Oecologia (1999) 120:235±241

Ó Springer-Verlag 1999

Ana H. Ladio á Marcelo A. Aizen

Early reproductive failure increases nectar production and pollination success of late ¯owers in south Andean Alstroemeria aurea

Received: 23 June 1998 / Accepted: 3 May 1999

Abstract Fertile ramets of bumblebee-pollinated Alstroemeria aurea, a clonal perennial native to the temperate forests of the southern Andes, produce single terminal in¯orescences that may bear two or more temporally non-overlapping whorls of ¯owers. While fruit set is commonly high (>80%) among early-opening ¯owers, it is usually low (<20%) among late-opening ¯owers within ramets. Using ¯owering ramets with two whorls of ¯owers, we examined experimentally the following related hypotheses. First, late ¯owers act as a reserve of ovaries, increasing their likelihood of setting seed when early fruits abort due to either pollen or resource limitation. Second, where early fruit abortion has occurred, plants may actively ensure pollination of late ¯owers by increasing their attractants and rewards. In a natural population, we simulated (1) lack of pollen deposition in early ¯owers, by excising their stigmas just before receptivity, and (2) resource limitation, by removing all the leaves from an experimental ¯owering ramet. Treatments were applied to individual ramets according to a 2 ´ 2 factorial design. We found that when early ¯owers failed to set fruit due to stigma excision, nectar secretion and particularly pollen receipt strongly increased in late ¯owers. Higher pollen deposition contributed signi®cantly to the observed ®ve-fold increase in seed output of late ¯owers. Fruit and seed set from early ¯owers were more negatively a€ected by defoliation than that from late ¯owers. Defoliation did not interfere with a ramet's capacity to increase late reproductive output when early reproduction failed. These results support the assertion that late ¯owers act as a reserve of ovaries helping a plant to cope with an

A.H. Ladio á M.A. Aizen (&) Departamento de EcologõÂ a, Universidad Nacional del Comahue, Centro Regional Bariloche, Unidad Postal Universidad, 8400 San Carlos de Bariloche, RõÂ o Negro, Argentina e-mail: [email protected], Fax: +54-944-22111

unpredictable environment. These results also suggest that plants may actively increase pollinator visitation by opportunistically increasing ¯ower rewards. Key words Late-opening ¯owers á Ovary reserve á Nectar secretion á Pollination á Alstroemeria aurea

Introduction Many plant species produce late-opening ¯owers that usually do not mature into fruits. Several non-mutually exclusive hypotheses have been advanced to explain this pattern. Among them is the proposition that late-opening ¯owers represent a reserve of ovaries to bu€er infrequent adverse conditions in terms of pollinator or resource availability, or fruit and seed damage, which might compromise early reproduction (Sutherland and Delph 1984; Sutherland 1986; EhrleÂn 1991, 1993; Guitian 1993, 1994). According to this adaptive hypothesis, the cost associated with an apparently wasteful production of ¯owers is o€set by the bene®t of spreading the risk of ovary mortality over time. An increase in reproductive success of late ¯owers when early ¯owers fail either naturally or experimentally to set seed is commonly found (e.g., Lee 1988; Solomon 1988; Herrera 1991; Guitian 1993; Harriss and Whelan 1993; Niesembaum 1996; StoÈcklin 1997), which supports the evolutionary interpretation that surplus ¯ower production provides a mechanism of maternal reproductive reassurance. Resource preemption by successfully pollinated early ¯owers is assumed to represent the proximate mechanism of frequently observed late-¯ower reproductive failure (Stephenson 1986; Lee 1988; Diggle 1995; Brunet 1996). Only when early ¯owers fail to reproduce might resources be free for late ¯owers to set seed, suggesting that seed production in such plants is resource-limited overall. Little is known, however, of how resources are partitioned between pre- and postzygotic functions in late ¯owers (but see Ashman 1992; Wolfe 1992). In animal-

236

pollinated species, plants may, to some extent, increase reproduction by investing in ¯ower attractants (e.g., size, fragance) and rewards (e.g., nectar, pollen) that might, in turn, a€ect pollinator visitation and pollination levels. If an unpredictable pollinator environment is indeed one of the key factors involved in the evolution and maintenance of surplus ¯ower production, plants might then bene®t from mechanisms that ensure pollination of lateopening ¯owers in the case of early reproductive failure. Although several studies have now examined how late reproductive success might depend on the fate of early-blooming ¯owers, it is unknown whether late reproduction is, at least in part, enhanced through plant manipulation of pollinator visitation, and not just as a direct consequence of resource shifting to fruit and seed production. Here we provide evidence suggesting that this is the case in the bumblebee-pollinated Alstroemeria aurea Graham (Alstroemeriaceae), a perennial herb native to the temperate forests of the southern Andes. Flowering ramets of this species can produce compound in¯orescences with two or more temporally distinctive whorls of ¯owers (Aizen and Basilio 1995). We conducted a ®eld experiment to study how pollination and reproductive success of secondary, lateopening ¯owers in compound in¯orescences of A. aurea varied when we simulated lack of pollen deposition in primary, early-opening ¯owers and hindered photosynthetic capacity by removing leaves of ¯owering shoots. We also evaluated whether impaired early reproduction enhanced ¯oral features in late ¯owers that might increase pollinator visits (¯ower size and nectar production) and pollen capture eciency (stigma size).

Materials and methods Study species and area A. aurea is a clonal, self-compatible plant that can reproduce vegetatively by rhizome branching and fragmentation, and sexually by seed. Despite the occurrence of vegetative propagation, clones are dicult to identify in the ®eld because of continuous seedling recruitment within established patches (Puntieri 1991). Each growing season, an underground rhizome produces a series of vegetative ramets and usually one ¯owering ramet bearing a terminal simple or compound umbellate in¯orescence. Flowers are odourless, large (~5 cm), long lived (8±10 days), and protandrous with a 4- to 5-day male phase that is followed by a ~1-day neuter phase, and a 3- to 4-day female phase. The start of the female phase is marked by the separation of the three stigmatic branches (Aizen and Basilio 1995). Nectar, the main ¯ower reward for which bumblebees actively forage in this plant, is produced at the base of two nectariferous tepals that are streaked with reddish nectar guides (Aizen and Ra€aele 1996; Aizen and Basilio 1998). Fertilized ovaries become capsules containing 1±25 ballistically dispersed seeds. In¯orescences consist of a whorl of two to seven primary ¯owers. In ramets producing compound in¯orescences, primary pedicels branch bearing one secondary ¯ower each. Plants with tertiary or higher-order ¯owers are rare. The proportion of ramets with simple or compound in¯orescences is extremely variable among populations (from <20% to >80%; M.A. Aizen, unpublished work). Flowers belonging to the same whorl open, change sexual-phase, and senesce synchronously. In compound in¯ore-

scences, secondary (hereafter late) ¯owers open directly after primary (hereafter early) ¯owers senesce (Aizen and Basilio 1995). The ®eld experiment was conducted in a natural population of A. aurea occurring on the eastern slope of Cerro Otto approximately 500 m away from the city boundary of San Carlos de Bariloche (41°8¢S, 71°19¢W) in the southern lake district of Argentina. This population occurs in an open Austrocedrus chilensis forest. In this population of Alstroemeria aurea, shoots producing compound in¯orescences were the dominant ¯owering ramet type (>80%). The native Bombus dahlbomii and the exotic B. ruderatus, a recent invader of European origin, were the main ¯ower visitors and pollinators. Experimental design and sampling In mid-December 1995, we tagged a total of 114 ¯owering ramets producing early and late ¯ower buds in equal numbers. Sampled ramets were distributed in three dense ¯owering patches 20±50 m apart. For each ramet, we measured stem length and counted the number of early and late ¯ower buds. We randomly assigned to each ramet one of the following four combinations of experimental leaf and stigma removal treatments: (1) no manipulation, (2) leaves removed, (3) stigmas removed, and (4) both leaves and stigmas removed. In the leaf removal treatment, we cut all the leaves of a given ¯owering ramet, including those subtending the in¯orescence, c. 2 weeks before anthesis. Defoliation of a ¯owering shoot represents the removal of 10±20% of the total leaf area of a plant (i.e., a ¯owering shoot and associated vegetative shoots). In the stigma excision treatment, we carefully clipped the three stigmatic branches of all the primary (early) ¯owers using a pair of ®nepointed forceps. This treatment was performed in the middle of the male phase c. 2±3 days before the ®nal elongation of the style and spreading out of the stigma that marks the start of stigma receptivity. Other than preventing pollen receipt, this treatment did not have any other apparent e€ect on early ¯ower characteristics such as size and nectar secretion (see Results), or date of anthesis and ¯ower longevity (A. Ladio, personal observation). We estimated ¯ower size by measuring the length of the two nectariferous tepals of each of two haphazardly-chosen early and two late ¯owers per ramet (see also Aizen and Ra€aele 1996). Daytime nectar secretion rate was measured by bagging in¯orescences with ®ne-mesh mosquito netting between 0900 and 1000 hours after emptying the standing nectar from either two early or late ¯owers per whorl. Secreted nectar was sampled 4 h later by inserting a 4-ll micropapillary tube repeatedly in the two nectaries of each of the two ¯owers until no nectar could be extracted. In this species, variation in nectar volume rather than in sugar concentration is the main determinant of variation in nectar sugar content (Aizen and Ra€aele 1996; Aizen and Basilio 1998). Because nectar secretion varies in relation to sexual phase (Aizen and Basilio 1998) and because our study focused on the female function (i.e., pollen receipt and seed set), nectar production was measured in the middle of the female ¯ower phase. Excluding pollinators for a 4-h daytime period might have reduced natural levels of pollen deposition by about 10%. We collected all stigmas of each in¯orescence (except those of the primary ¯owers in stigma-excised in¯orescences) from every focal ramet 1±2 days after ¯ower senescence, and just before the natural shedding of the style. Stigmas were mounted on a slide and stained with Alexander's solution (Alexander 1980). Pollen grains were counted under a stereoscopic microscope at 400´. Despite self-compatibility, A. aurea avoids self-pollen deposition at the ramet level by its synchronous protandry (Aizen and Basilio 1995). Therefore, pollen receipt in this species is completely pollinatormediated and the amount deposited is a good surrogate of pollinator visitation (Aizen and Basilio 1998). We also measured the length of each of the three stigmatic branches to the nearest mm and used the average length as a measure of pollen capture eciency. In this species, the amount of pollen deposition per bumblebee visit may be in¯uenced by stigma size (M.A. Aizen, unpublished work). We collected all mature fruits 5±6 week after

237 anthesis and about 1 week before seed dispersal. In the laboratory, we counted seeds and undeveloped ovules, and weighed all seeds from each single fruit as a group. Individual seed mass was estimated by dividing total seed mass by the number of seeds. Data analysis We analyzed the e€ect of leaf and stigma removal treatments on ¯oral and reproductive characteristics of early and late ¯owers, separately, by means of a 2 ´ 2 factorial ANOVA, including ¯owering patch as a blocking factor. The basic model included the ``leaf removal'' factor (®xed, 2 levels), the ``stigma excision'' factor (®xed, 2 levels), the interaction between them, and the blocking factor (random, 3 patches). The interaction between the ®xed factors and the blocking factor was not included in the ®nal model (see Mead 1988; Newman et al. 1997). Dependent variables were tepal length, stigma branch length, nectar production rate, number of pollen grains/stigma, number of ovules/¯ower (estimated as number of seeds+undeveloped ovules), seed output (number of seeds/ ¯ower), fruit set (number of fruits/number of ¯owers), seed set (number of seeds/fruit), and individual seed mass. Dependent variables were averaged over all ¯owers (either early or late) of an in¯orescence. Because of stigma excision, we only included the leaf removal factor in the analysis of stigma size and pollen receipt, and of fruit and seed variables of early ¯owers as those ¯owers failed to set fruit. Ovule number of early, stigma-excised ¯owers also was not estimated. We used ANCOVA to test for treatment e€ects on seed output after accounting for di€erences in pollen receipt. Relationships between average seed set and pollen receipt across ramets were described by simple linear regressions. Data were analyzed using PROC GLM (SAS 1988). Because data sets were not balanced, we based signi®cance tests for treatment factors on type III sums of squares (SAS 1988). Least-squares means and standard errors are reported in Fig. 1 because numbers of observations per category were unequal.

Results Flowering shoots averaged (x ‹ 1 SD) 72.5 ‹ 13.81 cm in length and bore 4.2 ‹ 0.95 ¯owers per whorl. Ramets assigned to the four treatment groups were similar in these characteristics (F3,103 = 1.09, P = 0.36 for shoot length, and F3,103 = 0.28, P = 0.84 for number of ¯owers/whorl) despite the existence of signi®cant inter-patch variation (F2,103 = 7.42, P < 0.001 for shoot length, and F2,103 = 16.3, P < 0.0001 for ¯owers/whorl). Thus, we assumed that ramets assigned to the di€erent treatment categories were similar with respect to other characteristics and that any di€erence in the measured reproductive traits arose as a consequence of the treatments. Prezygotic variables Shoot defoliation negatively a€ected ¯ower size as estimated by tepal length of both early and late ¯owers (Fig. 1a). This e€ect was weak in general but tended to be slightly stronger for late than for early ¯owers (Table 1a). On the other hand, stigma excision of early ¯owers did not have any e€ect on ¯ower size for either early or late ¯owers (Fig. 1a, Table 1a). Defoliation also negatively and signi®cantly a€ected the size of the stigma of late, but not of early, ¯owers (Fig. 1b, Table 1b).

Fig. 1 Least-squares means ‹ 1 SE for each of the 2 ´ 2 combination categories of defoliation and stigma excision treatments for the di€erent measured pre- and postzygotic characteristics of early- and late-opening ¯owers: a ¯ower size (tepal length), b stigma size (stigma branch length), c nectar production rate (volume secreted/4 h), d pollen receipt (no. pollen grains/stigma), e ovule number (no. ovules/¯ower), f seed output (no. seeds/¯ower), g fruit set (no. fruits/ ¯ower), h seed set (no. seeds/fruit), and i seed size (individual seed mass) (Control no manipulation, Defoliation leaves removed only, Excision stigmas removed only, Both both leaves and stigmas removed). E€ects due to defoliation are evaluated by comparing the Control vs. Defoliation, and Excision vs. Both means, and those due to stigma excision by comparing the Control vs. Excision and Defoliation vs. Both means. F-values associated with the defoliation and stigma excision factors are shown in Table 1

There was no evidence that either shoot defoliation or stigma excision a€ected nectar production of early ¯owers. On the other hand, stigma excision of early ¯owers signi®cantly increased nectar production of late ¯owers by an average of 36% (Fig. 1c, Table 1c). Stigma excision of early ¯owers also was associated with a large and highly signi®cant increase in pollen receipt of late ¯owers (Table 1d). Late ¯owers of ramets in which early reproduction was experimentally impeded received twice as many pollen grains as ramets in which early ¯owers were allowed to receive pollen and set fruit (Fig. 1d). We found an overall signi®cant relationship between nectar production and pollen receipt for late ¯owers

238 Table 1a±i ANOVA results for e€ects of leaf removal, stigma excision of early ¯owers, and their interaction on pre- and postzygotic characteristics of early- and late-opening ¯owers. Flowering patch was included as a blocking factor. F-values were estimated from

type III sums of squares (SAS 1988). See Fig. 1 for treatment means and variable de®nitions. Only the defoliation factor was included in the analysis of stigma size, pollen receipt, ovule number, and fruit and seed variables of early ¯owers

Prezygotic Variable/Source

Postzygotic Early ¯owers df

a Tepal length Defoliation (D) Excision (E) D´E Block Error

1 1 1 2 91

b Stigma length Defoliation (D) Excision (E) D´E Block Error

1 ± ± 2 46

c Nectar production Defoliation (D) Excision (E) D´E Block Error

1 1 1 2 87

d Pollen receipt Defoliation (D) Excision (E) D´E Block Error e Ovule number Defoliation (D) Excision (E) D´E Block Error

MS 47.79 1.22 17.51 30.34 13.74

Late ¯owers F

df

3.48(*) 0.09 1.27 2.21

1 1 1 2 82

0.005 ± ± 0.062 0.017

0.27 ± ± 3.69*1

1 1 1 2 85

0.189 0.285 0.213 0.180 0.166

1.14 1.72 1.28 1.09

1 1 1 2 76

1 988.2 ± ± ± ± 2 7011.3 46 3247.2

0.30 ± ± 2.16

1 1 1 2 84

1 ± ± 2 57

0.05 ± ± 0.17

1 1 1 2 34

0.65 ± ± 2.18 12.77

MS

Variable/Source Early ¯owers F

df

MS

Late ¯owers F

df

MS

F

7.01*1 ± ± 2.48(*)

1 1 1 2 97

14.84 305.71 3.87 1.13 6.33

2.78(*) ± ± 0.59

1 1 1 2 97

0.172 1.42 6.110 50.42*4 0.009 0.07 0.100 0.83 0.121

6.65*1 2.40 2.32 0.06

f Seed output Defoliation (D) 1 Excision (E) ± D´E ± Block 2 Error 61

0.074 0.015 0.002 0.013 0.012

6.13*1 1.24 0.20 1.08

g Fruit set Defoliation (D) 1 Excision (E) ± D´E ± Block 2 Error 62

0.192 0.481 0.058 0.009 0.096

2.01 5.03*1 0.60 0.10

h Seed set Defoliation (D) 1 Excision (E) ± D´E ± Block 2 Error 57

31.75 ± ± 12.84 7.75

4.09*1 ± ± 1.66

1 1 1 2 34

0.97 2.42 1.73 5.68 5.85

0.17 0.41 0.29 0.97

1.03 22.59*4 0.47 1.64

i Seed mass Defoliation (D) 1 Excision (E) ± D´E ± Block 2 Error 57

4.28 ± ± 29.86 9.42

0.45 ± ± 3.17*1

1 1 1 2 34

7.78 3.11 0.73 5.48 14.76

0.53 0.21 0.05 0.37

95.54 34.52 33.36 0.87 14.36

1425.8 31387.6 647.2 2280.6 1389.3 5.95 0.04 10.78 11.98 13.88

88.99 ± ± 31.51 12.69 0.232 ± ± 0.049 0.083

2.34 48.28*4 0.61 0.18

0.43 0.00 0.78 0.86

(*) 0.05 < P < 0.10, *1 P < 0.05, *2 P < 0.01, *3 P < 0.001, *4 P < 0.0001

(r = 0.252, n = 80, P < 0.05) but not for early ¯owers (r = 0.054, n = 45, P = 0.73). We did not detect any treatment e€ect on ovule production for either early or late ¯owers (Fig. 1e, Table 1e). There were no signi®cant Defoliation ´ Stigma excision interactions for any of the prezygotic variables analyzed (Table 1a±e). Postzygotic variables We detected several treatment e€ects on seed production (Table 1f±i). While shoot defoliation negatively a€ected the number of seeds produced by early ¯owers, there were no detectable defoliation e€ects on the number of seeds produced by late ¯owers (Fig. 1f, Table 1f). Defoliation weakly reduced both the number of fruits set by early ¯owers and the number of seeds set by those fruits (Fig. 1g±h, Table 1g±h). Seed output of late ¯owers in ramets in which early ¯owers were allowed to receive pollen and thus set fruit was almost nil (<1 seed/ ¯ower), independently of whether the ramet had been

defoliated or not (Fig. 1f, Table 1f). Moreover, there was no apparent e€ect of defoliation on late seed output in ramets where early reproduction was prevented (due to stigma excision) and late reproduction was successful (F1,33 = 1.02, P = 0.32). Stigma excision of early ¯owers greatly increased the number of seeds produced by late ¯owers (Fig. 1f, Table 1f), which was due to a highly signi®cant increase in fruit set (Fig. 1g, Table 1g) rather than to an increase in the number of seeds set per fruit (Fig. 1h, Table 1h). We did not detect any treatment e€ect on the size of the seeds produced for either early or late ¯owers (Fig. 1i, Table 1i). There were no signi®cant Defoliation ´ Stigma excision interactions for any of the postzygotic variables analyzed. Relationship between pollination and seed output While the mean number of seeds set per ¯ower was not related to the number of pollen grains deposited on the

239

stigmas of non-manipulated early ¯owers (r2 = 0.018, n = 49, P = 0.36; Fig. 2a), there was a highly signi®cant relationship between mean seed output and pollen receipt for late ¯owers across ramets (r2 = 0.374, n = 87, P < 0.0001; all treatment categories included; Fig. 2b). Slopes of late seed output vs pollen receipt regressions among the 2 ´ 2 categories of shoot defoliation and stigma excision were relatively homogeneous (F3,78 = 2.15, P = 0.10). However, there was a trend towards a steeper increase in late seed output with increasing pollen receipt for ramets in which early reproduction was experimentally prevented compared to ramets in which it was not (data pooled across defoliation categories; F1,82 = 5.39, P < 0.05; Fig. 2b). We found that late ¯owers of ramets in which stigmas of early ¯owers were excised still set many more seeds than those in which early ¯owers were not manipulated after

statistically accounting through ANCOVA for di€erences in pollen receipt between these two groups (leastsquares means ‹1 SE = 3.73 ‹ 0.33 vs. 0.95 ‹ 0.41 seeds/¯ower; F1,81 = 25.2, P < 0.0001). On the other hand, the regression between late seed output and pollen receipt for ramets in which stigmas of early ¯owers were excised (Fig. 2b) may be used to estimate how much an increase in pollination in late ¯owers (Fig. 1d) could have contributed to their observed increase in reproductive success (Fig. 1h). If late ¯owers of ramets in this category were receiving an average of (x ‹ 1 SE) 42.8 ‹ 5.23 pollen grains, as found in late ¯owers of ramets in which early reproduction was not prevented (Fig. 1d; data pooled across defoliation categories), instead of 81.1 ‹ 6.13, as actually observed, they would have been expected to set only 2.8 seeds. The observed average of 4.5 ‹ 0.42 seeds set per late ¯ower in ramets where early ¯owers failed to set fruit represented a highly signi®cant increase in late reproduction that might be attributed to increased pollination (t35 = 3.99, P < 0.0005).

Discussion

Fig. 2 Scatter plots of mean number of seeds per ¯ower vs. pollen receipt for a early (primary) ¯owers and b late (secondary) ¯owers. Di€erent symbols are used to identify ramets receiving the same defoliation and stigma excision treatment combination. Lines represent the least-squares regression equations: a y = 6.47 + 0.0084x (r2 = 0.018, n = 49, P = 0.36), and b, dashed, y = )0.114 + 0.0138x (r2 = 0.127, n = 53, P < 0.01) for ramets with untouched early ¯owers (i.e., Control and Defoliation categories pooled) and, continuous, y = 0.946 + 0.0430x (r2 = 0.285, n = 35, P < 0.001) for early ¯owers of which the stigmas were excised before receptivity (i.e., Excision and Both categories pooled)

The results of our experiment show that when early, primary ¯owers of compound in¯orescences of A. aurea failed to set fruit, the number of seeds produced by late, secondary ¯owers increased more than ®ve-fold. This change in late reproduction success is due to a much higher probability of late ¯owers setting fruit when early ¯owers fail, rather than as a consequence of increases in ovule production or in the number of seeds set per fruit. The reallocation of limiting resources that have not been usurped by early ¯owers into late fruit and seed set is the prevailing hypothesis that has been advanced to explain increases in late seed output when early reproduction fails, a response that has been reported now for a diversity of species (reviewed in Lee 1988; Stephenson 1992; Diggle 1995). However, our study shows not only strong increases in late seed output in ¯owering ramets of A. aurea, but also a sharp increase in the pollination of late ¯owers. This is, to our knowledge, the ®rst study to show changes in pollen receipt by late ¯owers when early reproductive losses occur. How do these shifts in pollination levels occur, and what is the signi®cance of increased pollination for late seed output? Strict dichogamy in A. aurea in both early and late ¯owers (Aizen and Basilio 1995) prevents within-¯ower self-pollen deposition. Thus, pollen transfer is completely pollinator-mediated in this plant species. We found that when early ¯owers failed to set seed, nectar production of late ¯owers exhibited a signi®cant increase. An increase in nectar production might, in turn, a€ect bumblebee visitation resulting in the observed increase in pollen deposition. A positive response of pollinator and pollination levels to nectar secretion rates seems to be a common theme in plant-pollinator interactions (reviewed in Rathcke 1992). Although we did not monitor visitation frequencies in this study, there

240

is evidence that ¯owers of A. aurea with higher nectar production receive more bumblebee visits than lownectar-secreting ¯owers (Aizen and Basilio 1998). Furthermore, bumblebees actively reject individual ¯owers of this plant species based on their nectar content, hovering over but not visiting low-rewarding ¯owers (M.A. Aizen and M. Lozada, unpublished work). Increased rates of pollination could be meaningless for maternal reproductive success if late reproduction were resource-limited, rather than pollen-limited (Willson 1979; Bierzychudek 1981; Burd 1994). In this species, pollen loads lower than 60 grains per stigma will usually lead to severe pollen limitation of fruit initiation and total seed output (Aizen and Basilio 1998). In this study, less than 10% of the stigmas of early ¯owers screened received such a few number of pollen grains, suggesting that pollen limitation of seed output in early ¯owers rarely occurs. On the other hand, more than 50% of the stigmas of late ¯owers received fewer than 60 pollen grains, suggesting that seed output in late ¯owers is commonly pollen-limited. Accordingly, a signi®cant relationship between mean seed output and pollen receipt in late ¯owers, but not in early ¯owers, might provide evidence that pollination usually limits late reproduction but not early reproduction (Fig. 2). How seed output in late ¯owers is related to increases in pollen receipt, however, seems to depend on the fate of early ¯owers. We found that when early ¯owers set fruit, an increase in pollination of late ¯owers apparently leads only to weak increases in late seed output. Conversely, when early reproduction fails, late reproduction seems to become highly pollen-limited, so that increased rates of pollination strongly a€ect seed output (Fig. 2b). Nevertheless, the observed di€erences in pollination of late ¯owers between ramets that set and did not set early fruits did not fully account for the large di€erences these two type of ramets exhibited in late seed output. Hence, late reproduction in this plant species could be actually limited by both pollen and resources (Haig and Westoby 1988; Campbell and Halama 1993; Casper and Niesembaum 1993; Corbet 1998). The in¯uence of pollen limitation would only become apparent when resources are not preempted by early fruit maturation. In contrast to the proposed strong cascading e€ects associated with early stigma excision, defoliation of ¯owering shoots only had weak-to-moderate e€ects on the pre- and postzygotic variables we studied. The small changes in late ¯ower and stigma size, however, were not associated with signi®cant changes in either late ¯ower pollen receipt or seed output (see also Aizen and Ra€aele 1996). These results also imply that decreases in size su€ered by these ¯oral parts were not large enough to a€ect to any signi®cant extent bumblebee attraction or the ability of stigmas to capture pollen from a bumblebee's body. Leaf removal moderately a€ected the number of seeds produced by early ¯owers, but again these decreases in both fruit and seed set could not be ascribed to changes in pollination levels. In general, seed output in

A. aurea seems to be well bu€ered against damage to leaves close to the in¯orescence (Aizen and Ra€aele 1996), which contrasts with ®ndings for other species in which the removal of even small portions of leaves subtending in¯orescences may lead to strong e€ects on reproduction (Marquis 1988, 1992; Horton and Lacey 1994; but see Obeso and Grubb 1993). It might well be that rhizomes, specialized starch-storing roots, or even associated vegetative shoots (Bayer 1987) are important sources of carbohydrates and nutrients for reproduction in A. aurea in addition to the resources provided locally by the ¯owering shoots themselves. Our results, however, suggest that photosynthates produced by, or nutrients stored in, ¯owering shoots subsidize early reproduction more than late reproduction (but see Emms 1996). Additionally, it is important to note that leaf removal did not interfere with the capacity of a ¯owering shoot to shift to late reproduction when early reproduction failed, as evidenced by nonsigni®cant treatment interactions for any of the variables analyzed (Table 1). The implications of these weak e€ects of defoliation are that investment in late reproduction seems not to be directly dependent on whole-plant resource availability but strictly on the performance of early ¯owers. The relationship between early and late reproduction could be regulated and, to some extent, ``constrained'' by in¯orescence architecture (i.e., a constraint ascribed exclusively to ¯ower position within in¯orescences; Diggle 1995). The secondary branching of late ¯owers from early ¯ower pedicels in A. aurea might condition them to become less powerful resource sinks than early ¯owers, perhaps because of reductions in vascular tissue. After all, late ¯owers of A. aurea do not seem to fully compensate for early reproductive losses, as only ~60% of the seeds produced naturally by early ¯owers are set by late ¯owers when early reproduction fails (Fig. 1f). Nonadaptive architectural constraints, however, might represent an unavoidable side e€ect of any adaptive regulatory mechanism of late reproduction that depends on early reproduction success (Wolfe 1992; Diggle 1995). Despite these constraints, and without discarding other non-mutually exclusive functions of late ¯owers in A. aurea (see Brunet 1996), the patterns we found are consistent with the view that late-opening ¯owers represent a reserve of ovaries to bu€er an unpredictable pollinator environment or other factors that might compromise early plant reproduction (Sutherland and Delph 1984; Sutherland 1986; EhrleÂn 1991, 1993; Guitian 1993, 1994). Particularly, our results provide evidence that plants are capable of plastic ``manipulations'' of ¯ower rewards, to increase the chance of reproduction in temporally variable environments. Acknowledgements We thank Carole Brewer, Pamela Diggle, Peter Feinsinger, Javier Puntieri and two anonymous reviewers for helpful comments and suggestions. This research was funded by the Universidad Nacional del Comahue (Grant B-036) and the Consejo Nacional de Investigaciones Cientõ ®cas y TeÂcnicas of Argentina (CONICET).

241

References Aizen M, Basilio A (1995) Within and among ¯ower sex-phase distribution in Alstroemeria aurea (Alstroemeriaceae). Can J Bot 73:1986±1994 Aizen M, Basilio A (1998) Sex di€erential nectar secretion in protandrous Alstroemeria aurea (Alstroemeriaceae): is production altered by pollen removal and receipt? Am J Bot 85:245± 252 Aizen M, Ra€aele E (1996) Nectar production and pollination in Alstroemeria aurea: responses to level and pattern of ¯owering shoot defoliation. Oikos 76:312±322 Alexander MP (1980) A versatile stain for pollen, fungi, yeast and bacteria. Stain Technol 55:13±18 Ashman TL (1992) Indirect cost of seed production within and between seasons in a gynodioecious species. Oecologia 92:266± 272 Bayer E (1987) Die Gattung Alstroemeria in Chile. Mitt Bot Staatssamml Muench 24:1±362 Bierzychudek P (1981) Pollinator limitation of plant reproductive e€ort. Am Nat 117:838±840 Brunet J (1996) Male reproductive success and variation in fruit and seed set in Aquilegia caerulea (Ranunculaceae). Ecology 77:2458±2471 Burd M (1994) Bateman's principle and plant reproduction: the role of pollen limitation in fruit and seed set. Bot Rev 60:83±139 Campbell DR, Halama KJ (1993) Resource and pollen limitations to lifetime seed production in a natural plant population. Ecology 74:1043±1051 Casper BB, Niesenbaum RA (1993) Pollen versus resource limitation of seed production: a reconsideration. Curr Sci 65:210±214 Corbet SA (1998) Fruit and seed production in relation to pollination and resources in bluebell, Hyancithoides non-scripta. Oecologia 114:349±360 Diggle PK (1995) Architectural e€ects and the interpretation of patterns of fruit and seed development. Annu Rev Ecol Syst 26:531±552 EhrleÂn J (1991) Why do plants produce surplus ¯owers? A reserveovary model. Am Nat 138:918±933 EhrleÂn J (1993) Ultimate functions of non-fruiting ¯owers in Lathyrus vernus. Oikos 68:45±52 Emms SK (1996) Temporal patterns of seed set and decelerating ®tness returns on female allocation in Zigademus paniculatus (Liliaceae), an andromonoecious lily. Am J Bot 83:304±315 Guitian J (1993) Why Prunus mahaleb (Rosaceae) produce more ¯owers than fruits. Am J Bot 80:1305±1309 Guitian J (1994) Selective fruit abortion in Prunus mahaleb (Rosaceae). Am J Bot 81:1555±1558 Haig D, Westoby M (1988) On limits to seed production. Am Nat 131:757±759 Harriss F, Whelan RJ (1993) Selective fruit abortion in Grevillea barklyana (Proteaceae). Aust J Bot 41:499±509 Herrera J (1991) Allocation of reproductive resources within and among in¯orescences of Lavandula stoechas (Lamiaceae). Am J Bot 78:789±794

Horton M, Lacey EP (1994) Carbon integration in Plantago aristata: the reproductive e€ects of defoliation. Am J Bot 81:278± 286 Lee TD (1988) Patterns of fruit and seed production. In: Lovett Doust J, Lovett Doust L (eds) Plant reproductive ecology: patterns and strategies. Oxford University Press, New York, pp 179±194 Marquis RJ (1988) Intra-crown variation in leaf herbivory and seed production in striped maple, Acer pensylvanicum L. (Aceraceae). Oecologia 77:51±55 Marquis RJ (1992) The selective impact of herbivores. In: Fritz RS, Simms EL (eds) Plant resistance to herbivores and pathogens: ecology, evolution and genetics. University of Chicago Press, Chicago, pp 301±325 Mead R (1988) The design of experiments. Cambridge University Press, Cambridge Newman JA, Bergelson J, Grafen A (1997) Blocking factors and hypothesis tests in ecology: is your statistics text wrong? Ecology 78:1312±1320 Niesembaum RA (1996) Linking herbivory and pollination: defoliation and selective fruit abortion in Lindera benzoin. Ecology 77:2324±2331 Obeso JR, Grubb PJ (1993) Fruit maturation in the shrub Ligustrum vulgare (Oleaceae): lack of defoliation e€ects. Oikos 68:309±316 Puntieri JG (1991) Vegetation response on a forest slope cleared for a ski-run with special reference to the herb Alstroemeria aurea Graham (Alstroemeriaceae), Argentina. Biol Conserv 56:207± 221 Rathcke BJ (1992) Nectar distributions, pollinator behavior, and plant reproductive success. In: Hunter MD, Ohgushi T, Price PW (eds) E€ects of resource distribution on animal-plant interactions. Academic Press, New York, pp 113±138 SAS (1988) SAS/STAT user's guide, release 6.03 edn. SAS Institute, Cary Solomon BP (1988) Patterns of pre- and postfertilization resource allocation within an in¯orescence: evidence for interovary competition. Am J Bot 75:1074±1079 Stephenson AG (1986) Lotus corniculatus regulates o€spring quality through selective fruit abortion. Evolution 40:453±458 Stephenson AG (1992) The regulation of maternal investment in plants. In: Marshall C, Grace J (eds) Fruit and seed production. Aspects of development, environmental physiology and ecology. Cambridge University Press, Cambridge, pp 151±171 StoÈcklin J (1997) Competition and the compensatory regulation of fruit and seed set in the perennial herb Epilobium dodonaei (Onagraceae). Am J Bot 84:763±768 Sutherland S (1986) Patterns of fruit-set: what controls fruit-¯ower ratios in plants? Evolution 40:117±128 Sutherland S, Delph LF (1984) On the importance of male ®tness in plants: patterns of fruit-set. Ecology 65:1093±1104 Willson MF (1979) Sexual selection in plants. Am Nat 113:777±790 Wolfe LM (1992) Why does the size of reproductive structures decline through time in Hydrophyllum appendiculatum (Hydrophyllaceae)?: developmental constraints vs. resource limitation. Am J Bot 79:1286±1290