Plant material.

A biparental mapping population with 66 F₁ individuals was developed from crosses made during 2012 and 2013 and subjected to no prior selection before use in the current study. The female parent, S. meyeri ‘Palibin’ (10-0209), acquired from Blue Heron Farms Nursery (Corvallis, OR), was pollinated with S. pubescens ‘Penda’ Bloomerang^® Purple plants from Garland Nursery (Corvallis, OR). Plants were grown in containers from 2013 to 2017 and field-planted at the Lewis Brown Horticulture Farm (Corvallis, OR).

Phenotypic segregation.

Phenotyping of remontancy was conducted during Summer 2017 and Summer 2018, when plants were 4 and 5 years old. The phenotype for initial blooming was recorded in mid-May of each year to confirm that each plant had reached reproductive maturity and prevent phenotyping a plant as nonremontant when it may still be juvenile. Phenotyping of remontancy was repeated three times for each year between mid-July and mid-Aug. 2017 and 2018. Levels of remontancy were also recorded. Plants without any flowers were recorded as nonremontant. Plants with one to three inflorescences were recorded as semi-remontant. Plants with more than three inflorescences were recorded as remontant. Several segregation ratios were examined by chi-square (χ²) tests.

Genotype and marker-trait association analysis.

A total of 20,730 SNP markers were selected and filtered (Lattier, 2017). SNPs were identified using the TASSEL-GBS discovery software pipeline (Glaubitz et al., 2014). Principal component analysis (PCA) and marker–trait association analysis were performed using GAPIT (Lipka et al., 2012). Different PCA depths from three to eight and several statistical models, including Multivariate Linear Models (MLM), Generalized Linear Model (GLM), Regular and Compression MLM (CMLM), and enriched CMLM (ECMLM) were used for marker-trait association analysis.

Phenotype and segregation.

Of the 66 F₁ plants, only 57 had reached sufficient size to be phenotyped in 2017. In 2018, all 66 F₁ plants were phenotyped: 27 were remontant, 13 were semi-remontant, and 26 were nonremontant (Fig. 1A). Phenotypes from the 2 years showed a good consistency (χ² = 0.56, df = 2, P = 0.65). Of the 57 plants phenotyped in both years, 38 had the same phenotype. Although 19 plants (33.3%) showed different phenotypes between 2017 and 2018, 13 shifted to increased remontancy, including six plants moving from semi-remontant to remontant, four plants moving from nonremontant to semi-remontant, and three plants moving from nonremontant to remontant. Six plants exhibited a reduction in remontancy between years, including two plants that moved from remontant to semi-remontant and four plants that moved from semi-remontant to nonremontant. No plants categorized as remontant in 2017 shifted to nonremontant in 2018. Considering that many plants may not have reached sufficient maturity to express their true remontancy phenotype, only phenotype data from 2018 were used for marker-trait-association analyses.

View Full Size

Marker–trait association among a total of 66 hybrid plants of Syringa meyeri ‘Palibin’ × S. pubescens ‘Penda’ Bloomerang^® phenotyped for reblooming (remontancy) with 20,730 single-nucleotide polymorphism (SNP) markers used in the GAPIT analysis (Lipka et al., 2012). Results from phenotyping and analysis include the following: (A) genotype distributions of remontancy, nonremontant (NR), semi-remontant (SR), and remontant (R), among the population; (B) principal component analysis (PCA) eigenvalues from 1 to 10; (C) quantile-quantile (QQ) plot analysis; and (D) the method of arbitrary functions (MAF) plot.

Citation: Journal of the American Society for Horticultural Science J. Amer. Soc. Hort. Sci. 145, 2; 10.21273/JASHS04847-20

• Download Figure
• Download figure as PowerPoint slide

All possible segregation ratios including single- and two-gene models were tested, including two possible two-gene segregation models. The first two-gene model had an expected segregation ratio of 6:4:6 for nonremontancy, semi-remontancy, and remontancy (χ² = 5.44, df = 2, P = 0.76). Another possible segregation ratio, 3:5 for nonremontancy and the sum of semi-remontancy and remontancy (χ² = 0.05, df = 1, P = 0.82), was also a two-locus segregation model that aggregates semi-remontant and remontant phenotypes together as one “remontancy” category. For possible single-gene segregation ratios of remontant semi-remontant and nonremontant, the probability values of χ² goodness-of-fit tests of 1:1 and 1:3 were 0.082 and 0.007, respectively. The results showed that the phenotype fit two-gene models better than single-gene models based on χ² goodness-of-fit tests; however, at this juncture, it is not possible to indicate if this is due to dominance or recombination. Furthermore, with our small population size it is not possible to completely rule out a single-gene model.

Marker-trait association.

After trimming the genotype and phenotype datasets, 20,730 markers for 66 plants were included in the genome-wide association study (GWAS) analysis. In the marker-trait association analysis, results of the analysis with different PCAs were very similar; the same markers were discovered with slightly different probability values. Because of their similarity, only a PCA with a result of 8, which might obtain the most information, was reported (Fig. 1B). Results of different statistical models, including MLM, GLM, CMLM, and ECMLM, were similar; therefore, only the results of MLM are presented. We further used a quantile-quantile (QQ) plot and method of arbitrary functions (MAF) plot to examine the MLM model fitness for three levels of the trait of interest (remontant, semi-remontant, and nonremontant). The top one to three markers may be associated with remontancy (Fig. 1C and D). Without using marker location information, GAPIT found two SNP markers, TP70580 and TP82604, associated with remontancy with P < 0.0001. TP70580 was a T/C SNP marker, and both parents were heterozygous T/C. Genotypes of C/C, T/C, and T/T of TP70580 were found in the F₁ population. TP82604 was a T/A SNP marker. The genotype of the remontant parent, S. pubescens Bloomerang^®, was T/T and the genotype of the nonremontant parent was T/A. Genotypes of T/A and T/T of TP82604 were found in the F₁ population. For convenience, we expressed TP82604 and TP70580 as an AB two-loci model. T/A and T/T alleles of TP82604 are presented as Aa and aa, and C/C, T/C, and T/T of TP70580 alleles are presented as BB, Bb, and bb in this study. Genotypes of the remontant parent S. pubescens Bloomerang^® and the nonremontant parent S. meyeri ‘Palibin’ were Aabb and AaBb, respectively. The genotypes of seedlings were AABb, AaBb, AAbb, aaBb, Aabb, and aabb (Fig. 2).

View Full Size

Genotypes and phenotypes among a total of 66 plants resulting from the cross of Syringa meyeri ‘Palibin’ × S. pubescens ‘Penda’ Bloomerang^® included in the study without prior selection. Values are the percentages of plant phenotypes (with an original number) in each genotype. The order of the number from top to bottom represents plants with phenotypes of remontant, semi-remontant, and nonremontant. Genotypes Aa and aa represent T/A and T/T at single-nucleotide polymorphism (SNP) marker TP82604; BB, Bb, and bb represent C/C, T/C, and T/T at SNP marker TP70580 marker. To illustrate the phenotype percentage trend among genotypes, genome types aaBb and AaBb are shown twice in the figure.

Citation: Journal of the American Society for Horticultural Science J. Amer. Soc. Hort. Sci. 145, 2; 10.21273/JASHS04847-20

• Download Figure
• Download figure as PowerPoint slide

Genetic model of remontancy in lilacs.

By observing the phenotypes of each genotype, both A and B loci affect remontancy by epistasic and additive interaction. The A locus (TPTP82604) is critical to remontancy and shows a semi-dominant or dominant effect. Of recessive homozygous aa plants, only 17% were nonremontant and 83% were remontant. For plants with the Aa genotype, 79% of plants were nonremontant and 21% of plants were remontant. The relative impact on levels of remontancy at the A locus is aa > Aa. Similar results were observed with TP70580, with a comparatively smaller effect on remontancy. For plants with the genotype BB, 73% were nonremontant and 27% were semi-remontant. No plant with genotype BB was remontant. For plants with the Bb genotype, 42%, 29%, and 29% of plants were nonremontant, semi-remontant, and remontant, respectively. In contrast, for plants with the bb genotype, individually, 17%, 10%, and 72% of plants were nonremontant, semi-remontant, and remontant, respectively.

The results showed that the genotypes of the two loci had a cumulative effect on remontancy. None of the plants with genotype AABb was remontant. Furthermore, 91% of plants with the genotype aabb were either remontant or semi-remontant. For other genotypes, including AAbb, AaBb, Aabb, and aaBb, the frequency of remontant, semi-remontant, and nonremontant plants varied, but the frequency of remontant and semi-remontant plants increased as more recessive alleles were present at the two loci (Fig. 2).

Our results indicated that there are two SNPs associated with remontancy segregating in our population of lilacs; however, the precise mode of genetic control is still unknown. Additional research with larger populations and/or the use of reference genome sequences will help build a fuller understanding of the genetic control of remontancy in lilacs. In this study, the phenotype segregation ratio fits in the two-gene model better than single-gene models, and the character of remontancy trait was more like a quantitative trait; however, the two discovered markers were linked (χ² = 31.88, df = 5, P < 0.0001). Therefore, even though the two markers found by this study were significantly associated with remontancy, we cannot confirm whether this was due to dominance or recombination.

Two markers suitable for selection of remontant lilacs.

The two SNP markers, TP70580 (A locus) and TP82604 (B locus), were strongly associated with remontancy and could be used for MAS. Simultaneously using these two recommended markers could efficiently predict remontancy (Fig. 2). The additive effect of the two markers on remontancy was observed such that as recessive alleles at the two loci accumulate, the more likely the phenotype is remontant. In our population, 91% of plants with the genotype aabb were either remontant (87%) or semi-remontant (4%). Of the two candidate loci, the A locus showed a stronger effect on remontancy. If one locus was homozygous recessive and the other was heterozygous, then the relative impact of the two loci were as follows: 84% of plants with genotype aaBb were remontant; however, only 50% of plants with Aabb genotype were remontant.

In addition to the accumulated effect of the two loci, some level of epistasis between the two loci might be present by the following mechanism: recessive homozygous aa determines whether a plant exhibits any level of remontancy (on or off), and the b allele of the semi-dominant B locus determines the level of remontancy (semi-remontant or remontant). Comparing the phenotype proportion of aaBb and aabb supported the presence of epistasis. High percentages of both aaBb (84%) and aabb (91%) plants were semi-remontant or remontant; however, 87% of aabb plants were remontant and only 42% of aaBb plants were remontant.

Using a three-level phenotype and two-loci model effectively explained remontancy in our population. Selection targeting the greatest expression of remontancy is a desired strategy because of the improved stability of the trait in these plants. Nursery producers have reported instability or unimpressive flower production during the second flush in some remontant cultivars, likely due, at least in part, to environmental influence or production practices. Plants that have stable, consistent, and high rates of remontancy are valued. Misleading low levels of remontancy may lead to false-positive phenotyping. Low levels of remontancy were observed in dwarf lilac cultivars generally regarded as nonremontant according to nursery producers (personal observation). Some inconsistent phenotypes between 2017 and 2018 were observed, including eight plants that shifted between semi-remontant and nonremontant phenotypes.

The double recessive aabb is the best genotype for selecting remontant phenotypes and we suggest simultaneously using the two markers for the best results. Although the remontancy level prediction could be affected by a variety of causes, including the environment, other possible minor genes, and distance of the genes to discovered markers, 87% of the aabb plants were remontant. If selection is made by only one marker, then many semi-remontant plants will be falsely selected. For example, using SNP marker TP70580 exclusively would lead to selection of 71% aa plants exhibiting some level of remontancy, but more than 50% of those were semi-remontant. When selecting plants with valuable remontancy, the breeder should only select those plants with a remontant type that is stable.

GBS for ornamental plant breeding.

Like many other minor crops, the lack of a reference genome and the shortage of resources limit breeding strategies for woody ornamental plants. In the present research, heterozygosity and interspecific hybridization that are often challenging to molecular marker development were present in our population. However, GBS successfully identified high-density genetic markers for MAS without relying on a reference genome. The complexity and lack of a reference genome are common in most clonally propagated ornamental plants. In addition, because of the long lifecycle or long juvenile phase, MAS for early-stage selection could reduce costs and increase efficiency. Our research demonstrated the potential practicality of GBS for breeding projects of such ornamental plants. Two SNP markers from GBS were found to relate to remontancy in the dwarf lilac population and appear sufficient for early-stage selection.

Further research is needed to make the two SNP markers practical. Theoretically, the two-marker screening method for remontancy could be used in a larger population from the same crossing combination or, potentially, in populations with parents with a similar genotype and phenotype. However, sequencing all plant materials can be expensive and impractical. In addition, the same alleles may not be present. Therefore, converting the SNP marker to a polymerase chain reaction (PCR)-based marker is necessary to make these two markers useful for MAS. There are two possible strategies: finding SSR markers near the SNPs or directly designing SNP primers. Without a reference genome, finding an SSR marker in close proximity to the SNP marker is difficult. However, the raw sequence data might be used in a draft genome assembly. With a draft reference genome sequence, discovering genes of the sequence near the SNPs is not impossible using GBS alone. However, directly designing primers with PCR products containing the SNP site by using sequence walking might be another viable method. With this method, only short sequences near the SNPs are required to design the primers. Another method to convert the discovered SNPs into PCR-based markers is the use of a high-resolution melting (HRM) marker to directly detect the genotype of the SNP by designing primers around it (Mehta et al., 2013). If an HRM is not available, then designing a pair of primers that have a 5′-end on the SNP marker to generate a new PCR-based dominant marker is another option. However, scoring a dominant marker result can be problematic because homozygous recessive and failed PCR would both yield the same apparent results, leading to false-negative results.

Remontancy in perennial plants.

In the present study, two recessive loci were identified as selection targets for remontancy in dwarf lilacs. This result implied that at least two genes (or loci) control dwarf lilac remontancy; however, the genes that control lilac remontancy remain unknown. Studies of the mechanism of remontancy in other perennial plants have found two models of genetic control: 1) Terminal Flower 1 (TFL1) deficiency (Freiman et al., 2012; Kotoda et al., 2006; Wang et al., 2012; Zhu et al., 2015), and 2) a multigene model in which vernalization, gibberellic acid (GA), and photoperiod response genes all might be involved (Jia et al., 2014; Zhou et al., 2013). TFL1 maintains juvenility and prevents floral initiation in Arabidopsis thaliana (Hanano and Goto, 2011). Reduced TFL1 function is related to remontancy or continuous flowering in several perennial plants like apple [Malus ×domestica (Kotoda et al., 2006)], pear [Pyrus communis (Freiman et al., 2012)], rose [Rosa (Iwata et al., 2012; Wang et al., 2012)], strawberry [Fragaria ×ananassa (Iwata et al., 2012)], and daylily [Hemerocallis (Zhu et al., 2015)]. Multigene regulation may regulate remontancy of tree peonies [Paeonia sp. (Zhou et al., 2013)], and a tropical fruit species, Dimocarpus longan (Jia et al., 2014).

The multiple gene model best explains dwarf lilac remontancy. The results of our research indicated that two major loci with epistasis and additive interactions contributed to the remontancy of the hybrid dwarf lilac population. Without a reference genome, predicting or identifying the exact gene(s) that result in remontancy of hybrid dwarf lilacs is difficult. The two loci could be associated with either two homologous TFL1 genes or two other genes in a multiple-gene control model. However, when considering prior findings for other perennial plants, we believe that remontancy in lilac is controlled by multiple genes. First, reports of rose, strawberry (Iwata et al., 2012; Wang et al., 2012), and daylily (Zhu et al., 2015) indicated that only one TFL1 homolog affects remontancy but others did not. Second, remontancy in dwarf lilac appears to be quantitative instead of qualitative and interactions of the two loci were observed. In the TFL1 deficiency model, remontancy was a simple qualitative trait. In our case, the two loci showed different impacts on remontancy and the QQ plot analysis implied that more minor genes might exist. These phenomena indicated there might be other unknown factors involved in remontancy rather than a single-gene model. We believe the multiple genetic control model explains remontancy in dwarf lilacs better than a single-gene model. However, further research and a reference genome are needed to fully reveal the genetic control of remontancy in lilac.