Male and female reproduction of the offspring under controlled conditions were not affected by abortion level

(1)

Embryo selection, abortion and inbreeding depression.

I. Cynoglossum officinale (Boraginaceae).

Summary

Selective abortion of embryos with low potential fitness later in life might free resources for higher quality offspring and thereby increase the fitness of the maternal plant. We tested the hypothesis that a rise in abortion levels leads to an increase in offspring quality. Our experi- mental design was based on the observation that embryo abortion in Cynoglossum officinale increases during the flowering season. Plants were pollinated with self- and cross-pollen during the flowering season and we followed the survival, growth and reproduction of the resulting offspring.

Offspring produced when abortion levels are low (early in the flowering season, mean seed set per flower 1.23), survived significantly shorter in the field than offspring produced when abortion levels were higher (late in the flowering season, mean seed set per flower 0.36). After one year, 19.7% of the offspring from low abortion levels survived and 26.7% of the offspring from high abortion levels. After surviving a winter, the offspring from low abortion levels had a total leaf length of 13.6 cm compared to the total leaf length of 21.3 cm of offspring from high abortion levels. Male and female reproduction of the offspring under controlled conditions were not affected by abortion level. Assuming that plant size (total leaf length) is proportional to plant fitness, and combining results of survival and growth, the life- time quality of offspring produced at low abortion rates is estimated to be 47% of the life-time quality of offspring from high abortion rates, more than a two-fold difference.

On average, seed set was not statistically different for self- and cross pollination. In some maternal parents, self-pollen was more successful than outcross-pollen in producing seeds, while in others the reverse was true. Selection among different cross-pollen donors was less pronounced. Offspring produced by self pollination had a significantly higher mortality than the offspring produced by outcrossing. After one year, 23% of the offspring from cross- pollination survived, while 20% of the offspring from self-pollination survived. Growth and reproduction of the offspring were not affected by pollination type. Differences in offspring survival could not be related to the pollination type (self or cross) that sired most seeds on the maternal plant. These results suggest that the genes affecting embryo survival are different from the genes affecting offspring quality later in life. Selection of offspring may occur within rather than between the two pollination types.

C. Melser and P.G.L. Klinkhamer.

(2)

Introduction

Selective abortion of embryos of relatively low quality can free resources for higher quality offspring and thereby increase the fitness of the maternal plant if reproduction is resource limited. This hypothesis of selective embryo abortion has received considerable attention (e.g.

Darwin 1883, Janzen 1977, Stephenson 1981, Marshall & Ellstrand 1988, Marshall & Folsom 1991). The hypothesis assumes a negative relationship between the probability for a particular offspring of being aborted and its potential fitness later in life if not aborted. Such offspring selection may act among offspring sired by different pollen donors, or among offspring sired by a single pollen donor.

A decrease in abortion level will give rise to a less severe selection of offspring (more offspring survives) and a subsequent decrease in average offspring quality, if embryo abortion is a mechanism by which offspring of the highest quality is selected. To test this, in some stu- dies abortion levels were artificially decreased by random destruction or removal of part of the ovules. The remaining ovules had a lower abortion rate than ovules in undamaged control flowers. With an artificial decrease in abortion level, offspring quality decreased significantly for Lotus corniculatus (Stephenson & Winsor 1986), Crypthanta flava (Casper 1984, 1988), Phaseolus coccineus (Rocha & Stephenson 1991) and Cynoglossum officinale (Chapter 3).

Casper (1988) gives three reasons why the results of such experiments should be taken with caution. Firstly, the treatment itself might damage the remaining ovule or seed, and therefore results in offspring of lower quality. Secondly, with removing prematurely reproductive struc- tures, one might upset initial source-sink relationships and plant-resource levels. Thirdly, for- cing a flower to allocate resources to an ovule that normally would not have matured, might itself result in an inferior seed. The flow of nutrients to nourish an embryo might be smaller through normally unused veins and might result in an inferior embryo. In order to study selective embryo abortion, one should preferably use a design with undamaged plants. As Levri (1998) points out, temporal differences in seed set in the absence of pollen limitation provide an opportunity to study different levels of embryo abortion and their effect on offspring quality. For C. officinale, it was indeed shown that the fraction of embryos aborted increased strongly during the flowering season (Chapter 2).

Another prediction of selective embryo abortion is that if pollen donors have equal access to ovules and are selected for their high quality offspring, then the number of seeds sired by a pollen donor should be positively correlated with the fitness of the resulting offspring. Such a positive connection between siring success and offspring quality has indeed been found in Asclepias speciosa (Bookman 1984) and Raphanus sativus (Marshall &

Whittaker 1989). After controlled hand pollinations, pollen donors differed significantly in number of seeds sired (Bookman 1984, Marshall 1988). The pollen donor that sired most seeds also sired the largest seeds, from which the largest offspring grew (Bookman 1984, Marshall & Whittaker 1989). Note, however, that these outcomes may be explained by differences in pollen germination or pollen competition rather than by selective abortion of seeds. However, in Discaria americana (Medan & Vasellati 1996), the number of seeds sired by a particular pollen donor was not related to the germination success of the seeds it sired.

Selection among pollen donors might be most pronounced between self pollination and outcross pollen donors, especially if recessive deleterious alleles are frequent within the

(3)

population. If offspring suffers from a strong inbreeding depression, large investments of resources in this progeny could be avoided if the offspring from self-pollination is discarded in order to free resources for higher-quality offspring from outcross pollination. However, offspring from self-pollination passes two copies of the genome to the next generation. For maternal plants with a low genetic load, it may pay to retain this offspring (Schultz & Willis 1995, Chapter 4). In another study by Bertin and Peters (1992), differences in offspring quality were more pronounced between self- and cross-pollination than among different cross-pollen donors. In Campsis radicans, the favored pollen donor (self- or cross-pollen) sired the more vigorous offspring (Bertin & Peters 1992).

Few is known about offspring selection that can act within individual pollen donors.

By segregation of alleles in the meiosis to form the haploid pollen, each heterozygous locus will cause variation in the genetic composition of the pollen within one individual pollen donor. Selection on offspring quality may then also reflect differences among pollen from one single pollen donor (Korbecka et al. in prep). In this paper we will use Cynoglossum officinale to test the selective embryo hypothesis while a parallel paper will present data on Echium vulgare.

In earlier experiments with C. officinale in a natural population (Chapter 2), the em- bryo abortion at the start of the flowering season was estimated at least as 16.4%. This fraction increased along the flowering season to 81.8%. The abortion level over all flowers of the plants was estimated as at least 39% (Chapter 2). Moreover, after destruction of three of the four ovules in each flower of C. officinale, seed set in the remaining ovule is significantly increased, compared to the seed set per ovule in the control group, which shows that otherwise viable embryos are aborted (Chapter 3). These seeds that otherwise would have been aborted, had a significantly lower survival, as predicted by the hypothesis of selective embryo abortion.

The model presented in that study, shows that a relatively small increase in seed quality due to selective embryo abortion can give rise to selection for a considerable decrease in seed to flower ratio (Chapter 3). With this follow-up study, we had two goals: first to study selection on self- or cross-pollinations, and second to study the effect of selective embryo abortion on lifetime quality of the resulting offspring, without damaging the plants, flowers or ovules. We pollinated plants, measured seed production and determined offspring survival and reproduction to answer the following questions: — Do higher abortion levels increase offspring quality? — Is selection among offspring more pronounced if abortion levels are high? — Does a different siring success in C. officinale differ between selfed and outcrossed embryos?

— Does siring success in C. officinale differ among outcross pollen donors? — Does inbreeding depression decrease offspring quality? and if so, is embryo selection directed against selfing?

Material and methods

Species

Cynoglossum officinale (L) is a rosette forming, monocarpic, self-compatible perennial. From the main flowering stem, cymes diverge at which flowers develop sequentially. Each day one to two new flowers open at each cyme and each flower remains open for about two days. The most common flower visitors at our study area are bumblebees (de Jong & Klinkhamer 1989).

(4)

The hermaphroditic flowers have five anthers and four ovules. The four ovules are symmetri- cally arranged in a square and may all develop into seeds. Natural populations of C. officinale in Meyendel, the Netherlands are not pollen limited (de Jong & Klinkhamer 1989, Chapter 2).

Average seed set in these populations is circa one seed per flower (i.e. the seed to ovule ratio equals 0.25; de Jong & Klinkhamer 1989). The estimated selfing rates in the field ranged from 0 in small plants to 70% in larger plants (Vrieling et al. 1999). All plants for this study were collected at Meyendel at least 150 m apart.

Pollination method

We present data of two experiments. In the first experiment ("single donor"), flowers from the maternal plants were emasculated with forceps before the flower opened and the pollen would be released from the anthers. The flowers for self-pollinations were not emasculated and pollinated by gently pressing the flower at the moment that the pollen was being released from the anthers to make it touch the stigma. For cross-pollination an anther with ripe pollen from another plant was placed in the middle of the emasculated flower, where it touched the stigma.

In the second experiment ("selective abortion and offspring quality"), flowers that would be self-pollinated, were also emasculated. A small selection of flowers on the maternal plants (approximately one out of ten) were not emasculated, to provide the pollen for the self-pollinations on the same plant. Those flowers were not used to receive pollen. Pollinated flowers were marked with a small drop of paint to identify the pollen donor. All plants changed posi- tions daily to avoid environmental effects. After pollination of the last experimental flowers on each plant, the later opening flowers were pollinated with a random sample of pollen for at least three days, so that seeds of the last experimental flowers also developed in the presence of younger seeds.

Experiment 1. Single-donor

This experiment was designed to study selection between self- and outcross pollination, and selection among different pollen donors. Seven different maternal plants were used for self- pollination or cross-pollination with four different paternal plants, eleven plants were involved in total. Plants were collected in early spring of 1995. All pollinations were carried out in a growth chamber within a 15-day period. Day and night temperatures were, respectively, 20^°C and 15^°C ± 1^°C, and relative humidity varied between 60% and 85%. One maternal plant received different pollinations with pure pollen of each individual cross-pollen donor and self- pollen: 18-26 flowers from each plant were pollinated by each donor, giving a total of about 99-125 treated flowers per maternal plant. Five to six weeks after the last pollinations, the number of developed seeds per flower was counted and seeds were weighed.

The number of seeds per flower was tested for maternal plant, pollination type (either self- or cross-pollination), the interaction between maternal plant and pollination type, cross- pollen source and the interaction between maternal plant and cross-pollen source. As we used one plant per maternal genotype, maternal effects can be contributed to either genetic differences between genotypes, or different conditions of the individual plants. Effects are tested two-sided, unless stated otherwise. The analysis of the binomial distributed data of the seed numbers per flower used a Generalized Linear Models procedure with a logit link function (McCullagh & Nelder 1989, SAS Institute 1993). The effects of the later entered factors in the

(5)

analysis are adjusted for effects of the earlier entered factors (SAS PROC GENMOD, type I).

Non-significant factors were excluded from the final analysis, except when they take part in a significant interaction (Crawley 1993). Significant main factors of interest were tested for differences with a contrast statement. Interactions were examined for significance by a X² test on the estimated parameter in the model.

Seed mass was tested for maternal plant, effects of pollination type, the interaction between maternal plant and pollination type and paternal effects with an analysis of variance.

Experiment 2. Selective abortion and offspring quality

This experiment was designed to study selection between self- and outcross pollination, and differences in offspring quality both between self- and outcross pollen donors and between different abortion levels. The design of this experiment is based on the notion that the abortion level in C. officinale increases during the flowering season (Chapter 2). When few seeds per flower are produced, the abortion level is high (Chapter 2) and potentially this can lead to a strong selection among developing embryos. With a relatively low abortion level in the beginning of the flowering season, early flowers are therefore expected to show weaker selection than late flowers. Consequently we expect late flowers to produce offspring under higher selection and with higher fitness later in life. Furthermore we wanted to examine if selfing leads to a depression in offspring quality. For plants that do select for or against selfed offspring, we want to test if such selection is more pronounced in a period with a high abortion level.

Pollinating the parental generation

To determine selection between self- and outcross pollination at different abortion levels, we pollinated the parental generation with either self- or outcross pollen during the flowering season, and analyzed the number of seeds per flower. Thirteen different maternal plants were placed in a tent of gauze to exclude potential pollinators, while keeping environmental conditions as natural as possible (May 1997).

The design of the experiment is given in Figure 1. Flowers were pollinated with either self-pollen or a mixture of cross-pollen. The mixture of cross-pollen was made by collecting the five anthers of five flowers of five unrelated pollen donors. A fresh mixture was made every two to three hours. The composition of the outcross pollen mixture remained constant during the experiment. Maternal plants were pollinated each day during a four-week period. In the statistical analysis, time was divided in two periods. The division in two periods was marked by the date at which the drop in mean seed number per flower was the largest for each maternal plant individually. In this way we can compare the hazard functions of two categories (low and high abortion levels) in the survival analysis. At least 17-40 flowers per maternal plant were pollinated with self-pollen, and equal numbers were pollinated with cross-pollen, randomly assigned among the position on the plant and flower opening date in the first time period (within 12 days). Another 19-36 flowers per maternal plant were pollinated with self-pollen and 12-38 flowers with cross-pollen in the second time period (within 15 days). This gives a total of 98-126 flowers for each maternal plant. After seed ripening, the number of fully developed seeds per flower was counted and the seeds were weighed.

(6)

The statistical analysis of the binomially distributed data of the seed numbers per flower (seed present or absent in each ovule of a flower) was performed as in the single-donor experiment. The number of seeds per flower was tested for the time period of pollination, maternal plant, the pollination type (either self- or cross-pollination), the interaction between maternal plant and pollination type, and the three-way interaction between maternal plant, pollination type and time period. To detect if selection between pollination types is more pronounced with high abortion levels, we performed this analysis also for the two time periods separately and determined the number of individual maternal plants for which the difference between self- and cross-pollination was significant (X² test on the estimated parameter in the model). with a X² test. Seed mass was tested for maternal plant, effects of pollination type and time period with an analysis of variance.

Offspring quality

To determine the effect of selection and inbreeding depression on offspring quality, we cultivated the offspring and examined survival and reproductive success of the offspring. The seed coat of the seeds was sliced carefully at the top to break dormancy, and in total 1126 seeds were germinated from 13 January onwards on filter paper. Each seed was placed in a separate cell of a small tray. Day and night temperatures in the growth chambers were 20⁰C and 15⁰C respectively, and relative humidity ranged from 60% to 85%. After germination, 939 seedlings were planted in pots of 9 cm diameter between 21 and 31 January 1998.

13 Maternal plants

High abortion and self-pollination High abortion and cross-pollination Low abortion and self-pollination Low abortion and cross-pollination

Survival and growth of the offspring

445 offspring to Meyendel

March 1998 - October 1999

1126 seeds germinated Offspring cultivated in growth chambers

Male reproduction of the offspring

42 offspring

Hand pollination on 10 maternal

plants counting seeds sired

May 1999

Female reproduction of the offspring

54 offspring

Open pollination in garden, counting seeds and flowers

May 1999 P generation

F₁ generation

Figure 1

A schematical representation of the experimental design in experiment 2 (selective abortion and offspring quality).

(7)

Survival and growth

To measure survival and growth of the offspring under natural conditions, 445 plants were transplanted into the field in Meyendel at 9 March 1998, when plants had four to five leaves.

The habitat was a small valley in vegetated sand dunes. The area was fenced to exclude large herbivores. Plants were placed 30 cm apart from each other. Plants were watered regularly during the first three weeks, to reduce initial mortality after the transfer. Survival was censused initially weekly, decreasing in frequency to once every three weeks in October and November of 1998. During winter C. officinale looses its leaves above ground so that no census could be made. From April 1999 onwards, the census was retaken and continued once every two weeks until the end of the experiment in October 1999, when only 52 plants were still alive. No offspring flowered in 1999. The number of leaves and the length of the longest leaf of the surviving plants were determined in March 1998 before transplantation into the field, and after transplantation in May 1998, September 1998 and July 1999.

A survival analysis was performed with Cox' model (Lagakos 1992) in SPSS from the start of the experiment onwards. Time period in the P generation and pollination type in the P generation were included as categorical covariates. The confidence intervals for the estimated parameters in Cox' model cannot be directly translated into a confidence interval for the survival functions. We compared hazard functions for two combinations: offspring produced after cross-pollination or self-pollination and offspring from the first time period and from the second time period in the P generation. To see if selection within the maternal plants between self- or outcrossed offspring was related to offspring mortality later in life, we also performed a separate survival analysis for those maternal plants showing a significant difference in seed production between self- and cross-pollination in both time periods; we compared the hazard functions between the pollination type that sired most seeds and the pollination type that sired fewest seeds. The two time periods appeared to differ not only in abortion level, but also in mass per seed. Seed mass may influence offspring germination, growth and survival, thus counteracting any effect of offspring selection. To estimate the lifetime effect of abortion level (time period) without the confounding effects of seed mass, the survival analyses were repea- ted with seed mass as a continuous covariate. Percentages of surviving offspring were obtained by the values of the baseline functions in the survival analyses.

An index for the total leaf length was calculated by multiplication of the number of leaves per plant and length of the longest leaf. This total leaf length is highly correlated with the dry mass of the plants (Wesselingh et al. 1993). Differences in total leaf length index were analyzed with an analysis of variance with time period and pollination type as main factors and seed mass and day of germination as a covariate for the four dates in March, May, September 1998 and July 1999.

Reproduction

We also kept plants in the growth chambers and used these for an additional estimation of offspring reproductive success, because we anticipated that plants in the field would remain small and no offspring would flower in 1999. Moreover, to assess male reproductive success without using molecular methods, controlled pollinations are required. Plants were transplanted in 1.5 l. pots. All plants survived in the growth chamber. The temperature of the growth chamber was decreased to 5^°C for 18 weeks at November 1999 to induce flowering.

(8)

Thereafter, the day and night temperatures were raised to, respectively, 20^°C and 15^°C. Plants started to flower at the first of May 1999.

Male reproductive success of the offspring (siring seeds)

Intentionally, we wanted to measure siring success for F1 plants of all four combinations (two time periods, two pollination types) for each of the 13 maternal plants of the P generation.

Due to limited availability of some categories, this resulted in a total of 42 plants instead of 52 plants. Those 42 plants served as pollen donor on 10 maternal plants from the same plant col- lection. The time consuming hand-pollinations limited the number of plants we could include.

To reduce the chance of including a maternal plant with an unusually low seed production due to genetic factors, we selected maternal plants from outcrossed offspring from the second time period, because we expected these to give the highest seed production. Each maternal plant in the F1 generation used in this design had a different mother in the P generation. On each of the maternal plants, 2-3 flowers were pollinated with pure pollen from each of the 42 pollen donors. This made a total of 84-126 flowers per maternal plant, while each pollen donor sired seeds on 22 - 29 flowers. The mean number of seeds sired per pollinated flower was analyzed with an analysis of variance, and tested for effects of the maternal plant in the P generation, the pollination type in the P generation and the time period in the P generation.

Female reproductive success of the offspring (seed production)

To assess female reproductive success, plants were transferred to the experimental garden in Meyendel for open pollination in May 1999. The adult plants remained in their original pot to prevent root damage by transplanting. Pots were dug into the ground to prevent desiccation.

Flowers and seeds were counted on 54 plants. Those plants were as evenly distributed as possible among the 13 maternal plants of the P generation, the two pollination types in the P generation, and the two time periods in the P generation. The number of flowers per plant, the number of seeds per plant and the average number of seeds per flower were tested with an analysis of variance for effects of the maternal plant in the P generation, the pollination type in the P generation, and the time period in the P generation.

Results

Experiment 1. Single-donor Seed number per flower

Averaged over all seven maternal plants in this experiment, the mean number of seeds per flower was 1.02 (se=0.046; n=766). Maternal plants differed in the number of seeds per flower (Table 1, Fig. 2). There was no main effect of pollination type (self- or cross- pollination). This means that, averaged over all maternal parents, pollination with either self- or cross-pollen did not significantly affect the number of seeds per flower. However, the interaction between maternal effect and pollination type did significantly affect seed number per flower (Table 1, Fig. 2): in some maternal parents, self-pollen was more successful than outcross-pollen, while in others the reverse was true. This difference between self- and cross-

(9)

pollination was not significant however within individual maternal plants. Pollen donors differed significantly in siring success (Table 1), due to one relatively successful pollen donor (Fig 3). Maternal plants and cross-pollen donors did not interact significantly, indicating that relative siring success of the donors was not statistically different over the maternal plants.

Mass per seed

Mass per seed was affected only by maternal plant (Table 2). Seeds from self-pollination did not differ in mass from seeds from cross-pollinations. Seed mass did not differ among offspring from cross-pollen donors.

Number of seeds per flower

0

self-pollination cross-pollination

a

Maternal plant

b c d e f g

2

1

Number of seeds sired per flower

0

h i j k

1 2

Paternal plant

*

Figure 2

Mean number of seeds per flower (s.e.) produced by seven maternal plants after self- and cross-pollination.

Figure 3

Mean number of seeds per flower sired (s.e.) after self-pollination and single donor pollinations with one of four different cross pollen donors on the seven maternal plants.

Table 1

Generalized Linear Model (see Materials and methods) of number of seeds per flower in the single- donor experiment with a logit link function (GENMOD procedure), with the date of pollination, maternal effect, pollination type as self- or cross-pollination and paternal effect as main effects. The interaction between maternal plant and paternal plant was insignificant and excluded from the final analysis.

Source df F value p value

Date of pollination 1 4.72 <0.0001

Maternal effect 6 4.00 0.0006

Pollination type 1 0.03 0.8527

Maternal effect x pollination type 6 3.26 0.0036

Paternal effect 3 3.54 0.0144

Maternal effect x paternal effect ns

(10)

Experiment 2. Selective abortion and offspring quality Seed number per flower

The experiment aimed at a decrease of seed number per flower, representing an increase of the abortion rate during the flowering season. Such an increase indeed occurred. The average number of seeds per flower decreased from 1.23 (n=741, se=0.047) in the first time period to 0.36 (n=724, se=0.030) in the second time period (Table 3, Fig. 4 and 5 A), a decrease in seed number per flower of 70.7% ((1.23-0.36/1.23). The decrease was apparent in all plants for both pollination types, with one exception. Plant K produced more seeds per flower after self pollination in the second time period compared to self pollination in the first time period. As in the analysis of the single-donor experiment, there was a significant maternal effect, no significant effect of pollination type (self- or cross pollination), and a significant interaction between maternal plant and pollination type (Table 3, Fig. 4). Although there was an overall significant interaction between maternal genotype and pollination type, relatively few individual plants showed a significant difference between self and cross pollination. Two individual maternal plants (F and J) showed a significant difference between self- and cross-

Table 3

Generalized Linear Model (see Materials and methods) of number of seeds per flower with a logit link function (GENMOD procedure), with the position of the cyme along the main flowering stem, time period (i.e. low or high abortion level), maternal effect and pollination type as self- or cross- pollination as main effects. The interaction between time period and pollination type was insignificant and excluded from the final analysis.

Position of cyme 1 22.04 <0.001

Time period 1 235.41 <0.001

Maternal effect 12 4.61 <0.001

Time period x pollination type ns

Table 2

Analysis of variance (type I) of seed mass in the single-donor experiment with maternal effect, pollination type (self- or cross-pollination) and paternal effect as main factors.

Maternal effect 6 42.74 0.005

Paternal effect 3 2.15 0.090

(11)

pollination if we tested seed numbers per flower over both time periods. Plant F produced more seeds per flower after cross-pollination, while plant J produced more seeds per flower after self-pollination. The difference between self- and cross-pollination was largest for plants F and J in the second time period, when abortion rates were high (figure 4). For both time periods the overall effect of self- or cross-pollination on number of seeds per flower was similar, and the insignificant interaction between time period and pollination type is excluded from the final analysis. We wanted to examine if abortion rate affects the strength of selection between self- and cross-pollination. Therefore we analyzed the two time periods separately.

We were interested to know if in the first time period, where abortion rate was relatively low, fewer individual maternal plants would have a significant difference between self- and cross- pollination (expressed as a significance for either pollination type or interaction between maternal effects and pollination type) than in the second time period when abortion rate was high. Tested for the first time period only, the interaction between maternal plant and pollination type was significant (p=0.0093), although no individual maternal plant had a significant difference in number of seeds per flower between self- and cross-pollination. In the second time period, the interaction between maternal plant and pollination type was stronger (p<0.0001), and two individual plants (B and M) showed a significant difference between self- and cross-pollination with most seeds per flower produced after self-pollination.

2

1

0

1

2

second time periodfirst time period

A B C D E F G H I J K L M

Maternal plant self-pollination

cross-pollination

Figure 4

Mean number of seeds per flower (s.e.) produced by thirteen maternal plants after self-pollination or pollination with a mixture of cross pollen in two different time periods.

(12)

0

low abortion &

self-pollination 0.5

1

low abortion &

cross-pollination

high abortion &

self-pollination

high abortion &

cross-pollination 1.5

0.2 0.4 0.8 1.0

Time (days)

0 200 400 600

0.6

0.0

high abortion & cross-pollination high abortion & self-pollination low abortion & cross-pollination low abortion & self-pollination

Percentage survival

0.2 0.4 0.8 1.0

Time (days)

0 200 400 600

0.6

0.0

high abortion low abortion

Percentage survival

0

low abortion &

self-pollination 10

low abortion &

cross-pollination

high abortion &

self-pollination

high abortion &

cross-pollination 20

30

Total leaf length (n * length longest leaf; cm)

A

C

B

D

Number of seeds sired per pollinated flower 0

low abortion &

self-pollination

low abortion &

cross-pollination

high abortion &

self-pollination

high abortion &

cross-pollination 1

2

Number of flowers

0

low abortion &

self-pollination

low abortion &

cross-pollination

high abortion &

self-pollination

high abortion &

100 150

Number of seeds

0

low abortion &

self-pollination

low abortion &

cross-pollination

high abortion &

self-pollination

high abortion &

200

50 150

0

low abortion &

self-pollination

low abortion &

cross-pollination

high abortion &

self-pollination

high abortion &

cross-pollination 1

2

E

G

F

H

Figure 5:

Measures of reproductive success for four different categories: seeds produced in the first or second time period (low or high abortion level) and produced after pollination with self pollen or a mixture of cross-pollen.

A. Mean number of seed per flower produced in the P generation (s.e.). B. Calculated survival percentages of the offspring. Time=0 is the start of germination at 13 January 1998. Percentages are calculated at the overall mean value of the covariate seed mass. C. Survival function for offspring produced at high or low levels of abortion. The survival function is plotted at the overall mean value of the covariate seed mass. Time=0 is the start of germination at 13 January 1998. D. Total leaf length index (number of leaves * length longest leaf; s.e.) of the offspring in July 1999. E. Mean number of seeds (s.e.) sired per pollinated flower of the offspring. F. Mean flower production (s.e.) of the offspring. G. Mean total seed production (s.e.) of the offspring. H. Seed number per flower (s.e.) of the offspring.

(13)

Offspring quality Mass per seed

Mass per seed differed significantly among maternal plants (Table 4). Seeds from self-pollination (22.71 mg) weighed significantly less than seeds from cross-pollinations (23.74 mg).

The interaction between maternal plant and pollination type was not significant, indicating that seeds produced on all maternal plants were affected equally by pollination type. On average, seeds produced in the second period (higher abortion level) had a significantly lower mass (21.46 mg) than seeds from the first period (lower abortion level) (23.67 mg).

Survival of the F1 generation

The survival analysis included the effects of pollination type (self- or cross-pollination) and abortion level (time period). The percentages of surviving offspring one year after germination are given in table 5. Offspring survival could be affected by both seed mass and abortion level. Firstly we therefore correct for seed mass by including it as a continuous covariate. Seed mass affected survival positively (p<0.0001). The overall survival of offspring produced at high abortion rates (second time period) was significantly better (p=0.0147, Fig. 5 B and 5 C) than that of offspring produced at low abortion rates (first time period). This difference was most pronounced in offspring from cross pollination (fig. 5B). After one year, 27% of the offspring of the second time period survived, while 20% of the offspring of the first time period

Table 4

Analysis of variance (type I) of mass per seed in the experiment of selective abortion and offspring quality, with maternal effect, pollination type (self- or cross-pollination) and time period (i.e. high or low abortion level) as main factors.

Maternal effect 12 62.42 <0.0001

Time period 1 80.84 <0.0001

Table 5

Percentages of surviving offspring in the experiment of selective abortion and offspring quality after one year (t=365 days), for two abortion levels (time period), and two pollination types (self- or cross-pollination), with and without seed mass as a covariate.

No covariate Seed mass as covariate

Time period 1 Self pollination 23.6 23.4

(low abortion) Cross pollination 26.1 24.5

Time period 2 Self pollination 25.7 26.8

(high abortion) Cross pollination 40.2 42.0

(14)

survived. Thus, relative to the second time period, the survival of the offspring from the first time period was 26% ((0.267-0.197)/0.267) lower. Secondly, when seed mass is not included as a covariate, one could expect that the lower mass of the seeds produced late in the flowering season would counteract the effect of the higher selection. However, offspring produced at high abortion rates (second time period) still had a lower mortality rate than offspring produced at low abortion (first time period) if we excluded seed mass as a covariate (one- sided test, p=0.0399, Fig. 5 B).

The offspring produced after selfing had a significantly higher mortality than the offspring produced after outcrossing (p=0.0497; Fig. 5 B). After one year, 23% of the offspring from cross-pollination survived, while 20% of the offspring from self-pollination survived.

Thus, relative to cross-pollination, the survival of the offspring from self-pollination was 15%

((0.232-0.198)/0.232; fractions obtained from the baseline function of a separate analysis with only pollination type included) lower. If we included seed mass as a covariate, this effect of pollination type on survival of the offspring dissappeared (p=0.1055), indicating that the negative effects of self-pollination were largely caused by a reduced seed mass.

Growth of the F1 generation

At none of the censused dates, seed mass had a significant effect on the growth of the offspring. After the winter, at the last date of census in July 1999, the offspring produced at a lower abortion level (the first time period, n=50) had a 35.9% ((21.3-13.6)/21.3) smaller total leaf length (Fig. 5 D) than offspring produced at a high abortion level (second time period, n=15). This difference was significant (p=0.0365). Calculating with the extremes of the 95%

confidence intervals of the total leaf length, the differences in size ranged between 33.3% and 41.6%.

Initially, offspring from self-pollinations had a significantly smaller total leaf length compared to offspring from cross-pollinations. However, in July 1999 the significance of this difference dissappeared, probably due to low power, because considerably fewer individuals were present (n=65; Fig. 5 D).

Table 6

Analysis of variance (type I) on number of seeds sired per pollinated flower in experiment 2 (selective abortion and offspring quality) with maternal effect in the parental generation, pollination type as self- or cross-pollination in the parental generation and time period (i.e. high or low abortion level) in the parental generation as fixed factors.

Maternal plant in P generation 12 1.71 0.163

Pollination type in P generation 1 0.49 0.493

Maternal plant x pollination type 12 2.05 0.095

Time period in P generation 1 0.03 0.871

(15)

Reproduction of the F1 generation

Male reproductive success of the offspring (pollen quality and siring seeds)

For the analysis of the seed numbers sired by the offspring, we were interested in the effects of maternal plant of the P generation, pollination type in the P generation and abortion level of the P generation (time period). None of these three factors had a significant effect on the number of seeds sired by the offspring (Table 6, Fig. 5 E). The interaction between maternal plant and pollination type was not affecting seeds sired by the offspring.

Female reproductive success of the offspring (producing flowers and seeds)

We measured the total number of seeds and average seed number per flower as measures of female reproductive success. In addition we measured the total production of flowers. These plants were initially cultivated in growth chambers, and were not different in size. None of the three measures were significantly influenced by maternal plant of the P generation, pollination type in the P generation, or abortion level in the P generation (Table 7, Fig. 5 F, G and H).

Relation between seed set and survival

To test if the most successful pollination type indeed produced offspring with the highest survival we analyzed the data in two ways.

First, we looked at the survival curves of the offspring of the most successful pollination type. Under the hypothesis of selective embryo abortion, the pollination type that produced more seeds, was expected to produce offspring that survive relatively longer. We performed this analysis for the two maternal plants (F and J) from the P generation that showed a significant difference between self- and cross-pollination over both time periods. For plant F cross pollination was more successful in siring seeds, and for plant J self-pollination was more successful with siring seeds. Pollination type coded as most or least successful type was included in a survival analysis of the offspring with seed mass as a covariate. The most successful pollination type in the P generation did not sire offspring that survived longer (p=0.6683). If we perform this survival analysis with seed mass as a covariate for all maternal plants of the P generation, the pollination type that sired most seeds in the P generation did not produce offspring that survived longer compared to the pollination type that was less successful (p=0.9112).

Secondly, we compared the inbreeding depression (calculated as δ=1-Ws/Wc) for seed set with the inbreeding depression in offspring survival. Plants with a positive inbreeding co- ëfficient suffer from inbreeding depression. This means for seed set that these plants produce less seeds after self pollination than after cross pollination. Plants with a negative inbreeding coëfficient for seed set produce more seeds after self pollination compared to cross pollination. The inbreeding coëfficient for seed set of the P generation varied widely among individual plants and ranged from -1.57 to 0.44 (Fig. 6). Maternal plants with a positive inbreeding depression in seed set were expected to have a positive inbreeding depression in survival of the offspring. The inbreeding coëfficient for survival after 145 days of the offspring varied from -1.80 to 0.68 (Fig. 6). Plants with a positive inbreeding coefficient for survival of the offspring, produced offspring after selfing that survived shorter than offspring

(16)

from outcrossing. Plants with a negative inbreeding coefficient for survival, produced selfed offspring that survived longer than offspring from outcrossing. However, there is no significant correlation between inbreeding coefficient in seed set of the P generation and inbreeding coefficient for survival of the offspring. This means that the pollination type that was most successful in producing seeds not always gave rise to the offspring of the highest quality. This calculation is not corrected for differences in seed mass.

Discussion

Embryo abortion and offspring quality

Small seeds produced at the end of the flowering season have higher fitness than larger seeds produced at the beginning of the season. Although perhaps at first sight surprising, this result is in line with the hypothesis of selective embryo abortion. In our experiment, a 70.7%

decrease of the seed number per flower during the flowering season caused an increase in offspring survival of 26% after one year. The seedlings were initially cultivated in growth chambers, where selection pressure is presumably low compared to the natural habitat of the

Figure 6

The relation between individual inbreeding coefficient in seed set in the P generation and individual inbreeding coefficient for percentage surviving offspring after 145 days.

Table 7

Analyses of variance (type I) on total flower production, total seed production and mean number of seeds per flower of the offspring, with maternal effect from the P generation, offspring from self- or cross pollination and time period (i.e. high or low abortion level) of the P generation as factors.

Flower production Seed production Mean number of seeds per flower

df F value p

value

F value p value

Maternal plant in P generation 11 1.02 0.458 0.81 0.630 2.10 0.054 Pollination type in P generation 1 0.25 0.624 0.02 0.894 0.41 0.529 Maternal plant x pollination

type

11 0.59 0.819 0.57 0.840 1.22 0.320

Time period in P generation 1 1.86 0.183 0.89 0.352 0.30 0.863

(17)

sand dunes. Seedlings of a low quality might have survived in the growth chamber and after planting in the field that otherwise would have died when they would have germinated in the natural habitat. Our estimation of differences in offspring survival is therefore presumably conservative. In addition, the growth of the surviving offspring after the winter differed by 35.9%. The differences in size might reflect differences in flowering probability (Wesselingh et al. 1993) and hence influence the number of offspring that will reach the reproductive stage at Meyendel. Flowering probability was not included in this experiment, because mostly it takes several years for C. officinale to reach the reproductive stage (Wesselingh et al. 1993).

Indeed, at the end of the experiment no plant flowered in the field and the flowering probability is not included in the final calculations. Abortion level did not affect the size of the plants that were cultivated in the growth chambers. Once these plants reached the reproductive stage, there was no additional effect of abortion level on offspring fitness. If we assume that the differences in offspring survival after one year will remain similar in time, and if plant size (total leaf length) has a proportional relation with total fitness, then the differences in performance in survival and growth can be multiplied to estimate differences in life-time fitness. The quality of plants produced at low abortion levels is then estimated to be 47.4%

((1-0.26)*(1-0.36)) of the quality of plants produced at high abortion levels. Part of the differences was obscured by the fact that in our experimental design abortion level and seed mass are coupled. In the second time period, abortion levels are high and seeds weigh less.

Because offspring of smaller seeds have a higher mortality rate, we corrected our data for seed mass. But even without such a correction, offspring produced in the second time period survived longer than offspring produced in the first time period (tested one-sided). In an earlier experiment with C. officinale (Chapter 3), with a somewhat different experimental design, we found a similar effect. In this experiment we removed three of the four ovules of all flowers on the experimental plants. In the remaining ovule, seed set was doubled compared to ovules on the control plants. The seeds from the experimental plants with relatively low abortion levels, were heavier than seeds from the control plants. Nevertheless, with or without correcting for seed mass, the offspring from seeds of plants with a higher abortion rate had a significantly higher survival with a difference of 18.9% and 14.4% respectively. In this previous study, we modeled offspring quality in relation to number of seeds per flower (Chapter 3). For the calculations, we included a trade-off between flower and seed production.

For mathematical reasons, it appeared that within a very broad range (from 4 to c. 0.6 seeds per flower), large differences in the seed to flower ratio are accompanied by small differences in total seed production. A relatively small increase in offspring quality would already compensate for the lower seed production when abortion rates are high. This indicates that a very small increase in average offspring fitness due to selective embryo abortion, may easily give a selective advantage leading to a considerable decrease in number of seeds per flower.

Selective seed abortion is not the only evolutionary hypothesis to explain a low seed to ovule ratio (e.g. Stephenson 1981). However, the discussion about the relative importance of other explanations is beyond the scope of this chapter and has been given more attention elsewhere.

In Chapter 2 we concluded that although bet-hedging (Stephenson 1981) may explain part of the overproduction of ovules, it is unlikely to be the only evolutionary explanation for the 'surplus' of flowers produced in C. officinale. In Chapter 8 we compare the relative importance of the effects of sex-allocation on the ESS (evolutionary stable strategy) number of seeds per

(18)

flowers with the effects of selective seed abortion. The evolutionary hypothesis of sex allocation explains a large part of the overproduction of flowers (to increase the male function) in hermaphroditic species and selective seed abortion adds there only a relatively small effect. However, in male sterile individuals a male function is not present and sex- allocation can not explain an overproduction of flowers. It appeared that for these individuals, selective seed abortion can explain a low seed to ovule ratio.

Offspring selection

Averaged over the maternal plants of C. officinale, self pollination yielded an equal number of seeds compared to cross pollinations, although the significant interaction between maternal effect and pollination type revealed the presence of a selective mechanism. On eleven maternal plants in our two experiments, selfing produced the greatest number of seeds, while in nine maternal plants cross-pollination produced most seeds. However, in only few plants this difference between self- and cross-pollination was significant within the plant. In an experiment with Echium vulgare we applied one single donor per flower and found a similar interaction between maternal plant and pollination type (Chapter 4 and 7). This indicates that the occur- rence of individual plants favoring selfing over outcrossing might indeed be a more general phenomenon.

One of the cross-pollen donors was more successful in siring seeds compared to the other cross-pollen donors. Another pollination experiment with C. officinale does not confirm this pattern (Melser, unpubl. data) and we are cautiously by interpreting this difference in siring success. The absence of an interaction between maternal effect and cross-pollen donor suggested no other selection among maternal plants and different cross-pollen donors in this experiment. Selection of offspring may then rather be within than among embryos from different pollen donors.

We would have expected that with increasing abortion levels, the selection would be more pronounced, but our data do not present good evidence for this. The few maternal plants that individually show a difference between self- and cross-pollination, do indeed increase their pollination preference in the second time period. Averaged over all plants, however, no clear picture arises.

Inbreeding depression

Selfed seeds of C. officinale weighed less, and selfed offspring had a lower survival. This lower survival was mainly due to the decrease in seed mass. The difference in percentages of surviving offspring after one year between self- and cross-pollinations, was estimated at 14.7%. In the greenhouse, offspring from selfing and outcrossing had equal male and female reproductive success. This is in striking contrast with another species of the Boraginaceae, Echium vulgare (Chapter 5 and 7), in which we found a considerable late-acting inbreeding depression at the reproductive stage of the offspring. Offspring produced by selfing produced less seeds per flower as a maternal parent, and sired less seeds on other plants as a paternal parent. The selfing rate of E. vulgare under natural conditions was estimated between 0% for small plants and 30% for larger plants (Rademaker 1999). The selfing rate for C. officinale in natural populations was estimated from 0 up to 70% for large plants (Vrieling et al 1999).

This comparison suggests that selfing rates in C. officinale are higher than in E. vulgare. A

(19)

breeding history with a higher selfing rate could have purged deleterious alleles (Husband &

Schemske 1996). Inbreeding depression is then less pronounced in populations of C. officinale compared to E. vulgare.

Selection against selfed offspring?

Inbreeding depression can be calculated per individual maternal plant. We estimated individual inbreeding coefficients for seed set and survival of the offspring. As Schultz and Willis (1995) argued, individual inbreeding depression varied largely within a population. We expected that individuals that produced most seeds after self pollination, would also produce offspring that performed well in the rest of their life. However, our data showed that an individual with a low inbreeding depression in seed set, did not necessarily produce offspring after selfing that survive longer. Thus the level of individual inbreeding depression in seed set did not predict the performance of the offspring of the two pollination types later in life.

The life-time inbreeding coefficient in C. officinale was smaller than 0.5. With a trans- mission of more than one genome to the next generation after selfing, there is also no clear advantage for this species to select against selfing. This might explain why selection against selfing was not present. The increase in offspring quality with higher abortion levels in our second experiment could be caused by selection among different cross-pollen donors. In addition, even within one individual plant, pollen grains have different genotypes, caused by segregation of heterozygote alleles after meiosis. Selection on offspring quality might therefore even act within individual pollen donors (Korbecka et al, in prep.).

In conclusion, higher levels of embryo abortion resulted in higher survival and better growth of the offspring in C. officinale. Other than through increased plant size, reproduction of the offspring remained unaffected by abortion level of the maternal plant. Inbreeding depression act on survival of the offspring, mainly through the effects of reduced mass of selfed seeds.

The most successful pollination type (self- or cross pollination) on the maternal plant did not produce offspring with the highest quality. The genes affecting the differences in number of seeds between self- and cross-pollination, apparently were different from the genes affecting offspring quality later in life, irrespective of the level of individual inbreeding depression. Se- lection might act within different pollination types or within pollen donors rather than between self- and cross pollination.

Acknowledgements

We are grateful for the help in the field by Hans de Heiden and for assistance in the greenhouse by Henk Nell. Martin Brittijn kindly drew the figures. Tom de Jong, Eddy van der Meijden and Peter van Tienderen gave valuable comments on the manuscript.