THE weight of the seed in the tomato is to a large extent determined by the genetical -

VOL. XXXV, No. OCTOBER, 199 ON THE FACTORS AFFECTNG THE MEAN SEED WEGHT OF TOMATO FRUTS BY L. C. LUCKWLL, PH.D. Department of Botany, The Victoria University of Manchester^. NTRODUCTON THE weight of the seed in the tomato is to a large extent determined by the genetical - constitution of the tine. Lines homozygous for the factor D (tall, as opposed to dwarf habit) have relatively large seeds of mean weight between and mg., whilst lines homozygous for the corresponding recessive (d) have relatively small seeds whose mean weight lies usually between and mg. The wild species of Lycopersicum, which are closely related to the tomato, have smaller seeds than any of the cultivated varieties of L. esculentum, those of L. pimpinellifolium, for example, having a mean weight of only i mg. Although determined primarily by the genetic constitution, the weight of the seed exhibits wide variation within the genotype. Random samples of seed collected in different years from the same pure line are frequently found to differ significantly from one another in their mean weight, as are also samples collected during the same season from different lots of plants of the same constitution. Much of this variation is no doubt due to differences in the external environment, since under normal conditions of cultivation no special precautions are taken to keep these external factors constant. More difficult to explain, however, are the variations in the mean weight of seed between fruits collected at the same time from a single plant, or from different plants of the same genotype, growing under identical conditions. As the figures in Table show, these differences may be considerable, frequently being of the order of 6 % of the mean seed weight. Certain preliminary observations made in 196 had indicated that these variations in seed weight within a pure line might be accounted for by variation in such factors as the number of fruits developing on the truss, the number of seeds developing in the fruit, and the position of the truss on the plant, but the data collected during that year were not extensive enough to allow any definite conclusions to be drawn. The observations were therefore repeated on a more extensive scale during 197 and the results of this investigation are presented in this paper. t will be observed from Table that the variations in seed weight are accompanied by corresponding variations in the dry weight of the embryo, so that the ratio of mean seed weight to mean embryo weight for different fruits is practically constant. The weight of the embryo in the tomato is, in fact, very highly correlated ^ A large part of the experimental work was done while the author was working in the Department of Botany at the University of Bristol. New Phyt. xxxviii, 1

l8 L. C. LUCKWLL with the weight of the seed (Faberge, 197; Luckwill, 197, unpublished), so that the observations on variation in seed weight here recorded apply equally well to embryo weight. This fact is of considerable importance in view of the possible role of embryo size in the determination of the final size of the plant (Blackman, 1919; Ashby, 197). Table. Variations in seed weight and embryo weight between fruits collected on the same day from a single plant of line 15, homozygous for d, p, o, r, s, y, n {MacArthur, 19). Each mean is based on twenty-four observations No. of fruits on truss No. of seeds in fruit Mean Seed \vt, mg. S.E. T>rf wl. of embryo, mg. Mean j S.E. Ratio seed wt./ embryo wt. 1 9 88 76 117 98 5 1-6 -571 1-1 1-159 -89 - -9-5 -77-8 -555-5 -15-76 -16 - -5-8 -89-8 -76-79 -7. EXPERMENTAL METHOD The material employed in this research was derived from a commercial strain of tomato known as "Blaby", and was homozygous for the factors D, P, O, R, Y, S, a, (MacArthur, 19). Altogether a dozen plants were grown. These were raised in pots in the greenhouse until about the middle of June, when they were transplanted, two to three feet apart, in a plot out-of-doors. Each plant was allowed to develop five laterals which were trained up bamboo canes, all other laterals being removed as soon as they appeared. The growth of these five laterals was stopped after the formation of the sixth truss offiowers. The number of fruits developing on each truss was controlled by removing all the remainingflowersand buds as soon as the requisite number of fruits had set. The number of fruits allowed to develop on a truss varied from one to eleven, and sufficient trusses were trimmed in this manner to provide about twenty fruits in each fruit number group, the distribution of trusses with varying numbers of fruits among the twelve experimental plants being entirely at random. The fruits began to ripen early in August and collecting was continued until the end of October, the number of fruits collected each month in the different fruit number groups being shown in Table. Table. Number of fruits collected Month No. of fruits on truss 5 6 7 8 9 1 Total Aug. Sept. Oct. Total 5 1 1 9 6 ; 5 19 1 16 1 7 7 6 9 6 8 5 1 1 1 1 15 7 7 7 76 19

The mean seed weight of tomato fruits 18 Each fruit on gathering was given a serial number which was entered in a book, together with the corresponding plant number, truss number, number of fruits on the truss, approximate position of the truss on the plant (high or low), and the date of harvest. The fruits were picked as soon as they were ripe, and after weighing, the seeds were extracted from each fruit, enclosed in muslin bags and washed in running water to remove mucilage. They were then spread out on sheets of blotting paper to dry, after which they were stored in packets and were later counted and weighed, all the seeds in a single fruit being weighed together to the nearest o-i g. and the mean seed weight of a fruit being taken as total seed weight/seed number. Altogether data were collected in this manner from 19 fruits.. PRESENTATON OF DATA (i) Preliminary analysis The primary data are naturally far too extensive for publication so that only a table of means is given here. n Table the mean seed weights, mean seed numbers, and mean fruit weights are classified according to (i) the number of fruits developing on the truss, and () the position of the truss on the plant, the lowest three trusses being designated "low" and the upper three trusses "high". The mean seed weights recorded in Table exhibit considerable variation, and in order to establish the significance or otherwise of this variation the seed weight data were subjected to statistical analysis. The analysis, which is presented in Table V, is based on twelve observations in each fruit number group of which six came from a "high" truss and six from a "low" truss, so that altogether 1 observations were used in the preliminary analysis, giving 11 degrees of freedom. The "error" variance was computed from the sums of squares within the fruitnumber-truss-position groups and the significance of the observed variances was determined by the application of Fisher's "z" test (Fisher, 198). Table. Mean fruit weights, seed weights, and seed numbers, classified according to the number of fruits on the truss and the position of the truss on the plant No, of fruits on truss Fruit weight (g,) Low High Low Seed no. High Seed weight (mg,) Low High 5 6 7 8 9 1 9-8 5-1 6i-7-7-7 51-5 79-8-9-7 7-7 5 9 9' - 9-1 1-5-5 69-7-7 5-66- 1-9- 61 97-O 5-6-7 6-115- 7- Si-i 65-55-6 66-5 5-1 - 7-5 5-88-5 11-9 69-8 16-6 79-8 8- -5-5 -8-5 -9-1 -6 ' - -8-9 -59-5 -5 - -8 - -, -88-1 -7 Means 8-1 9-65- 75-5 -7 -

18 L. C. LUCKWLL Table V. Analysis of variance of mean seed weight Variance due to: D.F. Sums of squares Variance " z" o-oi z Position of truss No. of fruits on truss nteraction Error O O no -1 8-697 1-75 8- o-oooi -87-17 -766 i-i6s - -6-6 Total 11 17-77 The analysis shows clearly that the position of the truss on the plant has no effect on the mean weight of the seed produced, there being no significant variance between the seed weight of fruits from "high" and "low" trusses. The variance in mean seed weight between fruits from trusses on which different numbers of fruits have been allowed to develop, however, is far too large to be accounted for by errors of random sampling and is therefore statistically significant. () Further analysis of data The preliminary analysis considered above has established the existence of significant differences in mean seed weight between fruits from trusses bearing different numbers of fruits, and it now remains to elucidate further the causes of these differences. This has been done by the technique of correlation and regression. Six correlation tables were constructed between the four variables in question, each table having 19 entries grouped into convenient classes. Correlation and regression coefficients between the six pairs of variables were calculated and their significance tested by the apphcation of Fisher's "t" test (Fisher, 198), with the exception of correlations involving fruit number as one of the variables, which could not be tested in this manner owing to the fact that this variable is not normally distributed, there being an approximately equal number of observations in each fruit number group (Table ). This objection, however, does not apply to regressions in which fruit number is the independent variable or to partial correlations in which fruit number is the eliminated variable. The coefficients of correlation and regression, together with the corresponding values of "^", are given in Table V. t can be seen from Table V that the regression of mean seed weight on number of fruits per truss is nearly seven times its standard error and is therefore very highly significant. The corresponding regressions of fruit weight and seed number on number of fruits per truss are both less than twice their standard errors and are therefore of no statistical significance. The reduction in the number of fruits developing on a truss by the excision of some of the fiowers and buds may therefore be expected to bring about an increase in the mean weight of the seed in the remaining fruits. The weight of the fruit itself, however, or the mean number of seeds per fruit, does not appear to be infiuenced by this treatment. t can also be seen from Table V that there is a low but significant correlation between mean seed weight and the number of seeds in the fruit. The correlation

The mean seed weight of tomato fruits 185 Table V. Correlation and regression coefficients between seed weight, fruit number, fruit weight and seed number in the tomato Variables correlated a h n Correlation r t Regression of a on 6 h t o-oi t Seed wt. Fruit wt. Seed no. Seed wt. Seed wt. Fruit wt. Fruit no. Fruit no. Fruit no. Seed no. Fruit wt. Seed no. 17 17 17 17 17.7-16 -86-168 --1867-175 -857-8 -58 - --57-6 1-67 1 5-55 6-76 1-5 1-6 -8-6 -15 coefficient is negative, which means that high seed numbers tend to be associated with low seed weights. There is further a low positive correlation between fruit weight and seed weight which, however, is just significant on the 1% point of "i". Finally there is a very high positive correlation between the number of seeds developing in the fruit and the fruit weight. These correlations are discussed more fully in a later section of this paper. () Partial correlations and regressions The above analysis has shown that there are at least three factors (viz. fruit number, seed number, and fruit weight) which are significantly correlated with the mean seed weight in the tomato. The data in Table V, however, give little information as to the magnitude of the effects of these three factors on seed weight owing to the various interrelationships which exist between the factors themselves. n order to estimate the effect of each factor acting independently of the rest, the technique of partial correlation has been used. First order partial correlations were first calculated giving the correlations between successive pairs of variables with one of the remaining two variables eliminated, and from these partial correlations of the second order were calculated giving the true correlation between any two variables independent of variations in the remaining two variables. The general form of the equation used in the calculation of partial correlations of the first order was: where r^a, etc., represent the total correlation coefficients between various pairs of variables. The equation for the calculation of second order coefficients is similar, ^ \ /T r Vl' The results of these calculations are given in Table V. Three second order correlations have been extracted, all of which are highly significant. t is also possible from the sums of squares and products computed from the correlation tables to isolate the corresponding partial regression coefficients to satisfy an equation of the type y = 61*1 + bx + 6X,

i86 L. C. LucKwiLL Table V. Partial correlation and regression coefficients between the following variables 1 = mean seed weight. = no. of fruits on truss. = mean fruit weight. = mean seed no. Variables correlated a b Variables eliminated n l6 l6 l6 l6 l6 l6 15 S 15 Correlation?- --9-876 -56-98 O'5598 --167 - --519 O'55O t 1-9-9 9-1 8-1 Regression of a on A b -87-59 -117 t -58 1-1- 1 t ~" where x^, x^ and x^ represent the deviations from their respective means of fruit number, seed number, and fruit weight, y is the corresponding deviation of seed weight from its mean, and b^, 6 ^^'^ ^ ^re the three partial regression cofficients. These coefrcients have been calculated and are given in Table V, together with the corresponding values of "t". All are very highly significant. V. NTERPRETATON OF RESULTS (i) Seed weight and fruit weight The strongest of the three second order partial correlations is that between fruit weight and seed weight, which has a value of -55 (Table V). t seems clear that this correlation arises as a result of the common effect on both seed and fruit of fluctuations in various environmental factors during development. n the present experiment the plants, as far as was possible, were given uniform cultural treatment, but as the fruits were collected over a period of three months there was bound to be considerable variation in the conditions of temperature and water supply under which different fruits developed. The data show, however, that seed weight is less influenced by such variations than is fruit weight, which is probably due to the fact that the ovules develop within the fruits where the environment -is relatively more constant than that in which the fruits themselves develop. The mean weight of all the fruits collected is about 5 g.; the mean seed weight is -5 mg. (Table ), so that for a change of fruit weight of i g. we should expect a change of seed weight of -65 mg. if both were affected proportionally. The actual regression of seed weight on fruit weight, however, is only -117 mg. (Table V), which is about one-sixth of this value.

The mean seed weight of tomato fruits 187 () Seed weight and seed number The negative partial correlation of -519 between the number of seeds developing in the fruit and the mean seed weight is more difficult to interpret and the present research yields no information as to the causes of this relationship. t seems probable that it is a nutritional effect similar to that which gives rise to the negative correlation between birth weight and litter size in mammals (Kopec, 19), but it may possibly be dependent upon some spatial relationships in the young fruit, when the ovules are closely packed on the placenta. n this connexion it is interesting to note that Harris has found a similar, though lower, correlation between seed weight and the number of seeds developing in the pod in the bean (191) and also in Staphylea and Cladastris (1911). The regression of seed weight on seed number in this experiment is -59 mg. P^^ seed, which, taking into account the large variations in seed number which occur in the tomato, is very considerable (see Table V). Confirmation of this relationship deduced from partial correlations and regressions is given by the results of another experiment carried out in 197. n this the data were all collected from a single plant and this was grown in a greenhouse so that environmental fluctuations were reduced to a minimum. Variation in seed weight due to differences in fruit number were also eliminated by allowing only a single fruit to develop on each truss. Altogether 7 fruits were collected from this plant, the seed number and mean seed weight of each was determined, and correlation and regression coefficients were calculated. The results are given in Table V and it will be observed that the regression coefficient of - o-oo66 agrees closely with the partial regression calculated from the main body of data. Truss no. Table V. Correlation between seed number and mean seed weight in subsidiary experiment Number of seeds Total seed wt. mg. Mean seed wt. mg. 1 57-19 5-8 77-9 6 68-6 5 91 58 9 Correlation between number of seeds and mean seed weight: r=--859; ^=75U,,^.6. 6=--66; t = T,-9o\ ^^ () Seed weight and number of fruits on truss The coeflicient of partial regression of seed weight on number of fruits per truss is - -87 mg. per fruit, which is somewhat lower than the total regression given in Table V, but is still highly significant. The correlation coefficient of -- also is somewhat lower than that first calculated and shows that the relationship between the two variables in question, although quite definite, is not a particularly close one. Like the correlation between seed weight and seed number just considered, it seems that this correlation must also be interpreted as an effect of nutrition, the removal of fruits from the truss resulting in an increased supply of food material to the remaining fruits and seeds. n this case, however, we might expect a negative 6 1 85-8 7 185 588 -

i88 L. C. LUCKWLL correlation to exist between fruit number and fruit weight as well as between fruit number and seed weight, but the fact that no such correlation was found does not necessarily invalidate this interpretation. t must be remembered that the water content of tomato fruits is high (6-7 %), whereas that of the seeds is low (less than 5 %), and for this reason it seems reasonable to suppose that a small increment in dry weight which would be readily detectable in the seed might pass unnoticed in the fruit itself or be masked by slight fluctuations in the water content. () The range of variation of seed weight This experiment has shown that the variation in seed weight (and embryo weight) within the genotype in the tomato may be classifled under one of three headings as follows : (1) that due to variations in the external environment, () that due to variation in the number of fruits developing on the truss, () that due to variation in the number of seeds developing in the fruit, and the partial regression coeflicients given in Table V enable the extent of the variation from each of these sources to be estimated. As has been pointed out on p. 186, any variations in the external environment in which the fruits developed will affect both the seed weight and the fruit weight, so that the partial regression of seed weight on fruit weight may be taken as a measure of the effect of such ffuctuations on the mean seed weight of the fruit. The regression coefficient is -117 mg. per g. and the maximum range of variation in fruit weight within any one seed number group (since seed number and fruit weight are correlated) in this experiment is about 6 g. This means that the maximum possible variation in seed weight due to inequalities in the external environment is 6 x -117 = -7 mg., which is about % of the mean weight of all the seeds collected. Similarly, the number of seeds in a fruit varies from i to 5 and the regression of seed weight on seed number is -59 i^g- PS"^ seed, so that the maximum variation in seed weight from this source is i -75 mg. or 5 % of the mean seed weight. A similar calculation for fruit number shows that a reduction in the number of fruits developing on a truss from to may be expected to bring about an increase in the mean seed weight of approximately 1%. As the flgures in Table V show, the maximum possible variation in mean seed weight from all three sources is of the order of 75 %. The maximum range of variation recorded in this experiment was about 55 %, which is well within the calculated limit. Table V. Factors correlated with the mean seed weight in the tomato Correlated variable No. of fruits on truss No. of seeds in fruit Fruit weight Range of variation - 1 1-5 6 Partial regression coefficient -87-59 -117 Max. range variation in seed weight (mg.) -16 1-75 -7 Max. range variation as % mean Total maximum variation = -9 75 Range of mean seed weight recorded in this experiment =- mg. to -1 mg. i-8mg. = 55 % of the mean seed weight. 1 5

The mean seed weight of tomato fruits 189 V. SUMMARY t is found that significant variations exist in the mean weight of the seed collected from different fruits of the same genotype in the tomato. These variations in mean seed weight are accompanied by corresponding variations in the mean dry weight of the embryo. The position of the truss on the plant is found to have no effect on the mean weight of the seed. t is established by the technique of correlation and regression that these variations in seed and embryo weight are attributable to three causes, viz.: (i) variation in the external environment, (ii) variation in the number of fruits developing on the truss, (iii) variation in the number of seeds developing in the fruit. The maximum amount of variation attributable to each of these three sources is estimated from the partial regression coefficients. wish to record my sincerest thanks to Mr E. J. Hatcher of Bristol for his invaluable assistance in collecting the data for this analysis, without which should have been unable to complete the experiment. REFERENCES ASHBY, E. (197). The determination of size in plants. Proc. Linn. Soc, Lond., 19, pt.. BLACKMAN, V. H. (1919). The compound interest law and plant growth. Ann. Bot., Lond.,, 5. FABERGS, A. C. (196). The physiological consequences of polyploidy. J. Genet., 65. FSHER, R. A. (198). Statistical Methods for Research Workers. London. HARRS, A. J. (1911). Seed weight in Staphylea and Cladastris. Torreya, 11, 165. (191). A quantitative study of the factors influencing the weight of the bean seed.. ntraovarial correlations. Bei. bot. Zbl., Abt. i, p. i. KOPEC, S. (19). On the influence exerted by certain factors on the birth weight of the rabbit. Anat. Rec. 7, 95. LUCKWLL, L. C. (197). On embryo size as a means of accounting for heterosis in tomato hybrids. Unpublished thesis. MACARTHUR, J. W. (19). Linkage groups in the tomato. J. Genet. 9, 1.