Precise spatio-temporal regulation of the anthocyanin biosynthetic pathway leads to petal spot formation in Clarkia gracilis (Onagraceae)

Summary

• Petal spots are widespread in Angiosperms and are often implicated in pollinator attraction. Clarkia gracilis petals each have a single red-purple spot that contrasts against a pink background. The position and presence of spots in C. gracilis are determined by the epistatic interaction of alleles at two, as-yet-unidentified loci.

• We used HPLC to identify the different pigments produced in the petals, and qualitative and quantitative RT-PCR to assay for spatio-temporal patterns of expression of different anthocyanin pathway genes.

• We found that spots contain different pigments from the remainder of the petal, being composed of cyanidin/peonidin-based, instead of malvidin-based anthocyanins. Expression assays of anthocyanin pathway genes show that Dfr2 has a spot-specific expression pattern and acts as a switch for spot production. Co-segregation analyses implicate the gene products of the P and I loci as trans-regulators of this switch. Spot pigments appear earlier in development due to early expression of Dfr2 and F3′h1. Pigments in the background appear later, due to later expression of Dfr1 and F3′5′h1.

• The evolution of this spot production mechanism appears to have been facilitated by duplication of the Dfr gene and to have required substantial reworking of the anthocyanin pathway regulatory network.

Introduction

A major objective of evolutionary developmental biology (‘EvoDevo’) is to elucidate the genetic changes that result in the evolution of novel characters. One unresolved issue is whether novel characters can arise through changes in the activity of genes that are already active in developmental programs or if they require recruitment of additional genes into those programs (Keys et al., 1999). In addition, while the evolution of novel morphologies is sometimes associated with gene duplication (Xiao et al., 2008, Parker et al., 2009), it is unclear how frequently duplication contributes to the emergence of novel traits. Because of the apparent simplicity and ecological relevance of wing spotting patterns, the developmental control of evolutionarily novel spots in animals, particularly butterflies, has been used to address these issues (Nijhout 1980, Carroll et al, 1994, Keys et al. 1999, Beldade et al. 2002, Reed & Serfas, 2004). These investigations have tentatively indicated that new wing spots have arisen by the modification of existing developmental programs.

In plants, petals with large spots— discrete depositions of visible pigments that contrast with the background coloration of the flower—occur in the families Liliaceae, Orchidaceae, Asteraceae, Papaveraceae, Fabaceae, and many others, indicating that they have evolved independently numerous times. Moreover, field studies have frequently demonstrated that spots are important in mediating interactions with pollinators (Jones, 1996; Johnson & Midgley, 1997; Van Kleunen et al., 2007; Goulson et al., 2009). Nevertheless, the genetic and developmental control of petal spots, and more generally, of petal color patterning, have seldom been investigated. Studies in model species such as petunia and snapdragon have clarified the genetic control of some patterns, namely vein-associated pigmentation (Schwinn et al., 2006; Shang et al., 2011) or variation in pigment intensity in different regions of the corolla (Jackson et al., 1992; Schwinn et al., 2006; Albert et al., 2011). However, our general understanding of how floral pigment patterns develop and are regulated at the molecular level, specifically in relation to petals spots, remains poor. In this report, we describe experiments that begin to elucidate the biochemical, genetic and developmental basis of petal spots in the California wildflower, Clarkia gracilis.

Clarkia gracilis (Onagraceae) is native to northern California, Oregon and Washington. It is the only polyploid in section Rhodanthos. Based on cytological analyses, (Abdel-Hameed & Snow, 1972) suggested that this allotetraploid derived from hybridization between two diploid species, one closely related to C. amoena ssp. huntiana and one related to C. lassenensis and C. arcuata. However, the parentage of C. gracilis has not been tested adequately by modern molecular systematic methods.

Clarkia gracilis is composed of four different subspecies that have somewhat overlapping ranges: C. g. ssp. albicaulis, C. g. ssp. gracilis, C. g. ssp. sonomensis, and C. g. ssp. tracyi. The subspecies differ in their floral morphologies, particularly with respect to the presence and position of petal spots. Subspecies albicaulis and tracyi have a single spot at the base of each petal, ssp. sonomensis has a single spot in the center of each petal, and ssp. gracilis lacks petal spots (Fig. 1). In all cases, mature petal spots are dark reddish-purple concentrations of pigment on a pale pink background. However, there is some variation within some subspecies: C. g. sonomensis and C.g. albicaulis can be found in either spotted or unspotted morphs.

Fig. 1.

Flowers of subspecies in the C. gracilis species complex. From left: C. g. ssp. tracyi, C. g. ssp. sonomensis, C. g. ssp. gracilis, and C. g. ssp. albicaulis.

The ecological significance of petal spots in C. gracilis has been studied in some detail. In mixed populations of central spotted and unspotted C. gracilis ssp. sonomensis, spotted plants have higher fitness, producing as much as 32% more seed than their unspotted counterparts in some growing seasons and sites (Jones, 1996). This difference might reflect pollinator preference, since petal spots were also shown to influence pollinator foraging behavior (Jones, 1996). In another species of Clarkia, C. xantiana, the maintenance of both spotted and unspotted morphs has been attributed to pollinator-based negative frequency-dependent selection (Eckhart et al., 2006).

Previous investigations have established that C. gracilis petal spots are under simple genetic control. Gottlieb & Ford (1988) identified two independent loci that interact epistatically to give rise to all of the morphs present in this species complex. One locus, P, has co-dominant alleles that determine the position of the spot within the petal: the P^B allele causes the production of basal spots, and the P^C allele causes the production of central spots. The other locus, I, affects the presence of basal spots: the I^A allele (or I, in (Gottlieb & Ford, 1988)) suppresses basal spots, while the I^P allele (i, in Gottlieb & Ford, 1988) allows basal spots to form. Neither allele at the I locus affects the formation of central spots. Gottlieb & Ford (1988) determined that C. gracilis ssp. gracilis generally has a P^BP^BI^AI^A genotype, whereas spotted plants of ssp. sonomensis often have the genotype, P^CP^CI^PI^P. However, a variable dominance relationship was also found between I^A and I^P suggesting that additional modifiers or alternative alleles are present in different backgrounds.

The identity of the pigments underlying this phenotype can provide important clues as to the genes that may be involved in spot formation. Species in the genus Clarkia produce anthocyanin pigments, mostly derivatives of malvidin and cyanidin, although not all species produce these pigments (Dorn & Bloom, 1984; Soltis, 1986). Production of pelargonidin and derivatives is rare in the genus, being known only in C. unguiculata (Robinson & Robinson, 1931), suggesting that the biochemical pathway branch leading to pelargonidin production is inactive in most species (Fig. 2). In C. gracilis flowers, the anthocyanidins malvidin, cyanidin and delphinidin (a precursor of malvidin) have been reported (Soltis, 1986).

Fig. 2.

Simplified schematic representation of the anthocyanin biosynthetic pathway.

Abbreviations for enzymes: CHS, chalcone synthase; CHI, chalcone isomerase; F3H, flavanone 3 hydroxylase; F3′H, flavanoid 3′ hydroxylase; F3′5′H, flavonoid 3′-5′ hydroxylase; DFR, dihydroflavonol-4-reductase; ANS, anthocyanidin synthase; FLS, flavonol synthase; UF3GT, UDP-flavonoid-3-glucosyl-transferase. Dihydroflavonols: DHK, dihydrokaempferol; DHQ, dihydroquercetin; DHM, dihydromyricetin.

Our objectives in this investigation were to identify genes associated with spot formation in C. gracilis and characterize their pattern of regulation. Specifically, we sought to answer the following questions:

1. What pigments are produced in C. gracilis flowers? Do pigments in spots differ from the rest of the petal? Do pigments in central spots differ from those in basal spots?

2. Is transcriptional regulation of specific anthocyanin biosynthetic genes involved in petal spot formation? If so, does any gene correspond to the spot position P locus?

Materials and Methods

Plant growth and crosses

Seeds from C. gracilis (Piper) A. Nelson & J. F. Macbride were germinated in vermiculite at 15°C in the dark, and transferred to a 1:1 metromix:peatmoss mixture once seedlings emerged. Seedlings were then transferred to 5-inch pots and were grown in a glasshouse (20–24°C) under natural light. Plants were periodically randomized to avoid possible light-induced changes in flower color.

Several individuals from various populations were grown, and only those from monomorphic (with respect to spots) populations were used as the parents for crosses. Two independent crosses were carried out to separate the P and I loci. For the ‘P’ cross, a central-spotted C. g. sonomensis was crossed to a basal-spotted C. g. albicaulis. For the ‘I’ cross, an unspotted C. g. gracilis was crossed to a basal-spotted C. g. albicaulis. Voucher information is available in Supporting Information Table S1. Pollen recipient flowers were emasculated several days before full stigma receptivity to avoid self-pollination. Stigmas were pollinated by touching them with a dehiscing anther from the designated sire. F1 individuals were selfed to obtain F2 progeny.

Pigment extraction and High Performance Liquid Chromatography (HPLC) analyses

Anthocyanins were extracted from F2 individuals derived from the ‘P’ cross mentioned above as well as from C. gracilis individuals derived from several natural populations. Seven F2 individuals were examined: two central spotted, two basal spotted, and three double-spotted. Five to ten F1 individuals (all double spotted) were pooled into one sample. Individuals from natural populations included three C. g. ssp. albicaulis, and six C. g. ssp. sonomensis plants (three spotted and three unspotted, see Table S1 for voucher information). Each sample contained either the top, center, or base of 4–5 adult flowers per plant (Fig. S1).

Anthocyanidins were extracted as in Harborne (1984). Pigments were separated using HPLC and identified using the retention times of commercially available standards (delphinidin, pelargonidin, cyanidin, peonidin, malvidin, and petunidin, Polyphenols Laboratory, Sandnes, Norway) that were run using the same conditions as the unknown samples (previously described in Smith & Rausher, 2011).

DNA extraction, gene amplification and sequencing

Genomic DNA from C. gracilis ssp. for gene isolations and phylogenetic analyses was extracted from fresh plant tissue using the CTAB method (Doyle & Doyle, 1987). Genomic DNA from C. amoena ssp. huntiana, C. lassenensis, C. arcuata, C. franciscana, C. rubicunda, and C. breweri were kindly provided by Kenneth Sytsma (University of Wisconsin-Madison; voucher information in Table S1). We identified the two homeologs of C. gracilis genes (arbitrarily called A and B) by including sequences from other species in section Rhodanthos in phylogenetic analyses. This was facilitated by C. amoena and C. lassenensis (the putative parents of C. gracilis) not being sister taxa, allowing for the homeologs present in C. gracilis to be identified more easily.

We used sequences available in Genbank to design degenerate primers to isolate partial sequences corresponding to the anthocyanin biosynthetic genes (F3h, F3′h, F3′5′h, Dfr and Ans). Primers and conditions used for each gene/species are available upon request. PCR bands corresponding to predicted amplicon sizes were transformed into E. coli cells. At least 10 colonies per ligation were sequenced, and additional clones were sequenced as needed. Sequencing reactions used the BigDye Terminatorv. 3.1 system following the manufacturer’s protocol (Applied Biosystems) and electrophoresis was performed at the University of Wisconsin-Madison Biotechnology Center.

Sequence and Phylogenetic analyses

Sequencherv. 4.7 (Gene Codes Corp.) was used to edit, assemble and align sequence fragments with further manual alignment in MacClade v. 4.05 (Maddison & Maddison,2000). To confirm gene identity, sequences were submitted to the Basic Local Alignment Search Tool at the National Center for Biotechnology Information. All phylogenetic analyses were performed in PAUP* v. 4.0b10 using maximum likelihood (ML, Swofford, 2002). Potential PCR recombinants were excluded from the analyses. For all datasets, the GTR model with estimated base frequencies was assumed. Support values were obtained by performing 1000 replicates of ML bootstrapping. Only coding sequences were used in analyses (alignments of included sequences provided in Notes S1.

mRNA expression analyses

Total RNA was extracted from petals using RNeasy Plant Mini kit (Qiagen), followed by a DNAse step (MBI Fermentas) and cDNA synthesis using ImpromII Reverse Transcription System (Promega), or Superscript III First Strand Synthesis (Invitrogen), all according to respective manufacturer’s protocol.

For qualitative gene expression experiments, F2 plants of three different phenotypes, central, basal, and double spotted, from the P locus cross, and of two different phenotypes (basal spotted or unspotted) from the I locus cross, were used. Each RNA sample consisted of either the top, base, or center of three to four flower buds from each plant and three plants for each phenotype. Buds were collected at a stage of development when spot pigments were first appearing (or in equivalently sized petals of unspotted plants), but before background color was visible (Fig. S2).

For quantitative gene expression experiments, plants of three different phenotypes (central-, basal-, and non-spotted), derived from the P and I crosses, were used. Buds representing four different stages of development were collected, with three replicates of each stage. ‘Stage’ was based on the emergence and quantity of pigments, because petal length proved to be an unreliable marker for age, as flower size and maturation times varied greatly, within and between individuals. Change in floral pigmentation has been used previously in Clarkia spp. to determine floral developmental stage (Pichersky et al., 1994). In C. gracilis, color develops in the following order: no color, central spot appears (if present), basal spot appears (if present), background color appears (Fig. S3).

Quantitative assays of gene expression were carried out using Clontech’s SYBR Advantage qPCR premix and run in a Stratagene Mx3000P instrument. Dissociation curves and sequences of qPCR products were obtained to verify amplification of only the desired product. Absolute quantification of mRNA transcript levels was determined using a standard curve, followed by normalization to Actin. qPCR runs were performed in duplicate. Primers used for RT-PCR and qPCR (Table S2) flanked introns, with the exception of F3h and Ans. PCR products were sequenced to confirm that the desired copy was being amplified.

Genotyping F2 individuals from P and I crosses at the Dfr2-A and Dfr2-B loci

Thirty-two and 31 F2 plants from the P cross were genotyped for Dfr2-A and Dfr2-B, respectively. Homeolog specific primers were designed to amplify a portion of exon 2 and intron 2 (Dfr2-A), and a portion of intron 2 (Dfr2-B). Fifty plants from the F2 generation from the I cross were genotyped at Dfr2-A and Dfr2-B by amplifying intron 3 and partial sequences of exon 3 and 4. The same Dfr2 primers used for qPCR analyses were used. All alleles were identified using fragment length polymorphisms, with the exception of alleles of Dfr2-B for the P cross, which were genotyped by the presence of a SNP using restriction enzyme digestion (Tables S3 and S4). Fragment length runs were completed in an ABI 3730XL DNA Analyzer at the Duke Genome Sequencing & Analysis Core Resource, visualized using the GeneMapper software (Applied Biosystems) and scored by eye. Fisher’s exact test was used to test for possible co-segregation between spot presence and Dfr2-A/B alleles.

Results

Crosses confirm the two-locus model and provide segregating families for further analysis

Although the genetics of petal spot inheritance had already been determined, we established populations of individuals with known genotypes to facilitate genetic and molecular analyses and to confirm previous genetic results. We developed two separate segregating F2 populations for the two independent loci, thus removing the epistatic interaction that could mask the genotype at the P locus.

Segregation patterns (Table 1) confirmed those previously reported (Gottlieb & Ford, 1988; Jones, 1996). In the P-cross (see Materials and Methods), 100% of F1 progeny were double-spotted. Segregation in the F2 population did not deviate significantly from the expected 1:2:1 ratio of basal spotted:double spotted:central spotted plants (chi-square test, P = 0.74) consistent with a model of a single locus with co-dominant alleles. According to this model P^BP^B plants have basal spots, P^CP^C plants have central spots, and P^BP^C plants have both a central and a basal spot.

Table 1.

Segregation for presence versus absence and position of petal spot in two separate crosses, in F1 and F2 generations of C. gracilis plants

	Spot Position
	n	Central	Basal	Double	Absent
P cross
F1	23		-	23	-
F2	82	18	23	41	-
I cross
F1	7	-	7	-	-
F2	101	-	77	-	24

P cross refers to central-spotted C. g. sonomensis basal-spotted C. g. albicaulis cross, while I cross refers to unspotted C. g. gracilis basal-spotted C. g. albicaulis cross.

In the I-cross (see Materials and Methods), all F1 individuals had basal spots, suggesting that the suppressing I^A allele was recessive to the spot-permitting allele I^P. In the F2 population, we saw variation in spot intensity, from a strongly defined spot to an ill-defined ‘smudge’, to unspotted petals. Because it was difficult to distinguish spots from smudges, plants showing any color at the base were coded as ‘basal spotted’ for statistical analyses. As shown in Table 1, using this scoring, a 3:1 pattern of spotted:unspotted plants could not be rejected (chi-square test, P= 0.77), suggesting that the I^P allele isolated is generally (but variably) dominant to the I^A allele. Genotypes for P and I and expected phenotypes under this genetic model are presented in Fig. S4.

It should be noted that, while in our crosses the I^P allele appears to be dominant, previous work has shown that alleles at the I locus can switch from dominant to recessive in different backgrounds, possibly due to other modifiers (Gottlieb & Ford, 1988). Further genetic studies are necessary to fully understand the phenotypic expression of this allele in different C. gracilis populations and subspecies.

HPLC analyses show that spots contain different pigments from the rest of the petal

While the pigment composition of entire C. gracilis petals had been previously determined through chromatography (Soltis, 1986), it was unclear whether the spots contained the same pigments as the background color. Using HPLC, we analyzed anthocyanidins present in mature petals from several C. gracilis individuals, spanning all spot phenotypes (see Materials and Methods). In all cases, irrespective of subspecies or spot phenotype, the distribution of anthocyanidins in the petal followed the same pattern. Petals of C. gracilis were found to include the anthocyanidins malvidin, cyanidin, and peonidin (Fig. 3, Table 2). These results contrast with a published report where delphinidin was found instead of peonidin (Soltis, 1986). The discrepancy between this study and Soltis (1986) may be due to the different methodologies used (HPLC versus thin-layer chromatography).

Fig. 3.

Anthocyanidins found in C. gracilis flowers. HPLC traces showing pigments in different sections of petals. Each panel (a–d) shows traces for sections of the petal (top, center, and base, in that order) for one of the four phenotypic classes: Basal-spotted (a); central-spotted (b); double-spotted (c), and; unspotted (d). Peaks corresponding to malvidin, cyanidin, and peonidin (indicated x, y, and z, respectively) were identified by comparison to known standards.

Table 2.

Pigment composition in C. gracilis petals with different spot phenotypes

		Phenotype
		Basal spot	Central spot	Central & Basal spot	No spot
Petal section	Top	0.171 (±0.050), 5	0.132 (±0.026), 3	0.205 (±0.102), 3	0.137 (±0.033), 2
	Center	0.227 (±0.078), 2	3.524 (±1.599), 5	2.717 (±0.179), 3	0.143 (±0.026), 2
	Base	4.55 (±1.020), 5	0.494*, 1	5.316 (±1.091), 3	0.355 (±0.115), 2

Numbers refer to the ratio of cyanidin+peonidin: malvidin (mean ± standard deviations), followed by number of individuals sampled. Amount of each pigment was inferred from the average peak height between 10.4–10.5 min. for cyanidin, 11.1–11.2 min. for malvidin, and 12.3–12.4 min. for peonidin.

^*All but one central-spotted individuals had spots that extended close to the petal base (on the abaxial, but not the adaxial surface), impeding dissection of spot-free tissue. Therefore, only one individual was used.

Cyanidin and peonidin were present in all petal sections that included spots and were never detected in petal sections that lacked spots (Fig. 3, Table 2). Likewise, petal sections that lacked spots, only having background color, contained only malvidin. The central regions of central-spotted flowers sometimes contained small amounts of malvidin, consistent with the presence of some background color in the central sector. In contrast, the basal regions of basal-spotted flowers contained very little or no malvidin, consistent with the almost complete lack of background color near the petal base (Fig. S1).

Given these findings, we conclude that, regardless of spot position, anthocyanins in C. gracilis petal spots are derived from cyanidin and peonidin, two similar anthocyanidins that differ in that peonidin has a methyl substitution on the B-ring. In contrast, the background anthocyanins present in the petal appears to be malvidin-based, an anthocyanidin that has two methyl substitutions on the B-ring, and derives from a different branch of the anthocyanin pathway (Fig. 2).

Clarkia gracilis carries multiple copies of anthocyanin biosynthetic genes

In the anthocyanin pathway the branch-point between cyanidin/peonidin and malvidin occurs after the F3H-mediated synthesis of DHK (Fig. 2). This points to the regulation or action of F3′h and/or F3′5′h as possible determinants of spot formation. Another possible candidate for the synthesis of different pigments is Dfr. Numerous studies show that single or very few amino acid changes in DFR can lead to changes in substrate specificity that redirect flux down a different branch of the pathway (Johnson et al., 2001; Fischer et al., 2003; Shimada et al., 2005; DesMarais and Rausher, 2008). Consequently, we focused on documenting whether differences in the patterns of regulation of F3′h, F3′5′h and Dfr are associated with the different petal spot phenotypes of C. gracilis. In addition, we also examined F3h and Ans, which are involved in earlier and later steps of pigment biosynthesis, respectively, to serve as controls and provide a more comprehensive understanding of patterns of expression of anthocyanin biosynthetic genes.

We conducted PCR amplification of partial gene sequences using multiple degenerate and specific primers, designed to conserved regions of the genes encoding the targeted enzymes, and cloned numerous PCR products in an attempt to identify all homologous gene copies in C. gracilis and, when possible, in related diploid species. We analyzed sequences for the presence of stop codons to assess gene functionality and used phylogenetic analyses to confirm gene identity (Fig. S5). However, it should be noted that we obtained only partial sequences for most genes from most species.

Two copies of F3h, F3′h, and F3′5′h and three copies of Dfr were found in diploid species in Sect. Rhodanthos, while a single copy of Ans was found. For F3′h and F3′5′h, only one copy appeared to be functional in the diploids: one copy of each gene had a premature stop codon or a frameshift mutation. For F3h, one copy from C. amoena and one from C. arcuata appeared to be divergent (as evidenced by longer branches, see Fig. S5). There were no indications from the sequences to suggest that any Dfr genes lack function (Fig. S5). We focused our efforts on isolating the putative functional or non-divergent copies of these genes from C. gracilis, and recovered one copy (two homeologs) corresponding to the putative functional F3h and F3′h, and all three copies (and 2 homeologs for each copy) of Dfr. We isolated both homeologs of the functional copy of F3′5′h, and one homeolog of the nonfunctional F3′5′h gene, which was confirmed to have an early stop codon. We isolated a single sequence corresponding to Ans, but is unclear whether only one homeolog exists, or whether both homeologs are present but do not differ in sequence within the short fragment obtained.

Spatial expression of Dfr2 shows a tight correlation to spot location in C. gracilis petals

If spot formation depends on location-specific expression of any of the anthocyanin biosynthetic genes, it follows that such gene(s) would be expressed mainly in areas where spots are formed. To examine spatial expression patterns, we performed RT-PCR on RNA extracted from three different petal sections (top, middle and base) from young buds, from plants with central (P^CP^C), basal (P^BP^B), or double-spotted (P^BP^C) flowers. Copy-specific primers were used for all of the putatively functional genes, with homeolog-specific primers being used in C. gracilis when that proved possible.

As shown in Fig. 4, the gene encoding the most upstream enzyme examined in the anthocyanin pathway, F3h, and one of the two copies of the gene encoding the next enzyme, F3′h1-A, showed consistent, robust expression in all sections of all petals. F3′h1-B transcript was not found in floral tissue, despite multiple attempts using several primers, suggesting it is either not expressed or limited to other tissues. Genes encoding enzymes further down the pathway, such as F3′5′h, Dfr1, and Ans (hereafter, ‘downstream genes’) showed variable expression that did not correlate with genotype or spot location. This variation most likely reflects slight differences in the age of petals. For instance, the most consistent pattern among the downstream genes is that their levels tend to be lower in central spotted petals than in either basal-spotted or double spotted petals. This is likely due to the age of the floral buds. Central spots appear earlier in development than basal spots, which means that central spotted flowers were harvested at a slightly earlier stage of development than the other genotypes (Fig. S3). This is significant because it suggests that, at this early stage of development, downstream genes are not fully and consistently activated.

Fig. 4.

Spatial expression of anthocyanin biosynthetic genes in C. gracilis petals. (a) Expression patterns for three individual plants for each of the P-locus phenotypes. The same nine individuals were used for all genes, and they are placed in the same position for each panel. Ovals represent petals, and filled circles represent spot locations. T, C, and B refer to top, central, and basal sections of the petals. F3h, flavanone 3 hydroxylase; F3′h, flavanoid 3′ hydroxylase; F3′5′h, flavonoid 3′-5′ hydroxylase; Dfr, dihydroflavonol-4-reductase; Ans, anthocyanidin synthase; (b) Expression patterns for Dfr2 in three spotted and three unspotted plants from a F2 population segregating for I.

The two homeologs of one gene, Dfr2, had a pattern that was strikingly distinct from all the other genes surveyed, a pattern that is correlated with the presence of the spot (Fig. 4a). Dfr2 appeared to be expressed at high levels in sections of the petal that contained a spot, but not in unspotted sectors. In central-spotted individuals Dfr2 was expressed at high levels only in the center of the petal, but was expressed only at the base of basal-spotted petals. Similarly, double-spotted individuals showed high expression of Dfr2 in both central and basal sections of the petal. The low levels of expression that were sometimes seen in sections adjacent to the spot can be attributed to imprecise sectioning of the buds, since they are only a few millimeters in length when dissected. The sole exception to this pattern is that one central spotted individual, showed a brighter band for Dfr2 in the top as opposed to the central section of the petal (Fig. 4a). This could be partly due to initial RNA quality, as the top section for that petal yielded a brighter band than the central sections for the other genes surveyed, including Actin. The spot-specific pattern of expression of Dfr2 in petals (and lack in unspotted petals) was confirmed in nine other C. gracilis plants grown from seed derived from wild populations (3 unspotted and 2 central spotted C. g. sonomensis, 3 basal spotted C. g. albicaulis, and 1 basal spotted C. g. tracyi, Fig. S6). Therefore we conclude that Dfr2 (and only this gene) shows a spot-specific expression pattern.

The role of Dfr2 in spot formation was further evaluated by examining the expression of this gene in individuals segregating for the I locus. In basal-spotted plants (I^PI^P or I^PI^A), Dfr2 is expressed at high levels in the basal section whereas unspotted siblings (I^AI^A ) show very low or no Dfr2 expression (Fig. 4b). This corroborates the pattern seen in F2s from the P locus cross, suggesting that Dfr2 is necessary for spot production. It also indicates that suppression of the basal spot by the I locus is genetically upstream of Dfr2.

Dfr2 deviates from other downstream genes in showing expression early in petal development

The variable expression of genes encoding downstream enzymes suggested the possibility that temporal regulation plays a role in petal spot development. This seems plausible since petal spots, especially central spots, are visible several days before the appearance of background color (Fig. S3), implying that genes required for spot development might be expressed earlier than genes that are only required for the production of background color. To investigate this possibility, we used qPCR to obtain an expression time-course for Dfr and for the functional copies of F3′h1, F3′5′h1, which are situated at the branch-point between cyanidin/peonidin and malvidin biosynthesis in developing buds. We assayed all three copies of Dfr and both homeologs for all genes except the B copy of F3′h1, which did not appear to be expressed in petals.

As shown in Fig. 5, two genes, F3′h1-A and Dfr2, are expressed early in development, having high expression levels at the first time points (as the spot appears) and declining expression in more mature buds (which are developing background color). Although F3′h1-A shows a similar pattern in all three phenotypes, Dfr2 expression is reduced or absent in flowers that lack spots, consistent with the results obtained from non-quantitative PCR (Fig. 4). In contrast to F3′h1-A and Dfr2, F3′5′h1 and Dfr1 were expressed later in petal development, having low expression levels in early time points, but higher expression later in development.

Fig. 5.

Quantitative analysis of gene expression of anthocyanin biosynthetic genes through C. gracilis petal development. Photos above the graphs show a representative petal for each of the four developmental stages defined for the purpose of this experiment. From L to R: basal, central, and unspotted. All buds are shown in the upright position (apex at the top, base at the bottom). From left to right, basal spotted plants: spot first appearing, spot well defined, background beginning to show, mature petal. Central spotted plants: no color, spot first appearing, spot well formed, background color beginning to appear. Unspotted plants: young colorless, older colorless, background first appearing, mature petal. Y-axes show expression of genes normalized to Actin. F3′h, flavanoid 3′ hydroxylase; F3′5′h, flavonoid 3′-5′ hydroxylase; Dfr, dihydroflavonol-4-reductase. Columns represent averages across three replicates (except last time point for central spotted, n=2), error bars indicate standard error.

Finally, one gene Dfr3 shows no clear pattern, being expressed at very low levels in all tissues except for a peak in late, central-spotted flowers. The latter could be an experimental artifact (note error bars). The success in amplifying Dfr3 in RT-PCR experiments confirms that this gene is expressed, but the level of expression appears to be substantially lower than the other Dfr gene copies, indicating that this gene may lack a function in pigment production in the petal.

Co-segregation analysis suggests that the P and I loci correspond to trans-regulators of Dfr2

Qualitative and quantitative gene expression data indicate that cyanidin/peonidin-based petal spots in C. gracilis correlate with the early and localized expression of Dfr2, coupled with the early expression of F3′h1-A. Under the assumption that localized, early expression of Dfr2 is the proximate cause of petal spot development, then the P and I loci must act by modulating the expression of Dfr2, and likely correspond to upstream regulators of Dfr2. However, it is possible that either P or I could correspond to Dfr2 itself if it had undergone a cis-regulatory change that resulted in altered expression.

To verify whether Dfr2 alleles co-segregate with spot location and/or presence, we genotyped several F2 individuals from both the P and I crosses, for both homeologs of Dfr2 (Table 3). The genotypes at neither of the Dfr2 homeologs co-segregated perfectly with spot location in the P cross, suggesting that the P locus does not encode Dfr2, but rather is a trans-regulator of Dfr2. However, the p-value for co-segregation between Dfr2-B and P approached significance (P = 0.0843), suggesting that perhaps there is some genetic linkage between P and Dfr2-B. Larger numbers of offspring would be necessary to confirm this hypothesis. Similarly, neither of the Dfr2 homeologs co-segregated with the absence/presence of basal spots in F2s from the I locus cross, suggesting that I is also a trans regulator of Dfr2.

Table 3.

Analysis of co-segregation of alleles at the DFR2-A and DFR2-B loci with phenotypes obtained from the I and P crosses

Homeolog	Genotype	I Cross			P Cross
		Phenotype		P	Phenotype			P
		S	U		B	D	C
Dfr2-A	A1A1	11	2	0.591	3	5	1	0.9314
	A1A2	19	5		5	8	5
	A2A2	12	1		2	2	1
Dfr2-B	B1B1	10	4	0.2365	3	1	0	0.0843
	B1B2	18	3		0	6	2
	B2B2	14	1		6	8	5

Different alleles of Dfr2-A/B were identified using SNPs and indels in introns 2 or 3.

Abbreviations: S = spotted; U = unspotted; B = basal-spotted; D = double-spotted; C = central spotted. P are from Fisher’s exact test. Dfr, dihydroflavonol-4-reductase.

Discussion

A model for the development of pigmentation patterns in C. gracilis petals

Based on the results presented here we propose a model for petal pigment development in C. gracilis that can explain the production of cyanidin/peonidin-based anthocyanins in spots and malvidin-based anthocyanins in the background (Fig. 6). This model invokes transcriptional regulation of anthocyanin biosynthetic genes as a function of time and space. Although post-translational regulation may occur, it is not necessary to explain the observed data. For this model, we assume that Dfr3 is expressed at such low levels that it plays no significant role in petal pigmentation. Likewise we will ignore the putatively non-functional copies of F3′h and F3′5′h.

Fig. 6.

Model of pigment pattern development in C. gracilis flowers. Early pigment production in spots is explained by F3′h expression throughout the petal and Dfr2 expression in spots, which leads to dihydroquercetin (DHQ) production throughout the petal and cyanidin/peonidin production in spots only. Later in development expression of F3′5′h and Dfr1 converts DHQ (and any residual dihydrokaempferol (DHK)) in the petal background to dihydromyricetin (DHM) and then malvidin. F3′h, flavanoid 3′ hydroxylase; F3′5′h, flavonoid 3′-5′ hydroxylase; Dfr, dihydroflavonol-4-reductase.

According to the model (Fig. 6), most of the core enzyme-coding genes of the anthocyanin pathway (Chs, Chi, F3h, F3′h, Ans, and Uf3gt) are activated early in development and expressed throughout petal tissue. In addition, Dfr2 is activated early but is expressed only in the area destined to become a pigmented spot, implying that it is under different regulatory control. By contrast, neither F3′5′h1 nor Dfr1 are expressed at this time. Because Dfr is required for pigment production, the spatial restriction of Dfr2 explains why pigment is produced only as a spot. Further, because F3′5′h1 is not activated early, only anthocyanins derived from dihydroquercetin (DHQ) (i.e., cyanidin and peonidin) are produced in the spot domain.

At a later stage of development, the core enzymes are still active. This may represent continuous expression of these genes, or a second wave of expression. At this point, F3′h1-A and Dfr2 expression has ceased. Activation of both F3′5′h1 and Dfr1 throughout the developing petal means that the complete set of enzymes is present, causing the background pigment to be deposited. The expression of F3′5′h1 causes this pigment to be derived via DHM rather than DHK or DHQ, explaining the production of malvidin in the background. Thus, the difference in pigment type between background and spot, as well as the spatial restriction of the spot, is explained by the different temporal expression of F3′h1-A/Dfr2 versus F3′5′h1/Dfr1.

Regulation of enzyme-coding genes

In most species that have been examined, the core enzyme genes of the anthocyanin pathway (those depicted in Fig. 2) are coordinately regulated by common sets of transcription factors (Cone et al., 1986; Paz-Ares et al., 1986, 1987; Ludwig et al., 1989; Goodrich et al., 1992; de Vetten et al., 1997; Elomaa et al., 1998; Quattrocchio et al., 1998, 1999; Walker, 1999; Borevitz et al., 2000; Spelt et al., 2000, 2002; Carey et al., 2004; Schwinn et al., 2006; Gonzalez et al., 2008; Lin-Wang et al., 2010; Niu et al., 2010; Albert et al., 2011). In some taxa, such as Ipomoea and maize, all core genes are expressed at similar times in development and are activated by a single set of transcription factors (Cone et al., 1986; Paz-Ares et al., 1987; Ludwig et al., 1989; Morita et al., 2006). In other species, such as Petunia and Antirrhinum, genes are activated in two blocks, with those coding for upstream enzymes activated first, and those coding for downstream enzymes activated later, perhaps by a different set of transcription factors (Martin et al., 1991; Goodrich et al., 1992; Jackson et al., 1992; Quattrocchio et al., 1998; Spelt et al., 2000; Schwinn et al., 2006).

The pattern of regulation exhibited by C. gracilis differs from these canonical patterns. Although we have not demonstrated it directly, the production of anthocyanins in the petal implies that upstream and downstream genes, represented by F3h and Ans, are expressed from the time anthocyanins are first produced in spots to the time they are last produced in the remainder of the petal—although the existence of sub-blocks within these genes cannot be ruled out. However, Dfr, F3′h, and F3′5′h exhibit more restricted temporal distributions, with Dfr2 and F3′h1-A being expressed only in the early stages of development and Dfr1 and F3′5′h1 being expressed only later in development. This pattern suggests that factors regulating these genes differ from those regulating the remaining core enzymes, and that the factors regulating Dfr2 and F3′h1-A differ from those regulating Dfr1 and F3′5′h1. Moreover, two patterns indicate that Dfr2 is regulated differently from all other enzyme coding genes. First, only Dfr2 exhibits a spatially restricted expression domain, confined to the regions that develop spots. Second, variation at the P and I loci affect only expression of Dfr2. Because neither of these loci cosegregates with Dfr2, they likely correspond to transcription factors or cofactors that are unique to this gene. These patterns imply that the production of spots has required a substantial reworking of the anthocyanin regulatory network. We next discuss the evolutionary implications of this reworking.

Possible substrate specificity in different copies of DFR

DFR2 and DFR1 appear to only encounter one substrate, DHQ or DHM, respectively, and, while the model we propose for formation of cyanidin/peonidin-based spots does not require evolution of substrate specificity among the different copies of DFR, it is possible that it has occurred. In C. gracilis, the two different copies differ in numerous sites, including in the active site, where they differ by nine amino acid substitutions. One such substitution, at residue 133 (Fig. S7), has been implicated in substrate specificity in other species (Shimada et al., 2005; DesMarais and Rausher, 2008; Johnson et al., 2001). Possible evolution of substrate specificity could be explored by biochemical characterization of these enzymes combined with more fine-scale temporal expression studies.

Evolutionary implications

Although the progenitors of C. gracilis have petal spots, as do other species in the section Rhodanthos, our analysis of spot formation in C. gracilis sheds light on aspects of how petal spots originally evolved in this group. Two major types of regulatory change appear to have been required for the evolution of spots. The first was a spatial restriction of the expression domain of Dfr2. It is reasonable to assume that pigment deposition throughout the petal was the ancestral state because species that are situated in the phylogeny as sister to the rest of the genus, namely C. breweri, C. concinna, and C. pulchella, are all pink and unspotted. Although it is unclear when the duplication of Dfr occurred, it is likely that initially both paralogs exhibited the same broad spatial expression pattern. Restriction of Dfr2 expression to the areas of spot formation subsequently occurred. Because Dfr2 seems no longer to be activated by the anthocyanin transcription factors that activate the other core anthocyanin enzymes, we suspect its cis-regulatory region evolved to respond to a new set of activators that are expressed at the time and position at which spot formation initiates. The gene products of the P and I loci are candidates for these new activators, since they influence expression of Dfr2 but not expression of the other core enzyme-coding genes.

The second major type of regulatory change necessary for the evolution of cyanidin-based spots separated the timing of expression of Dfr2 and F3′h on the one hand from Dfr1 and F3′5′h on the other. By itself, divergence in spatial expression domains of the two Dfr paralogs might produce a spot of increased pigment intensity, which could explain the existence of Clarkia species that produce spots but only malvidin in their flowers (Soltis, 1986), but cannot account for cyanidin/peonidin spots on a malvidin background, as seen in C. gracilis. Our analyses suggest that temporal separation of the expression of F3′h and F3′5′h is largely responsible for the production of cyanidin/peonidin-based anthocyanin in spots.

In most plant species that have been examined, both F3′h and F3′5′h are active simultaneously and result in the production of malvidin/delphinidin derivatives (Gerats et al., 1982; Tornielli et al. 2009, Hopkins & Rausher 2011). We therefore presume this was the pattern in ancestors of C. gracilis that lacked spots (only malvidin is produced in the unspotted flowers of C. breweri and C. concinna, Soltis, 1986). Cyanidin, peonidin and their derivatives can only be produced if F3′5′h activity is greatly reduced or eliminated. C. gracilis has apparently achieved this by delaying the expression of F3′5′h so that, at the time of spot development, only F3′h is expressed. By contrast, at the time of background pigment development, only F3′5′h is expressed, allowing the accumulation of malvidin derivatives. Finally, restriction of the expression of Dfr1 to the late stage of development was required to prevent cyanidin- and peonidin-based anthocyanins from being synthesized in the petal background.

Our analysis of the genetic control of spot pattern formation has revealed that the gene Dfr2 acts as a switch for spot production. Activating this gene completes the anthocyanin pathway and allows pigment deposition in spots. The evolution of this switch mechanism appears to have been facilitated by duplication of the ancestral Dfr gene, which allowed one paralog to evolve to serve as a spot regulator. Based on the broad occurrence of petal spots in Clarkia, we infer that the Dfr gene duplication occurred early in the radiation of the genus. Subsequent evolutionary remodeling of the regulatory network for both copies of Dfr, as well as for F3′h and F3′5′h, resulted in the novel ability to finely regulate the spatial pattern of contrasting colors in the petals of many Clarkia species. Further work in this group, particularly temporal and spatial expression assays of anthocyanin pathway genes in petals from species throughout the genus will help to illuminate the order of events that led to the evolution of petal spot.