Vitis hybrid cv. Tamiami is an F₁ hybrid resulting from the cross V. aestivalis “Fennell 6” × Vitis vinifera “Malaga.” In 2002, the segregating modified-BC₁ (mBC₁) 02-3512 family with “Tamiami” resistance was developed by crossing B37-28 (“Tamiami” × V. vinifera) × C56-11 (V. vinifera), resulting in 94 full-sibling progenies. The cross was repeated in 2007, resulting in “Tamiami” derived family 07-3509 with 150 full-sibling progenies. To validate quantitative trait locus (QTL) results in an independent family, a third cross was made “Tamiami” × V. vinifera “Solbrio,” resulting in 300 full-sibling progenies. C56-11 and Solbrio are susceptible V. vinifera breeding parents in the USDA- Agricultural Research Service (ARS) grape breeding program at Parlier, CA.

As previously described, the progeny of 02-3512 were grown at USDA, ARS San Joaquin Valley Agricultural Sciences Center, Parlier, CA, on their roots at 4 m × 0.5 m spacing and were cane pruned on a single T-trellis (Ramming et al., 2012). No fungicides were applied. Highly susceptible V. vinifera cv. Ruby Seedless plants were planted in every 15th vine as an inoculum source and to check for the amount of natural GPM infection. Once the plants started fruiting (in their 3rd growing season), natural GPM severity assessments were performed between July and October for 9 years, often with two ratings per year: an “early” rating after susceptible “Ruby Seedless” vines had >50% foliar coverage and a “late” rating 2 or 2 months later. Foliar disease severity ratings were recorded based on percent of foliage covered by GPM colonies: 1 = no visible infection; 2 = very few small colonies; 3 = <50% coverage; 4 = >50% coverage. Analogous ratings were recorded for three other tissues: rachis, berries, and stems.

To evaluate resistance more quantitatively AUDPC values were calculated for the same planting in 2016 and 2017 based on bi-weekly ratings using a modified Horsfall-Barratt scale (Horsfall and Barratt, 1945). Specifically, leaves, rachis, berries, and stems each were evaluated on a 1 to 9 scale: 1 = no disease, 2 = 0 <2.5%, 3= 2.5 <10%, 4 = 10 <20%, 5 = 20 <25%, 6 = 26 <45%, 7 = 45 <60%, 8 = 61 <80%, 9 = 81 <100%. For each vine, AUDPC was calculated by taking the average disease rating of two sequential time points multiplied by the time interval between those points and calculating the total sum across all time intervals (Shaner and Finney, 1977).

Progeny of 02-3512 was screened for resistance to GPM as young vines grew from dormant cuttings in USDA-ARS, Parlier, CA greenhouses as described previously by Ramming et al. (2012). Vines were grown in 6.35 cm² × 17.8 cm high Anderson pots (Anderson Die and Manufacturing, Portland, OR). Two GPM susceptible V. vinifera “Ruby Seedless” vines were placed in the middle of each tray to provide the natural inoculum source surrounded by 13 randomly placed test vines. Symptoms were evaluated when 70% of susceptible control “Ruby Seedless” leaves exhibited conidiating colonies. After the first evaluation in September 2006, the epidemic was allowed to progress further for the second evaluation in November 2006. The presence of mycelia was confirmed by microscopy. Incidence (percentage of leaves with one or more GPM colonies) and severity (percentage of foliage covered by GPM colonies) scores were recorded separately, and then were averaged.

To phenotype resistance in controlled experiments as described previously in detail by Cadle-Davidson et al. (2016), one leaf from each of four replicate shoots was sampled at the 3rd node and represented as two replicate surface-sterilized disks on eight agar trays, each a replicate block containing all progeny and check samples. E. necator isolate NY90 (originally from V. vinifera “Chardonnay” in Burdett, NY; Barba et al., 2015) was inoculated at 2 × 10⁵ conidia/ml in 0.001% Tween-20 using a handheld atomizer (Preval, Coal City, IL). Inoculated samples were maintained at 23°C in a 12-h photoperiod for 9 days until processing. Samples were cleared in 3:1 v/v ethanol:acetic acid until the tissue was completely bleached, then stained with Coomassie Brilliant Blue R-250 (Bio-Rad Laboratories, Richmond, CA), and mounted for hyphal transect data collection using a compound microscope. Hyphal transects count the number of interceptions of individual hyphae crossing one of two axial transects (vertical and horizontal) of the entire disk at X 200.

DNA of 02-3512 and 07-3509 families was extracted from 1 cm or smaller leaves using QIAGEN DNeasy 96 Plant Kits with 3% w/v PVP40 added to the lysis buffer. DNA of “Tamiami” × “Solbrio” and the rhAmpSeq diversity panel were extracted from 1-cm or smaller leaves by Intertek AgriTech (Intertek ScanBi Diagnostics, Elevenborgsvägen 2, Alnarp, Sweden) using an automated magnetic bead pipeline with sbeadex kit and a PVP-amended buffer provided by LGC (Teddington, United Kingdom). DNA was used without within-plate normalization but was quantified by QuantiFluor fluorescence as per the instructions of the manufacturer (Promega, Madison, WI) and diluted to target about 30 to 60 ng of template DNA, based on the average concentration of eight quantified samples. rhAmpSeq amplification enrichment using the 2000 marker panel was conducted following the protocol of the manufacturer (IDT-DNA, Redwood City, CA), but with half-volume reactions (Zou et al., 2020). In brief, the first polymerase chain reaction (PCR) used 14 cycles with annealing temperature at 61°C for each sample. The PCR products were diluted at 1:20 and indexed with Integrated DNA Technologies, Inc (IDT), Redwood City, CA, USA indexing primers using 24 cycles with an annealing temperature at 60°C. The indexed PCR products were pooled, cleaned with Agencourt AMPure beads, quantified, and sequenced on an Illumina (Illumina, San Diego, CA, USA) NextSeq (2 × 150 bp) sequencer.

rhAmpSeq sequencing data were analyzed using the analyze_amplicon.pl Perl script (https://github.com/avinashkarn/analyze_amplicon/blob/master/analyze_amplicon.pl; Zou et al., 2020) to obtain haplotype variants (per marker) across all vines in each family, to generate the haplotype to genotype (hapgeno) file. Monomorphic markers and markers with >75% missing data in the hapgeno file were manually removed from further analyses. Finally, using a custom Perl script, haplotype_to_VCF.pl (https://github.com/avinashkarn/analyze_amplicon/blob/master/haplotype_to_VCF.pl; Zou et al., 2020), the four most frequent haplotype alleles for each marker (representing the four possible haplotypes within a heterozyogous diploid bi-parental family) in the hapgeno file were converted to a variant call format (VCF) file, where each haplotype allele of a marker was converted to a pseudo A, C, G, or T allele, for further marker validation analyses that are discussed hereafter. Genotype data are available in NCBI BioProject PRJNA281110.

For each grapevine family, the VCF file was imported into Trait Analysis by aSSociation, Evolution and Linkage (TASSEL) version 5.2.51 software (Bradbury et al., 2007) for genotype imputation using the default parameters of the LD-kNNi imputation plugin, also known as LinkImpute (v1.1.4) (Money et al., 2015) as previously described by Zou et al. (2020). Post-imputation, vines with >90% missing data were removed from the analysis. Quality control focused on the removal of vines with genotypes appearing to be due to self-pollination, pollen contamination, or other contaminants. Post-imputed and filtered markers were used to calculate a genome-wide pairwise identity-by-state (IBS) distance matrix for each family in TASSEL software using 1–IBS, followed by multidimensional scaling (MDS) analysis (Bradbury et al., 2007), and the first two principal coordinates were graphically depicted using the R statistical software. Vines not grouping with full-sibling progeny were removed from linkage mapping analysis (Supplementary Table 1). In addition, based on parental alleles, for each progeny vine, we calculated the percentage of loci inconsistent with Mendelian expectations using bcftools plugin mendelian (cite), and the outlier vines with high-mendelian inconsistencies (Supplementary Table 2) removed from the downstream analyses.

Genetic maps were constructed in Lep-MAP3 v.0.2 (LM3; Rastas, 2017) as previously described by Zou et al. (2020) using the VCF file of the post-imputed and filtered markers and curated full-siblings, with the following LM3 modules: (1) ParentCall2 to call parental genotypes; (2) the resulting output was filtered by using Filtering2 (parameter dataTolerance = 1.00E-6) with markers filtered out based on monomorphism or a two-sided χ² (chi-squared) test (testing if the allele ratio has significantly deviated from the expected mendelian ratio, at the above tolerance thresholds); (3) SeparateChromosomes2 module was used to identify linkage groups; and (4) OrderMarkers2 to compute the sex-specific parental genetic distances of the markers in each linkage group using 20 iterations per group. Correlations of genetic and physical distances of individual markers per chromosome were plotted to evaluate the consistency of the maps, genome organization, and structural variation.

The phased output data from OrderMarkers2 were converted into genotypes (1 1, 1 2, 2 1, 2 2) using map2genotypes.awk script and further into R/qtl 4-way cross-format, where,1 1 = AC = 1, 1 2 = AD = 2, 2 1 = BC = 3 and BD = 2 2 = 4. QTL mapping was performed in the statistical software R version 3.6.3 (2020-02-29) using the qtl package (Broman et al., 2003) focused on the sex-averaged genetic map, assuming a common recombination rate in both the parents. Genotype probabilities were calculated using calc.genoprob with step = 0 (probabilities were calculated only at the marker locations) and assumed genotyping error rate of 1.0E-4. The scanone function was implemented to perform standard interval mapping with Haley-Knott regression (Haley and Knott, 1992). Logarithm of the odds (LOD) significance thresholds were determined by 1,000 permutation tests at an alpha of 0.1 to detect minor QTLs. QTL support interval was determined calculating 1.5—LOD support intervals by using lodint. The percentage of variance explained in the context of a full additive model was calculated using fitqtl.

Pearson Correlation was calculated for AUDPC severity ratings among tissues in 2016 and 2017 using the PerformanceAnalytics package, and distributions were plotted using the ggplot2 package in R software version 3.6.3 (2020-02-09). Best linear unbiased estimates (BLUEs), also known as least squares, were calculated to avoid shrinkage toward the mean using the glm function for the vines for each trait in Trait Analysis by aSSociation, Evolution and Linkage (TASSEL) version 5.2.51 software, where years were included in the model as main effects and the effects of vines as fixed.

The presence/absence of the resistance locus REN11 among the germplasm was determined by the Bayesian hypothesis testing, with a null hypothesis that no resistant allele is present at the REN11 locus (Zou et al., 2020), and an alternative hypothesis that each allelic site at the REN11 locus contains the resistant allele. We converted the genotype to a 0/1 code, where 0 stands for the susceptible allele, and 1 stand for the resistant allele. Then, the hypothesis can be written as H0 ~ B (n0, p0) and H1~ B (n1, p1), where B (n, p) is a binomial distribution, n denotes the total number of allelic sites in this locus, p stands for the probability of the sites with desired alleles. The ratio of the likelihood of the alternative hypothesis to the null hypothesis is indicated by the Bayes Factor (BF). Above 1, larger BF values indicate greater support that the resistant alleles are present in this locus. The ranked log (BF) follows a sigmoidal distribution with REN11+ individuals having large values, REN11– individuals having small values and between them a steep slope in ranked log (BF) among full siblings with recombinations at the locus and related accessions lacking the complete REN11 haplotype. The maximum differential of ranked log (BF) corresponded with BF values in the range of 4,094–10,266. This range included two individuals: Vitis shuttleworthii 2366 (BF = 5,685) and resistant progeny Y533_224 (BF = 7,727), so a BF threshold of 7,000 was set between them.

Following natural infection in the vineyard, GPM severity rating distributions were skewed right regardless of whether the data were recorded on a 1–4 severity scale or AUDPC (Figure 1). The mode for the 1–4 severity scale was 1, indicating most commonly vines had no visible infection on all tissues observed. Similarly, for AUDPC, the mode for all tissues indicated vines most frequently had minimal GPM on any tissue, but leaves were slightly less skewed than other tissues. Susceptible vines had quantitative variability in disease severity on most tissues (Figure 1). Quantitative AUDPC was highly correlated among rachises and berries (r = 0.90–94) and well-correlated among leaves and those tissues (r = 0.61–0.84). In contrast, while stems had a similar intensity of disease severity as compared to the other tissues, stems were not as well-correlated with other tissues (r = 0.28–0.53).

The curated genetic map consisted of 1,058 rhAmpSeq core genome markers, forming 19 linkage groups after filtering for non-informative and distorted markers (Supplementary Tables 3, 4). The genetic map sizes of the parents ranged were 1,084.2 cM for B37-28 and 1,151.4 cM for C56-11. As with previous rhAmpSeq maps, both the maps were highly correlated to the physical map from PN40024 12X.v2 (Zou et al., 2020), with an average genome-correlation coefficient (r) of 0.92 and covering 93.4% of the genome physical distance (Figure 2, Supplementary Table 3).

We analyzed the genetics of resistance in the vineyard, greenhouse, and laboratory. A locus from B37-28 on the chromosome (Chr) 15 near 13–15 Mbp was consistently detected on all vineyard-rated tissues in most years, explaining up to 96.2% of the phenotypic variation on the berries (Table 1, Figure 3, Supplementary Table 5). The locus was assigned the name REN11. Only in the first year of the study (2006, a year with data for only 49 vines, Supplementary Table 5) did REN11 not significantly predict foliar vineyard severity; in 2006, Chr 1 near 7–10 Mbp was the only significant locus, explaining 38.0% of phenotypic variation on leaves.

Three other loci were detected in multiple environments. On leaves, proximal Chr 6 (0.50–2.03 Mbp) explained 31.5% of the phenotypic variance late in 2010 and was epistatic with REN11, but distal Chr 6 (17.68–21.33 Mbp) explained less variance in 2011 and 2013 as well as on the single case when it significantly affected rachis or stem severity. In all the significant cases for Chr 6, a specific haplotype from each parent was required for significance, with that haplotype combination predicting increased susceptibility. Two loci from B37–28 repeatedly explained significant variation in stem severity: Chr 8 (3 ratings) and 9 (6 ratings). In quantitative AUDPC ratings, these two loci and REN11 had additive effects in shifting mean stem GPM severity lower (Figure 4).

Six other QTL were detected in a single environment: two of these were from the Parlier greenhouse ratings that contributed to the hypothesis tested in this study (Ramming et al., 2012), each being significant at a different stage of the epidemic. Chr 2 was the only QTL with reduced severity inherited from C56-11, and explained differences late in the greenhouse epidemic, 2 months after the initial ratings. Epistatic interactions were only significantly detected in 1 year each for leaves, berries, and stems.

In 2016 and 2017, a detailed AUDPC monitoring was conducted, and results reflected the observations in most years that REN11 on Chr 15 significantly and strongly predicted disease severity on leaves, rachis, and berries, but additional loci were involved in resistance on stems (Figure 5). REN11 was independently validated in the “Tamiami” × “Solbrio” AUDPC (Figure 5).

To narrow in on the genetic basis of REN11 resistance, we identified seedlings with recombinations within the QTL in relation to AUDPC phenotype from 2016. rhAmpSeq markers delimited the region to 13.7–15.3 Mbp on Chr 15, with the marker 15_13822901 being the most predictive of disease severity (Figure 6). This region of the reference PN40024 12X.v2 genome contains 149 genes including an NLR (nucleotide-binding, leucine-rich-repeat) resistance gene (VIT_215s0048g00680) that was previously shown to be upregulated upon inoculation of susceptible grapevines (Lin et al., 2019).

The resistance alleles shown in Supplementary Table 6 can be used directly for the marker-assisted selection by AmpSeq. To identify additional germplasm with the REN11 resistance haplotype, we screened the existing VitisGen database of 20,651 vines for these desirable resistance alleles. Based on a BF threshold of 7,000, 396 vines were identified as having the REN11 locus. These included the resistant progeny described in this study, 28 breeding lines (of which 25 are known to be related to Tamiami and two are from a breeder who frequently used Tamiami), and four interspecific hybrid accessions from the ARS repository at Winters, CA (Supplementary Table 7). These three breeding lines and four accessions represent a false positive rate of <0.034% (7 of 20,651). However, the four accessions differ in their rhAmpSeq haplotype at five markers (Chr15_12844486, Chr15_13063122, Chr15_15543151, Chr1515944126, and Chr15_16066225) flanking the REN11 locus defined in Figure 6, suggesting that they do not have the REN11 resistance allele. Thus, while the 18 haploblock markers presented in Figure 6 work well for the statistical prediction of REN11 and identification of recombinant siblings, these five markers work well for eliminating rare false positives in a diversity panel.

To narrow in on the genetic basis of stem resistance, we identified seedlings with recombinations within the QTL. Because of the complex inheritance of the stem resistance, this approach failed to define the Chr 8 locus (Figure 7). For six key recombinant vines in relation to July 2014 phenotype, rhAmpSeq markers delimited the Chr 9 region to 17.0–19.5 Mbp (Figure 7), which is adjacent to an NLR locus in PN40024 12X.v2.