**3. Discussion**

We investigated the distribution of color polymorphic individuals in three rock populations of *P. limprichtii* within Huanglong District and combined chemical detection and transcriptomic analysis to isolate the main pigment compounds and candidate genes that determine flower color intensity. We present a putative biosynthesis pathway and discuss the regulatory mechanisms of color formation.

To explore color variation formation factors, we used the CIELAB evaluation system to distinguish rose-purple, pink, and white flower color, and then counted the number of individuals of each of these phenotypes in each rock population. In all three populations the color distribution pattern was nearly consistent, with a color ratio of rose-purple 6:pink 3:white 1. This distribution pattern is very rare in natural color polymorphic populations. Studies have shown that intraspecific flower color variation is often attributed to genetic drift, pollination-mediated selection, environmental conditions, or herbivory [13,29–31]. In our experiments, the populations grow across a small range, with the whole population occurring on similar rocks and exposed to the same climatic conditions. Environmental elements such as temperature, drought stress, and ultraviolet radiation were therefore not considered to be crucial promoters of color variation. Pollinator-mediated selection plays an important role in color variation, especially for deceptive pollination species in which competition for pollinators in sympatry promotes flower color divergence [32], and shifts in pollinators also contribute to the macro-evolution of flowers color [22]. Thus, we inferred that the flower color polymorphism within these populations might have been induced by pollinators. Color polymorphism may be a consequence of pollination competition or specific adaptations to pollinators, and pollinator behavior exerts strong selection stress on color variations. Some research also showed adaptive selection for pigmented flowers because colored flowers are less likely to be disrupted by herbivories than colorless ones [3]. According to our field observations, we found that white-flowered individuals were more susceptible to damage than individuals with pigmented flowers, and that white flower petals and cores were severely foraged when blooming. We thus inferred that the dominant pigmented color was beneficial to

avoid herbivory, and reduce damage by herbivores to the population. When individuals were damaged they also suffered reduced attractiveness to pollinators, which is not conductive to the stability and development of the population. This explains the distribution pattern of the number of rose-purple flowers in the population. There is another view that flower color may not be the primary goal of natural selection, nor the initial choice of pollinators. Indeed, the biosynthetic precursors of pigments not only display color variations, but also serve other physiological functions [14]. Studies have shown that secondary metabolites associated with plant defense functions share the same biosynthetic pathway, the flavonoid synthesis pathway, correlating pigment with defense ability [33]. Therefore, colorful-flowered individuals were more resistant to some adversities. In summary, the phenomenon that rose-purple flowers were frequent and white ones rare within the population may be mediated by the pollinators and herbivores, and also related to survival adaptability of the *P. limprichtii*. The 6:3:1 distribution pattern of color polymorphism might be a reproductive strategy for the population to maintain the maximum population density, but further evidences should be investigated.

In theory, the transition from pigmented to white flowers could involve any mutations that block one or more steps in the anthocyanin pathway. This includes loss-of-function mutations in any pathway enzyme-coding genes, as well as the cis-regulatory genes that influence any of the pathway enzymes [34,35]. In our study, expression analysis identified several obvious differentially expressed genes in the petal which were down-regulated in white samples compared to pigmented samples, but metabolite detection found that Cy- and Del-derivatives existed in both white petals and pigmented petals, indicating that the color variations, especially the white petals, do not lacking any steps of the anthocyanin pathway. The cis-regulation of transcription factors is a crucial element to promote color divergence. This result is similar with the white color formation in *Primula vulgaris* which is caused by different genes expression pattern rather than loss- of-function mutations leading to the lack of anthocyanin [36].

For ABP-related genes, we isolated 21 transcripts which encode ten enzymes. Seven of the ten were flavonoid synthase genes, including *PlCHS*, *PlCHI*, *PlF3H*, *PlF3 H*, *PlF3 5 H*, *PlDFR*, and *PlANS*; one was a proanthocyanidin synthase gene, *PlANR*; one was a flavonol synthase gene, *PlFLS*; and one was an anthocyanin synthase gene, *PlUFGT*. Most of them are multi-gene families, only *PlF3H*, *PlF3 5 H*, *PlFLS*, and *PlANR* are single copy. To clearly illustrate the catalyzation steps of the ABP, we regarded Cy-related and Del-related biosynthesis processes as independent branches [8]. Thus we did not have to measure the content of each anthocyanin compound. Each branch of anthocyanin synthase was considered to make an equivalent contribution to pigmentation. Considering that both the Cy-related and Del-related branches may share the majority of enzymes, here, we selected ABP-related genes on the Cy-related branches to analyze their expression patterns.

The expression analysis of early step structural genes revealed a high level of *PlCHS* and *PlCHI* expression in white petals and a low level in pigmented petals, suggesting that white petals can produce a large amount of naringenin but cannot eventually flux this into anthocyanin synthase. Meanwhile, the expression levels of *PlF3 H*, *PlF3 5 H*, *PlDFR*, *PlANS*, and *PlUFGT* in pigmented flower petals was high, compared with white ones, and we inferred that downstream structural genes make a large contribution to coloration. Analysis also showed that *PlFLS* was significantly more upregulated in white petals than in pigmented petals. *FLS* encoding enzymes lead substrate into the flavones and flavonols pathway [37]. It has been suggested that the competition between the anthocyanin synthesis pathway and the flavone and flavonols pathways mainly results in substrate competition between *FLS* with *DFR*, while the *FLS* enzyme strengthens the metabolic flux toward the flavonols and limits anthocyanin accumulation [38]. This situation has also been reported in other species, such as in *Paeonia ostii*, a higher expression of *PoFLS4* in the nearly white flowers promotes dihydroflavonols transition into flavonols [11]. In onions, enhanced *AcFLS* could maximize flavonol production in the sheath [39]. Finally, in *Muscari armeniacun*, the conversion of substrate between *FLS* and *DFR* facilitates the elimination of blue pigmentation [8]. Thus, we confirmed *PlFLS* as one of the candidate genes for white color formation in *P. limprichtii*. The up-regulation of *PlDFR* in pigmented flowers is closely

accompanied by a decrease of *PlFLS*; hence, more dihydroflavonols flow into anthocyanin production in pigmented flowers. From our analysis, the expression patterns of *PlANS* and *PlUFGT* are correlated with color intensity, and they showed their highest expression levels in the rose-purple flowers and their lowest in the white flowers. The ANS (Anthocyanidin synthase) encoding enzyme catalyzes the conversion of colorless leucocyanidin into colored anthocyanin [40], and anthocyanin is further glycosylated by different UFGT (UDP flavonoid glucosyl transferase) encoding enzymes that convert the anthocyanidins to different anthocyanin derivatives, exhibiting the final color [41]. We therefore speculate that *PlANS* and *PlUFGT* are two crucial genes that determine color intensity in *P. limprichtii*.

It has been revealed that the MBW protein complex, a combination of R2R3-MYB and bHLH transcription factors, along with WD40 proteins, play an important role in regulating the transcription of structural genes [41–45]. The activities of R2R3-MYB factors have distinct roles in determining the action of the complexes, either to promote or inhibit the transcription of anthocyanin biosynthesis genes [46,47]. By combining R2R3-MYB phylogenetic and co-expression network analyses, we isolated *PlMYB10*, which was homologous with the *AtMYB11*, *AtMYB12*, and *AtMYB111* belonging to S7 in *Arabidopsis* and that have been demonstrated to contribute to the regulation of genes that account for anthocyanin accumulation in all tissues [27,28]. The expression pattern was high expression in nearly white flower petals, gradually reducing in flowers with increasing color intensity. It also corresponded with the color polymorphism phenotypes. In other species, overexpression of *AmMYB330*, a negative regulator of the flavonoid biosynthesis, has been proven to inhibit phenylpropanoid metabolism in transgenic tobacco (*Nicotiana tabacum*) plants [48]. The co-expression network showed that *PlbHLH20* and *PlbHLH26*, along with *PlWD40-1* have a strong relationship to *PlMYB10*. It is likely that this potential MBW complex *PlMYB10*/*PlbHLH20*/*PlWD40-1* or *PlMYB10*/*PlbHLH26*/*PlWD40-1* may serve as a repressor responsible for the variation in color intensity in *P. limprichtii*. Co-expression network revealed *PlFLS* is the most likely target gene interacting with *PlMYB10.* Previous studies have verified that R2R3-MYB can regulate the expression pattern of *FLS* through the overexpression of *PsMYB114L* (from *Paeonia su*ff*ruticosa*) in *Arabidopsis* [49], which is consistent with our results. Thus, we tentatively speculate about the ABP of *P. limprichtii* (Figure 9). Further studies examining sequencing differences in these candidate genes and *PlMYB10* are necessary to assess our speculations. Metabolic substance quantification is also necessary to confirm the leading pigments in *P. limprichtii.*

**Figure 9.** Tentative pathways for *P. limprichtii* color variations. The colored bar is the value of log2 (RPKM + 1), represented using the depth of color, with purple representing the up-regulated expression genes and pink representing the down-regulated expression genes. RPKM means the reads per kb per million reads mapped. CHS, chalcone synthase; CHI, chalcone-flavanone isomerase; F3H, flavanone-3-hydroxylase; F3'H, flavonoid 3'-hydroxylase; F3'5'H, flavonoid 3'5'-hydroxyla; DFR, dihydroflavonols 4-reductase; ANS, anthocyanidin synthase; UFGT, UDP flavonoid glucosyl transferase; FLS, flavonol synthase; The three gene complex consist of a MYB, bHLH and WD40 in most angiosperm.
