Cleistogamy as a bet-hedging strategy in Oxalis acetosella, a perennial herb


Henrik Berg, Department of Ecological Botany, Uppsala University, Villavägen 14, S-752 36 Uppsala, Sweden (tel. + 46 184712853; fax + 46 18553419; e-mail


1 Phenology and reproduction were studied in three populations of the cleistogamous herb Oxalis acetosella during three growing seasons, in order to see how the balance between chasmogamous (CH) and cleistogamous (CL) reproduction varies with temporal and spatial environmental variation and with plant size. The numbers of CH and CL flower buds, flowers, immature capsules and mature capsules were counted per ramet, ramet sizes were estimated, and mature capsules were collected for seed counts.

2 Both CH and CL flower numbers were positively correlated with ramet size expressed as leaf number, but the correlation was much stronger in the CL phase. CL flower production also varied significantly between years and sites. Thus, CL production in O. acetosella was not independent of resources and climatic variation.

3 When the effects of year and site were taken into account, the probability of an individual flower developing into a mature fruit was not higher in the CL phase than in the CH phase.

4 CL production was affected by fertilization success in the CH phase. Ramets with one or more CH flowers left unfertilized generally produced more CL buds than ramets with all their CH flowers fertilized. The former group also tended to have more seeds per CL capsule.

5 Since reproductive success in the two phases varies in different temporal and spatial patterns, cleistogamy in O. acetosella is considered to be a bet-hedging strategy optimizing reproductive output in fluctuating environments.

6 The findings of this study are in conflict with the general view on cleistogamy as a fail-safe mechanism for back-up seed production, unaffected by variations in resource supply and environmental conditions.


A cleistogamous flower is structurally modified for autogamy: it never opens, and self-pollination occurs in a bud-like stage that is much reduced compared with a ‘normal’ flower. Nectar and odours are absent, petals are rudimentary or completely missing, stamens are often reduced in both number and size, pollen grains are few and the pistil is reduced in size (Darwin 1877, pp. 310–345; Goebel 1904; Ritzerow 1908; Campbell et al. 1983; Ellstrand et al. 1984; Ruiz de Clavijo & Jimenez 1993). This condition is quite widespread and has been reported from 287 species in 56 angiosperm families (Lord 1981).

‘True’ cleistogamy, as defined by Lord (1981), occurs when a single species or an individual plant produces both chasmogamous (open, potentially cross-pollinated) and cleistogamous flowers. This is the type of cleistogamy addressed here and it will be referred to simply as ‘cleistogamy’, although there are other common types of cleistogamy among angiosperms (Lord 1981; Campbell et al. 1983).

Lloyd (1984) divides the variation strategies of plants into five principal classes. Under this scheme, alternation between chasmogamous and cleistogamous flowering is classified as a ‘conditional choice’, i.e. a plastic response triggered by some external or internal cue. Schoen, however, describes it as a ‘multiple strategy’, at least for those situations where both flower types occur during the same flowering season (Schoen 1984; Schoen & Lloyd 1984). In species that show temporal separation of cleistogamous and chasmogamous flowers within a season, the production of the two morphs is likely to be initiated and regulated by different environmental cues (e.g. photoperiod, temperature or soil moisture; Bergdolt 1932; Uphof 1938; Brown 1952; Evans 1956; Langer & Wilson 1965; Campbell et al. 1983; Masuda & Yahara 1994), or by shifts in the plant's own status (size, age; Lloyd 1984) Thus, cleistogamy could still be regarded as a conditional choice (sensuLloyd 1984; Redbo-Torstensson & Berg 1995). Generally, cleistogamous flowering is favoured under adverse and stressful conditions, when both pollinators and resources are likely to be scarce (Schemske 1978; Lord 1981; Campbell et al. 1983; Le Corff 1993). Seasonally regulated cleistogamous species may also be able to adjust the production of the later flower type according to reproductive success in the earlier one (Redbo-Torstensson & Berg 1995).

The respective advantages of chasmogamous and cleistogamous reproduction have been discussed. Out-crossing offers the benefits of (i) genetically variable offspring, reducing the risks of inbreeding depression, and of extinction in the case of habitat changes (Stebbins 1957; Beattie 1976; Solbrig 1976), and (ii) fitness gains through pollen donation (Schemske 1978; Schoen & Lloyd 1984). These benefits have often been referred to, but are difficult to demonstrate. Selfing, on the other hand, provides (i) safe reproductive back-up under conditions unsuitable for out-crossing (Darwin 1877; Schemske 1978; Campbell et al. 1983); it may also help to (ii) preserve locally adapted genotypes (Stebbins 1957; Beattie 1976; Solbrig 1976), (iii) prevent the introduction of maladapted genes (Waller 1984), and (iv) eliminate the ‘meiotic cost’, which can reduce parental fitness by 50% (Solbrig 1976; Waller 1979). In addition, cleistogamous flowering reduces the energetic costs of pollinator attractants and rewards, androecium and male gametes (Solbrig 1976; Schemske 1978; Waller 1979), and sometimes protects the reproductive structures from unfavourable environmental conditions and predation (Campbell 1982; Campbell et al. 1983). Due to their independence of pollen vectors, cleistogamous flowers often have a higher fruit- and seed-set than chasmogamous ones (Schemske 1978; Schoen & Lloyd 1984).

Cleistogamy has also been well covered from a morphological viewpoint (Goebel 1904; Ritzerow 1908; Madge 1929; Bergdolt 1932; Lord 1982), as well as from a genetic one (Clay 1982; Campbell et al. 1983; Schoen & Lloyd 1984). However, few authors have addressed the questions of the benefit to an individual plant of producing both cross- and self-fertilized flowers and of the relative contribution of each flower type to overall individual or population fitness, taking into account temporal and spatial variation. Is selfing just a cheap and always reliable compensation for unsuccessful out-crossing, or do the two reproductive modes complement each other by having their optima under different environmental conditions? Likewise, few studies have been performed on the relationship between a plant's resource status and production of the two flower types, at least in perennials.

The purpose of this study was to test two hypotheses. Hypothesis 1: cleistogamy is a ‘bet-hedging’ strategy, where two alternating flowering modes maximize reproductive output in environments that fluctuate between conditions favourable and unfavourable for each mode (Schemske 1978; Campbell et al. 1983; Lloyd 1984; Masuda & Yahara 1994). Hypothesis 2: the resource environment affects chasmogamous flower (CH) production via the plant's size – a plant has to reach a certain size to afford CH flowers, while cleistogamous flower (CL) production is more size- and resource-independent (Schemske 1978; Waller 1980; Wilken 1982; Schoen 1984; Schmitt et al. 1987; Ruiz de Clavijo & Jimenez 1993; Le Corff 1993). To test these hypotheses, studies were conducted on the relationships between plant size and CH/CL flower production, fruit-set and seed-set in Oxalis acetosella L., a perennial herb in which the CH flowers occur prior to the CL ones during the growing season. As in many other cool-temperate, early flowering plants, CH fruit- and seed-set is likely to be pollinator-limited (Schemske et al. 1978). The data were collected during three growing seasons 1994–96, in three natural populations of O. acetosella.

Materials and methods

The study species

Oxalis acetosella (Oxalidaceae) is a low-growing perennial herb with a shallow, creeping rhizome rooting adventitiously and bearing swollen petiole bases that act as storage organs. It has extensive clonal growth, forming patches. Oxalis acetosella is winter-green and capable of growth under low light conditions, and it is sensitive to drought and strong sunlight. In Sweden, it is a common understorey plant of shady coniferous and deciduous forests throughout the country, except in the far north (Lagerberg 1948, pp. 1018–1020; Packham 1978).

All flowers are hermaphrodite and solitary and are borne on peduncles arising from the petiole bases. The CH flowers, which are produced in spring, are actinomorphic, pentamerous and slightly protandrous, with five styles and 10 stamens. Petals are white with purple nectar guides, and flower visitors include thrips, dipterans, bees, bumblebees and beetles (Packham 1978; Redbo-Torstensson & Berg 1995). The CL flowers, occurring in summer and early autumn, are greatly reduced in size, have very short peduncles, and are often hidden in moss or plant litter. The androecium is rudimentary with relatively few pollen grains, and sometimes the anthers of the five outer stamens are aborted. Pollen grains germinate within the anther, and the pollen tubes grow through the anther wall into the stigma. There are also reductions in the stigma and the ovary (Darwin 1877; Bennett 1880; Ritzerow 1908; Uphof 1938; Lagerberg 1948; Packham 1978). The fruit of both flower types is a capsule containing up to 10 seeds. Seeds are ballistically dispersed and can be thrown more than a metre (Hultén 1958, pp. 335–336; Packham 1978).

The study areas

In spring 1994, three sites were chosen for the study of phenology and reproduction in O. acetosella. All sites have a more or less continuous cover of the study species. Site 1 is situated 2 km south of Uppsala in central Sweden (59°49′60″ N, 17°41′20″ E). It is a relatively open and dry, south-west-sloping forest dominated by deciduous trees, e.g. Betula pendula and Quercus robur. Dominant field layer species include Anemone nemorosa, Oxalis acetosella, Hieracium sect. Silvaticiformia and Poa nemoralis. Site 2 is located approximately 300 m east of site 1 (59°49′60″ N, 17°41′23″ E). It is part of an old pine plantation and is quite closed, has mesic conditions and is dominated by middle-sized Pinus sylvestris. The field layer is dominated by Dryopteris filix-mas, Anemone nemorosa, Oxalis acetosella and Hieracium sect. Silvaticiformia. Site 3 is situated 10 km north-east of Uppsala (59°54′30″ N, 17°45′00″ E). It is a shady, damp spruce forest, dominated by Picea abies but also containing species such as Pinus sylvestris and Corylus avellana. Dominant field layer species are Anemone nemorosa, Oxalis acetosella, Vaccinium myrtillus and Deschampsia flexuosa. The moss layer is quite thick. This site is cooler than the others and has a shorter growing season.


At each site, 10 permanent quadrats were established. The positions of plots were not randomized, since they had to have a good Oxalis cover, including some reproductive ramets, but the plots were considered to be representative of the populations. Each quadrat was marked with two plastic tubes, driven into the ground, onto which a metal frame with a 2.5 × 2.5 cm mesh grid could be attached. The grid formed a coordinate system of 10 × 10 micro plots, and the observation area in each frame was thus 25 × 25 cm. This small plot size was chosen since it was a convenient working area and still contained sufficient numbers of ramets.

In early May 1994, 15 ramets in each plot were tagged with small pieces of plastic straw in individual colours, slipped over one of the petioles. The grid and tagging made it easy to follow individual ramets over long time periods. Most of the selected ramets had CH flowers by this time, but non-flowering ramets were also tagged to see whether they could produce CL flowers later. During remainder of this season and during the following two field seasons, from late April to mid-October, each plot was investigated two–four times a month, most frequently in June–August. Each CH and CL flower bud, flower, immature capsule and mature capsule on the tagged ramets was recorded. When a ramet died it was replaced by another, so that the number of tagged ramets in each plot remained at 15 (except for one plot, where often only 13 or 14 ramets could be found). Mature capsules on tagged ramets were collected for seed count. The number of leaves on each ramet was counted three times a year in early, mid- and late summer, in order to estimate the relationship between plant size and flower production.

During the field seasons of 1995 and 1996, all O. acetosella ramets in each plot were assessed as CH, CL or non-flowering. This was to estimate the percentage of CH and CL flowering ramets in each population.

Statistical analyses

For most analyses we used generalized linear models for analyses of deviance (McCullagh & Nelder 1989). The response variable number of flowers was considered to have a Poisson error distribution. When analysing the relative contribution of the two flower types to total number of mature fruits, the response variable was considered to have a binomial error distribution. When analysing fertilization, maturation and reproductive rate the response variable was set to 1 for success and 0 for failure. The relationships between fertilization and maturation rate and flower type, year and site were then analysed using a logistic regression model. This model is very similar to the one used above, with a binomial error distribution, except for the use of another link function (GENSTAT 5; Lawes Agriculture Trust 1995).

The effect of each explanatory variable was measured by the change in deviance caused by its inclusion in the model. The dispersion factor in all analyses was close to one, and the mean deviance (deviance/d.f.) of each variable was referred to a chi-square distribution for significance testing. Plots of residuals were always examined against fitted values, but in no case did this lead to the assumption of error distributions other than those mentioned.

The relationship between ramet size and flower production was analysed using simple linear regression. The response variable was square-root transformed.

Time of flower bud emergence and fruit development time (the time period from observation of bud to capsule dehiscence) were considered to have a Poisson error distribution. The relationship between these variables was analysed with simple linear regression.

The generalized linear model analyses were carried out using GENSTAT 5, release 3.2 (Lawes Agriculture Trust 1995). Other statistics were conducted with systat (Wilkinson 1992).



CH flower buds were observed between April and May, and flowers opened in mid- to late May, approximately 1 week later at site 3 than at the other sites. Fruits could be seen from early June onwards. Capsules dehisced between June and July, slightly earlier at site 1 than at the other sites. CH flowering and seed dispersal were thus rather synchronous among ramets.

The first CL buds (6% of total) had already emerged at the same time as the CH buds, but the majority (80%) appeared in June–August, and the last ones in late September (Fig. 1). Mean time of bud emergence varied between sites, being earliest at site 1 and latest at site 3 (F(2,1063) = 5.72, P < 0.005), but there was also an interaction between year and site (F(4,1057) = 4.35, P = 0.002). The earliest CL capsules dehisced in late June, the majority in July–September, and the last ones in late October. Mean time from observation of bud to capsule dehiscence was longer for CH flowers than for CL flowers (1.9 vs. 1.5 months, median = 2.0 vs. 1.3 months, Mann–Whitney U-test = 62670.5, n = 596, P < 0.001), but the variation was much smaller in the CH phase than in the CL phase. In the CH phase, fruit development time varied between years (longest in 1994, shortest in 1995; F(2,184) = 36.21, P < 0.001) and between sites (longest at site 3, shortest at site 1; F(2,184) = 38.67, P < 0.001). In the CL phase there was a slight but significant positive relation between fruit development time and phenology (r2 = 0.074, d.f. = 1, P < 0.001), probably due to lower temperatures towards the end of the growing season, but it also varied between years (longest in 1996, shortest in 1995; F(2,406) = 8.05, P < 0.001).

Figure 1.

The seasonal percentage distribution of CH (chasmogamous) and CL (cleistogamous) flowers in Oxalis acetosella. Data are pooled for three sites near Uppsala, central Sweden 1994–96.

Flower and fruit production

A total of 1978 flower buds was produced by the ramets investigated during the three growing seasons, of which 31% were chasmogamous and 69% cleistogamous. The mean number of CH buds/CH flowering ramet was 1.2 (maximum = 4, n = 525) and mean number of CL buds/CL flowering ramet was 2.0 (maximum = 8, n = 692).

Flowering frequency

The percentage of flower-producing ramets in the populations was significantly higher in 1996 than in 1995 (37% vs. 29%, χ2 = 16.75, d.f. = 1, P < 0.001). It also differed between populations, being highest at site 2 and lowest at site 3 (F(2,2762) = 126.97, P < 0.001). The percentage of ramets in the populations producing CH flowers was 15% in 1995 and 19% in 1996 (χ2 = 6.06, d.f. = 1, P < 0.05). It also differed between populations, being highest at site 2 and lowest at site 3 (F(2,2761) = 77.67, P < 0.001), and there was an interaction between year and population (F(2,2758) = 8.50, P < 0.001). The percentage of ramets with CL production varied in the same pattern between years (1995, 26%; 1996, 32%; χ2 = 12.81, d.f. = 1, P < 0.001) and between populations (F(2,2762) = 121.18, P < 0.001), but the variation was larger than in the CH phase (Table 1).

Table 1.  An overview of chasmogamous (CH) and cleistogamous (CL) flower, fruit and seed production in three populations of Oxalis acetosella near Uppsala, central Sweden 1994–96. Site 1, deciduous forest; site 2, pine forest; site 3, spruce forest. Numbers of flowers and mature fruits per ramet are mean numbers per flower-producing ramet (SD within parentheses). Flower fertilization rate is the ratio of immature fruits/flower, fruit maturation rate is the ratio of mature fruits/immature fruit
Site 1Site 2Site 3Site 1Site 2Site 3
Percentage of flowering rametsNo dataNo dataNo dataNo dataNo dataNo data
Number of flowers per ramet1.1 (0.35)1.2 (0.47)1.1 (0.38)1.9 (1.13)1.6 (0.96)1.2 (0.38)
Number of mature fruits per ramet0.2 (0.43)0.5 (0.59)0.5 (0.53)0.7 (0.69)0.4 (0.72)0.1 (0.28)
Flower fertilization rate0.810.690.710.790.790.53
Fruit maturation rate0.290.630.650.450.350.13
Mean number of seeds per fruit5.2 (1.33)No data3.6 (2.88)3.7 (2.39)4.1 (1.60)No data
Percentage of flowering ramets1724329396
Number of flowers per ramet1.2 (0.54)1.2 (0.40)1.1 (0.33)2.2 (1.25)2.3 (1.16)1.2 (0.59)
Number of mature fruits per ramet0.1 (0.26)0.4 (0.51)0.1 (0.33)0.9 (1.05)1.4 (1.01)0.5 (0.51)
Flower fertilization rate0.190.420.200.770.890.62
Fruit maturation rate0.310.710.500.540.680.63
Mean number of seeds per fruitNo data3.2 (1.14)4.2 (1.09)5.4 (2.97)5.1 (2.43)2.7 (1.50)
Percentage of flowering ramets133210344613
Number of flowers per ramet1.1 (0.35)1.2 (0.46)1.0 (0.00)2.0 (1.03)2.2 (1.45)1.5 (0.68)
Number of mature fruits per ramet0.3 (0.47)0.6 (0.60)0.5 (0.51)0.8 (0.81)0.9 (0.92)0.6 (0.63)
Flower fertilization rate0.580.540.810.680.750.66
Fruit maturation rate0.460.880.600.580.570.66
Mean number of seeds per fruit3.5 (3.51)3.6 (1.77)3.7 (2.12)4.4 (2.10)4.4 (2.23)3.6 (1.85)

Flower production in relation to ramet size

CH flower number was positively related to ramet size expressed as yearly mean leaf number (n = 525, P = 0.001), but leaf number explained only 2% of the variation. We even observed ramets with no leaves producing CH flowers. CL flower number was also positively related to ramet size, leaf number explaining 15% of the variation (n = 691, P < 0.001; Fig. 2).

Figure 2.

The relation between ramet size, expressed as yearly mean leaf number, and number of CH (chasmogamous) and CL (cleistogamous) flowers in Oxalis acetosella. Data are pooled for three sites near Uppsala, central Sweden 1994–96. Values presented are means with standard deviations.

Most ramets (c. 73%) in the populations with CH flowers also produced CL flowers, and there was a positive correlation between per ramet number of flowers of the two types (Spearman rank correlation = 0.333, n = 1482, P < 0.001). CL flowers only were produced by approximately 17% of the ramets in the populations.

Variation in flower production between years and sites

Mean number of CH flowers per ramet varied significantly between years (χ2 = 13.65, d.f. = 2, P < 0.001), but not between sites (Table 1).

Mean number of CL flowers per ramet varied significantly between years, being highest in 1996 and lowest in 1994 (χ2 = 32.62, d.f. = 2, P < 0.001; Table 1). It varied between sites as well, being highest at site 2 and lowest at site 3 (χ2 = 52.70, d.f. = 2, P < 0.001; Table 1).

CL flower production was also affected by fertilization success in the CH phase. An analysis of the effect of number of unfertilized CH flowers on CL flower number, after removing the effects of year, site and ramet size by including these variables first in the analysis, showed a significant positive relation (χ2 = 5.16, d.f. = 1, P < 0.025).

Total flower production per ramet

The total number of flowers per reproductive ramet varied significantly between years (χ2 = 43.08, d.f. = 2, P < 0.001) and between sites (χ2 = 26.43, d.f. = 2, P < 0.001). There was also a significant effect of fertilization rate in the CH phase (χ2 = 59.79, d.f. = 1, P < 0.001): the more unfertilized CH flowers per ramet, the higher the total flower number produced.

Flower fertilization rate

There was significant variation in flower fertilization rate (the ratio number of capsules/number of flowers) between years (χ2 = 14.20, d.f. = 2, P < 0.001), but not between sites. CL flowers had a significantly higher fertilization rate than CH flowers (χ2 = 84.54, d.f. = 1, P < 0.001). The interaction between year and flower type had a significant and substantial effect on fertilization rate (χ2 = 36.70, d.f. = 2, P < 0.001), and there was also a slight but significant effect of the interaction of site × flower type (χ2 = 6.00, d.f. = 2, P < 0.05). Analysing each flower type separately, the CH flower fertilization rate varied significantly between years but not between sites, while the reverse was true for the CL phase (Table 1).

Fruit maturation rate

Fruit maturation rate differed significantly between years (χ2 = 18.51, d.f. = 2, P < 0.001) and between sites (χ2 = 7.03, d.f. = 2, P < 0.05). It was significantly higher for CH fruits than for CL fruits (χ2 = 4.60, d.f. = 1, P < 0.05). However, there was a significant interaction of site × flower type (χ2 = 17.39, d.f. = 2, P < 0.001): CH fruit maturation rate varied significantly between sites, in contrast to CL fruit maturation rate (Table 1).

Reproductive rate

The probability of a flower developing into a mature capsule (the ratio number of mature capsules/number of flowers) varied significantly between sites (χ2 = 10.91, d.f. = 2, P < 0.005) but not between years. It was significantly higher for CL flowers than for CH flowers (χ2 = 12.57, d.f. = 1, P < 0.001). The interactions of year × flower type and site × flower type were both significant (χ2 = 25.13, d.f. = 2, P < 0.001; χ2 = 11.31, d.f. = 2, P < 0.005). In 1994 the probability of becoming a dehisced capsule was higher for CH flowers than for CL flowers, while the reverse was true in 1995. In 1996, the probability was equal for the two flower types. At site 1, CL flowers had a higher probability of producing seed than CH flowers in all years, while the relationship between the two flower types varied between years at the other sites (Table 2).

Table 2.  Reproductive rate (the ratio of mature fruits/flower) in three populations of Oxalis acetosella near Uppsala, central Sweden 1994–96, for chasmogamous (CH) and cleistogamous (CL) flowers. Site 1, deciduous forest; site 2, pine forest; site 3, spruce forest
Site 10.23990.36175< 0.05
Site 20.431050.28149< 0.05
Site 30.47730.0715< 0.005
Site 10.06690.42226< 0.001
Site 20.29820.60264< 0.001
Site 30.10100.3926  NS
Site 10.27480.40200  NS
Site 20.47930.43255  NS
Site 30.48310.4457  NS

The total number of mature capsules per reproductive ramet differed significantly between years (χ2 = 14.57, d.f. = 2, P < 0.001) and between sites (χ2 = 20.73, d.f. = 2, P < 0.001) (Table 3). The relative contribution of the two flower types to the total number of mature capsules differed significantly between years (χ2 = 48.35, d.f. = 2, P < 0.001) and between sites (χ2 = 28.33, d.f. = 2, P < 0.001) (Table 4).

Table 3.  Mean per ramet total number of mature fruits in three populations of Oxalis acetosella near Uppsala, central Sweden 1994–96. Site 1, deciduous forest; site 2, pine forest, site 3; spruce forest. The values are calculated from all reproductive ramets (SE within parentheses)
1994n1995n1996n P
Site 10.7 (0.08)1160.9 (0.09)1120.9 (0.09)107  NS
Site 20.7 (0.09)1141.5 (0.09)1231.2 (0.09)130< 0.001
Site 30.5 (0.07)720.4 (0.11)260.7 (0.08)57  NS
Table 4.  The percentage contribution of the cleistogamous phase to the total number of mature fruits in three populations of Oxalis acetosella near Uppsala, central Sweden 1994–96. Site 1, deciduous forest; site 2, pine forest; site 3, spruce forest
Site 1738696988692
Site 248868718271153
Site 333591116340

Seed production

CL capsules contained more seeds than CH capsules (mean = 4.6 vs. 3.8, median = 4.0 vs. 4.0, Mann–Whitney U-test = 5829.5, n = 242, P < 0.025). This difference was more pronounced in ramets where one CH flower had failed to become fertilized (ramets with more than one CH flower unfertilized were excluded from this analysis because of insufficient CL capsule material). These ramets had a higher number of seeds/ CL capsule than ramets with all their CH flowers fertilized (mean = 5.1 vs. 3.7, median = 5.0 vs. 4.0, Mann–Whitney U-test = 234.5, n = 53, P < 0.05). If the mean numbers of seeds per capsule were combined with the mean numbers of capsules per ramet for each year and site, this yielded the same total seed number for ramets with all their CH flowers fertilized as for those with one CH flower unfertilized (mean = 5.5 vs. 7.6; Fig. 3), although the latter group had a lower total number of capsules than the former (mean = 0.8 vs. 1.1, median = 1.0 vs. 1.0, Mann–Whitney U-test = 35501.5, n = 497, P < 0.001). Thus, ramets with poor CH fertilization did not suffer from decreased total reproductive output.

Figure 3.

Total number of CH and CL (chasmogamously and cleistogamously produced, respectively) seeds, and seeds of both types, in Oxalis acetosella ramets with 0 and 1 CH flowers unfertilized. Data are pooled for three sites near Uppsala, central Sweden 1994–96. Values presented (with standard deviations) are calculated from mean numbers of CH and CL fruits and seeds/fruit at each site in each year.

The seed number of CL capsules was also negatively correlated to the time of year of capsule dehiscence: the later the maturation, the fewer the seeds (r2 = 0.030, d.f. = 1, P < 0.05). There was no such correlation in the CH phase, where capsule dehiscence was much more synchronous.


There was a good correlation between leaf and CL flower number in O. acetosella, indicating that such flower production is affected by ramet size. However, this correlation was much weaker in the CH phase: very few of the investigated ramets (0.5%) produced more than two CH buds in one growing season. Production of CL flowers seems to be more size-dependent in this species. Many earlier studies have shown the opposite relationship, e.g. Waller (1980), Wilken (1982), Schmitt et al. (1987) and Ruiz de Clavijo & Jimenez (1993), but these have been made on annual plants. Jasieniuk & Lechowicz (1987) observed size-dependence of both flowering modes in the perennial O. montana, with a stronger dependence in the CL phase. Clay (1982) and Mattila & Salonen (1995) found CL flower production in the perennials Danthonia spicata and Viola mirabilis, respectively, to increase with plant size, whereas CH production was size-independent. Perhaps short-lived annuals, having only one growing season, have to secure reproduction through fast and cheap self-fertilization, while perennials can use stored resources to adjust the proportions between out-crossing and selfing (Jasieniuk & Lechowicz 1987). Moreover, in perennials, resource status is not a function of size of above-ground parts alone (Le Corff 1993).

The proportion of ramets with CH production varied significantly between years (1995 and 1996) and populations, while the CH flower number per ramet was much the same from year to year and from site to site. CL flower production differed significantly between years and sites, both on a population and a ramet scale. This implies that the CL phase is more dependent on weather and local climate than the CH phase, again in conflict with most theories. Jasieniuk & Lechowicz (1987) found that CL flower production in O. montana varied significantly between years and sites, and was highest at the sites most suitable for growth and reproduction. Bell & Quinn (1987), likewise, found allocation to CL reproduction to vary with soil moisture in the perennial Dicanthelium clandestinum. Contradictory trends are shown in the present study, however: the CL flower number per ramet was lowest in 1994, when the summer was unusually hot and dry in the Uppsala area, and it was also lower at site 3, which was the coolest and moistest site. The percentage of reproducing ramets in the population was the lowest in both flowering phases at this site. However, CL flower number was also negatively correlated to CH fertilization rate, which was highest in 1994. This was probably due to the low precipitation in May 1994, which favoured pollinator activity.

CH fertilization success differed between years but not between sites, while in the CL phase the relationship was the opposite. This suggests that the two reproductive modes are affected by environmental variation on different temporal and spatial scales. For the CH phase, pollinator activity may vary more due to large-scale weather conditions than to local climate, whereas CL flower survival is influenced to a higher degree by site conditions. Most of the non-surviving CL flowers seemed to have been killed by drought in the bud stage. The observation that fertilization rate is generally higher in CL flowers than in CH ones was expected, but the difference was not very large. At site 3, fertilization was actually better in the CH phase both in 1994 and 1996. Several authors have found a complete, or almost complete, fruit-set in the CL phase in contrast to a very poor CH performance in a variety of species (McNamara & Quinn 1977; Campbell 1982; Schoen 1984). It must, however, be kept in mind that failure in fertilization does not necessarily mean total reproductive failure for a CH flower, since it may still contribute pollen to other flowers. In contrast, a CL fertilization failure is a failure in both maternal and paternal function.

The response of fruit maturation rate to temporal and spatial variation was quite the reverse of that of flower fertilization rate: CH maturation differed between sites, CL maturation between years. When maturation was combined with fertilization success to calculate a reproductive rate, there was a small advantage for the CL phase, but this disappeared when the variation between years and sites was included. Thus, the overall probability for a flower bud to develop into a dehisced fruit is not higher in the CL phase than in the CH phase. The well-established view that cleistogamy is a fail-safe mechanism for reproduction, virtually uninfluenced by external factors (e.g. Schoen 1984), apparently does not hold for O. acetosella. Rather, the two reproductive modes complement each other through different responses to varying environmental conditions. By producing two different flower types, a plant most certainly yields more offspring than it would do by producing just one type over the same time period. This is supported further by the facts that the investigated ramets in two populations produced fewer mature CL fruits than mature CH fruits in 1994, and that the between-year variation in the mean total fruit number per ramet within populations was quite small despite the variations in CH and CL reproduction.

The higher mean seed number of CL fruits compared with CH fruits agrees with previous studies, e.g. Levin (1972), Ellstrand et al. (1984) and Jasieniuk & Lechowicz (1987). Here too, the difference between the two reproductive modes is not remarkable.

The relationship found between CH fertilization success and CL production is in agreement with the observations of Redbo-Torstensson & Berg (1995). Ramets with one or more CH flowers unfertilized had significantly more CL flowers than ramets with all their CH flowers fertilized, and approximately the same total seed number. Thus, an O. acetosella ramet is able to compensate for losses in seed-set due to poor fertilization in the CH phase by increasing the production of CL flowers and capsules. It is possible that the number of seeds per CL capsule is also increased compared to that seen in a ramet with better CH fertilization, although this hypothesis needs confirming by analyses of a larger sample of capsule material. The increase is probably achieved through reallocation of resources from CH fruit maturation to CL flower (and seed) production. This is supported by the fact that predation or death of immature CH capsules, which have already acquired additional resources, does not result in the same increase in the CL phase (Redbo-Torstensson & Berg 1995).

The observation that several CL buds appear simultaneously with the CH buds was unexpected, disagreeing with all previous literature on Oxalis (Bennett 1880; Lagerberg 1948; Hultén 1958; Packham 1978; Jasieniuk & Lechowicz 1987) and contradicting the resource allocation hypothesis. It is also in conflict with the assumption that production of the two flower types is triggered by different environmental conditions. One explanation might be that these CL buds have already been initiated in the preceding autumn, but did not have time to develop into flowers before the end of the growing season. Instead, they have become dormant until conditions become favourable. Small structures that might be flower buds have indeed been found on O. acetosella ramets in late autumn and in winter (P. Redbo-Torstensson and H. Berg, personal observations), but it is difficult to determine whether they are buds and, if they are, of which flower type. However, the majority of CL flowers were initiated after the CH flowers had either become fertilized or died. Therefore, the cleistogamous mode in O. acetosella can be viewed partly as a plastic response to complement out-crossing and secure a certain seed output (Redbo-Torstensson & Berg 1995).

In summary, the balance between chasmogamous and cleistogamous reproduction in O. acetosella is regulated by complicated interactions between plant size, weather conditions and local environment. The largest variation in per ramet flower number is in the CL phase, since most ramets seem to be unable to produce more than two CH flowers. Moreover, CL flower production increases when fertilization success in the CH phase is low, probably due to reallocation of resources within the ramet. Flower fertilization and fruit maturation do not vary in the same temporal and spatial pattern in the two phases, and the probability of a CL flower becoming a mature capsule is not higher than that of a CH flower. Thus, CL reproduction is not more secure than CH reproduction, and a population or an individual would not gain from being entirely cleistogamous (Lloyd 1984). Two main conclusions can be drawn:

1 ÜixCleistogamy in O. acetosella is a bet-hedging strategy, which optimizes reproductive output through two modes of sexual reproduction responding differently to temporal and spatial environmental variation. It also reduces the between-year variation in fruit number.

2 ÜixCleistogamous reproduction is not more resource-independent than chasmogamous reproduction in O. acetosella.

Further questions about cleistogamy in connection with the bet-hedging hypothesis concern the offspring of the two flower types. There is a possibility that the less costly selfed progeny are of lower quality than progeny of out-crossing, as a consequence of fewer resources put into the seed and/or of inbreeding depression. This could be expressed as a lower seed germination rate and/or reduced seedling survivorship in offspring derived from selfing compared with offspring of out-crossing (Waller 1979, 1984; Schoen 1983; Mitchell-Olds & Waller 1985; Jasieniuk & Lechowicz 1987), but if the difference is not too large it should still be advantageous to produce both offspring kinds. Another reason for maintaining two modes of reproduction could be to produce two kinds of diaspore, which differ in traits such as size, dispersal, habitat requirements and germination time (seed dimorphism; McNamara & Quinn 1977; Schemske 1978; Campbell et al. 1983; Lloyd 1984; Schoen & Lloyd 1984; Schmitt et al. 1985; Cheplick 1987). Both these hypotheses suggest that cleistogamy is a strategy for maximizing fitness in spatially and temporally heterogeneous environments, i.e. a bet-hedging strategy. We will test these two hypotheses in future studies.


We would like to thank Valerie Cochrane and Nigel Rollison for valuable comments on the manuscript, as well as Gregory P. Cheplick and Jeff Ollerton. We are grateful to Stefan Björklund, technician at the Department of Ecological Botany, who made the metal frame with grid. We are also grateful to Christina Munkert, for assistance in the field, and to the people in our departments for support and advice. The study was financed by the Swedish Natural Science Research Council (NFR).

Received 30 June 1997revision accepted 27 November 1997