Premise: Seed size is a functional trait that influences individual fitness but the genetic basis of

variation in individual seed size in wild species remains still largely unknown. The evolutionary

trajectory of seed size depends, in part, on whether seed size variation is determined by additive,

maternal, or non-additive genetic variance. The expression of these sources of variance in seed

size may be environment-dependent, reflecting genetic variation in phenotypic plasticity. In this

study, we used a quantitative genetic approach to reveal the genetic basis of seed size and its

plastic response to drought stress in Dithyrea californica.

Methods: We used a diallel mating design to estimate genetic and environmental variance

components for seed size in each of three greenhouse-raised populations sampled from

California and northern Mexico. We replicated diallels in two watering treatments to examine

genetic parameters and genotype x environment interactions affecting seed size. We estimated

general combining ability (GCA), specific combining ability (SCA), reciprocal effects (RGCA

and RSCA), and their interactions with water availability. We examined norms of reaction for

maternal and paternal families and for each maternal/paternal pair to reveal the magnitude of

genetic variation in phenotypic plasticity for seed size in each population.

Results: Seed size in the sampled populations of D. californica is determined primarily by the

combination of watering treatment, GCA and RGCA; parental identity alone does not

consistently affect seed size. Across genotypes, water availability did not have a significant

effect on seed size.

Conclusion: Genetic variation in phenotypic plasticity of seed size is greater than variation

among genotypes.

Study species—Dithyrea californica (Harvey) is a self-incompatible winter annual native to sand dune habitats in the Mohave and Sonoran Deserts. D. californica germinates in response to late-fall or early-winter rains. Once germinated, it grows as a basal rosette for approximately two months, flowering and producing seeds in late February to early April, depending on the timing of precipitation. Flowers are hermaphroditic and protogynous, and if unpollinated, they typically remain open for 3-4 days. The pistil becomes receptive before the petals open, after which the anthers dehisce simultaneously with petal opening. Anthers protrude slightly from the distal end of the corolla while pistils extend to a height that is no more than one quarter of the length of the petals, making the flowers easy to emasculate. Fruits consist of two subunits called “mericarps”, each of which contains one seed.

The geographic distribution of D. californica is broad, occurring in most arid sand dune systems in western North America (Northern Mexico and southwest USA). Dithyrea californica is an ideal species with which to investigate the selective dynamics of seed size in the field due to a special feature of the germinated seed. When a seed inside a mericarp germinates, the rim of the mericarp (the diameter of which provides a measure of seed size) remains attached to the root until plants become adults. This special feature has allowed us to track individual plants in the field and to relate seed size to both survival and fecundity, the main components of individual fitness (Larios et al., 2014; Larios and Venable, 2018). D. californica is also a suitable species in which to investigate quantitative genetic variation because it is easy to emasculate and is self-incompatible, preventing accidental pollinations when conducting controlled crosses to create pedigreed seeds.

Source populations—During the spring of 2017, seeds of D. californica were collected in three locations. The northernmost population is 800 km away from the southernmost location. A northern population located in the Mohave desert was collected from the Kelso dunes at the Mohave National Preserve in California, USA (34.88° N, -115.77° W, altitude: 676 m). A central population was located in the Sonoran Desert on a coastal dune near the town of Puerto Lobos, Sonora, Mexico (30.35° N, -112.81° W, altitude: 70 m). The most southern population is also located in the Sonoran Desert and was collected on a coastal dune in the town of Desemboque de los Seris in Hermosillo, Sonora, Mexico (29.51° N, -112.41° W, altitude: 10 m). Seeds were stored in an outdoor shed (protected from rainfall) but exposed to ambient temperature for two years in paper bags until the spring of 2019; this exposure to natural year-round conditions facilitated the loss of dormancy.

Breeding design—The experiment was performed in a greenhouse located at the University of California, Santa Barbara campus in Goleta, California, USA. In the greenhouse, we planted eight seeds representing each maternal line, one seed per pot, and selected two plants of each maternal line to be grown in two alternative watering treatments (one individual/treatment). Once plants became reproductive, we conducted controlled pollinations using 5 x 5 diallel breeding designs, performing reciprocal crosses and excluding self-pollinations. Because individuals of D. californica are cosexual, this breeding design allows us to estimate each individual’s effect on seed size as both a maternal (seed-bearing) and paternal (pollen-donating) plant.  We created three diallels from the “Kelso” population, 3 diallels from the “Seri” population and only one diallel from the “Lobos” population. The number of diallels per population was determined by the number of field-collected maternal families from which seeds successfully germinated in the greenhouse. Pollinations from a given pollen donor were performed on a pollen recipient’s single reproductive stalk so we could have  better control of the pedigree of resulting seeds. Pollinations were conducted by collecting entire dehiscent stamens and rubbing them against a receptive stigma using alcohol-sterilized forceps. We sterilized the forceps with rubbing alcohol after each flower was pollinated. To avoid self-pollen contamination, we emasculated all flowers prior to anther dehiscence by slicing off their tips prior to the flower bud opening (the short style prevented damage to the stigma).

Each diallel was replicated in two watering treatments (Fig. 1); here, replication meant that each of the five individuals in a given diallel had a maternal sibling in a paired diallel exposed to the other treatment. All pots were watered using a nutrient solution comprised of MiracleGro (Marysville, Ohio, USA); we diluted one tablespoon of fertilizer in 3.78 liters of deionized water, as suggested by the manufacturer.  Wet conditions were achieved by watering the pots every three days, which kept the soil saturated for the duration of the reproductive stage. In the wet treatment, plants were maintained at full turgidity throughout the experiment. Dry conditions were achieved by watering the pots each week with the nutrient solution and allowing the soil to dry until we observed the onset of wilting, but plants retained the capacity to return to turgidity after watering. Pedigreed seeds resulting from these controlled pollinations were identified by color coding each seed with small dots of acrylic paint to identify their maternal and paternal origin. Seeds were allowed to mature on the plant and were harvested when they became easy to detach.

Data analysis—The linear dimensions of each harvested seed were measured using the imaging software ImageJ (Schneider et al., 2012). D. californica seeds are shaped as flat ellipses; we estimated seed size as the area of one face of the ellipse, using the two major axes of each seed and the following equation:

where H is the length of the longest axis and W the length of the longest axis perpendicular to H.

To estimate the variance components of seed size we analyzed variation in ellipse area produced by crosses using the random diallel analysis that includes reciprocal crosses and no self-pollinations (diallel type III sensu Hayman, 1954; Griffing, 1956). From this analysis we estimated the general combining ability (GCA), the specific combining ability (SCA), the reciprocal general combining ability (RGCA), the reciprocal specific combining ability (RSCA), and the interactions of each of these variance components with the watering treatment, for each population separately. GCA refers to additive effects of genes (nuclear effects) and it is represented as the mean deviation between the mean seed size produced by both parents and the diallel’s grand mean (Hayman, 1954; Griffing, 1956). SCA refers to non-additive effects of nuclear genes (dominance and epistatic effects) and it is represented as the deviation of mean seed size of specific parental combinations from the parental seed size means. Reciprocal effects are deviations in mean seed size due to the direction of a cross between two parents and due to the effects of the parents’ functional gender on offspring genotype. In the Hayman (1954) model, reciprocal effects are partitioned into reciprocal general combining ability (RGCA), which is represented by the deviation in mean seed size produced by a parental genotype functioning as a male vs. as a female. The reciprocal specific combining ability (RSCA) is represented by the deviation in mean seed size between a given cross (between two parents) and its reciprocal. This model is defined by the following equation:

where Y_ijk are the individual seed sizes produced by the combination of dam j and sire k, m is the grand mean of seed size within a diallel, G_j and G_k are the GCAs of the dam j and sire k, S_jk is the SCA of the jth and kth parent combination, RG_j and RG_k are the RGCAs of the jth and the kth parents, RS_jk is the RSCA of the jth and kth parent combination, and E_ijk is the residual variance.

We analyzed the data using the R package ‘lmDiallel’ (Onofri et al., 2021) along with the ‘sommer’ package to estimate random effects of variance components (Covarrubias-Pazaran, 2016). We first constructed design matrices that indicated the identity of dams and sires for each pedigreed seed with the lmDiallel package functions GCA, SCA, RGCA and RSCA. We fitted generalized linear mixed-effects models where we used ellipse area (as a measure of seed size) as the response variable, watering treatment as a fixed explanatory variable, and GCA, SCA, RGCA, RSCA, GCA x water, SCA x water, RGCA x water, and RSCA x water as random effects. We tested the statistical significance of each random effect with likelihood-ratio tests comparing a model that contained all random predictor variables (the full model) with models from which we excluded the random variable to be tested. We tested for a change in mean seed size between watering treatments per population with likelihood-ratio tests between the full models that contained the watering treatment as a fixed factor and sub-models that did not contain the watering treatment as the fixed factor.

Additionally, we performed the diallel analysis on each population and watering treatment separately in order to estimate environment-specific variance components of seed size and the significance of maternal and paternal genetic contributions to seed size in each treatment.

We estimated narrow-sense heritability per population and per watering treatment as the additive genetic variance divided by total phenotypic variance (Falconer and Mackay, 2003).

            We also estimated broad-sense heritability as total genetic variance  divided by total phenotypic variance.

We estimated norms of reaction of seed size using the residual variation of ellipse area within each diallel cross. Residual variation was estimated for the dams and sires separately by subtracting the diallel mean seed size of each dam or sire from the mean seed size produced by each of the dams or sires within a diallel. We pooled all dam and sire residuals by population and represented the norms of reaction by population rather than by individual diallel. By examining the norms of reaction for each population, we can observe the environmentally induced effect of watering treatment on the mean seed size and phenotypic rank of each maternal and paternal genotype. We tested for a change in variance in seed size between watering treatments with Levene’s tests of homogeneity of variance per population where we compared the within-population variance among dams or sires in ellipse area of dry and wet treatments.

We examined the sex-specific performance of, and correlations between, cosexual individuals with respect to the size of seeds produced in each watering treatment. To analyze the correlation between sexes, we conducted general linear models to examine the bivariate relationship between the residual mean ellipse area produced by individuals as pollen donors (sires) and as pollen recipients (dams). Finally, to analyze the cross-environment correlation of sex-specific residual seed size, we conducted general linear models to examine the bivariate relationship between the residual mean ellipse area produced by individuals as pollen donors (sires) in the dry and wet treatments, and as pollen recipients (dams) in the dry and wet treatments. For each diallel matrix, we used the diallel mean ellipse area of all sires and dams to calculate sire- and dam-specific residuals. Then, all residuals for a given population were pooled prior to plotting the bivariate relationships presented here.