ddRADseq Applications for Petunia × hybrida Clonal Line Breeding: Genotyping and Variant Identification for Target-Specific Assays

Abstract

Molecular genotyping is a key factor for plant breeding programming and plant variety protection (PVP). However, its potential still remains to be elucidated when considering ornamental plants like Petunia × hybrida. In this study, a petunia breeding clone collection, including sister line groups, was genotyped through double digest Restriction-site Associated DNA sequencing (ddRADseq), and its genetic diversity and structure were studied. In addition to estimating the high genetic similarity observed among sister lines, this approach allowed the unique discrimination of each clone too. Molecular results agreed with genealogy data, supporting the assessment of genotyping effectiveness. In addition, the minimal number of variants able to uniquely discriminate and/or correctly cluster the experimental lines was investigated. The loci number could be reduced to eight to achieve line discrimination, and a method to identify the specific variant sets is presented. Conversely, to preserve the original clustering with minor adjustments, one hundred loci were required and were obtained through minor allele frequency (MAF) filtering. Moreover, analysis of the chromosomal distribution of variants revealed a predominant accumulation in distal regions. Genetic analyses were repeated considering only variants located in coding sequences and results were in agreement with what previously observed, disclosing the potential of the expressed regions for genotyping purposes. Eventually, the applied approach enabled the investigation of SNPs within genes putatively involved in traits of interest. Our findings encourage the adoption of high-throughput and cost-effective sequencing techniques for petunia genotyping aimed at achieving PVP, supporting new variety registration, and developing marker-assisted breeding (MAB) and marker-assisted selection (MAS) strategies.

1. Introduction

Petunia (Petunia × hybrida hort. ex E. Vilm.; synonym: Petunia × atkinsiana (Sweet) D. Don ex W. H. Baxter) is an annual herb belonging to the Solanaceae family, originating in South America [1], and it is well known worldwide for its ornamental use. A wholesale value of almost USD 160 million was reported in the US in 2020 for petunia ornamental varieties [2,3]. Commercial lines are derived from interspecific hybrids artificially obtained in the eighteenth century [4,5] between P. axillaris (Lamarck) Britton, Sterns and Poggemburg, and one or more species belonging the clade of P. integrifolia (Hooker) Schinz and Thellung [6,7]. Natural hybrids between the two species were never observed, in part due to different pollination syndromes. In particular, P. axillaris is pollinated by hawk moths, while P. integrifolia by bees [8,9]. Nevertheless, hybrids derived from artificial crosses can produce fertile progenies and commercial lines comprehend both seed-propagated F₁ hybrids and cutting-propagated clonal lines [1]. Although varieties are more often diploid, tetraploids have been observed and selected in breeding programs [10,11]. Petunia (2n = 2x = 14) is a facultative allogamous species, in which gametophytic self-incompatibility [12,13] and cytoplasmic male sterility [14,15,16] were extensively studied. It is considered a model plant [17,18], especially for the study of flower development and biochemistry [4], as well as for tissue culture practices [1]. In addition, the application of genetic engineering techniques has led to one of the first and most well-known cases of genetically modified petunia, the orange-flowered A1-DFR lines [19]. It is also worth noting that the RNA interference mechanism was discovered for the first time in this species [20,21]. Finally, molecular resources developed for petunia comprehend high-density linkage maps with different marker types [22,23,24] and the draft genome assembly of two of its main species P. axillaris and P. integrifolia [25]. Recently, a chromosome-scale genome assembly and annotation of P. × hybrida was developed by combining short- and long-read sequencing, optical mapping and Hi-C technology [26].

In recent decades, breeding goals for petunia have mainly included aesthetic–sensorial aspects, particularly in relation to flower morphology and phenology [2,27,28]. More recently, breeding has also dealt with new emerging needs, such as ecosystem service provision, adaptability and sustainability [29,30]. The evaluation of these traits through the exclusive use of morphological and phenological measurements is often impossible. Nevertheless, the official protocols for variety registration aimed at assessing distinctness, uniformity and stability (DUS), such as those of the European Community Plant Variety Office (CPVO) and International Union for the Protection of New Varieties of Plants (UPOV) [31,32], are still exclusively based on this type of descriptor. In petunia, 34 and 32 descriptors, including plant growth habit, leaf shape and flower density, are provided respectively, by EU and US authorities for DUS tests [33,34]. Since official registration is necessary to achieve plant variety protection (PVP) and to guarantee plant breeders’ rights (PBRs) [35], incorporating molecular data into the registered variety profiles, in conjunction with specific phenotypic measurements, could provide significant benefits. For example, a clear enhancement in a sustainability-related trait (e.g., tolerance to low nutrient levels or reduced phytosanitary inputs) assessed through appropriate phenotypic evaluations (e.g., value for cultivation and use (VCU) tests), may not be reflected by visible morphological changes. Hence, traditional DUS tests may be insufficient for a successful evaluation. In this case, DUS tests could be carried out combining phenotypic measurements and genetic diversity analyses based on variety molecular profiles (value molecular DUS (vmDUS)) [36]. However, the possibility to exploit molecular genotyping for official variety registration is still limited to few specific functional markers, or it is used only to facilitate traditional DUS test procedures [37]. Molecular genotyping can also be useful for directly helping PVP, thus facilitating the identification of plant materials of unknown origin, essentially derived varieties (EDVs) and commercial frauds [35], as well as for Marker-Assisted Breeding (MAB) and Selection (MAS) [38,39,40].

Double digest Restriction-site Associated DNA sequencing (ddRADseq) is a Reduced Representation Sequencing (RRS) technology exploiting the gDNA digestion with two restriction enzymes, in order to reduce the fraction of the genome to be sequenced [41]. ddRADseq is reported as an approach able to ensure reliable genotyping results and relatively low cost and time consumption. Therefore, it can be particularly useful for agricultural species lacking economic convenience in performing deeper and hence more expensive analyses [42,43,44], like the majority of ornamental plants. The technique was successfully exploited in several plant species for different purposes, such as structure evaluation of wild populations [45,46,47], germplasm conservation [48,49,50], Quantitative Trait Locus (QTL) identification [51,52], Genome-Wide Association Study (GWAS) [53], phylogenesis [54], Marker-Assisted Selection (MAS) and Marker-Assisted Breeding (MAB) [38,39,40]. Regarding ornamentals, ddRADseq have been used for genotyping in several species, like Bougainvillea spp., lantana, orchid and primula [52,55,56,57,58,59,60]. The thousands of variants generated by ddRAD may be reduced down to an informative core (SNP core set), which, in turn, can be used for the development of target-specific assays, like Fluidigm, Kompetitive Allele-specific PCR (KASP) or AmpliSeq assays [61,62,63,64,65,66]. This can lead to a reduction in costs and procedure times for future analyses, as well as to an easier comparison of data from different populations and crop cycles. Furthermore, this may help the detection of specific regions of interest to be targeted for functional studies and MAS scheme development [66,67,68]. However, in ornamentals, very few works have selected SNP core sets and created target-specific assays [69,70,71]. Understanding how and to what extent molecular profiles should be filtered to produce an effective SNP core set can be challenging, especially when dealing with breeding populations characterized by highly related lines with reduced genetic and phenotypic variability.

In this study, a genotyping analysis using a ddRADseq approach was performed in a petunia breeding population of clonal lines consisting of sister line groups. Assuming a high level of intragroup genetic uniformity, the potential of this genotyping technique in uniquely discriminating the clones and in providing genetic information for the population was assessed. The aim was to evaluate the suitability of ddRADseq as a starting point to develop procedures for variety registration, protection and creation through MAB strategies. In addition, we determined to what extent and in which way the polymorphism number could be reduced while remaining informative, thereby identifying an SNP core set, without altering sample clustering. Eventually, the possibility of exploiting only expressed regions for genotyping analysis by selecting variants exclusively located in Coding DNA Sequences (CDSs) was also evaluated.

2. Materials and Methods

2.1. Plant Materials

Leaf samples from 36 petunia clonal lines were granted by Gruppo Padana S.S. (Paese, TV, Italy), a nursery company specializing in the production, selection and collection of young plants from seeds and cuttings. The analyzed clones included 32 breeding lines (BLs) and 4 benchmarks (BMs). BLs were divided into 4 groups of sister lines, i.e., clones sharing at least one of the two parental lines. BMs consisted of already developed varieties available on the market and exploited to evaluate the distinctness of BLs. Since the population consisted of clones, a single replicate per line was exploited. Genomic DNA (gDNA) was extracted from 100 mg of leaf tissue through DNeasy Plant mini kit (Qiagen, Valencia, CA, USA), following manufacturer instructions. Genomic DNA quality and concentration were assessed using NanoDrop 2000c UV-Vis spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) and agarose gel electrophoresis with 1% agarose/1× TAE gel and 1× SybrSafe DNA stain (Life Technologies, Carlsbad, CA, USA).

2.2. ddRADseq Library Preparation and Sequencing

Double digestion Restriction-site Associated DNA sequencing (ddRADseq) libraries were produced following the protocol developed by Abed et al. [72], with some modifications. In detail, extracted gDNA was normalized to 20 ng/µL and a total of 200 ng per sample was used. PstI and MspI (Thermo Fisher Scientific) restriction enzymes were used, with 1 µL of each per reaction, for DNA digestion at 37 °C for 2 h, following the producer protocol. Adapter ligation was performed in 50 µL of reaction including 30 µL of digested DNA, 1 µL of T4 DNA ligase, 5 µL of 10× T4 ligation buffer (New England BioLabs, Ipswich, MA, USA), 5 µL of barcode adapters (0.1 µM) and common adapters (10 µM). Ligation was carried out with an incubation at room temperature for 2 h, followed by 20 min at 65 °C. Ligated DNA samples were grouped into two pools, each including 18 samples, and purified through QIAquick PCR and Gel Cleanup Kit (Qiagen) for the elimination of impurities, non-ligated adapters, small fragments and dimers. Size selection from 100 bp to 600 bp was performed with the CleanNGS magnetic beads (CleanNA, Waddinxveen, The Netherlands). Amplification of the obtained libraries was performed with Platinum SuperFi PCR Master MIX (Thermo Fisher Scientific), following the enrichment PCR protocol described by Abed et al. [72]. Amplified libraries were subjected to another step of purification with CleanNGS magnetic beads. Subsequently, they were quantified using the TapeStation System (Agilent Technologies, Santa Clara, CA, USA) with High Sensitivity D1000 DNA ScreenTape assays (Agilent Technologies) and diluted to 200 pM for the Ion Torrent sequencing. Libraries were loaded into Ion Chips 550 (Thermo Fisher Scientific) through the Ion Chef (Thermo Fisher Scientific) automated machine. Single-end sequencing was performed in two runs, one per pool, with the Ion GeneStudio S5 Plus System (Thermo Fisher Scientific) set to 200 bp of read length and 500 flows per run.

2.3. ddRADseq Data Processing

Raw reads were quality-checked by filtering out those with a Phred Quality score lower than 20, and then demultiplexed according to the adopted barcode recognition motif, through the Torrent Suite platform (Thermo Fisher Scientific) following the native implemented pipeline. After these steps, reads were mapped on the reference P. × hybrida v.01 genome assembly retrieved from the NCBI database (GCA_046563445.1) [26]. Mapping was performed exploiting the MEM algorithm of Burrows-Wheeler Aligner (BWA) v. 0.7.12 [73], followed by data conversion into BAM files through SAMtools and BCFtools v.1.13 software [74]. For the variant calling the gstacks and populations commands of Stacks v2.61 software [75] were used. gstacks was launched exploiting the ‘marukihigh’ model, with both a SNP discovery (--var-alpha) and a genotype call (--gt-alpha) with alpha thresholds of 0.10, while populations was run in order to keep loci present in at least 30% of individuals in a subpopulation (-r 0.30) and 65% of individuals across subpopulations (-R 0.65). Obtained SNPs were filtered through the poppr v2.9.6 package [76] for RStudio software v.4.2.3 (Posit, Boston, MA, USA), in order to retain only the loci present in all samples and with a minimal Minor Allele Frequency (MAF) of 0.1. The resulting SNPs composed the initial dataset for the genetic analyses.

2.4. Genetic Distance, Structure and Statistic Analyses

All analyses and graphic processing were carried out through RStudio, unless specified otherwise. Genetic Distance (GD) matrixes were produced through the Nei’s coefficient [77] considering all pairwise comparisons among sample SNP profiles and were made through poppr package’s nei.dist function. A Kinship-based INference for Genome-wide association studies (KING) matrix based on pairwise comparisons was calculated with the KING-robust method exploiting the PLINK2 [78] --make-king function [79,80]. Furthermore, 0.4 was chosen as a minimal threshold to hypothesize the same clone membership for a pair. The Unweighted Pair Group Method with Arithmetic Mean (UPGMA) dendrograms for sample clustering were created based on Nei’s coefficient with the aboot function from the poppr package, considering 1000 bootstrap replicates. Clustering visualization was carried out through dendrograms created with the plot.phylo function from the ape package [81]. Principal Component Analysis (PCA) based on the GD analysis was performed using the gl.pcoa function from the dartR package v.2.9.9.5 [82] Structure v.2.2 [83] software was used for Bayesian clustering to infer the sample genetic structure. The following parameters were set: founding groups ranging from 1 to 10, 10 replicate simulations per each K, burn-in of 20,000, final run of 100,000 Markov Chain Monte Carlo (MCMC) steps. The K values characterized by the highest likelihood were selected using Structure Harvester software v. 0.6.94 [84]. Calculated genetic statistics included: observed (Ho), expected (Hs) and overall (Ht) Heterozygosity; inbreeding coefficient within subpopulations (Fis) and within the total population (Fit); subpopulation fixation index (Fst); gene flow (Nm); average number of observed (na) and expected (ne) alleles per locus; Private Alleles (PA) and Polymorphic Loci (PL) percentage, average locus Polymorphic Information Content (AvgPIC) and MAF. Statistics were measured either with the hierfstat package [85] basic.stat function, the dartR package loc.metrics function, or with a manually developed R script. Analysis of Molecular Variance (AMOVA) between and within subpopulations was performed using the ade4 method with the poppr.amova function and the significance of the results was validated through a Monte-Carlo test considering 999 permutations.

2.5. SNP Subset Selection for Core Set and Location in CDSs

Further filtering of the initial dataset for the SNP core set identification was performed by setting alternative combinations of MAF, AvgPIC and Hs. The filtering continued until the obtained dataset led to a BL clustering consistently different from the one obtained using the entire SNP set. The genotype accumulation curve was created through the genotype_curve function of the poppr R package v.2.9.8, setting a 10 k times loci resampling. The definition of the minimal set of SNPs able to discriminate all 36 genotypes was obtained with the subsequent method: a custom R function randomly orders all the SNPs and then, starting from the first, selects the minimum set of SNPs able to discriminate all 36 samples, storing it in memory. The function reiterates the same process, and each time, if the identified set is smaller than the stored one, it substitutes it. A second custom R function parallelizes the process and independently repeats it to obtain different minimum discriminating SNP sets. In this analysis 100 minimum discriminating SNP sets were generated, each resulting from 100,000 random iterations. The custom code written for this publication is publicly available [86]. In parallel, Coding DNA Sequences (CDSs) were retrieved from the reference genome annotation and were intersected to the initial SNP dataset using the bedtools intersect command of the Bedtools v.2.30.0 software [84], allowing to retain and use variants exclusively present in the coding sequences for clustering analyses. To visually represent them and allow the comparison of their distribution with the gene density, a genome map was created using Circos software v.0.69-10 (Michael Smith Genome Sciences Centre, Vancouver, BC, Canada) [87]. Protein sequences relative to the CDSs harboring SNPs were aligned by BLASTx v.2.12.0 (NCBI, Bethesda, MD, USA) against the P. axillaris (v1.6.2) [25] and Arabidopsis thaliana (TAIR10.1) [88] proteomes. Hits were retained using a minimum coverage threshold of 65% and an E-value cutoff of 1 × 10⁻⁵. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways relative to the resulting putative orthologs were then retrieved and a Gene Orthology (GO) enrichment through ShinyGO (v0.85.1) [89] was performed.

3. Results

3.1. Genetic Distance and Structure Analysis

A ddRADSeq analysis was performed on genotype 36 petunia BLs, comprehending four sister line groups, and four benchmarks (BMs). After quality filtering and mapping, more than 80 M reads with an average of 2.2 M reads per sample were obtained. This value corresponded to the 85.7% of the 12.3 Gb sequenced in total. The mean demultiplexed and trimmed read length resulted was 173 bp (

Table 1. Sequencing results, after raw read quality filtering and mapping, of ddRADseq libraries relative to the 32 petunia clonal lines–divided in four subpopulations (BL1-BL4)–and 4 distinct BenchMarks lines (BM1-BM4). Averages are weighted for the number of lines per subpopulation (N). Mn bases: average number of bases (in millions) sequenced per sample; ≥Q20 Mn Bases: average number of bases (in millions) sequenced per sample with a Phred quality score ≥ Q20; Mn Reads: average millions of reads per sample, MRL: Mean Read Length.

Pop-ID	N	Mn Bases	≥Q20 Mn Bases	Mn Reads	MRL (bp)
PeHy-BL1	4	486.46	416.44	3.34	160
PeHy-BL2	7	360.03	304.84	2.39	173
PeHy-BL3	9	316.36	271.88	1.99	177
PeHy-BL4	12	267.85	230.08	1.68	178
PeHy-BM1	1	258.86	226.06	1.25	206
PeHy-BM2	1	104.60	91.09	0.50	208
PeHy-BM3	1	469.56	399.95	3.31	141
PeHy-BM4	1	551.17	468.70	3.91	141
Tot	36	12,291.01	10,528.17	80.01	–
Avg		341.42	292.45	2.22	173

; Supplementary Table S1).

Variant calling resulted in the identification of 42,236 SNPs located in a total of 15,655 loci. SNP filtering allowed the creation of a final dataset of 4349 variant sites in the 36 analyzed samples.

A GD analysis among samples through the Nei’s coefficient resulted in no pairwise comparison with complete genetic identity, and unique molecular profiles were produced for all the considered BLs. GD among samples ranged from 0.001 (PeHy-BL3-06 vs. PeHy-BL3-08) to 0.51 (PeHy-BL3-09 vs. PeHy-BM03), with an average of 0.23 ± 0.11 (Supplementary Table S2). Considering sister line groups as a whole, the lowest GD value, 0.07, was found comparing PeHy-BL1 and PeHy-BL2 subpopulations, while the highest, 0.29, was observed in the comparison between PeHy-BL2 and PeHy-BL3 (

Table 2. Genetic Distance (GD) with Nei’s coefficient and fixation index (st) matrix based on a pairwise comparison among four sister line groups, comprehending a total of 32 clonal lines.

GD				Cluster	Fst
				PeHy-BL2
			0.07	PeHy-BL1	0.23
		0.11	0.10	PeHy-BL4	0.23	0.22
	0.19	0.27	0.29	PeHy-BL3	0.50	0.45	0.34
PeHy-BL3	PeHy-BL4	PeHy-BL1	PeHy-BL2		PeHy-BL2	PeHy-BL1	PeHy-BL4	PeHy-BL3

).

The kinship matrix created with the KING-robust method led to an average value of 0.46 with a SD of 0.52. Three pairwise comparisons exceeded the 0.4 value, used as a threshold for the identification of individuals belonging to the same clone [79,80]: PeHy-BL3-01 vs. PeHy-BL3-07; PeHy-BL3-06 vs. PeHy-BL3-08; PeHy-BL4-03 vs. PeHy-BL4-12 (Supplementary Table S3).

In the UPGMA dendrogram based on the GD analysis, three principal branches were observed: one grouping exclusively PeHy-BL3 lines, a second with PeHy-BL4 lines, and the last comprehending PeHy-BL1, PeHy-BL2 and the four BMs. All bootstrap values at the bases of the branches reached the maximum value (100). Within the last cluster, PeHy-BM3 was highly separated from the rest of the group, while a branch grouped together the other three BMs and two further branches hosted respectively PeHy-BL1 and PeHy-BL2 lines. All the relative nodes were characterized by very high bootstrap values (Figure 1A).

The genetic structure analysis showed the most likely K value, namely the most probable number of common ancestors, to be K = 3 (Delta K = 10,407.3), followed by K = 2 (Delta K = 3068.5) (Supplementary Figure S1). In both cases, each BL was predominantly associated with a specific ancestral population, with the exception of PeHy-BL3-04 and PeHy-BL3-02, which had an admixed result. PeHy-BM3 and the PeHy-BL4-07–PeHy-BL4-11 group were also found to be admixed, but only for one of the two K values (Figure 1B). When assuming K = 2, PeHy-BL3 was the only group assigned to one of the two inferred ancestral populations, while all other BLs and the four BMs exhibited a shared ancestry. For K = 3, in addition to the separate clustering of the PeHy-BL3 group, a further division was observed between the PeHy-BL4 group and the remaining lines (namely PeHy-BL1, PeHy-BL2 and the four benchmark populations) (Figure 1B). Results were in agreement with what was observed with the UPGMA clustering, and therefore with the genealogy data. The PCA based on GD confirmed the findings observed in the structure analysis and the UPGMA dendrogram. The first two components alone explained 50.3% of the total variability. PeHy-BL3 and PeHy-BL4 clustered apart, while the two other sister line groups and the benchmarks fell into the same group and were found to be poorly distinguishable (Figure 2).

3.2. Genetic Statistics and AMOVA

The sample pool was studied considering several genetic parameters. Average Ho per BL resulted in 26.0 ± 8.5%, with 1.26 na and 1.15 ne. The highest Ho values considering sister line groups were observed for PeHy-BL4 lines (33.6 ± 6.3%), while the lowest values were scored for PeHy-BL1 lines (19.1 ± 3.4%) and PeHy-BM4 (16.3%). Average Fis was negative and comprehended between −0.2 and −0.3, underlining an overall moderate heterozygosity excess for BLs in relation to what would be expected for the respective subpopulations (

Table 3. Genetic statistics calculated from ddRADseq molecular profiles relative to the 4 petunia sister line groups (PeHy-BL1-BL4), with number of individuals (N); average number of observed (na) and expected (ne) alleles per locus; observed (Ho%) and expected (Hs%) Heterozygosity; inbreeding coefficient (Fis); percentage of Polymorphic Loci (PL%) of the total loci in the population; percentage of Private Alleles (PA%) of the total PAs in the population.

Pop ID	N	na	ne	Ho (%)	Hs (%)	Fis	PL (%)	PA (%)
PeHy-BL1	4	1.36	1.27	19.05	15.02	−0.27	63.75	6.38
PeHy-BL2	7	1.54	1.31	20.48	16.14	−0.27	68.54	14.89
PeHy-BL3	9	1.54	1.46	26.27	22.08	−0.19	73.14	33.96
PeHy-BL4	12	1.66	1.55	33.57	26.17	−0.28	81.88	12.75
Avg	8	1.53	1.53	24.84	19.85	−0.25	71.83	17.00
StDev	3	0.11	0.12	5.72	4.53	0.04	6.69	10.28

; Supplementary Table S2).

Fst ranged from 0.22 to 0.50, suggesting a high genetic fixation for sister line groups (Table 2). As a result of these two factors, a slightly positive inbreeding coefficient for BLs in comparison to the expected considering the entire plant collection was observed (Fit: 0.02), in addition to a limited gene flow (Nm: 0.05). The percentage of PA per sample, relative to the total number of PA in the population ranged from 0.4% for PeHy-BL4-12 to 20.1% for PeHyBM3. The other three BM lines also exhibited percentages of Private Alleles considerably higher than their BL counterparts, ranging from 7.33% to 19.49% (Supplementary Table S2). The PA percentages considering each BL as a whole ranged from 6.4% for PeHy-BL1 to 34.0% for PeHy-BL3. A minor deviation was noted considering PL percentage of the total population loci: from 63.8% (PeHy-BL1) to 81.9% (PeHy-BL4) (Table 3). The AMOVA carried out comparing molecular variance between and within subpopulations (sister line groups and BMs) demonstrated good reliability, showing a p-value of 0.001. The analysis revealed that a substantially larger proportion of the total observed variability was attributable to differences among subpopulations, rather than among samples within the same subpopulation. Respectively, the estimated variance quotes were 58.2% and 41.8% (

Table 4. Analysis of Molecular Variance (AMOVA) results calculated from ddRADseq molecular profiles relative to 32 petunia clonal Lines (in 4 BLs) and 4 BMs. Degrees of freedom (Df); Sum of Squares (Sum Sq); Mean of Squares (Mean Sq) and Estimated Variance (Est. Var.) are reported.

	Df	Sum Sq	Mean Sq	Est. Var.	Est. Var. (%)
Between Subpopulations	7	16,100	2300	490	58.19
Within Subpopulations	28	9855	352	352	41.81
Total	35	25,955	742	842	100

).

3.3. GD Analysis Comparison Between Total and Core Set SNP Profiles

The 4349 SNPs used for the GD analysis (Figure 3A) were filtered multiple times by acting on MAF, PIC and Hs values, to find the lowest number of variants (SNP core set) able to correctly cluster the BLs.

The minimum SNP amount by which a unique molecular profile was conserved for all BLs with this approach was 1051, which was obtained retaining only those variants with a MAF greater than 0.3. The UPGMA clustering and the bootstrap values remained unchanged for the majority of the previously highlighted groups (Figure 3B). Only the hierarchical position of the PeHy-BL1 cluster in relation to the PeHy-BL2 group and to the one including PeHy-BM1, PeHy-BM2 and PeHy-BM4, changed from the initial results (i.e., those obtained considering the whole SNP set) (Figure 3A). By increasing the MAF threshold to 0.45, the number of SNPs drops sharply to 103 with an average PIC of 0.3. Based on these settings, PeHy-BL3-06 and PeHy-BL3-08 presented complete genetic identity and minor clustering changes were observed (Supplementary Table S4, Figure 3C). The main difference found in the UPGMA grouping (compared to that obtained using the complete dataset) was a shift in the hierarchical positions of the PeHy-BM3 and PeHy-BL4 clusters. The bootstrap value at the base of the branch separating PeHy-BM3 from the rest of the accessions remained high (82.5), while the other basal nodes showed values lower than 50 (Figure 3C). Beyond the MAF threshold of 0.45, significant grouping distortions occurred. The same occurred when stringent filters related to PIC and Hs were applied to the original dataset (Figure 3D). A second analysis was performed, aiming to quantify and identify the minimal set of SNPs able to discriminate all 36 lines. A genotype accumulation curve was created with the initial dataset to determine the minimal variant number able to discriminate all BLs. 8 polymorphic loci met the target criterion, whereas the curve reached a plateau at approximately 30 loci (Supplementary Figure S2). An extensive search for the minimal discriminating SNP sets led to the identification of 5 sets composed by 8 SNPs. The pairwise GD analysis performed with each set did not show in any case complete genetic identity (Supplementary Table S5).

3.4. GD Analysis Comparison Between Total and CDS-Located SNP Profiles

The initial SNP set was intersected with the CDSs annotated in the reference genome in order to retain only those variants that were located within coding sequences. Genotyping and clustering the clones using exclusively these variants allowed to assess possible differences with the previous results derived from the same analyses. 1975 SNPs were retained, corresponding to 45.4% of the total dataset. Their positions resulted uniformly distributed across the 7 chromosomes. Within each chromosome, distal areas presented higher variant site density in comparison to pericentromeric zones, reflecting gene distribution (Figure 4A).

Pairwise GD comparisons led to values very similar to those observed considering the initial dataset (Avg: 0.22 ± 0.10) (Supplementary Table S6). The Ho, Hs and Fis values also remained largely unchanged. The UPGMA dendrogram led to a completely identical clustering to that obtained with the original dataset. Bootstrap values were slightly lower in the main nodes, but continued to be significant (>80) (Supplementary Figure S3). The PCA explained 50.4% of the total genetic variation and subpopulation differentiation matched that obtained in the previous results. In particular, PeHy-BL3 and PeHy-BL4 groups were strongly separated, while PeHy-BL1, PeHy-BL2 and BMs samples resulted undistinguishable (Supplementary Figure S4). From the GO analysis of the genes whose CDSs were found to contain SNPs, 8 subcategories resulted enriched in the “Biological process” category, including the protein refolding (GO:0042026), the transmembrane transport (GO:0055085) and the response to hormone (GO:0009725) processes (Figure 4B), whereas 19 subcategories were enriched in the “Molecular function” category, including nucleic acid binding (GO:0003676), structural molecule activity (GO:0005198) and protein kinase activity (GO:0004672) (Figure 4C). Following the KEGG investigation, 390 genes harboring SNPs fell into at least one of the pathways. Of these, 50 ortholog genes resulted involved in secondary metabolism, 6 in flavonoid biosynthesis, 5 in terpenoid production, 4 in phenylpropanoid production, 18 in cofactor biosynthesis, 48 in plant hormone signal transduction (Supplementary Table S7).

4. Discussion

Molecular genotyping exploiting different types of molecular markers proved to be very useful for breeding plan setting and management in several ways. It can help retrieving genetic information about populations, such as genetic diversity and Heterozygosity, planning crosses, reducing inbreeding depression, exploiting heterosis, and favoring traits of interest’s selection [39,90,91,92,93]. In addition, the production of molecular profiles through genomic marker analysis can support PVP and variety registration, simplify DUS tests, help germplasm management, and aid the identification of GMOs, unknown plant materials and EDVs [35,37,50,68,94]. In petunia, several genotyping studies have been conducted using different types of molecular markers, including RAPDs [95], AFLPs [96,97], SSRs [8,98], SRAPs [99] and CAPSs [8,100]. GBS was used for genetic structure studies on petunia natural populations, including the parental species used for the production of the commercial interspecific hybrids, P. axillaris and P. integrifolia [101,102]. SNP genotyping was also successfully carried out on breeding collections for linkage map production and QTL analysis of flower-related traits, such as flowering timing, pollinator syndrome, and double flower development. Informative variants were retrieved from transcriptome data [7,103], targeted-sequencing techniques [2,22,104] and other GBS-based protocols [101,102,105,106]. By contrast, the use of molecular assays for the genotyping of experimental lines for breeding purposes, variety registration and variety protection, remains an underexplored area in petunia.

The present study focused on molecular genotyping through ddRADseq of 36 petunia BLs and BMs. The genotyping analysis proved to be effective and useful for PVP-related purposes, assigning a unique molecular profile to all lines of interest and discriminating each genotype from the others. In addition, it allowed us to estimate the GD values within the collection and other genetic parameters that are of interest for breeding plan management. BLs were subdivided into four sister line groups. As expected, sister lines with a relatively high homozygosity level were also characterized by a high genetic uniformity. Some lines (i.e., PeHy-BL3-01 vs. PeHy-BL3-07; PeHy-BL3-08 vs. PeHy-BL3-06; PeHy-BL4-03 vs. PeHy-BL4-12) turned out to be distinguished by a very limited polymorphism amount (respective GD: 1.7 × 10⁻²; 1.2 × 10⁻³; 0.29 × 10⁻³) suggesting the possibility that they might represent the same clone. In this case, it is crucial to determine standard thresholds of molecular similarity above which two accessions should be regarded as belonging to the same clonal lineage [107]. Some studies in different plant species, e.g., Populus tremuloides, Piper spp. and Callistemon kenmorrisonii, developed methods that exploited molecular diversity calculated with different markers for this purpose [108,109,110]. Thresholds were determined by estimating sequencing error through replicate sampling and therefore the results were population- and marker- specific. However, the possibility of using several thousands of SNPs for genotyping might have minimized the sequencing error impact, allowing them to normalize analyses and hence define standard values for clone identification. The exploitation of a kinship coefficient, based on the probability calculation that a pair of randomly sampled alleles is identical by descent (IBD) (i.e., due to genetic inheritance rather than by chance) [111,112] may help achieve this goal. Indeed, kinship coefficients return value near 0.5 in case of clones in plants—or of homozygous twins when considering animals—while values between 0.25 and 0.5 identify full siblings. Due to sequencing errors or somatic mutations, a kinship coefficient obtained between actual clones tends to be slightly lower than 0.5; therefore 0.4 was suggested as an effective threshold [113]. Up to now, kinship matrixes in plant breeding studies have been mainly exploited to identify the most genetically different genotypes in order to plan crosses [114] or to carry out the correction for relativeness in GWASs [115]. KING-robust [79,80] was described in a study comparing several kinship estimation methods in six non-model Australian species, as the best framework in case of high population structure and inbreeding, setting a maximum locus missingness below 10% and a minimal minor allele frequency threshold of 10% [116]. Exploiting this method, all three very genetically similar BL couples, and only these, presented kinship coefficients higher than the threshold value of 0.4. However, the approach has to be tested in other petunia populations exploiting preliminarily assessed clones. Nevertheless, the production of unique genotyping profiles for all lines demonstrated the approach’s capacity for discriminating between highly related genotypes. The SNP ability to uniquely genotyping very similar lines in petunia was previously shown by GBS studies for QTL analysis exploiting Recombinant Inbred Lines (RILs) and Near Isogenic Lines (NILs) [2,22,104]. However, to our knowledge this is the first study to exploit and review a ddRADseq method for petunia genotyping. Conversely, the methodology was used in several other species of ornamental interest, including Bougainvillea spp., Persian buttercup, Ceiba spp., flowering dogwood, lantana, lavender, orchid and primula, for MAB-oriented [56,58,117], ecology of wild populations [55,59] or trait association [52,57,60] purposes. These studies are consistent with our results, providing thousands of high-quality SNPs and confirming the efficiency of ddRADseq for genotyping.

The UPGMA clustering, the PCA and the genetic structure analysis led to consistent results and confirmed the genetic uniformity between sister lines, grouping them in the same branch, area of the plot, or ancestral group, respectively. The link with the genealogical data supported the reliability of the genotyping analysis. On the contrary, variability among sister line groups was significant. This was suggested by the AMOVA (variability between subpopulations > 50%), the fixation index calculation (Fst > 0.25 overall and in every pairwise comparison between sister line groups) and the genetic structure. However, despite the PCA explaining a percentage of total variability > 50%, the PeHy-BL1, PeHy-BL2 and BMs groups’ results were undistinguishable and were assigned to the same putative ancestor group. Therefore, although showing the highest presence of PAs, BMs were not entirely differentiable at the genetic level from all the other lines. This implies that, in cases where the breeding aim is to obtain new varieties that are highly genetically distinguishable from those already present in the market, PeHy-BL1 and PeHy-BL2 lines may not match this goal. The average GD of these groups with the three more-similar BMs were, resulted respectively, 0.13 and 0.14. To determine GD thresholds which allow us to assess whether experimental materials are genetically different from already developed varieties is a key factor for breeding and variety registration. This is particularly important in cases of experimental materials presenting phenotypic distinctiveness for only a single descriptor, in order to assess whether they may have been directly derived from other varieties (defined initial varieties, IVs). In this case, new varieties should be considered EDVs, assigning royalties to the IV breeder [35,118,119]. It is not possible to define standardized thresholds and formulas that discriminate between EDVs, as their determination also depends on the proportion of total genetic variability inherent to each species. Nevertheless, the definition of species-specific values, together with the development of general operational guidelines for their calculation—to be integrated with phenotypic assessments—is of paramount importance for variety registration and PBRs. To date, competent authorities such as UPOV and CPVO did not establish mandatory species-specific thresholds and standard procedures, but have suggested voluntary methods with case studies in maize and French bean for general variety registration [34]. In addition, they endorsed the EDV thresholds determined by the International Seed Federation (ISF) for several species, i.e., perennial ryegrass, oilseed rape, cotton, lettuce and maize [120,121]. In these cases, thresholds were alternatively set following alternatively three principles, defined, respectively, as follows: “calibration”, (i.e., based on similarities of pairs of genotypes known to have close genetic relationships), “tail” (i.e., exploiting a given percentile of the distribution of similarities among independent genotypes) and “pedigree” (i.e., defining the parental contribution expected) principle [119,122]. In particular, these approaches led to the introduction of a threshold of 0.96 in lettuce based on the Jaccard genetic similarity coefficient, determined with AFLPs [120,122]. In maize with a set of 3072 SNPs, the values 0.91 and 0.95 of Roger’s genetic diversity were set to delimit the three cases of definite, potential and no EDV [123,124]. A recent study on pear set the same thresholds at, respectively, 90% and 96% of the genetic similarity with the simple matching (SM) coefficient found by analyzing samples with 310 multiple nucleotide polymorphisms (MNPs) [125]. However, up to now no study has investigated EDV molecular identification in petunia.

The collection under study showed a relatively low average level of Ho, especially if we consider its origin (breeding population of clones with a consistent structure) and its mode of reproduction (facultative allogamous species). This finding would suggest that the occurrence of crosses between closely related lines within the population may be possible, and hence that the population would be at least partially self-compatible [126,127,128]. Nevertheless, the inbreeding coefficient within subpopulations was generally low, highlighting a breeding selection effort to increase heterozygosis. The fact that the BL inbreeding coefficient within the entire population (Fit) did not considerably deviate from the equilibrium is the result of the high subpopulation differentiation, counteracted by low inbreeding coefficient within subpopulations. Heterozygosis levels (10–15%) even lower than those found in our study were reported in a genotyping analysis performed through RNAseq on petunia cultivars, mostly reported as F₁ hybrids. In the mentioned study, lines were representative of the total variation observable in the market considering the growth habit trait. Authors suggested a quite limited genetic diversity for the total commercial P. × hybrida germplasm pool [7]. Our findings supported the efficacy of ddRADseq in performing a genetic diversity analysis in petunia clonal populations. Although target-sequencing techniques methods are highly cost-effective and reproducible, as described below, genome-wide approaches such as ddRADseq for genotyping remain advantageous not only when prior marker information is lacking, but also when a more exploratory approach is needed. Indeed, the de novo SNP discovery facilitates unbiased analyses, like genetic structure and diversity, allows us to adapt the marker set to a specific population and is less sensitive to project design errors [41,129,130].

Given the relatively high genetic uniformity found for sister lines, assessing the minimal number of SNPs required to preserve the clustering patterns—representing the most valuable information for breeding purposes—was of particular interest. In order to also keep also a unique molecular profile for each BL, a final set of more than one thousand SNPs was revealed to be necessary and was obtained through MAF-based filtering. The clustering was overall respected: despite the presence of minor changes, no sister line separation or alteration of the BL group’s hierarchical order was observed. The need for a high minimum number of variants to simultaneously maintain line discrimination and clustering was expected, given the high within-group uniformity. When clustering preservation was not considered, a genotype accumulation curve showed that eight polymorphisms were sufficient to uniquely identify all the BLs. This may provide an indication that retaining clustering patterns weighs far more heavily than retaining discrimination when determining the number of variants to be used. However, preserving the unique identification of accessions, even if not their grouping, is useful for PVP purposes in order to detect fraud or unknown materials. With a manually developed method, five discriminating variant sets, each composed of eight SNPs, were identified, assessing the minimal locus number to characterize the collection. The custom method provided in this study is usable in other populations and species, in order to identify the minimal discriminating SNP sets and help genotyping analyses.

Keeping the same clustering effects described in the one-thousand dataset, and without observing major group disruptions, the variant reduction could be extended—using MAF filtering—down to approximately one hundred SNPs. In addition, the non-discriminated genotypes obtained with this set only included a couple of sister lines previously hypothesized as belonging to the same clone. Therefore, the dataset of one hundred variants obtained may be useful for both PVP- and MAB-related purposes. In particular, its most valuable application for breeding may lie in parental selection. In this case, indeed, crosses are designed between the most genetically differentiated genotypes in order to achieve greater phenotypic variability or to combine multiple traits of interest. Hence, clustering reliability is of pivotal importance. Studies aimed at selecting SNP core sets were performed in collections of several species of agronomic interest, including lettuce, corn, barley, cotton and tobacco [65,68,131,132,133], whereas in ornamental species such investigations have been reported for rose, Phalaenopsis spp. and orchids [69,70,71]. In particular, in polyploid rose, a core set of 18 PACE markers provided reliable fingerprinting across thousands of accessions from the German Genebank for Roses, while in Cymbidium ensifolium (L.) (Orchidaceae family) 11 KASP markers were selected and used for the genotyping of 83 commercial cultivars in China. In Phalaenopsis spp. a core set of SNPs—obtained from an initial dataset of thousands of variants with the following filtering parameters: MAF 0.1, max-missing 0.8, min-meanDP 3, and hwe 0.01—was effectively used to genotype 53 Chinese cultivars. The reliability of the results was assessed by observing the absence of differences in sample clustering with a Neighbor-Joining tree between the core SNP set and the initial dataset [69].

In general, core sets capable of discriminating between plant accessions are characterized by a few dozen markers. However, in the majority of the studies, only genetic diversity measurements were performed, without carrying out clustering analyses and hence not assessing the core set potential for MAB purposes. The SNP number is in fact consistent with that of our discriminating datasets. Furthermore, the accessions analyzed in the above-mentioned studies were part of germplasm collections, characterized by a genetic diversity far greater than that observed in our breeding population. This might explain why a high number of variants was required in our study for respecting the clustering of the main groups. In addition, in the other works PIC is considered a key parameter for defining a core set of markers to discriminate between different accessions (e.g., [68,71]). However, in our study we were able to identify a suitable SNP dataset by primarily adjusting the MAF thresholds. Our results would suggest that MAF might be more efficient for the discrimination of strongly related lines (i.e., characterized by low genetic diversity), whereas it may be less useful than PIC if the genetic differentiation degree of the entire population is higher. Nevertheless, further studies testing the approach in other petunia populations are required to extend these findings. SNP core sets can be used for the development of target-specific molecular genotyping assays, in order to reduce analysis costs, favor comparisons between populations and help identifying sequences potentially involved in trait of interest [61,62,63,64,66,134,135]. In particular, low-cost targeted panels include, among others, the following: PCR-based methods relying on allele-specific primers and fluorescence, e.g., KASP; multiplexed PCR with fluorescence microfluidic arrays, e.g., Fluidigm; and highly multiplexed PCR and Next Generation Sequencing (NGS) of the amplified products, e.g., AmpliSeq panels [66,136,137,138,139]. However, these platforms differ in scale, detection method, and multiplexing capacity, hence their suitability is subordinated to the specific breeding goal.

The selection of SNPs located in coding sequences led to retaining less than half of the initial dataset variants. This result is in agreement with what was reported for ddRADseq analyses exploiting the same restriction enzymes for library preparation [140]. Indeed, PstI and MspI are methylation-sensitive and therefore tend to favor the selection of coding sequences with greater probability than methylation-unsensitive enzymes. CDS-located SNPs were more present in the distal zones of chromosomes, as can be attested by the centromere positions reported for P. × hybrida chromosomes [26]. Variant density along chromosomes was in agreement with what was already reported for ddRADseq analyses with methylation-sensitive enzymes [49]. In addition, SNP distribution reflected annotated gene density, suggesting higher recombination rates and euchromatic features in distal regions, whereas pericentromeric zones are characterized by reduced presence of coding sequences, suppressed recombination and epigenetic silencing [141,142,143]. In petunia, recombination studies based on RIL populations were carried out before the release of a high-quality sequenced genome of P. × hybrida, and interpretations concerning recombination-suppressed regions were limited by the absence of a physical map [24,25,104]. Nevertheless, the inferences reported about the recombination and methylation of petunia chromosomes were corroborated in other Solanaceae species, i.e., tomato and pepper, in which also a higher abundance of transposable elements and repetitive sequences was also observed in the pericentromeric heterochromatic regions [144,145,146].

The finding that more than half of the retrieved SNPs in our collection were located in non-coding regions even using methylation-sensitive restriction enzymes is likely a consequence of the higher conservation degrees of CDSs due to functional reasons and their consequently lower levels of polymorphisms [147]. Despite this, the exclusive use of SNPs located in CDS regions for the genotyping of the petunia clonal lines led to results entirely comparable to those obtained with the initial dataset (considering the GD, UPGMA clustering, PCA and structure analysis) and to the genealogy data (i.e., similar genotyping outputs for sister lines). Hence, the obtained data assessed the suitability of the exclusive use of CDS-located SNPs for genotyping analyses in the petunia clonal population. In addition, the retrieval of CDS-SNPs enabled the investigation of variants within genes putatively associated with pathways of interest. Indeed, several putative CDSs related to the biosynthesis of terpenoids, flavonoids, phenylpropanoids, and other secondary metabolites, as well as plant hormones were found to harbor SNPs. The production and regulation of these compounds is of particular relevance, as they are involved in key biological processes, such as herbivore defense, flower opening and floral scent emission in petunia, both for the species itself and for its use as a model system to study these traits [148,149,150,151,152,153]. In particular, SNPs were identified in two PHYTOCHROME INTERACTING FACTORs (PIFs), PhPIF1a (Peaxi162Scf00052g01817; PHYG54490) and PhPIF3 (Peaxi162Scf00194g00213; PHYG60033), which have previously been characterized as playing a role in floral scent emission in petunia by transcriptionally activating key regulatory genes and biosynthetic enzymes involved in this process [154]. Overall, these results suggest that ddRAD sequencing represents a suitable approach for the identification of genetic variants putatively involved in trait manifestation, providing valuable candidates for subsequent functional analyses.

5. Conclusions

To summarize, ddRADseq was demonstrated to be effective in uniquely discriminating within an experimental petunia population including highly uniform sister line groups, in assessing their genetic relationship and in providing informative data for breeding purposes. A method to identify the minimal discriminating variant sets was developed. The selection of an SNP core set of one hundred variants able to correctly cluster the collection was achieved too, suggesting MAF as the best filtering parameter for a relatively uniform petunia population. In addition, the exclusive use of SNPs located in coding sequences for the population genotyping and genetic diversity analysis was demonstrated to be effective. CDS-SNPs, as annotated genes, were more present in the distal areas of chromosomes. Furthermore, the utilized approach seemed to be suitable to detect candidate variants for targeted in-depth analyses in order to assess the variant putative involvement in the traits of interest’s manifestation. In conclusion, results encourage the adoption of GBS techniques and target-specific assays for PVP to guarantee PBRs, for the registration of novel varieties whether coupled with DUS testing and other phenotype measurements, and for MAB approaches in petunia. The production of molecular profiles relative to experimental plant materials can help in achieving of innovative traits for cultivars. This aspect is particularly important nowadays for the ornamental sector, in order to select traits relative to sustainability for cultivation. The reduction in the agronomic input needs of plants leads to an enhancement of economic and environmental security. Further studies are needed in two goal directions: the enhancement of basic molecular knowledge relative to petunia, and the development of useful applications for breeders, such as assays able to predict traits of interest’s values. Eventually, GBS analyses need to be extended to understudied species, as is the case for the majority of the ornamental plants commercialized worldwide nowadays.