In this study, the optimum stage to truncate the dendrogram was based on the maximum genotypic variance (crG2), phenotypic variance (cr/) and heritability (h2). The maximum variances were either at stage 58 or at stage 124 for those 1 3 characters (Figure 3.6-3 . 1 8). The differences for heritability, and the genotypic and phenotypic variances between stage 58 and stage 1 24 were not significant, except in variance of genotype for seeds planrl and variances of phenotype for flowers planrl and seeds planrl (Table 3 . 1 1). Therefore, the stage 58 was defined the optimum cut-off point since the amongst-cluster mean squares was most significant relative to the within-cluster mean squares. Teow ( 1978) and Cullen ( 1 98 1 ) also used this truncation point (based on the minimum F probability for the ratio of
3. Meadowfoam 8 1
amongst cluster mean squares to within cluster mean squares). Lin and Butler ( 1 990) has reported that a cutoff point for the dendrogram was decided upon based on the F -ratio of the smallest dissimilarity index and the error estimate from ANOVA; that was, the cycle at which the calculated F-ratio exceeds the tabular F-value would be considered an appropriate cutoff point. Lin and Butler's method emphasized the internal (within group) homogeneity rather than the external (between groups) heterogeneity, thus, it may not be very suitable for grouping natural clusters, such as in taxonomy and our data set. A cutoff point was chosen based on an Euclidean distance (0.7) (Cerutti and Bigler, 1 995), and
pseudo F and t2 statistics calculated by SAS during the clustering procedure (Smith et al.,
1 995). Those methods, however, emphasized neither within group homogeneity nor amongst group heterogeneity.
On the basis of the results of pattern analysis, there were 190 phenotypes within 248 plants. Some phenotypes consisted of only one individual plant (Table 3 . 1 0), but the mean cluster-size for the rest was 7.8, 4.5, 2.8 and 2.0, respectively. Heterozygosity level for
L.
alba was found to be in the range of 12% and 27%, whereas percent polymorphic loci varied twofold (between 29 and 57%) among seven populations (Jain, 1 978). The average number of alleles per locus in
L.
alba was 1 .97 (Brown and Jain, 1 979). The population analyzed here could be expected to have considerable levels of heterozygosity, with diversity germplasm. In barley, the average numbers of alleles per locus were 5. 1 5, 2.75 and 1 .44, respectively for three stages of domestication, e.g wild barley, primitive Middle Eastern landraces and modem Californian cultivars (Allard, 1996). In transition from wild barley to promitive Middle Eastern landraces, numbers of alleles per locus decreased by about one half. During the domestication of maize, frequent (predominant) alleles contribute to adaptedness in many habitats and survive many cycles of selection in cultivation whereas alleles that are present in intermediate frequencies overall appear to be useful in some environments but not in others; rare alleles of allozyme loci appear to be of little value anywhere (Allard, 1 996). The data, therefore, provide strong evidence that plant breeding could lead to a reduction in allelic diversity (heterozygosity) during domestication of meadowfoam. The low levels of diversity found in the cultivated species compared with its wild progenitor has been recorded in many other cultivated plants (Doebley, 1 992).3. Meadowfoam 82
3.5.3 Genetic analysis
3.5.3.1 Genetic variances
A high level of significant genotypic effect of all characters (Table 3 . 1 2) proved the presence of considerable genetic variation for those attributes. The results suggested that meadowfoam has the genetic capability to respond to selection, but some characters may not evolve independently. Segregation with large variance in a progeny may increase the probability of progeny exceeding a critical value in selection. However, the variance
component cannot be separated from cr02 in case of one environment. Thus, cr02 may be
overestimated (Wricke and Weber, 1986) (because of the interaction between genetics and environment confounding).
The results showed that sampling errors associated with these estimates of the genetic contribution to phenotype were low (Table 3 . 1 2), but the errors in flowers planrl and seeds planrl after harvest were sufficiently large to be important in the design of future experiments.
3.5.3.2 Heritability
Heritability is a measure of the degree to which genetics contributes to the variation of a character within a population. The total genotypic variance may be partitioned into additive, dominance and epistatic variance components (Griffing, 1 956; Cockerham, 1 963; Baker, 1 986; Falconer, 1 989). Our results cannot distinguish between the contributions of dominance and epistasis to the broad-sense heritabilities because of a bee pollinator. Thus, the heritability estimates obtained in this study were broad sense.
The standard errors for these heritability estimates (Table 3 . 1 2) were of low value, indicating that the heritabilities were estimated with considerable precision. The precision of heritability can probably be achieved by either the appropriate and accurate measurement of characters or sufficient sample size (Hanshe et
al.,
1 966). The imbalanced experimental designs in this study may also improve the precision, because fewer3. Meadowfoam 83
experimental units are required for degrees of freedom to increase with imbalanced
relative to balanced designs (Knapp et aI. , 1 987). It is important that the limitations of
estimates of heritability are realised. While a mean is a first degree statistic, variance is a second degree statistic which is inherently less precisely estimated. Heritability, being the ratio of variances, not only shares all biological restrictions for estimating genetic
variances, but also shares the higher imprecision of second-degree statistics (Ewing et al.,
1 987).
The values of heritability varied from character to character in this study (Table 3 . 1 2).
Most characters had high heritability (h2 > 0.800), but for SETPC2 (early seed set) only
0.446. Chozin (1 990) reported that the broad sense heritability estimates of the meadowfoam lines varied in morphological characters (leaf hairiness, clump density, leaf size, budding time and flowering time), ranging from 0.01 to 0.72. He also found that the heritabilities for budding time and flowering time decreased from first self generation to second self generation, while the values for clump density and leaf size increased. Jain (1 979) reported that heritability estimates by standard mass selection theory (h2=�G/isp) ranged from 0.24 to 0.73 for flowering time in meadowfoam. The heritability estimates may imply that progress can be expected using proper selection techniques, though it may be slow in some character with low heritabilities, such as early seed set (SETPC2).
The variations of heritability in these characters were meaningful, because the experimental material was a wide gene pool drawn from 8 accessions (Table 3 . 1 ) and had diversity nature characters. These variations appeared clearly to depend much more on genes and therefore, they were potentially under selection. In general, selections directed towards the family or line performances would not be practical. Individual selection may be an effective and efficient method to improve meadowfoam genotypes. The selection combined the highest performing plants from the top performing lines seems to be the most reasonable choice. This is generally expected to be the most efficient selection strategy in any case (Falconer, 1 989), and so methods using such combined selection should lead to notable genetic advance in these plants. Such methods include line selection and line breeding (Allard, 1 960).
3. Meadowfoam 84
Jain ( 1979) pointed out that the earliest flowers are probably self-pollinated in meadowfoam, since bees are attracted to a plot only after several flowers had opened. Outcrossing rate, estimated by using two genetic marker loci, varied between 43% and 97% (Jain, 1 978). Typically, a plant would flower over a week's period so that intermating among neighbouring plants would not be complete but would occur frequently among and within plants overlapping in flowering duration. Thus, SETPC2 (early seed set) may relate to self-pollination. The result might suggest a potential for developing self-pollination lines in meadowfoam. Genetic control of relative seed set is suggested that successive generations for forced selfing could select for higher natural rates of autogamy for L. alba, although no major mutant or high heritability was noted (Jain, 1 978). Therefore, evolution toward autogamy would have to be slow.