Introduction

The use of marker-assisted selection (MAS) has expedited the improvement of economically important traits in grapevines (Vitis sp.) since its incorporation in breeding programs. Starting with restriction fragment length polymorphisms (RFLPs) used to distinguish individuals (Striem et al., 1990), the technology has grown to allow for the high-throughput screening of an ever-growing list of available loci using rhAmpSeq local haplotype markers, based on highly-multiplexed, amplicon markers targeting the Vitis core genome (Zou et al., 2020). Microsatellite markers such as simple sequence repeats (SSRs) have long been a widely adopted and highly useful step in this progression as they are multi-allelic by design and complement the high heterozygosity intrinsic to Vitis (Arnold et al., 2002). Not without their disadvantages, the broad use of SSRs in genetic mapping studies created opportunities for the comparison of alleles across breeding and mapping populations (Fischer et al., 2004). This was crucial for the identification and validation of loci influencing important phenotypes, including resistance to destructive plant diseases such as grapevine downy mildew.

Grapevine downy mildew, caused by the oomycete Plasmopara viticola, is among the most economically damaging diseases in vineyards in humid climates, such as the eastern United States and much of Europe (Koledenkova et al., 2022). Once grapevine downy mildew takes hold, potentially infecting all green tissues, producers often rely on extensive chemical applications for control. As this increases the risk of the pathogen developing fungicide resistance (Massi et al., 2021), many breeders have focused substantial efforts toward selecting genetic resistance within their released cultivars. This has led to the identification of over 35 Rpv (resistance to P. viticola) loci, each conferring resistance of varying strengths (Vitis International Variety Catalogue: www.vivc.de). Since V. vinifera, the most commonly cultivated species in wine and table grape production, is predominately susceptible (Cadle-Davidson, 2008), resistance is often introgressed from the North American and Asian Euvitis subgenus (Di Gaspero et al., 2012; Peng et al., 2024). One of the most selected resistance loci, Rpv3, resides on the lower arm of chromosome 18 and enacts a race-specific hypersensitive response when triggered by P. viticola (Casagrande et al., 2011).

The Rpv3 QTL was originally identified through the combined works of Fischer et al. (2004) and Welter et al. (2007) in the wine grape ‘Regent’ and later tied to the resistance in the hybrid ‘Bianca’ (Bellin et al., 2009). Di Gaspero et al. (2012) later performed an extensive survey of 580 grapevines and identified seven conserved haplotypes based on the locus flanking SSR markers UDV305 and UDV737. The most common haplotype found was Rpv3^299-279, now designated as Rpv3.1, was connected to the resistance found in both ‘Regent’ and ‘Bianca’. The second most conserved haplotype through ‘Munson’ (‘Jaeger 70’), Rpv3^null-297, is now known as Rpv3.2 after it was confirmed to confer downy mildew resistance through QTL mapping (Zyprian et al., 2016). The final conversed haplotype to receive an allelic designation, Rpv3^null-271, was named Rpv3.3 and descended from ‘Noah’ into the complex hybrid ‘Merzling’ (Vezzulli et al., 2019). Candidate genes for Rpv3, TNL-NBS-LRR (TNL) gene paralogs and a leucine-rich repeat kinase (LRRk), have been identified through controlled inoculation and gene expression analysis within individuals carrying Rpv3.1 (Foria et al., 2020; Wairich et al., 2022). Separately, a resistance locus identified in the V. aestivalis-derived ‘Norton’, Rpv27, co-locates with Rpv3 (Hammers et al., 2017; Sapkota et al., 2019) but is distinguished from Rpv3 through its haplotype’s absence from the conserved list in Di Gaspero et al (2012).

While Rpv3.1 is often found to confer the strongest level of resistance, followed by Rpv3.2 and then Rpv3.3, there is growing evidence for race-specificity on grapevines carrying Rpv3 alleles (Cadle-Davidson, 2008; Delmotte et al., 2014; Heyman et al., 2021; Paineau et al., 2022). The issue of race-specificity at this locus is further complicated by the fact that the allele designations are intrinsically tied to SSR markers with limited resolution. As pathogens evolve to overcome these resistance genes, SSR markers cannot be relied upon to capture sequence variation within individuals carrying the same Rpv3 allele, hampering the ability to track the co-evolution of pathogens and resistance genes at this locus. The potential discrepancies this introduces coupled with the technological advances in genomic sequencing and bioinformatics present the means and opportunity to improve upon the limited resolution of Rpv3 and Rpv27 currently provided by screening with SSR markers. Using the SSR-based allelic designations from Di Gaspero et al. (2012) (Rpv3) and Sapkota et al. (2019) (Rpv27), the PacBio long-read (Hon et al., 2020) genome assemblies of eight grapevines were used to compare the structural and candidate gene copy number variation among and between haplotype alleles. In doing so, the creation of more informative, higher resolution markers for MAS can be designed and utilized by breeders in selecting for improved resistance to grapevine downy mildew.

Materials and methods

1. Genome Assemblies

The genome assemblies of eight grapevines, ‘Catawba’, ‘Concord’, ‘Chambourcin’, ‘Horizon’, MN1264, ‘Norton’, NY84.0101.03 and PN40024, were used in this study. These individuals were included because they were known or implicated via SSRs as carriers for either specific Rpv3 alleles or Rpv27. Among the French-American hybrids, NY84.0101.03 is heterozygous for Rpv3.1, ‘Horizon’ and MN1264 both carry single copies of Rpv3.2, while ‘Chambourcin’ is heterozygous for both Rpv3.1 and Rpv3.2 (Di Gaspero et al., 2012; Clark et al., 2018; Patel et al., 2023; Reisch, 2024). ‘Norton’ is commonly described as being derived from V. aestivalis and is one of the known carriers of Rpv27 (Hammers et al., 2017; Sapkota et al., 2019). ‘Catawba’ and ‘Concord’ are related hybrids between Vitis labrusca and V. vinifera that date back to the early to mid-1800s. Specifically, ‘Catawba’ is an F₁ of V. labrusca and V. vinifera ‘Semillon’, while ‘Concord’ is the product of a ‘Catawba’ backcross to V. labrusca (Vitis International Variety Catalogue: www.vivc.de). While both were found to carry the SSR fragment lengths for Rpv3.3, neither could be tied to a known source through pedigree. Lastly, PN40024 is a highly homozygous V. vinifera, recently shown to originate from the repeated selfing of ‘Helfensteiner’, an F₁ of ‘Pinot noir’ (Velt et al., 2023) and is often used as the primary reference genome for grapevines.

Through the use of haplotype-resolved genome assemblies, the Rpv3 / Rpv27 region of sixteen haploid genome assemblies were characterized. Genome assemblies for ‘Catawba’, ‘Concord’, MN1264, ‘Norton’, and NY84.0101.03 were constructed at Cornell University while ‘Horizon’ was assembled at the University of California–Davis. Genome assembly followed the same methodology described in detail by Zou et al. (2023) for the Vitis x doaniana ‘PI 588149’ genome assembly. Briefly, SMRTbell Template Prep Kit SP v2 (PacBio, CA, USA) libraries were constructed from high molecular weight DNA that was filtered for DNA between 15–20 kbp using BluePippin (Sage Science, MA, USA). The libraries were then sequenced with a PacBio Sequel II using 1 SMRT cell by the DNA Sequencing and Genotyping Center at the University of Delaware. The resulting PacBio reads were then used to construct two phased assemblies using Hifiasm (v0.16.0) (Cheng et al., 2021). ‘Chambourcin’ and PN40024 had preexisting, publicly available genome assemblies that were used in this study. The ‘Chambourcin’ assembly constructed by Patel et al. (2023) was diploid with haplotypes separated into ‘primary’ and ‘haplotig’ assemblies and was accessed via the article’s supplemental data. PN40024, as an inbred accession, was constructed as a haploid genome assembly in its role as reference genome. However, including both high quality versions of PN40024, PN40024.v4 (Velt et al., 2023) and PN_T2T (Shi et al., 2023), allowed for a bioinformatic comparison of these versions in addition to the opportunity to update the physical positions of Rpv3 and Rpv27 in the latest versions of the reference genome. Both versions of PN40024 were downloaded from grapedia.org/genomes.

2. Characterizing the Rpv3 and Rpv27 Loci

As SSR markers are intrinsically tied to the allele classification and demarcation of Rpv3 and Rpv27, the primer sequences of associated SSR were used to validate the presence and the physical boundaries of these loci within each diploid assembly (Table 1).

Using BBMap (v39.03) (Bushnell, 2024) under default parameters, both forward and reverse primer sequences from six SSR markers were aligned to all sixteen haplotypes from eight assemblies, separately. Based on the predicted fragment lengths from BBMap, each haplotype assembly was then assigned to an Rpv3 allele or Rpv27 based on the combination of UDV305–UDV737 or VVCS1H077H16R1-1–UDV737 for Rpv3 (Di Gaspero et al., 2012) or Rpv27 (Sapkota et al., 2019), respectively. Fragment length combinations for Rpv3 allele assignment, as UDV305–UDV737, were: Rpv3.1, 299–279; Rpv3.2, null–297; and Rpv3.3, null–271 while the fragment length for Rpv27, as VVCS1H077H16R1-1–UDV737, was 99—294. An exact match was not required, however, given that small differences were expected between the in-silico fragment lengths derived through BBMap and the reported PCR amplification products for each allele. Three additional SSRs, UDV730, UDV734, and UDV736, were included as associated markers that span the gap between the diagnostic pairs described above.

In order to focus solely on Rpv3 and Rpv27, the physical positions of SSR markers VVCS1H077H16R1-1 and UDV737, the most physically distant pair, were used to isolate the region from the rest of each haplotype assembly using the ‘seqinr’ package (Charif & Lobry, 2007) to manually isolate the sequence in R (R Core Team, 2023). In most of the haplotype assemblies, the Rpv3 / Rpv27 region was contiguous. Extensions of 5 kb and 100 kb were added to the positions of VVCS1H077H16R1-1 and UDV737, respectively. A larger extension was included beyond UDV737 to include more of the neighboring NLR (Nucleotide-binding Leucine-rich Repeat) gene cluster found to reside nearby. For haplotype assemblies where the SSRs were split across multiple contigs, all-to-all genome alignments were performed using ‘nucmer’ with default parameters from MUMmer4 (v4.0.0) (Marçais et al., 2018). The discontiguous haplotype was aligned to a contiguous haplotype with the same allele. This allowed for the visual selection of contigs via ‘mummerplot’ and the formation of an Rpv3 / Rpv27 region for each haplotype assembly. Once the haplotype assemblies were narrowed down, AUGUSTUS (v3.4.0) (Stanke et al., 2008) was used for gene prediction, followed by functional annotation with diamond (v2.1.8) (Buchfink et al., 2021) and Blast2GO (v1.5.1) (Gotz et al., 2008) each under default parameters. NLR genes were annotated using NLR-Annotator (Steuernagel et al., 2020) with their functionality and completeness predicted using NLRexpress (Martin et al., 2022), again under default parameters. Visualization of the Rpv3 / Rpv27 regions, including positions of mapped SSR and annotated NLR genes was produced in Python (v3.9) using custom script incorporating tools from ‘pyGenomeViz’ (v0.4.4) (Moshi, 2024).

3. Candidate Gene Analysis

The candidate genes conferring Rpv3.1 downy mildew resistance were first identified by Foria et al. (2020) and later corroborated and expanded upon by Wairich et al (2022). These candidate genes were two TIR-NBS-LRR (TNL), ‘VIT_18s0041g01330’ (TNL1) and ‘VIT_18s0041g01340’ (TNL2), and a leucine-rich repeat kinase (LRRk), ‘VIT18s0041g01350’. Both articles demonstrated the role TNL2 (as paralogs TNL2a and TNL2b in Rpv3.1) played in triggering a defense response through expression analysis. Only Wairich et al. (2022), however, found evidence that TNL1 and LRRk may also play a role. All three gene models were included here for completeness in characterizing the locus. The reported gene model identifiers were based on a previous version of the PN40024 reference genome, however. In order to characterize the presence, functionality, and abundance of these genes, the gene model identifiers were updated to their PN40024.v4 counterparts through the PN40024 reference genome browser provided by grapegenomics.com (hosted by UC-Davis). These translations were: TNL1, ‘Vitvi18g02077’; TNL2, ‘Vitvi18g03133’; and LRRk, ‘Vitvi18g02080’. The genomic sequences for these three genes were aligned to each haplotype assembly using nucleotide BLAST (Camacho et al., 2009), (-max_target_seq 5 -evalue 1e-10) including the diploid Muscadinia rotundifolia ‘Trayshed’ assembly (Minio et al., 2022) as an outgroup. The high sequence similarity between TNL1 and TNL2 made it necessary to identify short, 60 bp sequences of high divergence that could be used to sort through the blast results. These sequences were identified using a combination of R (R Core Team, 2023) and alignment in MEGA11 (Tamura et al., 2021). These short sequences were then blasted back to the extracted alignment results of the initial blastn (-evalue 1e-10). Manual classification of the results to either TNL1 or TNL2 was based, in part, on the presence and quality of the alignment of these short sequences. Alignment quality was assessed based on the combination of percent identity and alignment length. Multiple sequence alignment of the resulting TNL1, TNL2, and LRRk sequences from each haplotype genome was followed by the construction of neighbor-joining trees using MEGA11 and visualized in iTOL (v5) (Letunic & Bork, 2021). In order to compare the sequence similarity among the LRRk from each haplotype, pairwise distances were calculated under default settings in MEGA11.

Results

1. Genome Assemblies

In order to have high quality representation of each haplotype allele at Rpv3 and Rpv27, assemblies of six new grapevines were constructed. These grapevines, ‘Catawba’, ‘Concord’, ‘Horizon’, MN1264, ‘Norton’, and NY84.0101.03 were included alongside ‘Chambourcin’ and PN40024 to account for alleles Rpv3.1, Rpv3.2, Rpv3.3, and the co-located Rpv27 as assigned based on diagnostic SSR. The primary or haplotype-resolved contigs for the six new diploid assemblies ranged in contig N50 from 4.9–21 Mb with complete BUSCO scores from 95.8–98.7% (Table 2).

The combination of SSR markers UDV305 and UDV737, or VVCS1H077H16R1-1 and UDV737 for Rpv27, were used to classify each haploid assembly with an Rpv3 allele according to the fragment lengths described in Di Gaspero et al. (2012). There were two assembled haplotypes of Rpv3.1, Chambourcin Primary and NY84.0101.03 Hap1. Rpv3.2 was assembled three times from Chambourcin Haplotig, Horizon Primary, and MN1264 Hap1. Potential haplotypes of Rpv3.3 based on SSRs were assembled from Catawba Hap1, Concord Hap2, and Horizon Haplotig. Norton Hap 2 was the only instance of Rpv27. An additional conserved haplotype from Di Gaspero et al. (2012) representing a V. vinifera background, UDV305: null–UDV737: 287, from PN40024.v4, PN_T2T, Concord Hap1, and Catawba Hap2. The final three haplotypes, MN1264 Hap2, Norton Hap1, and NY84.0101.03 Hap2 were distinct from each other and have yet to be characterized.

2. Characterizing the Rpv3 and Rpv27 Loci

Using the alignment positions of the six SSR markers, the Rpv3 / Rpv27 region was extracted from each haplotype assembly based on the two most physically separated markers, VVCS1H077H16R1-1 and UDV737. Across all haplotypes, this region ranged from 0.8 to 1.7 Mb (Table S1). Both examples of Rpv3.1 were contiguous in their assemblies and covered a physical distance of 1.36 Mb and 1.40 Mb on the Chambourcin Primary and NY84.0101.03 Hap1 assemblies, respectively. For Rpv3.2, both the Chambourcin Haplotig and MN1264 Hap1 assemblies were contiguous and spanned 1.3 Mb, while the Horizon Primary Rpv3.2 was not contiguous. This assembly was thusly aligned to Chambourcin Haplotig, another Rpv3.2, contig ‘vitin_chambo_h_18_02’. The Rpv3.2 from Horizon Primary was scaffolded using two contigs (‘VITHorizon_Primary000139F’ and ‘VITHorizon_Primary000043F’) spaced roughly 3 kb apart for a total length of 1.19 Mb (Figure S1). More variation in length existed among the three representatives of Rpv3.3. Catawba Hap1 and Concord Hap2 were both contiguous and 1.29 and 1.34 Mb in length. The third, Horizon Haplotig, was not contiguous and also lacked successful alignments for both VVCS1H077H16R1-1 and UDV730. Regardless, the Horizon Haplotig assembly was aligned to the Concord Hap 2 contig ‘h2tg000005l’ and scaffolded using four contigs with cumulative added spacing of 125 kb for a total length of 0.8 Mb (Figure S2). The single carrier of Rpv27, Norton Hap2, had a contiguous length of 1.3 Mb. The V. vinifera haplotype alleles were all contiguous and consistent in length, with the two versions of PN40024 at 1.52 Mb and both Concord Hap1 and Catawba Hap2 at 1.53 Mb. Of the remaining three haplotype assemblies, NY84.0101.03 Hap2 and Norton Hap1 were both contiguous yet varied in length at 1.43 Mb and 1.7 Mb, respectively. Lastly, the MN1264 Hap2 haplotype was not contiguous and due to its novelty within the study was aligned to the PN_T2T assembly contig ‘PN18’, resulting in a scaffolded region of 1.27 Mb consisting of two contigs, spaced 12.6 kb apart (Figure S3). The extracted genomic sequences for Rpv3/Rpv27 haplotypes, including those scaffolded for this analysis, are provided in Additional Data 1.

Visualization of the extracted Rpv3/Rpv27 regions of each haplotype, specifically the distribution of the SSR markers and predicted gene structure, allowed for the resolution of existing sequence variation within and among haplotype alleles. While the gene structure between individuals carrying the same Rpv3 allele differed, all haplotypes became more consistent as they approached UDV737, especially within allelic groups (Figure 1).

Figure 1. Rpv3 and Rpv27 region SSR alignments and annotated genes by SSR haplotype.

For Rpv3.1, the two haplotype assemblies showed strong similarities throughout the 157 and 159 predicted gene features in Chambourcin Primary and NY84.0101.03 Hap1 and are nearly identical between UDV736 and UDV737. For Rpv3.2, Chambourcin Haplotig and MN1264 Hap1 shared more structural similarity than with Horizon Primary. This was also reflected in their gene feature predictions, both containing 136 features compared to 121 for Horizon Primary. Lastly, Rpv3.3 was the least consistent, with each haplotype showing a unique distribution of their SSR and predicted gene structure. Concord Hap 2 had the most gene features with 145, then Catawba Hap1 with 132, followed by Horizon Haplotig at 63 features. Rpv27 maintained similar gene structure to the Rpv3 alleles nearer to UDV737 and included 145 gene feature predictions. The four V. vinifera haplotypes had reasonably similar structure and ranged between 167 and 173 predicted gene annotations. The remaining three haplotypes had 144 predicted gene features for NY84.0101.03 Hap2, 216 for Norton Hap 1, and 133 for MN1264 Hap2. An annotation file containing the regional SSR positions and predicted gene annotations is provided in Additional Data 2.

3. Candidate Gene Analysis

Three candidate genes have been found in connection to the resistance conferred by Rpv3.1. The first TNL, TNL2 was identified by Foria et al. (2020) as a pair of paralogs, TNL2a and TNL2b with the second TNL, TNL1, and a nearby LRRk added later by Wairich et al. (2022). The gene model identifiers were translated to the PN40024.v4 reference genome (Velt et al., 2023), functionally annotated, and aligned to the extracted Rpv3/Rpv27 regions. In every analyzed haplotype, the candidate genes clustered closely to UDV737, within 220 kb, with a single copy of the LRRk roughly 25 kb upstream (Figure S2).

Figure 2. Rpv3.1 candidate gene positions and predicted functionality by haplotype allele.

Pairwise comparisons of the LRRk from each haplotype revealed high amino acid sequence similarity, with a maximum divergence of just above 4 % (Table 2). The number, functionality, and position of the TNL genes relative to UDV737 were more varied across alleles compared to the LRRk. Consistent in both Rpv3.1, there was one complete copy of TNL1 roughly 178 kb away from UDV737, followed by a partial TN and two complete copies of TNL2 which correspond directly to the TNL2a and TNL2b described in Foria et al. (2020). Rpv3.2 was similar with Rpv3.1 in terms of gene frequency and functionality, however their organization differed. In all three examples of Rpv3.2, there were two complete TNL2 copies flanking a TNL1 homolog and a TNL1 predicted pseudogene, with the furthest TNL2 alignment roughly 166 kb upstream of UDV737. All three SSR-based haplotypes of Rpv3.3 differed in gene distribution and length. Catawba Hap1 and Horizon Haplotig both had two complete copies of TNL1, one complete copy of TNL2, and one TNL2 pseudogene. The Rpv3.3 allele from Concord Hap2 had the greatest number of TNL alignments, consisting of three complete TNL1, one complete TNL2, and two TNL2 pseudogenes. In all three Rpv3.3, a TNL2 pseudogene was the furthest from UDV737 at 159.1 kb, 195.1 kb, and 216.5 kb for Catawba Hap1, Horizon Haplotig, and Concord Hap2, respectively. Norton Hap2’s Rpv27 was most similar in TNL gene copies and functionality to the Rpv3.3 alleles from Catawba Hap1 and Horizon Haplotig. The four V. vinifera haplotypes and Norton Hap1 all had complete single copies of each TNL and were within 69.3–70.2 kb of UDV737. The final two haplotypes differed. MN1264 Hap2 had two complete copies each of TNL1 and TNL2 within 85.3 kb while NY84.0101.03 Hap2 had one complete copy of TNL2, one TNL2 pseudogene, and two copies of TNL1 within 190.5 kb of UDV737.

The sequences of the TNL1, TNL2, and LRRk gene copies from each haplotype were extracted from their assemblies, aligned, and used to produce neighbor-joining trees. The alignment results from both haplotypes of Muscadinia rotundifolia ‘Trayshed’ were also included to serve as outgroups. The TNL1 and TNL2 gene copies separated predictably when combined into an unrooted tree (Figure S4), and the Rpv3 alleles largely grouped together among the LRRk from each haplotype (Figure S5). The multiples of each TNL gene copy from Rpv3.1 and Rpv3.2 grouped together while Rpv3.3 was less consistent and grouped closely with Rpv27 (Figure 3A to 3B).

Figure 3. Rooted neighbor-joining trees of TNL1 (A) and TNL2 (B) gene copies across all haplotypes.

Among the TNL1 copies from each haplotype, the TNL1 from Rpv3.1 grouped most closely with the complete TNL1-like alignments from Rpv3.2 and those from Rpv3.3 and Rpv27 closest to UDV737 (Figure 3A). The partial TN from Rpv3.1 was also predictably more similar to TNL1 than TNL2 and grouped closely with the original TNL1. The haplotype from Norton Hap1 was more similar to the V. vinifera haplotypes than any other for both TNL1 and TNL2. While the TNL1-like pseudogene from Rpv3.2 grouped closely with the complete TNL1-like A’s from both Rpv3.3 and Rpv27, the inverse was true for TNL2 (Figure 3B). The TNL2-like pseudogenes from Rpv3.3 and Rpv27 were more similar to the TNL2-like A’s from Rpv3.2. Both paralogs of TNL2 in Rpv3.1, TNL2a and TNL2b were more closely grouped than the pair of TNL2-like alignments from Rpv3.2.

Discussion

Using the phased genome assemblies of eight diploid grapevines, the Rpv3 and co-locating Rpv27 loci were characterized to compare the structural variation within and among haplotypes. In doing so, this work showed that variation exists within individuals carrying the same Rpv3 allele, especially in Rpv3.2 and Rpv3.3, and demonstrates that the widely used diagnostic SSR markers lack the resolution necessary to capture variation at the locus. In comparing the copy number and structural variation of the Rpv3.1 candidate genes in each haplotype, the close relationship between the genes in Rpv3.3 and Rpv27 suggest the need for additional candidate gene analyses to validate the genetic mechanisms underlying their resistance. This work represents an important step in designing high density markers for improved coverage of this multiallelic resistance locus.

The concept of Rpv3 haplotypes introduced by Di Gaspero et al. (2012) provided pioneering insight into the challenges faced when characterizing resistance loci. While most haplotypes at a resistance locus confer susceptibility, grape scientists are now demonstrating functional allelic variation among resistant haplotypes that translates into quantitative and/or mechanistic differences in resistance phenotypes. This concept is even more impressive considering it was built upon a low-resolution marker platform, namely two SSRs spaced, at most, 1.7 Mb apart as shown in the current study. In quantitative phenotypic differences among Rpv3 haplotypes, Rpv3.1 typically offers the greatest resistance to grapevine downy mildew and therefore is the most selected allele (Di Gaspero et al., 2012; Foria et al., 2020). The resistances conveyed by Rpv3.2 and Rpv3.3 tend to be weaker (Foria et al., 2018), less reproducibly detected, and sometimes eliminated from breeding programs in favor of stronger alternative loci. Although Rpv27 may confer the strongest resistance among these haplotypes by effect size (Sapkota et al., 2019), it has not been as widely characterized.

We explored Rpv3 haplotypes more deeply to determine the level of variation within and among haplotypes both in structure and candidate gene sequence. For the genomes analyzed here, most of the alleles defined by Di Gaspero et al. (2012) are structurally conserved (at least moderately) across the SSR haplotype, and particularly in the 220 kb candidate gene region upstream of UDV737. The exception was Rpv3.3, which showed significant structural variation among individuals positive for this allele based on SSR haplotypes. While the Rpv3.1 and Rpv3.2 haplotypes were supported by pedigree connections to known sources, this cannot be said for ‘Catawba’, ‘Concord’, or ‘Horizon’ to Rpv3.3. While the potential for these connections is dependent on the accuracy of reported pedigrees, another potential explanation could be size homoplasy between haplotypes connected to V. labrusca and ‘Merzling’, where Rpv3.3 was first described. Size homoplasy of SSRs is the occurrence of amplification products of identical size in unrelated individuals which has been described in microsatellite markers and can result from mutations and recombination events within SSR flanking sequences (Grimaldi & Crouau-Roy, 1997). It has been documented but not well studied in grapevines (Battilana et al., 2013).

One hypothesis underlying this study was that the phenotypic variation observed among resistant haplotypes might be explained by the presence or sequence of their candidate genes. This hypothesis was supported by Rpv3.1, with divergent sequences relative to other resistant and susceptible alleles for TNL1, TNL2a, and TNL2b as shown by the neighbor-joining trees. Although the true inclusion of Rpv3.3 in this work is questionable, their candidate gene sequences clustering closely with those from Rpv3.2 may be reflected in the minor phenotypic differences between alleles. What was most surprising was that all Rpv27 TNLs clustered tightly with Rpv3.3, suggesting one possibility that the genetic basis of Rpv27 resistance may lie outside this TNL region given the stronger resistance conferred by Rpv27 compared to Rpv3.3. Another alternative still could be that the included Rpv3.3 SSR haplotypes bear more genetic similarity to Rpv27 irrespective of the SSR haplotype. As to how these observations can be tied to phenotypic variation seen among Rpv3 alleles requires further research.

Of the resistance alleles characterized, Rpv3.1 is the only one to have candidate genes validated through expression analysis and functional research (Foria et al., 2020; Wairich et al., 2022). Their presence in every haplotype and M. rotundifolia indicates that they are highly conserved, yet have been subjected to evolutionary events that resulted in their divergence into independent alleles. Foria et al. (2020) suggested an evolutionary model from a common ancestral haplotype leading to the PN40024 and Rpv3.1 haplotypes involving tandem segmental duplication and the movement of retrotransposons. Where the other haplotypes fit into their evolutionary timeline remains unclear along with the scope of influence these candidate genes have beyond Rpv3.1.

In making selections for improved grapevine downy mildew (P. viticola) resistance, plant breeders can rely on MAS to stack multiple resistance genes and expedite the release of new varieties and cultivars. While an individual SSR may typically have tens of alleles, they are based on sizes that can be attained convergently and may not capture informative underlying sequence variation. Across the Rpv3 haplotype defined by SSRs, selection appears more targeted toward one side of the locus, near the Rpv3.1 candidate genes and UDV737. This would suggest breeders have focused higher selection pressure on this roughly 160–220 kb region rather than the entire haplotype. For this to occur and for these Rpv3 haplotypes to be so widespread in breeding germplasm, selection for resistance in many breeding schemes historically relied on phenotype over MAS. Regardless, there is a demonstrated need for updated marker technologies whose positions are refined in closer proximity to the potential causal genes. The use of higher resolution markers for MAS would also enable grapevines breeders to track recombination events within the locus and how that relates to changes in resistance efficacy and race-specificity, and would exchange the resources available for the fine mapping of causal alleles (Zou et al., 2023). Core genome rhAmpSeq local haplotype markers typically have several hundred alleles across Vitis and could help to identify haplotype diversity missed by SSR analysis.

Conclusion

This study sought to characterize genome structural variation within the Rpv3 and co-located Rpv27 loci through the comparative studies of haplotype-resolved long-read genome assemblies. The alleles were categorized based on their diagnostic SSR markers and the copy numbers and locations of the validated Rpv3.1 candidate genes were compared among alleles. Structural and sequence variation within and among allelic groups suggested the effect of selection on the locus and demonstrated the need for the development of new markers that can distinguish each allele. In doing so, this work contributes to the knowledge of variation within Rpv3 and lays the foundation for future work in understanding this complex resistance locus.

Acknowledgements

We would like to thank Gabriele Di Gaspero for providing valuable insights and critical review of this research prior to peer review. We would also like to thank the grape breeding programs who provided leaf samples for genome sequencing: Bruce Reisch for NY84.0101.03, Matt Clark for MN1264, and Chin-Feng Hwang for ‘Norton’. An additional thank you to Erin Galarneau and the USDA National Plant Germplasm System for ‘Catawba’, ‘Concord’, and ‘Horizon’. Funding was provided by the USDA Specialty Crop Research Initiative (Award No. 2022‐51181‐38240) and the USDA Agricultural Research Service (Project 8060-21220-008-000D).