Keywords: hybridization, constraints, niche differentiation, adaptive evolution, speciation, Fleury hierarchy

Study Species

The fiveHelianthus species analyzed are diploid (n=17), self-incompatible annuals, all native to North America.Helianthus annuus andHelianthus petiolaris are widespread in the North American plains and western states and occupy mesic clay soils and drier sandy habitats, respectively (Heiser et al. 1969). Studies in present-day hybrid zones suggest that genotypes tend to be distributed according to habitats (Stebbins and Daly 1961;Rieseberg et al. 1999a;Carney et al. 2000).Helianthus petiolaris-like individuals tend to be found in higher, drier habitats with sandier soil, andH. annuus-like individuals are found in wetter areas. These species are the ancestral parents of three hybrid species:Helianthus anomalus,Helianthus deserticola, andHelianthus paradoxus (Rieseberg 1991;Rieseberg et al. 1991), all of recent homoploid hybrid origin. Current estimates for the time of origin are between 144,000 and 116,000 ybp forH. anomalus (Schwarzbach and Rieseberg 2002), 170,000 and 63,000 ybp forH. deserticola (Gross et al. 2003), and 208,000 and 78,000 ybp forH. paradoxus (Welch and Rieseberg 2002b). They are also ecologically differentiated, being restricted to desert sand dune, desert floor, and brackish marshes, respectively.

Plant Material

We collected achenes (one-seeded fruits) from one population of each species:H. annuus (Hwy. 24, 1/4 mile N of Hanksville, Wayne Co., UT; Rieseberg 1295);H. anomalus (Hwy. 160, 6 miles E of Mexican Water, Apache Co., AZ; Rieseberg 1282);H. deserticola (I-15, Frontage Rd., 6 miles E of Toquerville, Washington Co., UT; Rieseberg 1270),H. paradoxus (S side of I-40, 1/4 mile E of exit 85, Grants, Cibola Co., NM; Rieseberg 1370), andH. petiolaris (Hwy. 89, boundary of Glenn Canyon Recreation Area, Kane Co., UT; Rieseberg 1277). For each population, we derived achenes from 30–100 maternal plants collected at irregular intervals along a transect. We pooled the achenes from all heads from a given population and chose a random subset for propagation at the University of Georgia greenhouses. The analysis of a single population per species is justified because most genetic variation (i.e., G_st) is found within rather than among populations of annual sunflower species (Heiser 1954,1961;Schwarzbach and Rieseberg 2002;Welch and Rieseberg 2002b;Gross et al. 2003). Phenotypic variation (i.e., Q_st), however, often has a larger among-population component (Spitze 1993;Lynch et al. 1999;Schluter 2000). Therefore, our data are likely to underestimate the total species phenotypic variation of both the parental and hybrid species.

We derived synthetic hybrid populations from progeny from oneH. annuus (ANN1295: northern city limit of Hanksville, UT) and oneH. petiolaris (PET1277: Hwy. 89, 10 miles S of Page, AZ). For the BC₂PET, we backcrossed one F₁ hybrid from this cross with another individual of the PET1277. Due to high F₁ sterility, only 38 BC₁ progeny were produced. We then backcrossed these individuals with a third individual from PET1277 to yield the BC₂PET (hereafter BC₂PET). The BC₂ANN were produced in a similar manner except that the F₁ individual was backcrossed twice with ANN1295.

We germinated achenes following the protocol ofSchwarzbach et al. (2001). We placed 20 seedlings from each population in 25-cm-diameter plastic pots containing a 3: 1 mixture of sand/fritted clay (Turface, Profile Products, Buffalo Grove, IL). This mixture does not contain any nutrients or NaCl, which allowed us to add them in controlled quantities. Pots were arrayed in a randomized block design in the greenhouses, with 10 blocks and two individuals of each species per block. Each block also included 40 individuals from the BC₂ANN population (see above for crossing design), which is the focus of an ongoing quantitative trait locus (QTL) study (L. H. Rieseberg, D. M. Rosenthal, J. L. Durphy, and L. A. Donovan, unpublished study). We watered plants twice each day and fertilized with a time-release microencapsulated complete fertilizer (Osmocote Plus, Scotts-Sierra Horticultural Products, Marysville, OH). In addition, 8 days after planting, we treated the plants with 5.0 mM NaCl twice weekly for 3 weeks because previous studies indicated thatH. paradoxus responds to salt stress by accumulating sodium in its leaves (Welch and Rieseberg 2002a). The very dilute NaCl treatment was high enough to ensure detection of this response inH. paradoxus but low enough so that the remaining species were unaffected. Some plants died early in the experiment due to transplant shock and were excluded from analyses. In all, we analyzed data from 20 individuals ofH. annuus, 18 ofH. anomalus, 18 ofH. deserticola, 19 ofH. paradoxus, 19 ofH. petiolaris, and 383 BC₂ANN. Note that multivariate analyses necessarily excluded a few individuals because data was missing for some traits.

Traits Measurements

Twenty-three morphological and six life-history traits were measured (table 1). Morphological traits included seedling, vegetative, floral, and plant architectural traits. Floral traits comprised phyllaries, which are leaf-like bracts that underlie flowering heads, and ligules, which are the single florets found on ray flowers. We measured flower characters on fully expanded mature flowers in which 75% or more of the florets were mature. We measured leaf characters on a fully expanded nonsenescent leaf from each individual. We measured leaf area using a LI-COR 3100 area meter (LI-COR, Lincoln, NE) and estimated average leaf toughness with a push dynomometer (McCormick Fruit Tech, Yakima, WA). The six life-history traits included timing of developmental stages (BUDDAY and FLODAY), allocation to floral and vegetative traits (FLRNUM, FLOMASS, SHBIO), and height growth rate (HGR).

We also measured 17 physiological traits such as photosynthetic rate, water and nitrogen use efficiency, and nutrient status. Photosynthetic rate, stomatal conductance, and internal CO₂ concentration were measured on a single mature leaf from each plant using a LI-COR 6400 portable photosynthesis system concentration 360 ppm, air (CO₂ temperature 26°–28°C, photosynthetically active radiation 1,500 mmol m⁻² s⁻¹). Previous measurements indicated thatHelianthus did not exhibit midday stomatal closure (Schwarzbach et al. 2001); therefore, we measured plants between 10:30 A.M. and 4:00 P.M. Due to the large number of plants, measurements were made on several days, so all gas exchange data were covariate corrected for time and day (Dudley 1996). As inSchwarzbach et al. (2001), we used internal CO₂ concentration as an estimate of instantaneous water use efficiency (CI;Farquhar et al. 1989;Donovan and Ehleringer 1994). We determined seasonally integrated water use efficiency from stable carbon isotope ratios (δ¹³C) of dried leaf material (continuous flow mass spectrometry; University of Georgia Stable Isotope Soil Biology Laboratory), abbreviated DEL13C in this study. More negative δ¹³C values indicate reduced integrated water use efficiency for plants grown in similar environments.

To detect differences in nutrient status and use, we oven dried a mature leaf and then analyzed a portion of the leaf for percentage of N and C by Dumas combustion with a Carlo Erba NA 1500 elemental analyzer (Milan, Italy). We calculated photosynthetic nitrogen-use efficiency (PNUE) as the ratio of photosynthesis and N content per unit area of leaf (Field and Mooney 1986). We analyzed acid extracts from leaf tissue collected the day after the final 5 mM NaCl application for B, Ca, K, Mg, Mn, Na, and P on an inductively coupled plasma-atomic emission spectrophotometer (ICP-AES, University of Georgia, Chemical Analysis Facility).

Statistical Analyses

We visualized differences and similarities in trait variation in the synthetic hybrids as well as the parental and hybrid species by principal component analysis on character-by-character correlations (Sokal and Rohlf 1995). We used discriminant analyses (PROC DISCRIM;SAS Institute 2001) to develop a discriminant criterion (rule) that would differentiate among five classes: the two parental species,H. annuus (ANN) andH. petiolaris (PET), and the three hybrid species,H. anomalus (ANO),H. deserticola (DES), andH. paradoxus (PAR). Only those traits measured on both synthetic hybrid populations (n = 43) were used when building decision rules (table 1). We used nonparametric methods (k = 3, nearest neighbor option) to avoid problems associated with deviation from multivariate normality. We calculated Mahalanobis distances (a multidimensional equivalent of Euclidian distance) from the pooled covariance matrix and used these distances to assign individual group membership. We did not have sufficient individuals to have an independent data set to test the decision rule, so we tested the decision rules by using the same calibration data set, which necessarily gives a biased estimate of the error rate. Therefore, to evaluate the robustness of the discriminant function, we implemented a jackknife resampling method (PROC DISCRIM, CROSSVALIDATE option). This process removed the observation being classified from the density estimation before determining the classification error rate (SAS 2001).

To determine if we could re-create a parental-like or hybrid species-like multivariate phenotype, we classified the synthetic hybrid populations (BC₂PET or BC₂ANN) according to the decision rule (discriminating the five species) generated in the above analysis. In this way we could identify synthetic hybrid individuals as belonging to one of six possible categories: ANN-like, ANO-like, DES-like, PAR-like, PET-like, or OTHER. Although the synthetic hybrid lineages were grown in separate experiments, individuals from all five species were grown in both experiments. Moreover, we replicated the experimental conditions as closely as possible so that any variation between experiments would affect taxa similarly.

Since we could not evaluate all possible discrimination rules, we performed a subset of analyses that were chosen by considering both biological and statistical issues. The first discrimination rule based on all traits is the most conservative decision rule but is highly over parameterized. The second rule removes variables that are highly correlated such that when variables hadr² values >0.7 or <–0.7, one of the two variables was removed. In the third rule, only transgressive traits are included, as reported here (see table A1 in the online edition of theAmerican Naturalist) or in other studies (Rosenthal et al. 2002;Rieseberg et al. 2003). Finally, the fourth rule is based on a subset of traits selected in a stepwise discriminant analysis (PROC STEPDISC,SAS 2001). At each step, starting with no variables, the model is examined. Variables are added or removed from the model based on the significance level of anF-test from an ANCOVA (0 <P < .25 to add, 0 <P < .1 to remove). The variables already chosen act as covariates, and the variable under consideration is the dependent variable. When a variable in the model that contributes least to the discriminatory power of the model, as measured by Wilks’s λ, fails to meet the criterion to remain, it is removed. Otherwise, the variable not in the model that contributes most to the discriminatory power of the model is added. When all variables in the model meet the criterion to remain and none of the other variables meet the criterion to be added, the stepwise selection process stops (SAS 2001). The results of all four models are presented.

We compared the two synthetic hybrids’ phenotypic correlation structures by comparing their phenotypic correlation matrices (P matrix), using a hierarchical principal component analysis (Flury 1988). This method is useful because matrices can share more detailed relationships than simply being equal or unequal (Flury 1988). We determined whether matrices share similarities at different hierarchical levels by testing a series of hypotheses about matrix similarity against the null hypothesis that they are unequal or “jump-up” approach (Phillips and Arnold 1999). At the lowest level of similarity,p ×p matrices share one principal component (eigenvector) in common (PCPC 1). Hypotheses of increasing similarity are tested against the null hypothesis for up top − 2 partial principal components. Matrices can sharep − 2 principal components but have differing eigenvalues (CPC model). At the highest level of similarity (not equality), matrices differ only by a constant (proportional model). In this case, matrices share all their principal components (or eigenvectors) but the eigen values differ by only a single constant. We also compare matrices using a model-building approach advocated byFlury (1988). The criterion for best model is not statistical significance, but the overall best fitting model is chosen based on Akaike’s Information Criterion (AIC;Flury 1988). Akaike’s Information Criterion balances the goodness of fit (log likelihood ratio) against the number of parameters used to fit the model. We used the Common Principal Component Analysis Program, developed by Patrick Phillips, University of Oregon, Eugene (program available athttp://darkwing.uoregon.edu/~pphil/programs/cpc/cpc.htm).

We calculated post hoc pairwise means comparisons between the hybrid species and both parents using PROC MULTEST (SAS Institute 2001) and adjusted significance values for multiple comparisons using a sequential Bonferroni correction (Sokal and Rohlf 1995). We classified hybrid species traits as extreme (transgressive) when means significantly exceeded both parental values. We detected transgressive traits in the BC₂ANN population by comparing the number of BC₂ progeny that exceeded the high or low means of the parental populations by 2 or 3 SDs, with the numbers expected by chance.

Phenotypic Variation and Transgressive Trait Expression

The BC₂ANN trait values and ranges were first analyzed individually and compared with the parental and hybrid species. The BC₂ANN traits encompassed the most of the range of both parents (table A1). The BC₂ANN was positively and/or negatively transgressive for 29 (63%) or 32 (70%) of the 46 traits measured. The low and high percentages depended on whether the observed number of BC₂ANN with extreme traits exceeded the low or high parental mean by 2 versus 3 SDs.

The BC₂ANN trait values also encompassed most of the range of hybrid species traits values. However, some hybrid species exceed even the BC₂ANN mean for several traits (table A1). For example,Helianthus anomalus had three of 16 (18.75%) extreme traits beyond the range of BC₂ANN: PHYSHAP, PHYWID, and STEMDIA.Helianthus deserticola had two of nine traits (22.2%) beyond the range of BC₂ANN: PHYSHAP and PHYWID. Finally, four out of 12 (33.3%) ofHelianthus paradoxus traits exceeded the BC₂ANN range: LEAFTOUG, BUDDAY, FLODAY, and NA (seetable 1 for definitions of the abbreviations used in this article).

The hybrid species were also extreme relative to the parents for numerous traits (table A1).Helianthus anomalus differed significantly from one or both parents for 31 traits, 16 of which were extreme. ForH. deserticola, 30 traits were different from the parentals, 9 traits were extreme.Helianthus paradoxus was the most divergent with 33 significantly different traits, 13 of which were extreme (table A1).

Identifying and Re-creating Complex Multitrait Phenotypes

The multitrait phenotype of BC₂ANN was first compared with the five species. When all traits for BC₂ANN, the two parents, and the three hybrid species are considered simultaneously, the first three principal components of the PCA account for 47.2% of the variability in the data set (fig. 1). The BC₂ANN is highly variable; not surprisingly, much of the variation in BC₂ANN is explained by the same traits as in the recurrent parentHelianthus annuus. More importantly, there are several BC₂ANN individuals whose trait variation more closely resembles that ofHelianthus petiolaris,H. deserticola, and evenH. anomalus. However, no BC₂ANN individuals appear close toH. paradoxus (fig. 1). The BC₂PET population is similarly variable when compared to the five species, with the majority of the variation in BC₂PET explained by the same traits as in the recurrent parent,H. petiolaris (fig. 2). Moreover, BC₂PET appears to have even more individuals that resembleH. petiolaris,H. deserticola, andH. anomalus than BC₂ANN.

When the synthetic hybrids are compared to each other, the first principal component of the PCA with only BC₂ANN and BC₂PET accounts for nearly 43.2% of the variation in the data set (fig. 3). Morphological (size/shape) and physiological traits were important in explaining the variation between the synthetic hybrid populations. For example, BC₂ANN individuals have strong positive loadings for DISKDIA, PHYNUM, and LFAREA in the first principal component. In contrast, the BC₂PET population has strong negative loadings for PHYSHAP, LFLGTH, LFWDTH and PHOTO in the first principal component. Note that other principal components did not differentiate the synthetic backcross hybrid populations into distinct groups and are not shown.

Re-creating Hybrid Species’ Multitrait Phenotypes from Early-Generation Hybrids

Discriminant analyses of the two parental and three hybrid species were used to construct classification rules that we applied to each of the synthetic hybrid populations. The results for the classification of early-generation hybrids make sense when we look at the numbers classified into the recurrent parent. For example, on average, 88% (83–97) of the BC₂ANN were placed intoH. annuus, and 83% (79–92) of BC₂PET were placed intoH. petiolaris (table 2). Not surprisingly, fewer synthetic hybrids were classified into the nonrecurring parent phenotype. The BC₂ANN were classified asH. petiolaris about 10.1% (1.2–15.6) of the time, while none of the rules classified BC₂PET individuals asH. annuus-like.

Hybrid species-like phenotypes were much rarer among the synthetic hybrids than parental phenotypes, and their relative scarcity differed depending on the synthetic hybrid lineage. When analyzing the BC₂ANN population,H. anomalus-like individuals were recovered in all analyses about 0.7% of the time.Helianthus deserticola-like andH. paradoxus-like phenotypes were somewhat less common (0.5% and 0.4%). Unlike the other two hybrid species,H. paradoxus-like individuals were absent according to two of the four models (table 2). Overall, the decision rule error rates were low and similar (0%–1.9%) for all BC₂ANN decision rules. Classification error rates were higher because of ambiguity in assigning synthetic hybrids to parental classes; that is, most of the ambiguous cases were classified as partH. annuus and partH. petiolaris.

None of the rules classified BC₂PET individuals asH. annuus-like.Helianthus anomalus-like phenotypes were identified about 1% of the time.Helianthus deserticola-like phenotypes were much more common: on average, 13% of BC₂PET could be classified asH. deserticola-like.Helianthus paradoxus-like individuals were not observed in BC₂PET, regardless of the classification rule. Decision rule error rates were very low, and again, classification error rates were slightly higher. Here ambiguity was associated with BC₂PET cases that were petiolaris-like and/or deserticola-like. Nonetheless, BC₂PET were classified asH. deserticola-like unambiguously more than 50% of the time.

Phenotypic Correlations Matrices (P Matrix) in Synthetic Hybrids

Visual comparisons demonstrate that our first two testable hypotheses, matrix equality and proportionality, are rejected (fig. 4). If the matrices were identical, the points infigure 4 would all be on the dashed line (1: 1). Similarly, if the matrices were proportional, the points would be distributed along another straight line that passes through the origin but with a different slope (e.g.,Arnold 1992). Note that the regression intercept is very close to, but significantly different from, 0 (t = 5.24, df = 901,P < .0001). We tested the subsequent (lower) steps of matrix similarity using a hierarchical approach (Flury 1988). The jump-up approach indicates that the correlation structure of the synthetic hybrids share all their principal components in common and differ only in their eigen values (full CPC model;table 3). A model-building approach reveals a slightly different and more conservative estimate of similarity, with the best model being CPC (36) (table 4).

Trait Expression in BC₂ANN and BC₂PET

When considering traits individually, the proportion of BC₂ANN individuals that had transgressive trait values (63%–70%) is slightly higher than those reported for the reciprocal artificial BC₂PET population (Rieseberg et al. 2003) and very high when compared with other studies of wild plants. For example, a survey of 131 studies of segregating hybrids in plants reported that only 58% of traits examined were transgressive overall (Rieseberg et al. 1999b). Transgressive segregation was reportedly much lower in wild populations than in domesticated lines (38% vs. 92%), and crosses between outcrossing and/or mixed mating species was lower than for selfing species (39% and 92%, respectively). When we considered only outcrossing wild plants, a mere 14% of the traits were transgressive (Rieseberg et al. 1999b). Clearly, both BC₂PET and BC₂ANN have a very high number of transgressive traits for a wild, outcrossing plant (Rieseberg et al. 1999b). The unusually high rate of transgressive traits in both of these synthetic populations attests to the potential for synthetic hybrids to have individuals with multitrait phenotypes similar to the natural hybrid species.

Extreme traits in the ancient hybrid species do not necessarily have to have arisen immediately following hybridization. For example,Helianthus anomalus andHelianthus deserticola have significantly fewer ray florets (LIGNUM) than parental species, yet BC₂PET and BC₂ANN have only positively transgressive individuals (i.e., individuals with phenotypic values greater than either parent). The most recent estimates place the origin of the hybrid species approximately 60,000 ybp (Gross et al. 2003); therefore, some phenotypic differences (or extreme traits) almost certainly arose through selection on mutations arising after hybrid speciation. The response of BC₂ANN and BC₂PET populations to selection in the native hybrid species’ habitats was examined for each of the hybrid species (Lexer et al. 2003b;Gross et al. 2004;Ludwig et al. 2004). While those experiments did not measure selection on LIGNUM, they demonstrate that selection on synthetic hybrids in hybrid species’ habitats favors the hybrid species’ phenotypes in several instances (Gross et al. 2004;Ludwig et al. 2004).

Alternatively, it may be that some traits have other genetic or developmental constraints (Schlichting and Pigliucci 1998;Pigliucci 2003). It is interesting that the mean LIGNUM as well as associated floral traits (PHYNUM, DISKDIA) are virtually identical within the five species when compared between years (Rosenthal et al. 2002; this article). While it is not unusual that these floral traits covary, this is surprising given the variability we see among years for other traits. Their similarity from one year to the next suggests they are highly canalized and probably form a “correlation Pleiades” (Berg 1960). Such correlation and constraint should reduce the amount of variation we see for those characters in spite of hybridization.

Multitrait Phenotypes and Correlations

When considering all the traits simultaneously, some synthetic hybrids were classified as one of the three hybrid species. Overall, the classification results show that the complex multitrait phenotypes of the hybrid species can be re-created in early-generation hybrids. However, these are generally rare occurrences. For example, in all BC₂ANN cases, <2% of individuals were classified as similar to hybrid species. The proportion of BC₂PET recognized as hybrid species-like was higher (4.7%). However, the proportion of BC₂PET classified as hybrid species-like was radically skewed towardH. deserticola (7–18%). This makes sense becauseHelianthus petiolaris andH. deserticola are similar phenotypically in many respects. The high proportion ofH. deserticola-like hybrids in the BC₂PET population is also consistent with previous multivariate analyses that had difficulty differentiating between the two species (Rosenthal et al. 2002). Furthermore,H. deserticola trait values for flowering time, leaf size, and stem diameter are closer to BC₂PET than BC₂ANN when grown inH. deserticola habitat (Gross et al. 2004).

Given that we only evaluated two synthetic hybrid populations, the finding of even the occasional hybrid species-like individual is remarkable. Nonetheless, the predictions based only on the two crosses presented here underestimate the capability forH. annuus ×Helianthus petiolaris hybrid lineages to yield hybrid species-like individuals. As noted earlier, hybrid zones are likely to have numerous individuals from several types of early and later generation recombinant lineages (Cruzan and Arnold 1993) with a greater array of possible phenotypes and genotypes. Moreover, the relatively small sample sizes used here for the species may not reflect the entire range of each species’ phenotypic variation. Finally, the crossing design, necessary for the associated QTL mapping studies of BC₂PET (Rieseberg et al. 2003) and BC₂ANN (L. H. Rieseberg, unpublished data), is likely to underrepresent variation in either of the parental species. This implies that the discriminant criteria ultimately yielded a conservative estimate of hybrid species-like phenotypes.

Within and among population differences in phenotypic correlation matrices in general (i.e., integration in the broad sense) can be due to deterministic ecological and genetic factors in the short term and stochastic influences (genetic drift, mutation, and geneflow) in the long term (Waitt and Levin 1993;Armbruster and Schwaegerle 1996;Schlichting and Pigliucci 1998;Pigliucci and Hayden 2001). Past work suggests that both endogenous and exogenous selection (sensuArnold 1997) largely control genomic composition in stabilized sunflower hybrids (Rieseberg et al. 1996,2003;Lexer et al. 2003a) with exogenous selection having a slightly larger effect. The large size of parental chromosomal segments, consistent genomic composition, and tight linkage or pleiotropy among QTLs in synthetic or natural hybrid lineages (Rieseberg et al. 1995,1996,2003) implies that traits associated with those segments are likely to covary. The similarity of BC₂ANN and BC₂PET correlation matrices and the consistency with which we detected synthetic hybrids with hybrid species-like phenotypes reinforces the observations that deterministic processes underlie the initial formation of hybrid species multitrait phenotypes. In spite of seemingly strong genetic and genomic constraints, the small but apparent differences in trait correlations and apparent phenotypes may result in different responses of BC₂PET and BC₂ANN to selection (e.g.,Lexer et al. 2003b;Gross et al. 2004;Ludwig et al. 2004).

We note that the environmental/ecological factors can cause differences in the phenotypic (P) and genotypic (G) matrices (sensuLande and Arnold 1983). Since the synthetic hybrids here were grown in the same greenhouses but in different years, it is possible some of the similarities/differences we see in trait correlations have an environmental basis (Begin and Roff 2001;Pigliucci and Hayden 2001). However, the close similarities in the correlation matrices and phenotypic differences are consistent with expectation for closely related synthetic hybrids (Begin and Roff 2003).

Phenotypic Divergence and Lineage Age

While BC₂ANN and BC₂PET individuals were most commonly classified as one of the parental species,H. deserticola-like individuals were identified by all the decision rules and from both reciprocal crosses. Recent surveys have revealed four different CpDNA haplotypes across eightH. deserticola populations (Gross et al. 2003). One was unique toHelianthus annuus, another toH. petiolaris, and two more were equally distributed among the parental taxa. Similarly, the 18 nuclear loci analyzed in the same populations were evenly distributed between the parental taxa. The frequency and repeatability of recovery of theH. deserticola-like individual reflects the probable multiple origins ofH. deserticola. Helianthus anomalus-like individuals were also identified by every decision rule and in both synthetic hybrid lines. Chloroplast data for this species revealed four haplotypes, two of which were common to both parents (Schwarzbach and Rieseberg 2002). Furthermore, crossing data and meiotic analyses revealed three different karyotypic forms, which were closely correlated in geographic distribution with chloroplast DNA haplo-type. Analyses of genetic relationships amongH. anomalus microsatellite loci were more difficult to interpret. Nonetheless, the most parsimonious explanation is thatH. anomalus had multiple origins (Schwarzbach and Rieseberg 2002). The consistency with which we can recognizeH. anomalus-like orH. deserticola-like individuals by discriminant analysis reflects their purported multiple origins. This is also consistent with earlier work demonstrating the consistency with which the genome of these species can be reconstructed (Rieseberg et al. 1996,2003).

In contrast toH. anomalus andH. deserticola, synthetic hybrids resemblingHelianthus paradoxus were considerably more difficult to find in the BC₂ANN population and were never identified in the BC₂PET populations. Unlike the other two hybrid species, a survey ofH. paradoxus populations revealed only one CpDNA haplotype, which was only found inH. annuus populations. In fact, when CpDNA and nuclear microsatellite data are viewed together, they are consistent with a single origin forH. paradoxus (Welch and Rieseberg 2002b). The difficulty in recognizingH. paradoxus-like individuals in general (when compared toH. deserticola andH. anomalus) is consistent with a single origin hypothesis. The apparent absence of anyH. paradoxus-like plants in the reciprocal backcross with anH. petiolaris maternal individual is also noteworthy in light of the fact that it is fixed for anH. annuus haplotype.

Conclusions

Hybridization undoubtedly played an important role in the adaptive evolution of annual sunflowers. An important caveat is that we do not know the habitats, or the particular locations, where the initial hybridization events took place. However, re-creating the species’ complex phenotypes, each with a distinct and divergent habitat, suggests that hybridization today could generate individuals with phenotypes suitable for persistence in their respective environments. This is particularly encouraging since the initial hybridization event betweenH. annuus andH. petiolaris occurred at least 60,000 ybp (Schwarzbach and Rieseberg 2002;Welch and Rieseberg 2002b;Gross et al. 2003). Overall, the implication of this work is that several natural hybridization events similar to the two we generated are likely to have produced individuals with phenotypes suitable for colonizing habitats divergent from either parent within three generations.

1. Abbott RJ. Plant invasions, interspecific hybridization and the evolution of new plant taxa. Trends in Ecology & Evolution. 1992;7:401–405. doi: 10.1016/0169-5347(92)90020-C. [DOI] [PubMed] [Google Scholar]
2. Armbruster WS, Schwaegerle KE. Causes of covariation of phenotypic traits among populations. Journal of Evolutionary Biology. 1996;9:261–276. [Google Scholar]
3. Arnold ML. Natural hybridization and evolution. Oxford University Press; New York: 1997. [Google Scholar]
4. Arnold SJ. Constraints on phenotypic evolution. American Naturalist. 1992;140(suppl):S85–S107. doi: 10.1086/285398. [DOI] [PubMed] [Google Scholar]
5. Begin M, Roff DA. An analysis of G matrix variation in two closely related cricket species, Gryllus firmus and Gpennsylvanicus. Journal of Evolutionary Biology. 2001;14:1–13. doi: 10.1046/j.1420-9101.2001.00258.x. [DOI] [PubMed] [Google Scholar]
6. Begin M, Roff DA. The constancy of the G matrix through species divergence and the effects of quantitative genetic constraints on phenotypic evolution: a case study in crickets. Evolution. 2003;57:1107–1120. doi: 10.1111/j.0014-3820.2003.tb00320.x. [DOI] [PubMed] [Google Scholar]
7. Berg RL. The ecological significance of correlation pleiades. Evolution. 1960;14:171–180. [Google Scholar]
8. Buerkle CA, Morris RJ, Asmussen MA, Rieseberg LH. The likelihood of homoploid hybrid speciation. Heredity. 2000;84:441–451. doi: 10.1046/j.1365-2540.2000.00680.x. [DOI] [PubMed] [Google Scholar]
9. Carney SE, Gardner KA, Rieseberg LH. Evolutionary changes over the fifty-year history of a hybrid population of sun-flowers (Helianthus) Evolution. 2000;54:462–474. doi: 10.1111/j.0014-3820.2000.tb00049.x. [DOI] [PubMed] [Google Scholar]
10. Cheverud JM. A comparison of genetic and phenotypic correlations. Evolution. 1988;42:958–968. doi: 10.1111/j.1558-5646.1988.tb02514.x. [DOI] [PubMed] [Google Scholar]
11. Clausen J, Hiesey WM. Experimental studies on the nature of species. Carnegie Institute of Washington Publication 615; Washington, DC: 1958. [Google Scholar]
12. Cruzan MB, Arnold ML. Ecological and genetic associations in an iris hybrid zone. Evolution. 1993;47:1432–1445. doi: 10.1111/j.1558-5646.1993.tb02165.x. [DOI] [PubMed] [Google Scholar]
13. Devicente MC, Tanksley SD. QTL analysis of transgressive segregation in an interspecific tomato cross. Genetics. 1993;134:585–596. doi: 10.1093/genetics/134.2.585. [DOI] [PMC free article] [PubMed] [Google Scholar]
14. Donovan LA, Ehleringer JR. Potential for selection on plants for water-use efficiency as estimated by carbon-isotope discrimination. American Journal of Botany. 1994;81:927–935. [Google Scholar]
15. Dudley SA. The response to differing selection on plant physiological traits: evidence for local adaptation. Evolution. 1996;50:103–110. doi: 10.1111/j.1558-5646.1996.tb04476.x. [DOI] [PubMed] [Google Scholar]
16. Falconer DS, Mackay TFC. Introduction to quantitative genetics. Longman; Essex: 1996. [Google Scholar]
17. Farquhar GD, Ehleringer JR, Hubick KT. Carbon isotope discrimination and photosynthesis. Annual Review of Plant Physiology and Plant Molecular Biology. 1989;40:503–537. [Google Scholar]
18. Field C, Mooney HA. The photosynthesis nitrogen relationship in wild plants. In: Givnish TJ, editor. On the economy of plant form and function. Cambridge University Press; Cambridge: 1986. pp. 25–55 . [Google Scholar]
19. Flury B. Common principal components and related multivariate models: Wiley Series in Probability and Mathematical Statistics. Wiley; New York: 1988. [Google Scholar]
20. Gallez GP, Gottlieb LD. Genetic evidence for the hybrid origin of the diploid plant stephanomeria-diegensis. Evolution. 1982;36:1158–1167. doi: 10.1111/j.1558-5646.1982.tb05486.x. [DOI] [PubMed] [Google Scholar]
21. Grant V. The regulation of recombination in plants. Cold Spring Harbor Symposia on Quantitative Biology. 1958;23:337–363. doi: 10.1101/sqb.1958.023.01.034. [DOI] [PubMed] [Google Scholar]
22. Grant V. Character coherence in natural hybrid populations in plants. Botanical Gazette. 1979;140:443–448. [Google Scholar]
23. Grant V. Plant speciation. Columbia University Press; New York: 1981. [Google Scholar]
24. Gross BL, Schwarzbach AE, Rieseberg LH. Origin(s) of the diploid hybrid species Helianthus deserticola (Asteraceae) American Journal of Botany. 2003;90:1708–1719. doi: 10.3732/ajb.90.12.1708. [DOI] [PubMed] [Google Scholar]
25. Gross BL, Kane NC, Lexer C, Ludwig F, Rosenthal DM, Donovan LA, Rieseberg LH. Reconstructing the origin of Helianthus deserticola: survival and selection on the desert floor. American Naturalist. 2004;164:145–156. doi: 10.1086/422223. [DOI] [PMC free article] [PubMed] [Google Scholar]
26. Heiser CB. Variation and subspeciation in the common sun-flower, Helianthus annuus. American Midland Naturalist. 1954;51:287–305. [Google Scholar]
27. Heiser CB. Morphological and cytological variation in Helianthus petiolaris with notes on related species. Evolution. 1961;15:247–258. [Google Scholar]
28. Heiser CB, Smith DM, Clevenger SB, Martin WC. The North American sunflowers (Helianthus) Memoirs of the Torrey Botanical Club. 1969;22:1–218. [Google Scholar]
29. Hirai H, Taguchi T, Saitoh Y, Kawanaka M, Sugiyama H, Habe S, Okamoto M, et al. Chromosomal differentiation of the Schistosoma japonicum complex. International Journal for Parasitology. 2000;30:441–452. doi: 10.1016/s0020-7519(99)00186-1. [DOI] [PubMed] [Google Scholar]
30. Lande R. Quantitative genetic analysis of multivariate evolution, applied to brain-body size allometry. Evolution. 1979;33:402–416. doi: 10.1111/j.1558-5646.1979.tb04694.x. [DOI] [PubMed] [Google Scholar]
31. Lande R, Arnold SJ. The measurement of selection on correlated characters. Evolution. 1983;37:1210–1226. doi: 10.1111/j.1558-5646.1983.tb00236.x. [DOI] [PubMed] [Google Scholar]
32. Lewontin RC, Birch LC. Hybridization as a source of variation for adaptation to new environments. Evolution. 1966;20:315–336. doi: 10.1111/j.1558-5646.1966.tb03369.x. [DOI] [PubMed] [Google Scholar]
33. Lexer C, Welch ME, Durphy JL, Rieseberg LH. Natural selection for salt tolerance quantitative trait loci (QTLs) in wild sunflower hybrids: implications for the origin of Helianthus paradoxus, a diploid hybrid species. Molecular Ecology. 2003a;12:1225–1235. doi: 10.1046/j.1365-294x.2003.01803.x. [DOI] [PubMed] [Google Scholar]
34. Lexer C, Welch ME, Raymond O, Rieseberg LH. The origin of ecological divergence in Helianthus paradoxus (Asteraceae): selection on transgressive characters in a novel hybrid habitat. Evolution. 2003b;57:1989–2000. doi: 10.1111/j.0014-3820.2003.tb00379.x. [DOI] [PubMed] [Google Scholar]
35. Ludwig F, Rosenthal DM, Johnston JA, Kane N, Gross BL, Lexer C, Dudley SA, et al. Selection on leaf ecophysiological traits in a desert hybrid helianthus species and early generation hybrids. Evolution. 2004;58:2682–2692. doi: 10.1111/j.0014-3820.2004.tb01621.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
36. Lynch M, Pfrender M, Spitze K, Lehman N, Hicks J, Allen D, Latta L, et al. The quantitative and molecular genetic architecture of a subdivided species. Evolution. 1999;53:100–110. doi: 10.1111/j.1558-5646.1999.tb05336.x. [DOI] [PubMed] [Google Scholar]
37. McCarthy EM, Asmussen MA, Anderson WW. A theoretical assessment of recombinational speciation. Heredity. 1995;74:502–509. [Google Scholar]
38. Monforte AJ, Asins MJ, Carbonell EA. Salt tolerance in lycopersicon species. V. Does genetic variability at quantitative trait loci affect their analysis? Theoretical and Applied Genetics. 1997;95:284–293. [Google Scholar]
39. Murren CJ, Pendleton N, Pigliucci M. Evolution of phenotypic integration in Brassica (Brassicaceae) American Journal of Botany. 2002;89:655–663. doi: 10.3732/ajb.89.4.655. [DOI] [PubMed] [Google Scholar]
40. Phillips PC, Arnold SJ. Hierarchical comparison of genetic variance-covariance matrices. I Using the Flury hierarchy Evolution. 1999;53:1506–1515. doi: 10.1111/j.1558-5646.1999.tb05414.x. [DOI] [PubMed] [Google Scholar]
41. Pigliucci M. Phenotypic integration: studying the ecology and evolution of complex phenotypes. Ecology Letters. 2003;6:265–272. [Google Scholar]
42. Pigliucci M, Hayden K. Phenotypic plasticity is the major determinant of changes in phenotypic integration in Arabidopsis. New Phytologist. 2001;152:419–430. doi: 10.1046/j.0028-646X.2001.00275.x. [DOI] [PubMed] [Google Scholar]
43. Rieseberg LH. Homoploid reticulate evolution in Helianthus (Asteraceae): evidence from ribosomal genes. American Journal of Botany. 1991;78:1218–1237. [Google Scholar]
44. Rieseberg LH. Hybrid origins of plant species. Annual Review of Ecology and Systematics. 1997;28:359–389. [Google Scholar]
45. Rieseberg LH, Beckstromsternberg SM, Liston A, Arias DM. Phylogenetic and systematic inferences from chloroplast DNA and isozyme variation in Helianthus sect. Helianthus (Asteraceae) Systematic Botany. 1991;16:50–76. [Google Scholar]
46. Rieseberg LH, Vanfossen C, Desrochers AM. Hybrid speciation accompanied by genomic reorganization in wild sun-flowers. Nature. 1995;375:313–316. [Google Scholar]
47. Rieseberg LH, Sinervo B, Linder CR, Ungerer MC, Arias DM. Role of gene interactions in hybrid speciation: evidence from ancient and experimental hybrids. Science. 1996;272:741–745. doi: 10.1126/science.272.5262.741. [DOI] [PubMed] [Google Scholar]
48. Rieseberg LH, Whitton J, Gardner K. Hybrid zones and the genetic architecture of a barrier to gene flow between two sunflower species. Genetics. 1999a;152:713–727. doi: 10.1093/genetics/152.2.713. [DOI] [PMC free article] [PubMed] [Google Scholar]
49. Rieseberg LH, Archer MA, Wayne RK. Transgressive segregation, adaptation and speciation. Heredity. 1999b;83:363–372. doi: 10.1038/sj.hdy.6886170. [DOI] [PubMed] [Google Scholar]
50. Rieseberg LH, Raymond O, Rosenthal DM, Lai Z, Living-stone K, Nakazato T, Durphy JL, et al. Major ecological transitions in wild sunflowers facilitated by hybridization. Science. 2003;301:1211–1216. doi: 10.1126/science.1086949. [DOI] [PubMed] [Google Scholar]
51. Rosenthal DM, Schwarzbach AE, Donovan LA, Raymond O, Rieseberg LH. Phenotypic differentiation between three ancient hybrid taxa and their parental species. International Journal of Plant Sciences. 2002;163:387–398. [Google Scholar]
52. Rosenthal DM, Ludwig F, Donovan LA. Plant responses to an edaphic gradient across a active sand dune/desert boundary in the Great Basin Desert. International Journal of Plant Sciences. 2005;166:247–255. [Google Scholar]
53. SAS Institute. SAS/STAT user’s guide. Version 8.01. SAS Institute; Cary, NC: 2001. [Google Scholar]
54. Schlichting CD, Pigliucci M. Phenotypic evolution: a reaction norm perspective. Sinauer; Sunderland, MA: 1998. [Google Scholar]
55. Schluter D. The ecology of adaptive radiation. Oxford Series in Ecology and Evolution. Oxford University Press; New York: 2000. [Google Scholar]
56. Schwarzbach AE, Donovan LA, Rieseberg LH. Transgressive character expression in a hybrid sunflower species. American Journal of Botany. 2001;88:270–277. [PubMed] [Google Scholar]
57. Schwarzbach AE, Rieseberg LH. Likely multiple origins of a diploid hybrid sunflower species. Molecular Ecology. 2002;11:1703–1715. doi: 10.1046/j.1365-294x.2002.01557.x. [DOI] [PubMed] [Google Scholar]
58. Sokal RR, Rohlf FJ. Biometry. W. H. Freeman; New York: 1995. [Google Scholar]
59. Spitze K. Population structure in Daphnia obtusa: quantitative genetic and allozymic variation. Genetics. 1993;135:367–374. doi: 10.1093/genetics/135.2.367. [DOI] [PMC free article] [PubMed] [Google Scholar]
60. Stebbins GL. Self fertilization and population variability in the higher plants. American Naturalist. 1957;91:337–354. [Google Scholar]
61. Stebbins GL. The role of hybridization in evolution. Proceedings of the American Philosophical Society. 1959;103:231–251. [Google Scholar]
62. Stebbins GL, Daly K. Changes in variation pattern of a hybrid population of Helianthus over an eight-year period. Evolution. 1961;15:60–71. [Google Scholar]
63. Templeton AR. Modes of speciation and inferences based on genetic distances. Evolution. 1980;34:719–729. doi: 10.1111/j.1558-5646.1980.tb04011.x. [DOI] [PubMed] [Google Scholar]
64. Waitt DE, Levin DA. Phenotypic integration and plastic correlations in Phlox drummondii (Polemoniaceae) American Journal of Botany. 1993;80:1224–1233. [Google Scholar]
65. Waitt DE, Levin DA. Genetic and phenotypic correlations in plants: a botanical test of Cheverud’s conjecture. Heredity. 1998;80:310–319. [Google Scholar]
66. Welch ME, Rieseberg LH. Habitat divergence between a homoploid hybrid sunflower species, Helianthus paradoxus (Asteraceae), and its progenitors. American Journal of Botany. 2002a;89:472–478. doi: 10.3732/ajb.89.3.472. [DOI] [PubMed] [Google Scholar]
67. Welch ME, Rieseberg LH. Patterns of genetic variation suggest a single, ancient origin for the diploid hybrid species Helianthus paradoxus. Evolution. 2002b;56:2126–2137. doi: 10.1111/j.0014-3820.2002.tb00138.x. [DOI] [PubMed] [Google Scholar]