• artificial selection;
  • Bicyclus anynana;
  • egg size;
  • offspring fitness;
  • reproductive investment;
  • trade-off


  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

We artificially selected on egg size in a butterfly to study the consequences for fecundity, reproductive effort and offspring fitness. Correlated responses in either pupal mass, larval or pupal development time were virtually absent. Offspring size was positively related to fitness, but only partly traded off against fecundity. Rather, total reproductive effort (measured as fresh mass), egg water content and the decline of egg size with female age increased in the large-egg selected lines compared to either small-egg or control lines. Accounting for these effects showed that reproductive investment (in dry mass) was in fact similar across lines. Such mechanisms may enable increased investment in (early) offspring without a reduction in their number, revealing a much more complex picture than a simple trade-off between offspring size and number. Substantial variation among replicates suggests that there are different underlying mechanisms for change, rather than any single, unitary pathway.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Egg size is a particularly interesting trait in life-history evolution because it is simultaneously a maternal and progeny character: mothers determine egg size, which in turn can affect substantially the fitness of the progeny resulting from these eggs (Bernardo, 1996; Fox & Czesak, 2000). Thus, it is an ecologically and evolutionarily significant trait, and has consequently received considerable attention over recent decades (Roff, 1992; Einum & Fleming, 2000; Fox & Czesak, 2000; Heath et al., 2003). Much of the work on patterns of reproductive allocation has focused around the influential paper of Smith & Fretwell (1974). They assume a trade-off between offspring quantity and quality, that offspring fitness increases with investment per offspring, and that energy available for reproduction is a finite amount at any given time. Based on these assumptions their model predicts the evolution of an optimal egg size that maximizes maternal fitness, with mothers having the upper hand in this parent-offspring conflict (Einum & Fleming, 2000). Following Smith & Fretwell's paper, individual females have been usually considered to have no option but to trade off offspring quantity against quality.

In spite of supportive evidence for a trade-off between offspring quantity and quality and an increase in offspring fitness with increasing investment per offspring (Roff, 1992; Fox & Czesak, 2000), the general validity of the Smith–Fretwell model has been challenged. It has been criticized for being too simplistic and some of the basic assumptions remain under debate (Winkler & Wallin, 1987; Sinervo et al., 1992; Bernardo, 1996; Schwarzkopf et al., 1999; Einum & Fleming, 2000; Caley et al., 2001). In particular, the fundamental idea of an independent optimization of offspring size and total reproductive effort in a two-step process, with the decision about optimal reproductive effort being followed by the decision about the optimal partitioning of that investment into few large vs. many small offspring, has been questioned (Winkler & Wallin, 1987; Caley et al., 2001). Nevertheless, theory in this field has nearly universally modelled modifications in offspring size as being independent of the level of reproductive investment (Caley et al., 2001). However, at least one model predicts an evolutionary link between these two life-history traits (Winkler & Wallin, 1987), and there is indeed some limited support for this notion (Schwarzkopf et al., 1999; Caley et al., 2001; Czesak & Fox, 2003). Thus, even after more than three decades clearly more experimental tests of whether mothers producing large offspring inevitably have to sacrifice fecundity are needed, rather than making implicit or explicit assumptions.

One potentially powerful means to more rigorously examine relationships between egg size and other traits such as fecundity is artificial selection. However, this has only rarely been used in this context (Schwarzkopf et al., 1999; Czesak & Fox, 2003). Here we follow this approach by artificially selecting on egg size in the tropical butterfly Bicyclus anynana (Butler, 1879). Artificial selection yielded lines that differ substantially in egg volume (> 50%). We exploit that variation to explicitly address the following questions: (1) Is egg size traded off against egg number?; (2) Is reproductive investment independent of offspring size and/or number?; (3) Does offspring fitness increase with investment per offspring?; (4) Does selection on egg size yield correlated responses in pupal mass or development time?

Material and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Study Organism

Bicyclus anynana is a tropical, fruit-feeding butterfly with a distribution ranging from southern Africa to Ethiopia (Larsen, 1991). The species exhibits striking phenotypic plasticity (two seasonal morphs) as an adaptation to alternative wet–dry seasonal environments and the associated changes in resting background and predation (Brakefield, 1997; Lyytinen et al., 2004). A laboratory stock population was established at Leiden University in 1988 from over 80 gravid females collected at a single locality in Malawi. Several hundred adults are reared in each generation, maintaining high levels of heterozygosity at neutral loci (Van't Hof et al., 2005). Females from the stock population were used for this study.

Selection procedure

Starting from the above-mentioned outbred laboratory stock we derived lines selected for decreased egg size (hereafter ‘S’ for small egg size), increased egg size (‘L’ for large egg size) and unselected controls (‘C’). A total of 727 female butterflies were measured for egg size at generation 0, from which founding populations of 30 females each were selected at random to establish two unselected control lines. The remaining butterflies were randomly split into three groups from which three sets of replicate lines for the up and down direction were derived (using per line the 30 females with the most extreme phenotypes). In two of the sets, egg size was corrected for pupal mass by selecting on residuals, whereas in the third selection was purely on egg size and thus ‘unconstrained’ with respect to pupal mass (hereafter marked with ‘*’). The latter set was included to obtain at least an indication of whether not selecting on residuals would yield a strong correlated response in body size. This treatment could not be replicated due to time constraints but was considered necessary to check that our results were unlikely to have been severely influenced by the precise method of imposing selection. In subsequent generations between 100 and 160 females were measured per line, from which 30 were selected as parents (for C lines 30 females were selected at random). Selection intensities varied between 0.94 and 1.38.

Throughout, butterflies were reared in a climate room at 27 °C, high relative humidity (70%), and a photoperiod of L12:D12. Larvae were fed on maize plants and adults on moist banana. All female pupae were weighed on the day following pupation to the nearest 0.01 mg. After adult eclosion, females were individually marked and kept separate from males for 2 days. Afterwards females were set up for mating for 3 days with males, following which they were placed individually for 1 day in translucent plastic pots containing a fresh cutting of maize for egg laying. Eggs were subsequently collected and measured. Thus, all eggs measured were the first ones laid by an individual female within a 12 h light period on day 6 of the female life span, thereby effectively controlling for effects of female age (e.g. Karlsson & Wiklund, 1984; Brakefield et al., 1994; Braby & Jones, 1995). As the eggs of B. anynana are nearly perfect spheres, egg size was measured as cross-sectional area (mm2) using a digital camera (Leica DC200) connected to a binocular microscope (correlation between egg area and egg mass r ∼ 0.92; Fischer et al., 2002). The resulting images were analyzed using Scion Image public software (Scion Corporation, 2000). To calculate egg size for individual females, the mean of at least eight eggs (mean egg number: 15) per female was used.

Correlated responses to selection

Correlated responses to selection on egg size were examined in all eight replicate lines (3 L, 2 C, 3 S) as well as in F1s of pooled reciprocal crosses made between the replicates corrected for pupal mass of the L and S lines. The latter crosses were included to examine heterozygote vigour and potential inbreeding depression (Saccheri et al., 1996; Van Oosterhout et al., 2000). The eggs for this experiment were collected in generation 10. The resulting hatchlings were divided among four cages per line, with each cage containing ca. 40 larvae. All cages were kept in a single climate cell (27 °C, L12:D12, 70% relative humidity) and arranged in a randomized block design. For all individuals we measured larval and pupal development time as well as pupal mass (weighed on the day following pupation). Additionally, for all females longevity, lifetime fecundity and egg size over time were recorded. Females were kept individually in plastic pots as outlined above, with eggs being collected, counted and measured every other day until the death of the females.

To calculate reproductive investment we first converted the means for individual females in egg cross-sectional area into mean egg mass based on the equations given by Fischer et al. (2002). Subsequently we estimated reproductive investment in two ways: (1) as the product of the mean fresh mass of the eggs laid on oviposition days 1 and 2 and lifetime fecundity and (2) by multiplying daily mean dry mass of eggs with daily fecundity and summing up all values over the whole oviposition period. Although frequently used before (e.g. Schwarzkopf et al., 1999), the former estimate relies on fairly restrictive and partly unrealistic assumptions, which will be discussed in detail below. Data on egg dry mass were obtained by weighing fresh mass (on the same day the eggs were laid) and dry mass (after drying to constant weight for 48 h at 60 °C) of 64 individual eggs per selection direction (to the nearest 0.001 mg). These eggs were also used to analyse egg water content across selection directions.

Fitness assays

To investigate potential fitness consequences of variation in egg size we used new reciprocal crosses, being established within each selection direction in generation 11. We kept two independent lines per selection direction and one unselected control, and allowed three generations for recombination prior to fitness assays. As a second control we used the outbred stock population, resulting in a total of six lines. Except in one experiment (fitness experiment 2, see below), fitness was assessed on a population level by collecting random egg samples from the selection lines (> 100 ovipositing females each). Therefore, eggs were removed from maize cuttings (used as oviposition substrate) and randomly divided among treatment groups. The following assays were used.

Hatching success of individual eggs

Eggs were individually weighed to the nearest 0.001 mg and subsequently kept in Eppendorf tubes to score hatching success. The tubes were checked twice daily for hatchlings.

Egg hatching success of individual females

In this experiment, hatching success was investigated on a family level. Individual mated females were allowed to oviposit on two small Oplismenus plants, one of which was transferred to 20 °C whilst the other remained at 27 °C. Eggs were counted, remaining untouched on the plants, and hatching success subsequently monitored.

Egg hatching success in relation to relative humidity

To investigate egg-hatching success in relation to relative humidity, egg samples were placed on Petri dish bottoms, which were transferred to tightly sealed plastic pots (500 mL). Depending on treatment, these pots contained a ca. 1 cm layer of silica gel (<30% relative humidity), a layer of soaking wet paper towels (> 90%), or nothing (control; ∼70%). Ten replicates per line and treatment with 20 eggs each were used. Hatchlings were counted and removed daily until no more hatchlings were found.

Hatchling survival in relation to food quality and quantity

Hatchling survival was scored under three different feeding regimes. Eggs for this experiment were stored in plastic pots, which were checked daily early in the morning (i.e. soon after hatching) for hatchlings. There was no indication of a difference in development time between large and small eggs in a preliminary experiment (unpublished data). The hatchlings were either transferred to Petri dishes containing fresh maize cuttings (control), starved for 24 ± 2 h (i.e. fresh maize cuttings were added to the dish the following day), or transferred to senescent maize cuttings that had been cut 24 ± 2 h earlier. Food was replaced daily, and larval survival was scored on day 4, a time period sufficient to distinguish larvae that had been able to establish themselves successfully on the maize cuttings from those that had died. Sixteen replicates per line and treatment with 10 hatchlings each were used.

We used different environmental conditions in the fitness assays as described above because any differences in fitness are notoriously difficult to detect under benign rearing conditions. Thus, we included stressful conditions to maximize our chance of detecting differences in fitness. Among the above treatments low temperature (Fischer et al., 2003a,b), low relative humidity (increasing desiccation risk of eggs), starvation and offering food that was not fresh were considered stressful.

Statistical analyses

Realized heritabilities (h2) were calculated by fitting least-square regressions to egg size (relative to unselected controls) on cumulated selection differentials, with heritabilities being estimated as twice the absolute values of the slope of the regression lines (as selection was on females only). To test for differences in phenotype between selection directions we used analyses of (co-)variance (an(c)ovas). To account for the fact that individuals within lines are statistically not independent, replicate line was nested within selection direction (and, if applicable, cage was nested within replicate line; cf. Underwood, 1997). Unless otherwise stated, both the latter factors were treated as random effects, whilst all others were considered fixed effects. Because in analyses treating replicate line as a random effect, statistical power is exceedingly low (because of the low number of replicate lines), in the case of such analyses giving nonsignificant results they were repeated treating replicate line as fixed effect. The advantage is then that a repeat of a negative result indicates that this is not due to limited statistical power, while a positive one may at least indicate trends (though nonindependence of individuals within lines is not controlled for anymore). Throughout, pupal mass was added as covariate as appropriate. As the crosses between replicates 1 and 2 are not independent from the pure lines, they were not considered in any an(c)ovas. In contrast, the single ‘unconstrained’ lines were included even though they involved a different method of selection. This was done because there was no indication that these differed substantially from the other two replicates in the response to selection or any other life-history trait (Figs 1 and 2, Table 2). It is also noteworthy that no conclusion, and only a single marginally statistical result (variation among replicates in larval time n.s. when dropping the unconstrained replicate), would change when the ‘unconstrained’ lines are excluded from the analyses. Pairwise comparisons were performed using Tukey's HSD test. ancovas were also used to compare slopes of regression lines (Zar, 1999).


Figure 1. Response of egg size (for first eggs laid) over 11 generations of artificial selection for increased (diamonds, solid lines) and decreased (squares, dashed lines) egg size. Controls: triangles, solid line. Population means ± 1 SE are given for each generation. In two replicates per direction egg size was selected relative to pupal mass (i.e. it was selected on residuals), whilst selection was purely on egg size (‘unconstrained’) in the third ‘replicate’ (open symbols). (a) Absolute egg size data; (b) egg size relative to controls set to 100%.

Download figure to PowerPoint


Figure 2. Egg fresh mass (a, averaged for individual females over the whole oviposition period), lifetime fecundity (b), reproductive investment as fresh mass (c, product of mean initial mass and fecundity), and reproductive investment measured as dry mass (d, corrected for variation in egg size over time and differences in egg water content; see text) for small-egg selected (S, white bars), large-egg selected (L, black bars) and unselected control lines (C, hatched bars) of Bicyclus anynana. Given are population least square means + 1 SE. In two replicates per direction, egg size was selected relative to pupal mass, whilst selection was purely on egg size in the third ‘replicate’ (indicated by ‘*’). After completion of selection, crosses were made between replicates 1 and 2 within the large-egg (L) and small-egg (S) selected lines, respectively (indicated by ‘1 × 2’). Sample sizes per line range between 31 and 42 females.

Download figure to PowerPoint

Table 2.  Average larval development times, pupal development times and pupal masses for Bicyclus anynana across selection directions and replicate lines (SE in parentheses).
Dir.Repl.Larval time (days)Pupal time (days)Pupal mass (mg)n1n2
  1. Data were pooled across sexes (all line by sex interactions not significant). In replicates 1 and 2, egg size was selected relative to pupal mass, while this was not the case in the third ‘replicate’ (indicated by ‘*’). After completion of selection, crosses were made between replicates 1 and 2 within the large-egg (L) and small-egg (S) selected lines, respectively (indicated by ‘1 × 2’). There values suggest that none of the traits was seriously affected by inbreeding depression.

  2. C, control lines; n1, sample sizes for larval times and pupal masses; n2, sample sizes for pupal times.

C124.3 (0.20)6.3 (0.05)175.7 (2.47)14599
C225.0 (0.27)6.3 (0.05)170.0 (3.29)10283
L124.0 (0.17)6.7 (0.05)195.3 (2.76)152108
L223.9 (0.16)6.3 (0.05)179.5 (2.08)150102
L*22.8 (0.15)6.2 (0.04)188.7 (2.88)150121
L1 × 222.9 (0.15)6.3 (0.05)183.8 (2.39)160100
S122.7 (0.15)6.1 (0.03)181.3 (2.40)150105
S223.9 (0.18)6.3 (0.04)186.0 (2.46)152113
S*25.7 (0.30)6.2 (0.05)160.6 (3.30)11288
S1 × 222.5 (0.12)6.1 (0.02)186.7 (2.45)150116

To test for differences in the decline in egg size with female age across lines we used repeated measurement anovas. Factors included were selection direction, replicate line and time (with daily mean egg size of individual females as repeated measures). Here a significant time effect indicates a change in egg size over time, and an interaction between selection direction and time would indicate differences in the decline in egg size across directions (note that repeated measures anovas do not support random factors; thus all factors are treated as fixed effects).

Phenotypic correlations between two variables (e.g. female pupal mass and egg size) within selection directions were analyzed using Pearson's product moment correlation. Further, effects of an array of five traits on egg size and fecundity within selection directions were investigated using multiple regressions (stepwise forward addition of variables, Ridge regression; the use of nonstepwise multiple regressions yielded no qualitative change in the interpretations).

Effects of selection direction, replicate line and egg mass on hatching success of individual eggs was analyzed using a generalized nonlinear model with a binomial error distribution and a logit-link function (significance test: log-likelihood type 1). All other survival data were analyzed with nominal logistic regressions on binary data. Unless explicitly tested for the occurrence of significant interactions between factors, only significant interaction terms are given. Means are presented ±1 SE throughout.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Artificial selection on egg size

Artificial selection yielded highly divergent lines that differed after 11 generations by ca. 50% in egg volume across selection directions (based on eggs laid first; Fig. 1). A significant response was observed in both directions (L > C > S; Tukey HSD after anova on mean egg size per female; F2,3 = 56.1, P = 0.0042), with the response (averaged across replicates) being fairly symmetric (L lines: 23.8 ± 1.3% increase; S lines 27.3 ± 2.0% decrease). There was also significant though less pronounced variation among replicate lines (F3,557 = 26.1, P < 0.0001). Realized heritabilities (h2) were on average 0.39 (L1: 0.384, r2 = 0.98; L2: 0.386, r2 = 0.92; L*: 0.394, r2 = 0.91) for L and 0.44 (S1: 0.394, r2 = 0.85; S2: 0.436, r2 = 0.85; S*: 0.494, r2 =0.91) for S lines (all P-values of regressions <0.0001). The slopes of regressions fitted to egg size on cumulated selection differentials did not differ among replicates within selection direction or between directions (ancovas; all P-values ≥ 0.14 for interactions of direction or replicate line with cumulated selection differential).

Correlated responses in development time and pupal mass

Double-nested anovas with cage and replicate as random factors failed to reveal any significant effects of selection direction on larval development time, pupal development time or pupal mass. However, for all three traits there was significant variation among replicate lines (Table 1). Treating replicate line as a fixed effect in the anovas (everything else being equal) yields similar results for larval time, however, it shows significant variation for pupal time [L (6.4 ± 0.03 days, n = 452) > C (6.3 ± 0.03 days, n = 248) > S (6.2 ± 0.03 days, n =414); Tukey HSD after anova; F2,784 = 29.0, P < 0.0001] and pupal mass [L (mean 187.9 ± 1.5 mg, n = 452) > Control (173.4 ± 2.1 mg, n = 248) = S (177.4 ± 1.6 mg, n = 414); F2,1078 = 8.7, P = 0.0015] across selection directions. These results may indicate trends towards longer pupal times in the L lines and towards a divergence in pupal mass across selection directions. However, there was no clear pattern with selection direction in either trait, with overlap between L and S replicate line means occurring throughout (Table 2). Thus, there is only weak evidence for correlated responses in any of the three life-history traits considered above.

Table 1.  Nested analyses of (co-)variance for the effects of selection direction (fixed) and replicate line (random) on life-history traits of Bicyclus anynana.
  1. Cage (random), sex (fixed), and pupal mass (covariate) were added as appropriate. Throughout, replicate was nested within selection direction, and cage within replicate and direction. Significant P-values are given in bold.

Larval time
 Replicate (Dir.)5166.83.30.0216
 Cage (Dir., Repl.)2452.014.2<0.0001
 Direction × sex21.30.40.7012
Pupal time
 Replicate (Dir.)52.557.50.0002
 Cage (Dir., Repl.)240.342.40.0002
 Direction × sex20.795.60.0039
Pupal mass
 Replicate (Dir.)517491.26.50.0006
 Cage (Dir., Repl.)242728.55.4<0.0001
 Direction × sex2624.31.20.2941
Egg size
 Replicate (Dir.)50.02115.6<0.0001
 Pupal mass10.0128.90.0032
Lifetime fecundity
 Replicate (Dir.)519468.02.50.0339
 Pupal mass1218733.027.6<0.0001
Reproductive investment (initial egg size × fecundity)
 Replicate (Dir.)54546.33.20.0076
 Pupal mass141954.729.7<0.0001
Reproductive investment (dry mass)
 Replicate (Dir.)584.62.30.0471
 Pupal mass11154.331.1<0.0001

Throughout all analyses (Table 1), cage and sex were significant factors, with, regarding the latter, females being significantly heavier and having a longer larval but a shorter pupal development time as has been repeatedly shown for B. anynana (e.g. Fischer et al., 2003a).

Egg size-number trade-off

As expected, the females used to investigate correlated responses to selection laid eggs that differed substantially in size, with the difference in volume of first eggs amounting to, on average, 58.9% between L and S lines (L > C > S; Tukey HSD after anova; Table 1, Fig. 2a). A significant effect of the covariate pupal mass indicates that egg size tended to increase, if weakly and in case of the L lines not significantly, with increasing female pupal mass (L: r = 0.04, P = 0.637, n = 139; C: r = 0.30, P =0.014, n = 65; S: r = 0.23, P = 0.004, n = 147).

However, there was only limited evidence for a trade-off between egg size and lifetime fecundity (Fig. 2a, b; Table 1; effect of selection direction on fecundity n.s.). Treating replicate line as fixed effect in the anova, however, indicates that the S lines tended to lay on average more eggs (least square mean 302 ± 9, n = 109) than either C (268 ± 12, n = 65) or L lines (257 ± 9, n =103), with the latter two being statistically indistinguishable (Tukey HSD after ancova; selection direction F2,266 = 6.9, P = 0.0012). Likewise, removing the covariate pupal mass from the analysis presented in Table 1 (treating replicate line as random effect) yields a statistical trend (F2,3 = 4.3, P = 0.08; note that one could argue that it is not necessary to include pupal mass as it does not differ across selection directions). A significant effect of the covariate pupal mass indicates that egg numbers tended to increase with increasing female pupal mass (L: r = 0.20, P = 0.018, n = 139; C: r = 0.27, P =0.029, n = 65; S: r = 0.37, P < 0.0001, n = 147; slopes of the regression lines did not differ; ancova interaction term P = 0.17). To summarize, selection for smaller eggs tended to increase fecundity, but selection for larger eggs had no opposite effect in spite of the production of much larger eggs. Thus, the reproductive pattern in B. anynana seems to be more complex than a simple trade-off between number and size of eggs.

Further, there were significant phenotypic trade-offs or trends between egg size averaged over the whole oviposition period and fecundity within selection directions (both corrected for pupal mass of individual females using residuals; L: r = −0.32, P = 0.0001, n = 139; C: r = −0.24, P = 0.057, n = 65; S: r = −0.26, P =0.0013, n = 147). There was no indication that the slopes of the regression lines differed among selection directions (ancova; interaction term P = 0.65).

Correlated responses in reproductive investment

Reproductive investment was firstly estimated as the product of mean initial egg mass and fecundity, a measure often used in previous studies (e.g. Schwarzkopf et al., 1999). As above, the anova treating replicate as random factor revealed no significant effect of selection direction (Table 1), while treating replicate as fixed effect indicates that investment tended to be higher in L lines (least square mean 125.2 ± 3.8 mg, n = 103) than in S lines (107.7 ± 3.6 mg, n = 109), with the unselected controls intermediate (113.5 ± 4.9 mg, n = 65) and not significantly different from either of them (Tukey HSD after ancova; selection direction F2,266 = 5.7, P = 0.004; Fig. 2c). Again, removing the covariate pupal mass from the analysis presented in Table 1 yields a statistical trend (F2,3 = 4.3, P = 0.063).

These data suggest that the L lines may achieve a fecundity comparable to controls by increasing their reproductive investment. However, using the above estimate of reproductive investment relies on two critical assumptions, namely that (1) egg size is invariant over time (which is not true for butterflies and other organisms; e.g. Wiklund & Karlsson, 1984; Braby & Jones, 1995; Giron & Casas, 2003), or at least that the decline in egg size with female age is equal among females laying eggs that differ in initial size, and that (2) there are no differences in egg composition (e.g. Giron & Casas, 2003) across groups. Below we will show that these assumptions are not met in our study organism.

First, the decline in egg size with female age was overall stronger in L lines than in either C or S lines (Fig. 3). A repeated measurements anova on egg size revealed significant effects of selection direction (F2,160 = 423.0, P < 0.0001), replicate line (F5,160 =12.9, P < 0.0001), and time (F8,153 = 21.4, P < 0.0001). Most importantly there was also a significant interaction between selection direction and time (F16,306 = 4.4, P < 0.0001) confirming that L lines tend to lose proportionally more weight (ca. 11.2%) than S lines (ca. 3.5%; due to death of females this analysis was restricted to the first 17 oviposition days). It is striking that a single L line (L1, top line in Fig. 3) does not follow the general pattern and maintains a rather large egg size throughout. This particular line is quite exceptional, as it simultaneously realises the lowest fecundity, the lowest total reproductive effort (within L lines) and the highest mean egg size (averaged over oviposition period; Fig. 2a–d). Second, water content was found to differ significantly across selection directions [L (83.5 ± 0.2%, n = 63) > C (82.2 ± 0.2%, n = 64) >S (81.2 ± 0.2%, n = 63); Tukey HSD after anova; selection direction F2,3 = 12.9, P =0.034; replicate F3,184 = 4.1, P = 0.0075].


Figure 3. Mean fresh egg mass ( ± 1 SE) over time for large-egg selected (filled diamonds, solid lines), small-egg selected (open squares, dashed lines) and unselected control lines (open triangles, solid line) of Bicyclus anynana. In two replicates per direction, egg size was selected relative to pupal mass, whilst selection was purely on egg size in the third replicate (indicated by arrows).

Download figure to PowerPoint

These results show that variation in egg size over time and differences in egg composition (i.e. water content) need to be taken into account when estimating total reproductive investment. Doing so (by multiplying daily mean egg size with daily fecundity and summing up all values as well as correcting for differences in water content by using dry mass) reveals that reproductive investment was in fact similar across selection regimes, regardless of treating replicates as random (Table 1) or fixed effect (selection direction F2,266 = 0.56, P = 0.57; Fig. 2d). In each of the three selection regimes, total reproductive effort increased with increasing pupal mass (L: r = 0.22, P = 0.011, n = 139; C: r = 0.31, P = 0.012, n = 65; S: r = 0.42, P < 0.0001, n = 147; with similar slopes: ancova interaction term P = 0.18). None of the relationships given above were due to any associated differences in female longevity, as there was no difference across selection directions (L: 25.5 ± 0.7 days, n =103; C: 24.6 ± 0.8 days, n = 64; S: 25.6 ± 0.6 days, n =109; F2,3 = 0.04, P = 0.96).

Predictors of egg size and lifetime fecundity

A multiple regression analysis showed that, overall, pupal mass had the strongest effect on egg size (though still fairly weak; r2≤ 10%), followed by fecundity (Table 3a). Longevity had a rather strong effect on egg size in the L lines, an effect for which we have no explanation. The best predictor of lifetime fecundity was longevity, indicating that, within lines, longer-lived females laid more eggs (Table 3b). Pupal mass and egg size also affected fecundity. Thus the patterns revealed by multiple regressions fully confirm the trends obtained from the correlation analyses above.

Table 3.  Results of multiple regressions (stepwise forward addition of variables; Ridge regression; λ = 0.10; F > 1.0 for inclusion) for the effects of life-history traits on egg size (A) and fecundity (B) across selection directions (all whole model P < 0.02).
 Predictorβ (SE)rinline imageFP
  1. Given are standardized partial regression coefficients β, multiple coefficients of determination rinline image, F-value and significance level.

  2. L, large egg-selected (n = 134); S, small egg-selected (n = 147); C, control (n = 65). Significant P-values are given in bold.

(A) Egg size
LFecundity−0.465 (0.076)0.12719.14<0.0001
Longevity0.362 (0.077)0.26925.47<0.0001
Pupal mass0.170 (0.074)0.2903.890.0508
Larval time0.113 (0.076)0.3022.200.1405
SPupal mass0.372 (0.078)0.09915.760.0001
Fecundity−0.225 (0.078)0.1498.400.0044
CPupal mass0.323 (0.118)0.0835.580.0213
Fecundity−0.202 (0.118)0.1452.940.0918
Predictorβrinline imageFP
(B) Fecundity
LLongevity0.453 (0.069)0.17432.69<0.0001
Egg size−0.406 (0.069)0.30119.14<0.0001
Pupal mass0.202 (0.066)0.3499.660.0023
Pupal time−0.197 (0.069)0.3898.240.0048
SPupal mass0.372 (0.074)0.11318.32<0.0001
Longevity0.245 (0.070)0.17711.150.0011
Egg size−0.218 (0.074)0.2279.230.0028
Pupal time0.079 (0.071)0.2341.240.2670
CLongevity0.421 (0.104)0.22220.08<0.0001
Pupal mass0.249 (0.108)0.2643.760.0672
Egg size−0.188 (0.107)0.3003.740.0851

Fitness consequences of variation in egg size

Hatching success of individual eggs

An enhanced quality of larger eggs was first examined by scoring the hatching success of individually weighed eggs. The size of the eggs used for this experiment differed substantially across selection directions (L > C > S, with L and S differing by 50.7% in egg mass; Tukey HSD after anova; F2,3 = 133.1, P = 0.0012), and also to a lesser extent across replicate lines (F3,282 = 5.0, P =0.0023). Within lines, larger eggs were more likely to hatch (general nonlinear model; effect of egg mass on survival inline image = 19.3, P < 0.0001) with successful eggs, on average, 5.1% heavier than unsuccessful ones. Hatching success showed no differences across selection direction (inline image = 3.0, P = 0.22), but did differ among replicate lines (inline image = 16.2, P = 0.001; L1: 27.1%, L2: 62.5%, C1: 60.4%, C2: 41.7%, S1: 60.4%, S2: 54.2% of 48 eggs each).

To further analyse the very wide variation in egg hatching success among replicate lines (see above), we collected eggs from individual females and divided those among 20 and 27 °C. Selection direction (slightly lower in L than C or S; nominal logistic regression; inline image = 90.6, P < 0.0001), line (inline image = 332.5, P < 0.0001) and temperature (slightly higher at 27 °C; inline image = 35.3, P < 0.0001) affected egg hatching success. Overall, the data obtained (Table 4) were very similar to those from the above experiment, indicating that handling of eggs did not affect survival. Excluding females that produced no offspring reduced the differences across lines substantially, although the overall pattern remained (selection direction inline image = 10.1, P = 0.0064; line inline image = 44.3, P < 0.0001; temperature inline image = 39.0, P < 0.0001; Table 4). Consequently, differences across lines are largely, but not entirely, due to varying numbers of sterile females (i.e. females that only produce nonviable eggs), and are thus not closely related to egg size. They may indicate differences in the history of inbreeding although the crosses made between replicate lines consistently failed to demonstrate inbreeding depression for any life history trait (see Fig. 2, Table 2).

Table 4.  Mean egg-hatching success (%) for individual Bicyclus anynana females across selection directions (dir.), replicate lines (repl.) and two egg rearing temperatures.
Dir.Repl.Temp. (°C)All femalesFertile females
Mean (SE)nMean (SE)n
  1. Removing sterile females shows that differences in egg hatching success across lines are largely, but not entirely, due to infertile females.

C12063.8 (5.4)3168.2 (4.8)29
C12769.1 (5.6)3173.9 (4.9)29
C22040.7 (6.8)3165.7 (5.8)19
C22740.7 (7.3)3166.4 (7.2)19
L12024.7 (6.4)3158.6 (9.0)13
L12728.7 (6.6)3167.5 (6.7)13
L22053.7 (7.0)3172.4 (5.3)23
L22757.1 (7.1)3176.5 (5.0)23
S12059.4 (6.6)3172.0 (5.7)22
S12763.5 (7.1)3180.9 (5.3)22
S22038.8 (7.6)2958.6 (8.7)19
S22750.6 (7.6)2976.2 (5.5)19
Egg hatching success in relation to relative humidity

As above there was significant variation in egg hatching success across selection directions and replicate lines (nominal logistic regression; selection direction inline image =29.4, P < 0.0001; replicate line inline image = 136.3, P < 0.0001). Overall, hatching success was similar under high and intermediate relative humidity (except for one of the small-egg lines; Fig. 4a). At low relative humidity, however, all but one L line showed lower hatching success (humidity level inline image = 42.3, P < 0.0001). As expected from both their enhanced volume-surface ratio and higher water content, hatching success of larger eggs was less affected by low relative humidity than that of small eggs with the controls intermediate (interaction between selection direction and humidity level inline image =28.9, P = 0.0001; Fig. 4a).


Figure 4. Fitness consequences of differences in egg and hatchling size for large-egg selected (filled diamonds, solid lines), small-egg selected (open squares, dashed lines) and unselected control lines (open triangles, solid line). (a) Egg hatching success in relation to relative humidity (high: > 90%; intermediate: ∼70%; low: <30%). To control for differences in egg hatching success across lines, the ‘intermediate’ treatment was set to 100%. (b) Hatchling survival in relation to food quality and quantity (fresh: hatchlings were transferred to fresh maize cuttings soon after hatching; starved: hatchlings were starved for 24 ± 2 h after hatching, then transferred to fresh maize cuttings; old: hatchlings were transferred soon after hatching to senescent maize cuttings, cut 24 ± 2 h earlier).

Download figure to PowerPoint

Hatchling survival in relation to food quality and quantity

Hatchlings of the L lines also had a significantly higher survival probability under different feeding regimes than those of the C or S lines (nominal logistic regression; selection direction inline image = 45.4, P < 0.0001; replicate line inline image = 2.1, P = 0.55; Fig. 4b). Survival was generally higher on fresh food than either on senescent food or after 1 day of hatchling-starvation (treatment inline image = 401.3, P < 0.0001). The higher chance of establishment on the food plants for the L lines is presumably due to a larger size of the hatchlings (cf. Nakasuji, 1987; Braby, 1994), since head capsule width differed across selection directions in the expected direction [L (mean 0.740 ± 0.003 mm, n = 144) > C (0.681 ± 0.003 mm, n = 133) > S (0.612 ± 0.003 mm, n = 140); Tukey HSD after anova; selection direction F2,3 = 111.9, P = 0.0015].


  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Selection on egg size

Selection was clearly successful, yielding lines that differed substantially in egg size. The symmetric response to selection is consistent with egg size being under stabilizing selection in B. anynana. As suggested by similar slopes of the regressions fitted to egg size on cumulated selection differentials, the response to selection did not differ among replicates within selection direction or between directions. This means that selecting purely on egg size (without controlling for pupal mass) gave comparable results to selecting on residuals (Fig. 1), indicating that body size does not constrain changes in egg size in this butterfly. A realized heritability (h2) of about 0.4 is in broad agreement with other estimates of egg size heritability in B. anynana. Parent-offspring regressions gave estimates of 0.20 (Steigenga et al., 2005) and ca. 0.4 (Fischer et al., 2004), and a full- and a half-sib analysis showed heritabilities of ca. 0.2 respectively 0.1 (Fischer et al., 2004; Steigenga et al., 2005). Thus, heritability of egg size in B. anynana is low or moderate, as expected for a life-history trait (Roff, 1997).

Does selection on egg size yield correlated responses in pupal mass or development time?

Although the selection lines became highly divergent, there is no indication of a correlated response in larval time and no more than weak evidence for correlated responses in pupal time or mass. The trends towards longer pupal times and higher pupal masses in the L lines are basically due to single replicate lines. Thus, they may be better attributed to fixation of specific alleles just by chance rather than inherent differences across selection directions. It is especially noteworthy that even those replicate lines in which egg size was not selected relative to body size showed no consistent correlated response in body size (Table 2), challenging the assumption of a positive genetic correlation between egg and body size (Azevedo et al., 1997; Fox & Czesak, 2000). This result was not entirely unexpected since earlier studies found that phenotypic correlations between egg size and body size are extremely weak in B anynana (Fischer et al., 2002). Likewise, we found no support for the notion that larger egg and hatchling size reduces larval development time (see above; e.g. Sibly & Monk, 1987; Fox, 1994; Azevedo et al., 1996,1997; Yampolski & Scheiner, 1996).

Is egg size traded off against egg number?

Overall, we found only weak evidence for a trade-off between egg size and number. In spite of negative phenotypic correlations within selection directions, fecundity was not reduced in the L lines (except for L1), but tended to be higher in the S lines (Fig. 2b). Thus, our results give only weak support for the first assumption of the Smith–Fretwell model (see also Schwarzkopf et al., 1999; Czesak & Fox, 2003). However, our findings raise the issue of the mechanisms responsible for the ability to lay many large eggs. We were able to examine at least some of these, as discussed below.

Is reproductive investment independent of offspring size and/or number?

When using a ‘traditional’ estimate (product of mean initial egg mass and fecundity), reproductive investment tended to increase with increasing egg size, thus challenging another assumption of the Smith–Fretwell model. However, as we will demonstrate below, such a simple estimate does not yield reliable estimates of reproductive effort, at least not in terms of the energy allocated to reproduction. First, in our study organism, females laying large eggs show a steeper decline in egg size with age. Thus, using initial egg size will inevitably overestimate the reproductive investment of such females. Second, water content is higher in larger than in smaller eggs. Although the difference across lines is quite small (ca. 2%) and larger eggs are nevertheless energetically richer, this difference in egg composition reduces allocation of energy to reproduction in the L lines, and values based on fresh mass may overemphasize investment. The steeper decline in egg size with female age in our L-lines could be caused by an earlier depletion of resources (Wiklund & Karlsson, 1984; Giron & Casas, 2003), but they continue to maintain a larger egg size than the other lines throughout their life (Fig. 3). Taking all these factors into account suggested that reproductive investment (in dry mass) was actually closely similar across lines, suggesting that reproductive investment is indeed independent of offspring size.

Not considering egg water content does not mean that water does not contribute to fitness. Indeed this may be the case (at least under certain conditions such as low humidity; see below). However, water content is unlikely to be a limiting factor for reproduction in B. anynana, and is not important in terms of energy budgets (e.g. Schliekelman & Ellner, 2001). The question whether females actively determine the water content of their eggs (which then would be energetically costly) is highly relevant but currently unaddressed.

Does offspring fitness increase with investment per offspring?

Offspring fitness did increase with investment per offspring as larger eggs were more likely to hatch (within lines), had a clear fitness advantage in terms of hatching success at low relative humidity, and yielded hatchlings with larger head capsules enhancing survival probability under food stress (see also Fischer et al., 2003a,b). The fact that, across lines, there was no advantage of L lines in terms of a generally higher hatching or survival probability is not contradictory, as such comparisons include factors other than size per se (e.g. related to the selection history). In our case, for instance, varying numbers of sterile females would heavily confound such a comparison (Table 4). This difference across lines is unlikely to be caused by a failure to mate, as about 90% of butterflies mate within the first 2 hours of the three-day-mating period already, and because a similar proportion of sterile females was found among females which were monitored for matings (K. Fischer, unpublished data).

Note that a number of earlier studies on butterflies (in contrast to those on many other taxa; Roff, 1992; Azevedo et al., 1997) failed to indicate positive correlations between egg size and quite a number of life-history traits (such as development time and size at maturity; e.g. Wiklund & Karlsson, 1984; Karlsson, 1987 and references therein). We assume this to be a methodological artefact caused by either comparing eggs not sufficiently different in size to detect subtle differences or by using too beneficial conditions for fitness assays, under which any fitness effects are notoriously difficult to detect.

Conclusions: complexity rather than simplicity

Our results show that mothers producing large high-quality offspring do not inevitably have to sacrifice fecundity. Alternative mechanisms such as an increase in reproductive effort (fresh mass), a higher egg water content, and a steeper decline in egg size with female age may enable females that lay larger eggs to avoid detrimental consequences on lifetime fecundity at least in certain conditions. The associations with female age can lead to an increased investment in early eggs that may tend to represent the most important ones for female reproductive success in the wild (Wiklund & Karlsson, 1984; Begon & Parker, 1986; about 60% of the total egg load in B. anynana is laid within the first week of oviposition; Brakefield et al., 2001).

In spite of largely consistent responses to selection, there was substantial variation among replicates e.g. in fecundity and reproductive investment (Fig. 2). One replicate of the L lines (L1), for instance, actually traded off egg size against fecundity in accordance with the Smith–Fretwell model. Such differences in correlated responses are usually attributed to sampling error, gene/allele numbers and pleiotropy (e.g. Gromko et al., 1991; Gromko, 1995). This observation may suggest that some but not all genes for egg size are coupled with effects on fecundity or total reproductive effort. Anyway, there are obviously several, alternative possibilities rather than just a single unitary mechanism inevitably resulting in a trade-off. Thus, future models on reproductive resource allocation should consider such alternatives, in particular they should allow for an independent evolution of offspring size and reproductive effort (Winkler & Wallin, 1987) and possibly for variation with age.

Given the fitness benefits of larger eggs without a clear cost in terms of reduced fecundity, an obvious question to be answered is what actually keeps egg size relatively small. For the moment, we have no explanation, but, as stated above, our findings may hold under certain conditions only, in particular under beneficial feeding conditions. Thus, as has been repeatedly documented, a clear trade-off between egg size and number may only be visible when food is limited (Van Noordwijk & DeJong, 1983; Reznick et al., 2000; Zera & Harshman, 2001), while our animals were fed ad libitum. This, however, is unlikely to account for the differences across replicate lines, as it is difficult to image that, given the opportunity, some populations feed less than would be optimal.

Reproductive effort is a complex feature interrelated with e.g. egg size and composition, changes within the oviposition period, fecundity, longevity, resource acquisition, and pupal mass. Neglecting any of these potential sources of variation may considerably influence the interpretation of a given data set. In that respect our results are also of substantial methodological significance. Furthermore, subtle changes in many characters may sum up to a substantial effect, which may make it difficult to statistically validate any clear patterns (for example, given small changes in body size, egg size, egg composition, and fecundity, all affecting reproductive investment). Future studies should pay more attention to such potential complications, and doing so will hopefully gain new insights into the complex relationships between egg size, egg number and reproductive investment.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

We thank C. J. Breuker for writing a macro to automate egg measurements, and N. Wurzer and M. Lavrijsen for the supply of maize plants. Funding was provided by the German Research Council (DFG Grant no. Fi 846/1-1 and Fi 846/1-2 to KF) and The Netherlands Organization for Scientific Research (NWO 811-34.005 to BJZ).


  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References