Antirrhinum species placed in 3D genotypic space. ( A ) Mean A. tortuosum flower. ( B ) A. tortuosum flower warped to mean shape used to generate genotypic space (Fig. 1C). ( C ) Projection of (B) into genotypic space. ( D ) Image obtained when (C) is warped back to the mean A. tortuosum flower shape. ( E ) Projection of 19 Antirrhinum species into genotypic space. In each case, flower on the left shows the mean appearance, whereas flower on the right shows the appearance after projection into genotypic space [equivalent to (A) and (D) for each species]. ( F ) Cloud obtained for flowers from 19 species represented in genotypic space. Each point shows the position of a single flower projected into genotypic space. Examples of flowers from different positions in the genotypic space are illustrated. ( G and H ) Two different 2D projections of the cloud, with each species color-coded as in (E). 

Context in source publication

Context 1

... was largely accounted for by the ROS EL loci. These results allowed us to create an appropriate genotypic space for the F2. Digital images of representative flowers of four genotypes were warped to the same average flower shape. Principal component analysis on variation in pixel color at each position in the flower then allowed us to define a 3D genotypic space controlling flower color ( 15–17 ) (Fig. 2D). The role of flower color variation in natural populations was assessed by analyzing a hybrid zone between A. m. pseudomajus and A. m. striatum . Scoring 493 plants across the hybrid zone revealed a steep cline for flower color (Fig. 1B). Allelism tests on 14 plants from the contact zone with a range of phenotypes con- firmed that flower color was largely determined by ROS , EL , and SULF genotypes. For ROS , more extensive genotyping could be carried out by using molecular markers. The ROS locus comprises a tandem duplication of two MYB- related transcription factors, ROS1 and ROS2, with ROS1 having a greater role in flower color variation ( 13 ). We sequenced a 1.2-kb region of ROS1 , from the promoter to the start of the second exon. Sequences from 13 yellow and 15 magenta morphs from locations distant from the contact zone showed that ROS1 alleles fell into three major groups (haplogroups) (Fig. 3A). One haplogroup was specific to yellow morphs and was identical to the ros dor allele of A. majus , hypothesized to have been derived from the wild ( 13 ). The other two haplogroups were found only in magenta morphs. ROS1 sequences were used to design primers that allowed haplogroups to be distinguished by polymerase chain reaction. Genotyping 528 plants from the hybrid zone showed that the cline in ROS1 haplogroup frequency coincided with magenta flower color (Fig. 1C). Assuming that the hybrid zone arose from contact between previously separate yellow and magenta populations, the observed clines in flower color and genotype might have two ex- planations ( 18 , 19 ). One is that A. m. striatum and A. m. pseudomajus came into recent contact and the clines reflect a neutral mixing of alleles between the populations. Alternatively, there has been a longer history of contact, and clines reflect selection maintaining morph differences. To evaluate these possibilities, we ana- lyzed molecular variation at loci not involved in magenta and yellow morph differences. According to the neutral model, these loci should have a cline similar to that of ROS1 . The PALLIDA ( PAL ) and DICHOTOMA ( DICH ) loci were chosen because they are linked to ROS 16 centimorgan (cM) and 9 cM from ROS , respec- tively ^ , and sequences are available for primer design ( 20 , 21 ). Alleles were sequenced from 18 individuals on either side of the hybrid zone. Most PAL alleles fell into two major haplogroups (Fig. 3B). DICH alleles showed little haplogroup structure, although several DNA polymorphisms were detected. We genotyped 496 plants across the hybrid zone for the two PAL haplogroups and a polymorphism at DICH . No cline was observed for PAL or DICH , indicating that these genes were subject to different evolutionary factors than ROS1 (Fig. 1C). This was also supported by genotyping 16 A. m. striatum and A. m. pseudomajus populations distant from the hybrid zone (Fig. 1A and table S1). In all cases, the ROS1 haplogroup correlated with flower color, whereas PAL and DICH loci showed no such association. The simplest interpretation of these results is that spatial variation in PAL and DICH allele frequencies reflects historical gene flow between populations, whereas the ROS1 cline has been maintained by selection on flower color. The cline could be maintained, for example, if intermediate genotypes have lower fitness than the parental morphs ( 22 ). Thus, magenta and yellow morphs might represent distinct peaks on an adaptive landscape, whereas intermediate forms represent an intervening low fitness valley. However, this raises the problem of how the low fitness valley was traversed when the two morphs diverged from a common ancestor. To address this issue, we mapped the range of phenotypes exhibited by Antirrhinum species within the defined genotypic space (Fig. 2D). This was achieved by photographing several flowers from each species (Fig. 4A) and warping the images to the same flower shape (Fig. 4B). We then determined the position in the genotypic space that best approximated the color for each flower (Fig. 4C). The approximation was evaluated by warping the resulting image back to the initial flower shape and comparing it to the original image (Fig. 4D). Much of the var- iation in flower color was captured within the 3D genotypic space, consistent with previous studies showing that the ROS , EL , and SULF loci play important roles in color variation in the species group as a whole ( 11–13 ) (Fig. 4E). When flowers from 19 species were mapped into the genotypic space, they collectively formed a broad U-shaped cloud of points (Fig. 4, F to H). Flowers from each species formed smaller clusters within this broader cloud. Magenta A. m. pseudomajus flowers localized near one end of the cloud, whereas yellow A. m. striatum flowers were near the other end. Intermediate positions within the cloud corresponded to various other patterns and intensities of color. However, certain color combinations were excluded from the cloud, even though they were observed in F2 and hybrid zone populations. For example, orange flowers, having a broad spread of both yellow and magenta ( ROS el / ROS el; sulf / sulf ), were not within the cloud (Fig. 4F). The absence of this genotype in wild species could be ex- plained if individuals with orange flowers have lower fitness, perhaps because they are less at- tractive to pollinators ( 23–25 ). The role of pollinators in propagating A. m. pseudomajus and A. m. striatum is likely to be of central impor- tance because the species are self-incompatible, seed dispersal is limited (involving gravity or water runoff), and individuals typically survive for only 1 to 3 years. Taken together, our results suggest that magenta and yellow morphs did not evolve through intermediate genotypes giving orange flowers, but that instead evolution followed the route defined by the U-shaped cloud. According to this view, the cloud represents a region of high fitness, allowing flower color to evolve without incurring major fitness costs. However, when genotypes, such as magenta and yellow morphs, from distant parts of the cloud meet, they can generate progeny that lie outside the high fitness cloud, creating a barrier to exchange of flower color alleles. This would account for the observed steep cline at loci controlling color differences in the hybrid zone. A 2D slice through the U-shaped cloud, passing perpendicularly through its two arms, would yield an adaptive landscape with two separate peaks. The cloud therefore represents a high fitness path between what might otherwise seem like distinct peaks, showing how higher dimensional representations allow adaptive continuity and incompatibility to be more easily reconciled ( 2 ). cological theory ( 1 , 2 ) and field experi- E relationship ments ( 3 , 4 between ) have the revealed diversity a positive of plant species and the diversity of associated consum- ers. At least two mechanisms might explain this pattern. First, because approximately 90% of herbivorous insects exhibit some degree of host specialization ( 5 ), as plant species richness increases, so should the number of associated herbivore species. This resource specialization hypothesis has some theoretical support ( 1 , 2 , 6 ). Second, if aboveground net primary productivity (ANPP) increases as plant species richness increases ( 7 ), then more herbivore individuals, and therefore more species, will be supported by increases in available energy (this has been called the more individuals hypothesis) ( 8 ). An increase in the number of herbivore species by either of these mechanisms should support more predator species ( 9 ). Recent studies have shown that population genotypic diversity, like plant species diversity, can have extended consequences for communities and ecosystems ( 10–14 ). However, no studies to date have ex- plicitly linked intraspecific genotypic diversity, the structure of associated communities, and the potential mechanisms driving these patterns, such as energy availability. This paucity of studies exists despite numerous calls for such research within the literature regarding biodiversity-ecosystem function ( 7 , 15 ). We tested whether host-plant genotypic diversity determines the structure of associated arthropod communities and governs an ecosystem process, ANPP, that influences arthropod species richness. We manipulated the plot-level genotypic diversity (the number of genotypes per plot) of Solidago altissima , tall goldenrod, a common perennial plant throughout eastern North America. Twenty-one S. altissima ramets were collected from local S. altissima patches growing in fields near the study site, and each ramet was identified as a unique genotype by means of amplified fragment length polymorphism. From these 21 genotypes, we established 63 1-m 2 experimental plots, each containing 12 individ- uals and 1, 3, 6, or 12 randomly selected genotypes, mimicking the densities and levels of genotypic diversity found in natural patches of similar size. We censused arthropods on every ramet in each plot five times over the course of the growing season. In total, we counted 36,997 individuals of È 136 species. We estimated ANPP at the peak of the growing season using nondestructive allometric techniques ( 16 ). Total cumulative arthropod species richness increased with plant genotypic diversity. The number of arthropod species was, on average, 27% greater in 12-genotype plots than in single- genotype plots (Fig. 1), indicating that plant genotypic diversity was an important determi- nant of arthropod ...