefficient and p-value correlation matrix. The distances between the traits were then computed using the squared Euclidean norm based on similarity levels, using standardised data. A hierarchical cluster analysis was conducted in order to group the samples according to their morphological similarities. The patterns of the main variations and the most discriminant traits were analysed using PCA. Clustering was performed after Z-score standardisation of qualitative and quantitative variables using squared Euclidean distances and the Ward method (1963) by using only the selected key descriptors which present significant different across categories (location and place) and have high correlation between each other (Abdi & William 2010).

RESULTS
Results of this study were divided into sections which comprised of fruit characterisation by performing descriptive statistics, analyses of variance, pairwise comparison, PCA and clustering analysis.

Fruit Characterisation
Descriptive statistics i.e., mean, range, variance, standard deviation and coefficient variation results for 11 morphological traits are presented in Table 2. In fruit characterisation, high variations of few morphological descriptors were found among the 50 Harumanis accessions. The fruit weight varied extensively from 263.00 g to 670.33 g. Fruit size varied among the samples from 10.00 cm to 14.31 cm in length, 6.07 cm to 8.45 cm in diameter and 5.63 cm to 7.69 cm in thickness. Peel firmness ranged between 3.05 N s -1 to 7.24 N s -1 , flesh firmness ranged between 0.70 to 2.93 N s -1 and the Brix values of the fruits ranged from 14.6° to 18.23°. The hue angle varied from 68.09° to 81.09°. For mass composition, the highest peel, pulp and stone composition are 18%, 74% and 19%, respectively. Whereas, the lowest peel, pulp and stone percentage are 10%, 53%, and 9%, respectively. The hue angle varied from 68.09° to 81.09° as indicated in Table 2. Any morphological trait showing CVs greater than 20.00% indicates a high variability between the genotypes (Norouzi et al. 2017). The highest variations identified are flesh firmness (34.47%) and fruit weight (20.62%) followed by peel firmness (18.16%), peel percentage (14.94%) and stone percentage (14.23%). These high CV values indicate the existence of high range of selection for those traits (Mohammadi & Prasanna 2003). The lowest CV value is shown by the hue angle with a CV of 4.31%. The morphological traits with low variation are more homogeneous and can be repeated between accessions, and therefore may be considered as a stable feature (Khadivi 2018).

ANOVA and Pairwise Comparison
All of the data set were analysed using Kruskal-Wallis test due to the violation of normality assumption, except for fruit length, width, thickness and peel firmness were analysed using ANOVA across locations. Out of 11 morphological traits, nine morphological traits which are fruit weight, length, width, thickness, peel firmness, pulp firmness, peel percentage, stone percentage and hue angle showed a significant difference (p < 0.05) across different geographical region and tree ages from ANOVA and Kruskal-Wallis test. For location, there is a statistically significant effect of location on fruit weight, length, width, thickness and peel firmness at the  (Tables 3 and 5).
The results in a probability value which below the α level (p < 0.05) resulted the rejection of null hypotheses which means that there is presence of differences in the parameters across the categories (location and tree age), Tukey post hoc test in Table 4 and post-hoc Dunn's test in Fig. 1 with Bonferroni correction were performed to determine which means or median amongst a set of means or median differ from the rest (Bland & Altman 1995). Notice in Table 4, post hoc comparisons using the Tukey HSD test indicated that the mean score for Harumanis Santan (M = 151.567, SD = 33.319) weight is statistically significant different in positive effect at the p < 0.05 level when compared with Harumanis Simpang Empat. For length, the means difference was statistically significant in negative effect when Harumanis Simpang Empat became the (I). The length difference was slightly lower compared to Harumanis Chelong Balik Bukit, Paya Kelubi and Santan. Another fruit dimensional characteristic i.e. fruit width showed statistically significant difference in negative effect again between Harumanis Simpang Empat when compared to Harumanis Chelong Balik Bukit, Paya Kelubi and Santan. Nevertheless, only the comparison between Harumanis Simpang Empat and Harumanis Santan presented a significant difference at p-value of less than 0.05. Harumanis Paya Kelubi showed a positive mean difference of length trait when compared to Harumanis Alor Ara Timur with a mean score of (M = 0.928, SD = 0.313) indicated a greater length of Harumanis Paya Kelubi than Harumanis Alor Ara Timur. Meanwhile, for thickness, this test revealed that the thickness of Harumanis Simpang Empat is lower compared to thickness of Harumanis Chelong Balik Bukit and Harumanis Santan. Both showed statistically significant difference which implied by the positive mean difference of (M = 0.904, SD = 0.1524) when compared with Harumanis Chelong Balik Bukit, while the mean score when compared with Harumanis Santan was (M = 1.032, SD = 0.152) as these both locations became the (I). Harumanis Alor Ara Timur scored a negative mean difference (M = -0.488, SD = 0.152) of thickness when compared to Harumanis Santan with a significant difference of p < 0.05.
The results of Kruskal-Wallis rank sum test determined whether the morphological traits of peel percentage and hue angle are varied across different locations. According to Table 5, the Harumanis accessions peel percentage median of the five locations were 14.50%, 12.00%, 13.50%, 12.50% and 15.50%. The average rank exposed that Harumanis Simpang Empat group median was the highest, while Harumanis Chelong Balik Bukit group median was the lowest. This relationship can be clearly seen when performing the Dunn-Bonferroni pairwise comparison test by comparing the boxplot as illustrated in Fig. 1. Harumanis Simpang Empat attained the highest peel percentage because the boxplot displayed the highest median and Harumanis Chelong Balik Bukit attained the lowest peel percentage as the median was the lowest when compared to the peel percentage attained by Harumanis accessions from other locations. For hue angle, the Kruskal-Wallis test presented a significant result which indicated by the p-values less than 0.05 (Table 5). By referring to the hue angle boxplot, Harumanis accessions of Paya Kelubi acquired the most dense flesh colour due to the lowest median among the groups and the median was near the upper quartile ( Fig. 1).
Kruskal-Wallis test followed by Dunn-Bonferroni pairwise comparisons test was used in multiple comparisons to analyse the variation pattern of morphological traits of Harumanis across different age categories of young (5 to 10 years old), middle-aged (15 to 20 years old) and relatively old (33 to 35 years old) as the result violated the normality assumption as indicated by Shapiro Wilk test. Out of 11 morphological traits, seven traits presented significant differences with the p-value less of than 0.05 as indicated in Table 6. These results enabled the rejection of null hypotheses of equal distribution of parameters across tree age.     According to Table 6, the fruit weight average rank showed that the Harumanis accessions from middle-age (15 to 20 years old) trees differed the most amongst all observations. The highest Z value was attained by the Harumanis accessions from middle-age (15 to 20 years old) trees, which is 1.44. The positive Z value indicated that the Harumanis accessions of middle-age (15 to 20 years old) trees average rank were greater than the overall average rank. Table 6 indicated that the Harumanis accession from old trees presented lower median (M = 451.8 g) compared to middle-age (15 to 20 years old) trees and presented the most diverse fruit weight as the box plot was comparatively tall. The young (5 to 120 10 years old) tree of Harumanis accession are having smallest size due to lowest median value attained which was 361.5 g. Both young (5 to 10 years old) and middleage (15 to 20 years old) trees attained similar length for upper and lower whisker (Fig. 2), indicated the variation pattern of the fruit weight is similar in the age categories whilst, the whisker extended longer for old trees (33 to 35 years old) indicated the fruit weight for old tree Harumanis accession were varied more for young (5 to 10 years old) and middle-age (15 to 20 years old) tree Harumanis accession.
The fruit dimensional characteristics, fruit length, width and thickness showed similar pattern for those parameters as the Harumanis accessions from middle-aged (15 to 20 years old) trees presented the highest median and the lowest median was obtained by the Harumanis accessions from young trees (5 to 10 years old) (see Table 6 and Fig. 2). In spite of that, the young trees held the tallest box plot comparatively indicated the most diverse width and thickness.
The dimensional characteristic of fruit length exposed a contradiction finding as Harumanis fruit from young (5 to 10 years old) trees acquired the lowest median, followed by middle-aged (15 to 20 years old) trees and relatively old (33 to 35 years old) trees. For length, the lowest average rank of 11.8 cm was scored by young (5 to 10 years old) tree Harumanis accessions and the negative Z-score of -4.77 indicated that the group average rank was less than normal average rank. The box plot (Fig. 2) indicated by this group was comparatively short suggested that the overall accessions had lower differences of length with each other. Meanwhile, both middle-aged (15 to 20 years old) and old (33 to 35 years old) Harumanis trees presented not much differences by obtaining median scores of 12.35 cm and 12.41 cm, distinctively. The width box plot for middleage trees shows uneven size of the sections indicating uneven distribution of Harumanis accessions in this group into two different scales (Fig. 2).
The box plot illustrated a longer whisker in negative direction for both young (5 to 10 years old) and middle-age (15 to 20 years old) trees in fruits width denoting distribution with a negative skew, while the old (33 to 35 years old) trees attained longer whisker in positive direction suggesting a distribution with a positive skew. Fig. 2 illustrates a comparatively short box plot for thickness presented by middle-age (15 to 20 years old) trees which suggested a low variety of thickness between the accessions, while both young (5 to 10 years old) and old (33 to 35 years old) trees exposed a comparatively tall box plots indicated the accessions from these age categories are having quite differ thickness among each other.  Peel firmness revealed the highest median of 6.420 N s -1 for Harumanis accessions from young (5 to 10 years old) trees whereas the old (33 to 35 years old) trees showed the lowest median of 4.916 N s -1 . The Harumanis accessions from middle-age (15 to 20 years old) trees present a slightly higher median than the median of old (33 to 35 years old) trees. These results revealed that the highest peel firmness was acquired by Harumanis accessions from young trees (5 to 10 years old). This variation pattern is in line with the peel percentage findings as the young trees (5 to 10 years old) of Harumanis accessions are again having the highest median score and average rank (M = 15.00%; Ave Rank = 37.4). Nevertheless, the Harumanis from middle-age (15 to 20 years old) and old (33 to 35 years old) trees presented a not much different medians indicated similar peel percentage attained by the Harumanis from both categories.
Among the three different age categories, old (33 to 35 years old) trees presented the least median value of hue angle which was 74.50 indicated the most dense flesh colour attained from this group of Harumanis accessions. The old (33 to 35 years old) trees obtained a negative Z-score of -3.82, suggesting that most of the accessions scored below the average rank (Table 7). Based on the box plots, both Harumanis accessions from young (5 to 10 years old) and middleage (15 to 20 years old) trees illustrated a longer lower whisker which exhibited that most of accessions from these two categories were varied in negative quartile group.

Correlation Analysis
The bivariate Spearman's correlation coefficient indicated by Table 7 reveals a significant correlation between fruit dimensional characteristics, peel percentage and stone percentage as indicated in Table 7. Strong positive linear correlation (p < 0.05) were found between fruit weight with length (r s = 0.896), width (r s = 0.909) and thickness (r s = 0.896). Meanwhile, negative linear correlation were found between the peel percentage by weight (r s = -0.468), length (r s = -0.567), width (r s = -0.519) and thickness (r s = -0.519). Likewise, stone percentage also presented a negative linear correlation with weight (r s = -0.472), length (r s = -0.572), width (r s = -0.606) and thickness (r s = -0.579). There is also a weak significant correlation between fruit brix and hue angle with r s = -0.300 (p < 0.05) as the r value was distant from 1.   Table 7; * Correlation is significant at the 0.05 level. ; ** Correlation is significant at the 0.01 level

Principal Component Analysis (PCA)
A set of six diverse morphological traits were selected from the 11 traits, namely, weight, length, thickness, width, peel percentage and hue angle to be analysed using PCA. Julnes (2012) suggested PCA procedure to be carried out for multivariate continuous data where multicollinearity exist among variables with the absent of outliers. These assumptions have been checked and fulfilled before PCA procedure was implemented. Table 8 illustrates the total variation explained by each PC as well as the variables' contribution accordingly. However, the eigenvalues were used to select factors that discriminate the best between the entries. As a criterion to extract the main PCs, Kaiser criterion (eigenvalue greater than 1.00) was referred (Khadivi 2018). One significant component with eigenvalue greater than one was obtained that explained 67.72% of the total variation. However, by referring to the loading value, two PCs which summed up to 84.09% were chosen to explain the variation as loading value above 0.53 was considered as significant as mentioned in previous study (Ganopoulos et al. 2016). According to Table 8, the first PC has strong positive association with weight, length, width and thickness with total variance explained of 67.7%. The second PC accumulated 16.4% of variability and largely associated with hue angle and peel percentage. Loading plot indicated in Fig. 3 presents the coefficients of each variable for the first PC versus the coefficients for the second PC. The first PC in the horizontal direction presented by four morphological traits namely fruit weight, length, thickness and width, while the second PC summarised the two variables of peel percentage and hue angle. The loading plot of the first PC consists of morphological variables like fruit weight, length, width and thickness are close to 1.00. Meanwhile, in the second factor, peel percentage loading value is close to 0.5 and hue angle loading is close to -1.00. The loading plot able to illustrate the strong influence of variables towards the PC except for peel percentage as this trait showed the least influence towards the PC.
A PC biplot constructed shows the super imposed of the variables on the plot as vectors. This biplot graph (Fig. 4) was comprised of the loading (Fig. 3) and score plot (Fig. 5). It was used to compare the different accessions diversity (Habibpour et al. 2012). According to the graphs (Figs. 4 and 5), Harumanis accessions such as HM-Acc 10, HM-Acc 12, HM-Acc 16, HM-Acc 30 and HM-Acc 50 were positioned at the vertex of polygon and are the farthest from the vertex point of origin (Figs. 4 and 5), hence they are more diversified than others. Meanwhile, HM-Acc 4, HM-Acc 15, HM-Acc 27 and HM-Acc 27 were positioned very near to each other and very close to the point origin, hence they are suggested to be less diversified.

Clustering analysis
The clustering analysis was characterised using fruit weight, thickness, width, length, peel percentage and hue angle as shown in Fig. 6 and Table 9. Ward's dendrogram reflects the similarities and differences between Harumanis accessions according to the selected morphological characteristics which revealed maximum proportion of variation among the accessions. The 50 Harumanis accessions were mainly divided at the third node into five clusters with 10 (cluster 1), 8 (cluster 2), 9 (cluster 3), 11 (cluster 4) and 12 (cluster 5) accessions in different groups (Table 9 and Fig. 6). It was observed that the 50 Harumanis accessions were highly distributed into particular groups with no apparent location division. This can be clearly observed as multiple accessions were collected from different location such as HM-Acc 3 (Chelong Balik Bukit), HM-Acc 23 (Alor Ara Timur), and HM-Acc 40 (Simpang Empat) which were clustered together in the same group (cluster 4). Three clusters (cluster 1, 2 and 3) were comprised of fruits with similar fruit size of small and medium sizes. These clusters were differ distinctively in the flesh colour and peel percentage. Cluster 2 and cluster 4 fruit accessions reveal same two colour sets of fruit flesh (orange-yellow and yellow-orange), while cluster 1 was represented by one colour set of flesh color (orange-yellow). For peel percentage, highest peel percentage (14% to 18%) was observed in cluster 1 accession, whereas fruits in cluster 2 and 4 are having medium peel percentage value (cluster 2: 13% to 16%; cluster 4: 12% to 15%). Among five clusters, cluster 3 indicates Harumanis accessions with large size (> 500g), lowest peel percentage range (11% to 14%) and flesh with two different set of colours (orange and yellow-orange). Accessions in cluster 5 presented medium fruit size with the most diverse range value of peel percentage which is in between the minimum peel percentage till the maximum peel percentage (10% to18%). Table 9: Minimum and maximum value of five quantitative and one qualitative key descriptors used for morphological classification of 50 Harumanis mango samples separately for the five identified clusters.

DISCUSSION
Fruit characteristics were identified to have the strongest discriminating power in the previous studies and are useful for identifying particular landraces or morphological types (Sennhenn et al. 2014). CV among the Harumanis for several morphological traits such as fruit dimensional characteristics and flesh firmness showed high dispersion of the mean values and thus indicated high variability between the individuals ( Table 2). The phenotypic diversity may governed by geographical location of the accessions as reported by previous finding of Ziziphus fruits (Markovski & Velkoska-Markovska 2015). Such findings are in line with the previous reports as indicated by Begum et al. (2013;. The results of morphological traits like dimensional characteristics, peel percentage and hue angle variations were robustly linked at the margin of statistical significance (p < 0.05) to the different locations and tree age as indicated in (Tables 3, 4, 5 and 6; Figs. 1 and 2). Further, the result is explained by employing the ANOVA in combination with Tukey post hoc test (Tables 3 and  4) and Kruskal-Wallis in combination with Dunn's pairwise tests (Tables 5 and 6; Figs. 1 and 2) to point out the existence of morphological traits divergence across the two categories. These findings revealed the possible influence of distinct environmental factors such as human intervention ), pollinators (Sánchez-Lafuente et al. 2005 and climate (Pregitzer et al. 2013), while the different age categories may associated with different level of constituents such as canopy size, number of leaves, respiration rates, enzymatic activities, antioxidant activities, organic acids contents and mineral contents (Meena & Asrey 2018). This finding is supported by an earlier study of environment interactions of plant and soil which reported the significance of environmental factors towards the mediation of plant morphological characteristics (Pregitzer et al. 2013). The environmental factors which attributed by the different locations and tree ages (plant structure and function) are inter-correlated. The factors such as light intensity, temperature, carbon availability, water availability influenced the plant system and further affecting the fruit growth specifically on its productivity (Singh et al. 2014).
Taken together, the result showed that Harumanis accessions from Simpang Empat acquired the smallest fruit weight and size when compared to the Harumanis accessions from Chelong Balik Bukit, Paya Kelubi and Santan with an exception of Harumanis Alor Ara Timur as the post hoc comparisons failed to give a significant result (p < 0.05) for any dimensional characteristic between these two groups. Harumanis Chelong Balik Bukit, Harumanis Paya Kelubi and Harumanis Santan attained bigger size and fruit weight whilst, Harumanis Alor Ara Timur was classified as medium fruit size. The data of present study shows that Harumanis fruits of older trees (33 to 35 years old) were relatively bigger compared to fruits of young (5 to 10 years old) and middle-aged (15 to 20 years old) trees as indicated in Table 6. Harumanis accessions collected from Simpang Empat and Alor Ara Timur are having smaller sizes as trees from both areas are categorised as young (5 to 10 years old) and middle-aged (15 to 20 years old).

132
One of the possible factor is the larger canopy acquired by the older trees which lead to the optimum level of leaf-to-fruit ratio which enable the optimum uptake of the carbon sources, which further enable fruits to reach their optimum fruit size (Chacko et al. 1982;Reddy & Singh 1991). Contradictorily, young trees usually have smaller canopy, lesser number of leaves and all of these factors contributed to the lower photosynthesis rates in youth unexpanded leaves to obtain enough carbon sources and contribute to smaller size of fruits (Nakorn & Chalumpak 2016).
Other than that, previously it was stated that fruit sizes appeared to be controlled by multiple genes without complete dominance. At the same time, it is also concluded that fruit sizes were probably an expression of additive gene effects. This inheritance factor can be one of important potential factors contributing to the variation existence in Harumanis fruits sizes as indicated by descriptive statistic result for fruit weight which presented a high CVs of 20.69% which was greater than 20%. The value suggesting a great extent of fruit size variation as this trait was strong positively correlated with the fruit dimensional characteristics such as length, width and thickness ( Table 2). The inheritance factor contribution was strongly supported by the previous findings as it was noticed that high degree of genotypic coefficient variation (GCV) along with phenotypic coefficient variation in the traits like fruit weight, fruit volume, pulp: stone ratio and total carotenoids. Hence, indicated high heritability along with high genetic advance estimated for the fruit weigh and fruit volume. A study on genotypic and phenotypic variability in mango indicated high value of fruit weight (85.20%) and thickness (83.12%) heritability. The higher heritability indicated that either these characters were simply inherited characters governed by a few major genes or it is an additive gene effect. If this trait confirmed to be under polygenic control, selection of this trait will be effective for crop improvement (Majumder et al. 2012).
In the correlation coefficient analyses, only a few parameters are considered due to the submission towards the Spearman's assumption. From the findings, fruit dimensional characteristics, peel percentage and hue angle were selected as the most valuable key descriptors for the analyses due to the existence of significant differences across both categories (location and tree age) in which there are high correlation between the parameters and the variation composition of the Harumanis morphological traits can be explained. The fruit weight shows a strong significant positive linear correlation with fruit length, width and thickness. Meanwhile, brix values show negative correlation with hue angle. These findings are consistent with the previous results in which a strong positive correlation coefficient was also obtained between those traits in three species of Ziziphus genus (Ziziphus nummularia, Ziziphus spina-christi and Ziziphus oxyphylla) (Norouzi et al. 2017), sour cherry (Prunus cerasus) (Ganopoulos et al. 2016) and Elaeagnus angustifolia (Khadivi 2018). These significant correlations observed between phenotypic traits which contribute to fruit yield and quality will be so beneficial for any future plans of breeding programme (Ganopoulos et al. 2016. It also able to equip information on valuable traits in the evaluation of any accessions (Khadivi 2018).
According to Meena and Asrey (2018), higher respiration rate shown by fruits in older trees may attribute to higher peel thickness and peel percentage as both parameters have same palatability effects. The report opposed the findings of this study as the fruits of middle-aged (15 to 20 years old) and relatively old (33 to 35 years old) Harumanis trees attained lower peel percentage content compared to the peel percentage of young trees (5 to 10 years old) trees which attained the highest peel percentage. This result can be well analysed by performing the bivariate Spearman's correlation coefficient, as it clearly able to explain the parameters correlation as the result showed a strong negative linear correlation between the peel percentage with weight (r s = -0.468), length (r s = -0.567), width (r s = -0.519) and thickness (r s = -0.519) ( Table 7). These can be observed in the results of Harumanis fruits from orchard in Simpang Empat as it attained the lowest weight, thickness and highest peel percentage despite its young (5 to 10 years old) tree ages as presented by Tables 3, 4, 5 and 6. Other environmental factor such as human intervention which associated with the farming management system may attribute to these opposing results. Earlier, such finding was also reported on 'Amrapali' mango and one of the main factors identified is the lower level of potassium (K) in fruits from young trees which in turn disturbed the cell arrangement and resulted in higher peel thickness or peel percentage is mainly influenced by cell number, cell size and their arrangement in fruit tissue (Meena & Asrey 2018). Thus, a deep study dealing with some important chemical and biochemical aspects during both fruit development and ripening is needed.
The negative correlation (r s = -0.300 , p < 0.05) in Table 7 between fruit brix value and hue angle able to point out a non-linear relationship between these two variables. The fruit brix value will increase as the hue angle value decreases, in which brix value contributes to the sweet taste of the fruit and the lower value of hue angle may prone to orange colour region. A few previous works have addressed the non-linear behaviour of these two parameters especially on the climacteric fruits (Jha et al. 2010;Soltani et al. 2010). The findings were in line with previous study in which reported by Khan et al. (2015) as the mango pulp usually has a sweet taste and as soon as the fruit ripens the colour of pulp changes from yellow to orange. Earlier, it was stated that the regulation of these two parameters were associated with ethylene production especially for mango, as it was identified as climacteric fruit species with a well-characterised peak in ethylene production and respiratory activity at the onset of ripening (Seymour et al. 1993;Tucker 1993;Adams-Phillips et al. 2004;Zaharah et al. 2011). This burst in ethylene production stems from an up-regulation of ethylene biosynthesis genes at the onset of ripening, resulting in autocatalytic ethylene production and an up-regulation of the components of ethylene perception and signalling (Klee & Clark 2004). These changes act as a key signal for the initiation and coordination of ripening in all climacteric fruits, and have been shown to regulate key genes that control colour change, fruit softening, cell wall breakdown, pathogen defence, and nutrient composition (Hobson & Grierson 1993;Tucker 1993;Alexander & Grierson 2002;Giovannoni 2004). These findings were supported by other study which stated that any biochemical and physiological changes during fruit ripening are literally driven by the coordinated expression of fruit ripening-related genes. These genes encode enzymes that participate directly in biochemical and physiological processes. It is also stated that the ethylene production within the fruit activates many other enzymes resulting in physiological changes such as the change of colour from green to red and the softening of the fruit (Hossain et al. 2014). Surprisingly, it has been reported that the absence or presence of a respiratory climacteric on the creeper depends upon prevailing environmental conditions (Bower et al. 2002). This observation is correlate with another finding which showed that the respiratory climacteric is most likely not an absolute activator of the ripening process, but secondary and resultant to the ripening procedure (Kaushlendra et al. 2016). However, the result of brix value was not statistically significant across the locations and tree age with a p-value more than 0.05. The results suggested insufficient evidence to reject the hypothesis which revealed no significant variation of the parameters across both the location and tree age. Thus, these findings explained that there are no influence of environmental factor and tree age towards the Harumanis brix values.
However, flesh colour showed different result as the hue angle were robustly linked at the margin of statistical significance (p < 0.05) to the different locations and tree age. Thus, the result suggested the influence of environmental factor and different tree developmental stages to the flesh colour variation. Earlier, light has been identified as one of the important elements which able to regulate carotenoid metabolism in plants (Zhang et al. 2015). For example, ripe 'Keitt' mango from Bahia, Brazil (hot climate) is reported to possess more than twice the β-carotene content than those from São Paulo (moderate climate) (Mercadante & Rodriguez-Amaya 1998). β-carotene has been emphasised as the main component that is interrelated with flesh colour and resulting in vitamin A values (Golob et al. 2012). Practically, light intensity was correlated with temperature, particularly in the open-air condition (He et al. 2018) and it was also demonstrated that the temperature was significantly linear to the light intensity (Gou et al. 2015). The carotenoid accumulation is in parallel with the intensity of the colour which is reflected by the hue angle, h value. In this study, Harumanis accessions from Chelong Balik Bukit and Paya Kelubi which cover the Chuping area are having the densest flesh colour (orange and orange-yellow) with a mean rank score of 22.20 and 8.40 for Dunn's pairwise test respectively. Previous study on surface meteorological temperature observation data (January till June 2017) obtained from the Malaysian Meteorological Department involving two regions which are Chuping (representing orchards in Chelong Balik Bukit, Paya Kelubi, Santan and Alor Ara Timur) and Kangar (representing orchards in Simpang Empat) shows a consistent and only a slight difference in the temperature between the two regions with an average temperature of 28.4°C for Kangar and 27.7°C for Chuping (Yusuf et al. 2018b). Thus, it can be postulated that the temperature of Perlis region is not a crucial environmental factor in the flesh colour of Harumanis fruits (Yusuf et al. 2018b). Thus, other than environmental factor, tree age may possibly influence the carotenoid accumulation.
According to the result, the highest carotenoid accumulation explained by the lowest h value are strongly exhibited by the Harumanis fruits of relatively old trees (33 to 35 years old) from Chelong Balik Bukit, Paya Kelubi and Alor Ara Timur orchards compared to fruits of young trees (5 to 10 years old from Alor Ara Timur and Simpang Empat orchards) and middle-aged trees (15 to 20 years old from Santan orchards). This finding is in line with a previous study which reported that the highest carotenoids content was recorded in the pulp of fruits harvested from old trees (30 years old) (Meena & Asrey 2018). The linear relationship between carotenoid content and flesh colour is also reported in few previous studies. The fruit pulp was prominently orange in such early varieties of Mankurad and Furtado mangoes, mid-late variety of -RC-MS-1 (Bemcorado selection) mango, and also in late variety of Bardez Mussarat mango, which indicated the source for higher carotenoids contents (Desai & Dhandar 2000). The vegetative phase which is predominant in the early age of the plant usually resulted in higher sodium (N) content and lowered carotenoids synthesis and thus giving negative association with colour development in mango (Oosthuyse 1993;Martinez et al. 1997). Higher accumulation of calcium (Ca) was discovered in the fruits of 6 years old tree and might have further reduced the senescence process and thereby slowing the rate of chlorophyll breakdown and carotenoids synthesis in young tree fruits (Tingwa & Young 1974;Tirmazi & Wills 1981). However, the research on mango morphological traits performances according to the different tree age and environmental factor is still inadequate. Thus, further analyses on the differences Harumanis performances of morphological traits from different tree ages and the impact of specific environmental influence towards the traits performance would be a good approach to understand the underlying mechanism of the significant differences between those parameters stated. Other than that, it would be better if the colour trait performance of the accessions is assessed individually because the existence of morphological significant different results also may be governed by the dominants genes or number of genes. It is because of, though there was some dominance of light yellow colour to orange, the gene action was primarily additive both within and among loci for pulp colour (Iyer 1991). A few previous researches on the association of gene with the traits already performed and the findings showed significant association of gene loci with six traits of mango such as bloom, pulp colour, branch habit, ground skin colour, blush intensity and beak shape (Bally & Dillon 2018).
Further, those variables have been reduced by performing PCA analysis to reveal the general differences between the Harumanis accessions as numerical values and able to explain the variation among the individuals (Balkaya et al. 2010). The number of factors to be retained was determined according to the eigenvalues (Das et al. 2014). Therefore, in this analysis the first factor retains the information contained in 4.0638 of the original variables as given in Table 8. In this study, traits with high coefficients in the first and second (PCs) which are fruit weight, length, width, length, peel percentage and hue angle were considered important since these axes explain the biggest share of the total variation of 84.09% as indicated in Table 8. Though clear guidelines do not exist to determine the significance of a character coefficient, one rule of thumb is to treat coefficients > 0.6 as having a large enough effect to be considered as important (Jeffers 1967). The result is sufficient (84.1%) as it met a few of previous findings of PCA on mango which described the first ten components accounted for more than 86.07% of the total genetic variation in different mango varieties was accepted (Narvariya 2016) and the first six principal components axes took into account 75.12% of the total variance in the mango germplasm from the upper Athi river region of Eastern Kenya (Toili et al. 2016). In addition, the above-mentioned characters loaded high positively and negatively in the loading plots in Fig. 3, thus contributed more to the diversity and the selection of those characters is so effective as those traits able to differentiate the clusters greatly. In general, an identification of highly correlated morphological traits and the variation proportion contributed by selected parameters could be assessed by PCs analyses.
To better understand the overall diversity of the 50 Harumanis accessions, the data were analysed by cluster analysis that revealed the distribution of variety diversity and is displayed in Table 9 and Fig. 6. Categorising and characterising are necessary steps in the crop selection and breeding programmes (Markovski & Velkoska-Markovska 2015). The clustering analysis (Table 9 and Fig. 6) able to visualise the high distributional behaviour of the Harumanis accessions according to the selected morphological characteristics in the PCA analysis (weight, dimensional characteristic, peel percentage and hue angle). The clustering data shows that the Harumanis accessions were not grouped according to the location, opposing the previous reported by Sennhenn et al. in 2014. This finding was indicated by five clusters in Table 8 and Fig. 3, as multiple accessions collected from different locations were grouped in the same cluster such as Cluster 5, which consist of Harumanis accessions from Chelong Balik Bukit, Paya Kelubi, Alor Ara Timur and Santan. The 50 Harumanis accessions were effectively grouped together according to their close relationships associated with fruit morphological characteristics as indicated by the five clusters as shown in Table 9 and Figure  6. Overall, the clustering analysis results able to point out that the young trees are prone to produce fruits with small and medium sizes with a least dense flesh colour as the hue angle was quite high. These can be clearly seen in Cluster 2, as the cluster solely describe young tree performances. Meanwhile, middle-age and old trees mostly produced fruits with medium and large sizes with denser flesh colour as most of the accessions were grouped in the orange and orange-yellow colour sets.
Selection of morphological traits like fruit size and hue angle value would be effective for crop improvement study since they were possibly under polygenic control. However, previous study mentioned that the existence of variation in fruit weight under the same geographical region may be as a result of genotypic effect, cultivar and ecological condition (Markovski & Velkoska-Markovska 2015). In addition, since most of the plant characters of economic-important crops are governed by a group of genes and are highly influenced by environmental factor, hence it is difficult to judge whether the observed variability is heritable or due to environment. Thus, the optimisation of phenotypic variation into its heritable and non-heritable components for enhancement of genotypes collections is so important (Azam-Ali et al. 2006). Earlier, a study on the evaluation of the genetic variability for fruit quality in mango progenies showed the existence of genetic variability among progenies with differences in the progeny performance for the traits (Nayak et al. 2013).Thus, the individual analysis on the selected Harumanis which showed the desirable and economically traits would be a good resolution for future selection of potential parent and their subsequent prediction of progeny performance. Further, Harumanis crop breeding improvement can be assessed by selection of highly potential parent with highly commercial traits and comparing them against standard commercial varieties under replicate trials in combination with genetic study on the selected traits to confirm the distinctiveness and superiority. Therefore, the collection of desirable genotypes and a hybridisation programme may be initiated involving the evaluation of phenotypic diversity among genotypes as an important consideration for classification, utilisation of germplasm resources and breeding programme (Khadivi-Khub & Anjam, 2014).

CONCLUSION
Broad phenotypic diversity was detected within this cultivar with high distribution exists within the intra-cultivar of 50 Harumanis accessions from different locations and tree ages. This study, in overall, shows that different location which associated with environmental factors and tree age have robust significance on Harumanis morphological fruits variation. For instance, this study found that Harumanis fruits of older trees (33 to 35 years old) were relatively bigger compare to the fruits of young (5 to 10 years old) and middle-aged (15 to 20 years old) trees, whilst for flesh colour, the most dense flesh colour was exhibited by the Harumanis accessions of Paya Kelubi. The data further suggest that morphological traits (weight, dimensional characteristics, peel percentage and hue angle) are useful to be utilised for Harumanis descriptors establishment as these parameters able to differentiate the Harumanis cultivar effectively up to 84.1% total variation accounted in the PCA analyses. The current findings hopefully can be employed as a fundamental basis for the advancement of Harumanis crop breeding programme.