The interplay between ovule number, pollination and resources as determinants of seed set in a modular plant

Study site

Fieldwork was conducted from December to April of the 2015–2016 austral summer, in a commercial raspberry field in Northwestern Patagonia, Argentina, about 10 km east of the city of San Carlos de Bariloche (41°06′26.3″S; 71°12′11.7″W). The 0.4-ha field was cultivated with the Autumn bliss variety and was managed organically and irrigated artificially. In the NW Patagonia region, the managed honey bees, Apis mellifera, and the invasive bumble bee, Bombus terrestris, account for >90% of visits to raspberry flowers (Sáez et al., 2014), with the native halictid bee Ruizanthedella mutabilis and other native insects accounting for the rest of the visits (Morales, 2009; M. Strelin, 2015–2016, personal observation).

In the southern hemisphere, young ramets flower from January to mid-March, and their fruits mature from February to April. Typically, these ramets are pruned after fruiting and their dormant buds start regrowing the following spring, becoming old ramets. These old ramets flower during December and mature fruits during January.

Field sampling

We haphazardly chose 25 young ramets produced during the sampling season and 25 old ramets from the previous season distributed over the entire field. Each inflorescence produced by each of these ramets was identified and marked with a paper tag. Each inflorescence was then sketched, mapping the position and branching order of each flower. Flowers were classified based on the order of their subtending floral axes as primary, secondary, or tertiary. Higher order flowers were not included in this study because of reduced sample size. To assess the effect of pollinator visits on seed set, we selected 10 young and 10 old ramets and excluded pollinators by covering them with a wire-framed net mesh fixed to the ground. Pollinator excluded and control ramets were interspersed across the entire field. The other 15 young and 15 old ramets were left open to insect pollination. Fruits were harvested when the polydrupe was bright red and detached easily from the receptacle. Collected fruits were kept under shade for ≤4 h before being transported to the laboratory in San Carlos de Bariloche, where they were frozen at −80 °C until processing. From these fruits, we estimated ovule and seed number per flower based on the number of fully developed and persistent aborted drupelets. The latter can be recognized as thiny hair-like structures among the red, fleshy drupelets. Developed and aborted drupelets of each fruit were counted under a dissecting microscope at 10×. Total ovule number per flower was estimated by adding the number of drupelets and of aborted pistils, and seed set by dividing the number of drupelets by ovule number. A haphazardly-chosen subset of a total of 25 aborted and 25 developed drupelets from fruits of five ramets was dissected to ensure that they contained ovules, embryos or seeds, which was always the case. Raw data used in this study can be consulted in Table S1.

Data analysis

We first estimated variance components for the number of ovules and the number of seeds (=number of drupelets) among flowers within inflorescences, among inflorescences within ramets, and among ramets. Hierarchical variance analysis followed Gelman & Hill (2006) as implemented in the lmer function of R version 2.4.0 (R Core Team, 2017) (library: lme4; Bates et al., 2014). In the case of ovule number, the analysis was carried out for the young- and old-ramet categories, separately. In the case of seed number, this analysis was conducted separately for each of the four combinations of ramet age and pollination treatment.

Generalized linear mixed-effects models (GLMM) were then used to assess the hierarchical level (i.e., among flowers within inflorescences, among inflorescences within ramets, or among ramets) at which ovule number determines the proportion of formed seeds (objective 1); the effects of ovule number per flower, ramet age (young vs. old ramet) and pollination treatment (pollinator exclusion vs. open pollination) on seed set (objective 2); and the direct effect of flower hierarchy within the inflorescence on seed set and its indirect effect via ovule number (objective 3). Analyses considered “ramet” and “inflorescence within ramet” as random effects (Gelman & Hill, 2006). In all cases, we selected the most appropriate random structure based on AIC criterion, following the backwards model selection protocol suggested by Zuur et al. (2009). Models were implemented with the statistical software R version 2.4.0 (R Core Team, 2017) using the glmmadmb function (library: glmmADMB; Bolker et al., 2012) (objectives 1, 2 and 3) and the glmer function (library: lme4, Bates et al., 2014) (objective 3).

For objective 1, we assessed the effect of each of the three measures of ovule number on seed set per fruit. In addition to each flower’s ovule number, this analysis included the mean number of ovules per flower at the inflorescence and ramet levels. Since the number of ovules of an individual flower includes a component associated with its inflorescence mean, and the inflorescence mean includes a component associated with its ramet mean, ovule number at each level had to be corrected for variation in ovule number included in the level immediately above. Therefore, we first estimated mean values of ovule number per flower at the inflorescence and ramet levels, corrected by the number of observations for each inflorescence and ramet, by means of an intercept-only model, with a random intercept and two nested levels (flowers within inflorescences and inflorescences within ramets). This analysis was performed with the lmer function (library: lme4, Bates et al., 2014). The coefficients retrieved for each level are the sample-size corrected means of each inflorescence and ramet. Second, using these values we calculated the difference between the number of ovules of each individual flower and the corrected mean value of ovules per flower of its corresponding inflorescence, and the difference between the corrected mean value of the number of ovules per flower between each inflorescence and its corresponding ramet. In this way, we estimated independent components of ovule number that could be attributed to the individual flower (i.e., the difference in the number of ovules between the flower and inflorescence levels), the inflorescence to which this flower belonged to (i.e., the difference in the number of ovules between the inflorescence and ramet levels), and the ramet to which this inflorescence belonged to (i.e., the mean number of ovules per flower at the ramet level). These three hierarchical, independent variables were included as predictors of seed set using a generalized linear mixed model with a beta-binomial distribution of the dependent variable and a logit-link function. The chosen distribution accounted for both the nature and the presence of over-dispersion in the residuals of the response variable. Our model also considered interactions between plant age and ovule number at each hierarchical level (i.e., flower, inflorescence, and ramet), since plant resource status could presumably affect the relation between ovule number and seed set at any of these levels.

For objective 2, we used backward elimination, as described in Zuur et al. (2009), to assess the relative importance of the studied factors and their mutual dependence on seed set. The initial model included the three studied fixed factors (i.e., ovule number of individual flowers, without accounting for the mean number of ovules at the inflorescence level; ramet age; and pollination treatment) as main predictors and all their interactions. Because of over-dispersion of model residuals and the nature of the response variable (i.e., seed set), the analysis considered a beta-binomial distribution of seed set and a logit-link function. Likelihood-ratio tests of partial effects were conducted with the Anova function of the car R library (Fox & Weisberg, 2011).

To address objective 3, we carried out two analyses. The first determined whether ovule number varied with branching order within inflorescences (i.e., primary, secondary, and tertiary). This analysis, which considered a Poisson distribution of the response variable (i.e., ovule number) and a log-link function, included branching order, ramet age, and the branching order × ramet age interaction as fixed effects, and ramet and inflorescence within ramet as random factors. The second analysis, which considered seed set as the response variable, added flower order and its interaction with ramet age to the reduced model from the preceding analysis (i.e., objective 2). Likelihood-ratio tests of partial effects were conducted with the Anova function of the car R library (Fox & Weisberg, 2011).

Variance partitioning of ovule and seed number

The partitioning of variance in ovule number among flowers within inflorescences, among inflorescences within ramets, and among ramets depended on ramet age. For young ramets, the flower and ramet levels each accounted for ∼45% of the total variance in ovule number, with the remainder associated with inflorescences within ramets (Fig. 2). For old ramets, the flower level accounted for the largest portion of variation (∼50%) in ovule number, whereas the ramet level accounted for the least (∼20%).

Variance partitioning in seed number per fruit differed depending on ramet age and pollination treatment. For young ramets, variance in seed number paralleled that of ovule number, except that pollinator exclusion greatly reduced the among-ramet component, whereas open pollination increased it. For old ramets, the pattern of variation in seed number reflected that of ovule number only for ramets subjected to pollinator exclusion. For those exposed to pollination, seed number varied the most among ramets (Fig. 3).

Effect of modularity on the relation between ovule number and seed set (objective 1)

Even though we found an overall positive effect of ovule number on seed set (Fig. 4), the sign and significance of the relation between ovule number and seed set depended on the level of modularity. Variation in ovule number among flowers within inflorescences had a significant and positive effect on flower seed set (β = 0.0201; SE = 0.0054) (z = 3.76; P < 0.0005). On the contrary, variation in ovule number at the inflorescence level had a negative effect on seed set (β = −0.0457; SE = 0.0173) (z = −2.63; P < 0.01). Seed set was marginally but positively affected by ovule number at the ramet level (β = 0.0553; SE = 0.0333) (z = 1.66; P = 0.0972). There was neither a significant effect of plant age on seed set nor any significant interaction between plant age and the components of ovule number at the different levels of modularity in this model (Table 1).

Effects of ovule number, plant age and pollination treatment on seed set (objective 2)

For the 1,282 flowers we examined, an average of 73.3% of a flower’s ovules became seeds, with minimum and maximum values that ranged between 4.7% and 100%, respectively. The best-fitting model explaining variation in seed set included ovule number, pollination treatment, plant age and the ovule number × plant age interaction (Table 1). Of all the deviance explained by the fixed factors, 65.4% was accounted by ovule number, 18.4% by ramet age, 8.0% by pollination treatment, and 8.3% by the ovule number × ramet age interaction. Because of the strong positive effect of ovule number on seed set, the number of seeds, and thus of drupelets per fruit, increased disproportionally with the number of ovules (β = 0.02804; SE = 0.00508) (z = 5.62; P < 0.001). However, in a large number of fruits the total number of ovules per flower imposed a strict upper limit on the number of seeds produced (Fig. 4). Fruits produced by young and old ramets had an average of 62.4 and 44.1 seeds, respectively; with flowers from young ramets showing a seed set, on average, 31.5% larger than flowers from old ramets (z = 2.13; P < 0.05). Ramet age affected the positive density-dependence of ovule number on seed set, being stronger in old ramets (z = 2.17; P < 0.05) (Fig. 4). Flowers from plants excluded from and exposed to pollinators produced an average of 52.0 and 56.9 seeds, respectively; with flowers from plants exposed to pollinators showing a seed set, on average, 8.52% larger than that from plants that were excluded (z = 2.60; P < 0.01). Unlike ramet age, pollination treatment did not significantly affect the shape of the relation between seed set and ovule number (z = −1.28; P = 0.20) (Fig. 4).

Effect of branching order and plant age on ovule number and seed set (objective 3)

All fixed effects in the first model addressing objective 3, i.e., branching order, plant age and the branching order × plant age interaction, significantly influenced ovule number (Table 1). The number of ovules per flower greatly decreased from first to tertiary-order flowers (Fig. 5). Also, old ramets produced flowers with fewer ovules than young ramets (Fig. 5). The decrease in ovule number with increasing branching order was somewhat weaker in young than in old ramets (Fig. 5). First, second, and third order flowers in young ramets had, on average, 84.4, 80.0, and 78.3 ovules, respectively, whereas in old ramets had 66.5, 60.0, and 56.6 ovules, respectively. In contrast to the effect of flower branching order on ovule number, branching order did not significantly affect seed set after accounting for variation in ovule number (Table 1). Therefore any effect of plant architecture on seed set has to be interpreted as indirect, i.e., via ovule number.

Variation in seed set as a direct consequence of variation in ovule number

Our results indicate a clear distinction in the way resources are allocated to ovules and seeds. Whereas ovule number varied with branching order within the inflorescence (Fig. 5), the proportion of those ovules finally becoming seeds was determined more directly by ovule number rather than branching order (Table 1). Thus, ovule number might reflect the supply rate and total amount of resources provided during flower development; resource supply rate could, in turn, relate to the thickness of the vascular tissue that is expected to vary inversely with branching order (Wetzstein et al., 2013). Instead, seed set seems to be the result of the sink strength of an increasing number of developing embryos associated with a higher number of ovules. Embryos can have high endogenous levels of cytokinin and gibberellin that are involved in their nutrition (Finkelstein, 2010). As a consequence, a larger number of embryos could provide a stronger hormonal signal that would result in a developing fruit obtaining a disproportionate share of available resources (Stephenson, 1981). Therefore, whereas allocation to ovule number seems to be determined by a pure maternal feature like inflorescence architecture, allocation to seeds seems to be under an indirect maternal control through variation in ovule number.

A modular perspective of the relation between ovule number and seed set

The positive density-dependent effect of ovule number on seed set occurs basically among flowers within inflorescences and suggests that, within well-pollinated inflorescences, flowers with more ovules and embryos become stronger sinks for locally limiting resources (Fig. 1B). Extensive variation in ovule and seed number within inflorescences (Figs. 2 and 3) agrees with the view that differential resource allocation is strongest at this level of the modular organization of a raspberry plant (Guitian, 1994; Wesselingh, 2007; Herrera, 2009). At the inflorescence level, however, there was a negative density-dependent effect of ovule number on seed set, suggesting that inflorescences rely on a relatively constant pool of resources that must be allocated to developing embryos and their ancillary fruit structures. Hence, resources for seed development would become more limiting as the average number of ovules (and ensuing embryos) per flower at the inflorescence level increases. This latter result is consistent with the fact most carbon used for fruit development seems to be fixed by the inflorescence subtending leaf, with little subsidy from photosynthetic structures outside the inflorescence (Waister & Wright, 1989). Therefore, the findings we report here clearly support a modular perspective on how resources are compartmentalized and allocated to seed production based on the strength of source-sink relations associated with variation in ovule number.

The interplay between ovule number, resources and pollination

Our findings support the hypothesis that plant resource status is an important factor mediating the relation between ovule number and seed set. A reduction in ovule number and seed set observed in old ramets supports our assumption that resources allocated to reproduction are particularly limiting in ramets that have already produced fruit. Furthermore, old ramets exhibited a more marked positive density-dependent response of seed set to ovule number than young ramets (Fig. 4). This finding suggests that the sink effect of developing fruits with many embryos, produced by flowers with many ovules, becomes exacerbated under increasing resource limitation, which results in some fruits with many seeds and others with too few. Even though ovule number here affected seed set independently of branching order (Table 1), the reduced vascularity of higher order flowers (Wetzstein et al., 2013) might also constrain the supply rate of seed resources, thus exacerbating any effect related to ovule number per se. In the raspberry, resource-limited flowers on higher-order branches (which also have reduced ovule number) usually develop into fruit even if a reduced fraction of pistils develops into drupelets. Regulation of seed number in response to changes in the amount of resources through changes in seed set rather than fruit set (Ladio & Aizen, 1999; Ashman & Hitchens, 2000; Medrano, Guitian & Guitián, 2000; Kliber & Eckert, 2004) can be common among species producing apocarpic fruits, like raspberries, because seed resources can be allocated separately to each carpel (Wetzstein et al., 2013). Therefore, increasing resource limitation could increase variability in seed set among fruits (Fig. 4), and exacerbate the positive density-dependent effect of ovule number on seed set.

Resource-driven, density-dependent effects of ovule number on seed set did not diminish with increasing levels of pollen limitation, as expected if whole plants were physiologically integrated. This result challenges the classical and dichotomous view that seed set is either limited by the receipt of pollen or by available plant resources (works cited in Wesselingh, 2007; Knight, Steets & Ashman, 2006). Even under scenarios of adequate natural pollination, variability in pollinator availability in space and time and in flower placement and phenology should result in the inadequate pollination of a fraction, small or large, of all the flowers produced by a plant (Herrera, 2004; Richards, Williams & Harder, 2009). Because of plant structural limitations associated with anatomical and physiological modularity in raspberry (Waister & Wright, 1989), unused resources from poorly pollinated flowers cannot be reallocated to flowers receiving surplus pollen in different branches (Wesselingh, 2007). Therefore, persistence of the resource-driven relation between ovule number and seed set under increasing pollen limitation can then be understood in the context of resource compartmentalization.

A mosaic of reproductive limits

Considering that all the study factors play a role in limiting seed set in raspberry, we propose that seed production in modular plants can be frequently subject to a mosaic of reproductive limits, with flowers in some modules limited by pollen, others by resources, and others by ovule number. This latter overlooked factor clearly sets a sharp upper limit to flower seed production (Figs. 1 and 4). Even interactions can exist between these factors, as shown here that the magnitude of the density-dependent effect of the number of ovules could depend on the plant resource status. This mosaic of reproductive limits might also replicate at the within-flower level in plants with apocarpic ovaries, a flower sexual morphology that allows differential resource allocation and differential fertilization of ovules in different carpels. According to this integrative perspective, plants should be viewed as mosaics in which different factors, and interactions between them occurring within integrated physiological units, can limit the seed production of different plant modules. This perspective clearly defies the classical dichotomous view of “pollination or resources” as alternative limits of plant seed production (Wesselingh, 2007), and even the one provided by optimality models that predicts simultaneous limitation by both factors (Haig & Westoby, 1988). Instead, our perspective agrees with the view that individual plants can be seen as mosaics exhibiting high intra-individual phenotypic variation because a multiplicity of factors affect independently different modules (Herrera, 2009).