Towards Marker-Assisted Breeding for Black Rot Bunch Resistance: Identification of a Major QTL in the Grapevine Cultivar ‘Merzling’

Black rot (BR), caused by Guignardia bidwellii, is an emergent fungal disease threatening viticulture and affecting several mildew-tolerant varieties. However, its genetic bases are not fully dissected yet. For this purpose, a segregating population derived from the cross ‘Merzling’ (hybrid, resistant) × ‘Teroldego’ (V. vinifera, susceptible) was evaluated for BR resistance at the shoot and bunch level. The progeny was genotyped with the GrapeReSeq Illumina 20K SNPchip, and 7175 SNPs were combined with 194 SSRs to generate a high-density linkage map of 1677 cM. The QTL analysis based on shoot trials confirmed the previously identified Resistance to Guignardia bidwellii (Rgb)1 locus on chromosome 14, which explained up to 29.2% of the phenotypic variance, reducing the genomic interval from 2.4 to 0.7 Mb. Upstream of Rgb1, this study revealed a new QTL explaining up to 79.9% of the variance for bunch resistance, designated Rgb3. The physical region encompassing the two QTLs does not underlie annotated resistance (R)-genes. The Rgb1 locus resulted enriched in genes belonging to phloem dynamics and mitochondrial proton transfer, while Rgb3 presented a cluster of pathogenesis-related Germin-like protein genes, promoters of the programmed cell death. These outcomes suggest a strong involvement of mitochondrial oxidative burst and phloem occlusion in BR resistance mechanisms and provide new molecular tools for grapevine marker-assisted breeding.


Introduction
Grapevine is one of the most renowned fruit crops in the world. Europe grows the highest proportion of grapes (50%), followed by Asia (23%), the Americas (20%), Africa (5%), and Oceania (2%) [1]. Besides the making of noble products, such as wine, table grapes, and raisins, grapevine provides transformed products, such as juices, jams, and jellies of local interest, as well as wine industry by-products, such as must, marc distillates, marc pulp, tartaric acid, seed oil, and vinegar. All these commodities follow a growing trend and include grapevine among the multi-purpose crops. Most of the grapes produced worldwide are from cultivars of the Eurasian Vitis vinifera L. subsp. vinifera, while the rest are from interspecific hybrids with other American and East/Central Asian Vitis species [2,3].
Regrettably, the high susceptibility of vinifera varieties to most fungal diseases, such as downy mildew, gray mold, and powdery mildew [4], leads to the intensive use of fungicides in viticulture. To reduce the usage of chemicals, a potential solution that is being gradually adopted is offered by the development of novel cultivars resistant to the main pathogenic threats. Actually, genetic improvement for biotic stress resistance is a valuable strategy to embrace the principles of the European Green Deal-which will be greenhouse experiments, disease symptoms were visible after 14 days post-inoculation (dpi), while under natural field conditions, disease progression took longer (up to 21 dpi). The performance of the parents was stable, with 'Teroldego' always resulting fully susceptible and 'Merzling' highly resistant, and the phenotypic distribution was non-normal in all experiments (Shapiro-Wilk p < 0.05): for leaves and shoot internodes it was skewed towards susceptibility, while for bunches towards resistance ( Figure 2). All phenotypic data are reported in Table S1.
internode, FS) field inoculation in 2021. In the untreated vineyard, inoculated leaves presented also downy mildew lesions (Figure 1a). Two experiments were conducted on bunches (or clusters) in the two growing seasons, 2020 (field clusters, FC1) and 2021 (FC2). In each FC inoculation trial, the entire population was screened in duplicate (FC1a and b and FC2a and b). The evolution of BR lesions on the berry and symptomatic bunches is shown in Figure 1b,c. A five-step resistance rating scheme was used, i.e., (1) very low, (3) low, (5) medium, (7) high, and (9) very high (for resistance descriptors see M&M 4.2). Under the high pressure of the greenhouse experiments, disease symptoms were visible after 14 days post-inoculation (dpi), while under natural field conditions, disease progression took longer (up to 21 dpi). The performance of the parents was stable, with 'Teroldego' always resulting fully susceptible and 'Merzling' highly resistant, and the phenotypic distribution was non-normal in all experiments (Shapiro-Wilk p < 0.05): for leaves and shoot internodes it was skewed towards susceptibility, while for bunches towards resistance (Figure 2). All phenotypic data are reported in Table S1. Foliar lesions showed the typical circular shape, with a light brown to reddish color and rounded and darker borders. In the field, the co-occurrence of downy mildew (dark brown lesions) was evident. (b) Disease progression on the berry: after the appearance of a small dot, the lesion expanded concentrically, rapidly producing pycnidia; later, the berry dried up, turning into black mummy. (c) Clusters showing different degrees of resistance, from the left: very low (1), medium (5), high (7) (for resistance descriptors see M&M 4.2). Foliar lesions showed the typical circular shape, with a light brown to reddish color and rounded and darker borders. In the field, the co-occurrence of downy mildew (dark brown lesions) was evident. (b) Disease progression on the berry: after the appearance of a small dot, the lesion expanded concentrically, rapidly producing pycnidia; later, the berry dried up, turning into black mummy. (c) Clusters showing different degrees of resistance, from the left: very low (1), medium (5), high (7) (for resistance descriptors see M&M 4.2).

Genotyping and Linkage Mapping
The M × T segregating population was genotyped by means of the GrapeReSeq Illumina 20K SNPchip. The sample of one F1 individual with a call rate of 66.5% was excluded from the dataset, which was, hance, reduced to 146 individuals. Of the 18,071 available SNPs [30], 221 were manually edited. After SNP scoring, a total of 7175 robust SNPs (39.7%) were retained and 10,898 (60.3%) discarded, of which the majority-9312, that is 51.5% of the total number of SNPs-were monomorphic.

Genotyping and Linkage Mapping
The M × T segregating population was genotyped by means of the GrapeReSeq Illumina 20K SNPchip. The sample of one F1 individual with a call rate of 66.5% was excluded from the dataset, which was, hance, reduced to 146 individuals. Of the 18,071 available SNPs [30], 221 were manually edited. After SNP scoring, a total of 7175 robust SNPs (39.7%) were retained and 10,898 (60.3%) discarded, of which the majority-9312, that is 51.5% of the total number of SNPs-were monomorphic.
An integrated map and the related parental maps of 'Merzling' and 'Teroldego' were constructed with the entire marker dataset of 7175 SNPs and 194 SSRs (whole integrated map, wIN map), producing 19 linkage groups (LGs) with a minimum grouping logarithm of the odds (LOD) of 14. After excluding poorly performing markers, the wIN map presented 7358 markers, of which 3249 were unique, i.e., with no identical segregation pattern, and used to produce reduced (rIN, rM, and rT) maps with only non-redundant markers (Table 1). Segregation patterns were coded following the JoinMap ® segregation classification [31]. The complete and reduced datasets differed predominantly in the number of biallelic markers heterozygous only in one parent. In fact, the number of lm × ll and nn × np markers were reduced to a third, hk × hk resulted in a reduction of only a sixth, and the SSRs with segregation type ab × cd and ef × eg were all retained in the reduced dataset (Table S2). The description of the maps at single LG level is available in Table S3. . The black rot resistance trait did not follow a normal distribution in any of the experiments (Shapiro-Wilk p < 0.05). Both greenhouse (GL1, GL2, GL3, GL median and minimum) and field (FL, FS) shoot evaluations were skewed towards susceptibility, while cluster resistance trials (FC1, FC2, FC median and minimum) and sub-experiments (FC1a and b, FC2a and b) were skewed towards resistance. On the y axis the frequency of F1 individuals is shown, while the x axis indicates the degree of resistance: 1 = very low; 3 = low; 5 = medium; 7 = high; 9 = very high (for resistance descriptors see M&M 4.2). Abbreviations: FC = field cluster; FL = field leaf; FS = field shoot internode; GL = greenhouse leaf; M = 'Merzling'; T = 'Teroldego'.
An integrated map and the related parental maps of 'Merzling' and 'Teroldego' were constructed with the entire marker dataset of 7175 SNPs and 194 SSRs (whole integrated map, wIN map), producing 19 linkage groups (LGs) with a minimum grouping logarithm of the odds (LOD) of 14. After excluding poorly performing markers, the wIN map presented 7358 markers, of which 3249 were unique, i.e., with no identical segregation pattern, and used to produce reduced (rIN, rM, and rT) maps with only non-redundant markers (Table 1). Segregation patterns were coded following the JoinMap ® segregation classification [31]. The complete and reduced datasets differed predominantly in the number of biallelic markers heterozygous only in one parent. In fact, the number of lm × ll and nn × np markers were reduced to a third, hk × hk resulted in a reduction of only a sixth, and the SSRs with segregation type ab × cd and ef × eg were all retained in the reduced dataset (Table S2). The description of the maps at single LG level is available in Table S3.
The unique bins (loci) of the IN map were 1613, with an average inter-locus gap distance of 1.04 centimorgan (cM) and a map density of 0.96 bins/cM, while the T parental map presented a lower density of 0.56 bins/cM, with the higher inter-locus gap of 1.80 cM, due to the reduced number of loci per map unit (Table 1, Figure 3). Instead, the M parental map resulted the shortest, with 1452 cM and a density of 0.58 bins/cM. The lack of polymorphic markers in some regions determined the presence of major gaps in the maps. The integrated map presented the biggest gap of 9.5 cM on LG15. In the M map, LG12 showed a big lack of markers at the distal arm, with a gap of 15.7 cM. On LG18, the T map presented the biggest gap among all maps, of about 28 cM. Since the grouping of the integrated map is based on the recombination of alleles in the progeny, linked markers in the integrated map may not be linked in the parental ones. This occurred for LG15 in the M map, where the lack of polymorphism did not allow for linking the two portions of the linkage group (the distal one having only two bins), producing a shorter-and the shortest-linkage group of about 24 cM (Figure 3). A detailed correspondence between the whole integrated and parental maps is depicted in Figure S1. A higher quality of doubled-haploid (dh) maps were obtained using the Regression mapping algorithm, with Kosambi's instead of Haldane's mapping function, suggesting crossover interference. The dhM map (Table S3, Figure S2) was generated based on a minimum grouping LOD of 14; it measured 943 cM and comprised 998 unique markers, sorted in 791 loci, for an average inter-locus distance of 1.19 cM. As expected, the distal branch of LG15 grouped separately prematurely (LOD 3), determining the minimum value of 9 bins observed within all the developed maps. The major gap of 13.2 cM was present on LG6. The dhT map (Table S3, Figure S3) was based on a minimum grouping LOD of 8; it resulted 1346 cM long, counting 1137 markers and 917 unique bins, which determined a higher mean distance of 1.47 cM between loci.  LGs. Marker density is represented by a color scale from blue (lowest) to red (highest). Rendering adapted from LinkageMapView [32].

Quantitative Trait Locus Analysis
Due to the non-normal distribution of the phenotypic traits (Figure 2), Kruskal-Wallis (KW) non-parametric analysis was applied on the wIN map to highlight the most strongly associated markers. The analyses with GL and FC overall minimum and median values gave the same results within each experiment type; therefore, only GL min and FC min values were included in the following analysis.
To define resistance QTLs, Multiple QTL Mapping (MQM) analyses were performed on the resistant parent dhM map. The segregation of the berry color trait worked fine as a positive control for the QTL analyses, identifying the proper position of the locus BeCo [33]. All BR experimental datasets revealed only one single QTL on LG14. The significance always exceeded the most stringent p < 0.01 ( § § § § ) genome-wise (GW)-threshold (Table S4), but with a different peak position for the experiments on shoot (leaf and shoot internode) compared to bunches ( Figure 4, Table 2), and a minimum phenotypic variance explained LGs. Marker density is represented by a color scale from blue (lowest) to red (highest). Rendering adapted from LinkageMapView [32].

Quantitative Trait Locus Analysis
Due to the non-normal distribution of the phenotypic traits (Figure 2), Kruskal-Wallis (KW) non-parametric analysis was applied on the wIN map to highlight the most strongly associated markers. The analyses with GL and FC overall minimum and median values gave the same results within each experiment type; therefore, only GL min and FC min values were included in the following analysis.
To define resistance QTLs, Multiple QTL Mapping (MQM) analyses were performed on the resistant parent dhM map. The segregation of the berry color trait worked fine as a positive control for the QTL analyses, identifying the proper position of the locus BeCo [33].
All BR experimental datasets revealed only one single QTL on LG14. The significance always exceeded the most stringent p < 0.01 ( § § § § ) genome-wise (GW)-threshold (Table S4), but with a different peak position for the experiments on shoot (leaf and shoot internode) compared to bunches ( Figure 4, Table 2), and a minimum phenotypic variance explained (PVE) of 15.7%, hence making all of these major QTLs [34].
Due to the non-normal distribution of the phenotypic traits (Figure 2), Kruskal-Wal-lis (KW) non-parametric analysis was applied on the wIN map to highlight the most strongly associated markers. The analyses with GL and FC overall minimum and median values gave the same results within each experiment type; therefore, only GL min and FC min values were included in the following analysis.
To define resistance QTLs, Multiple QTL Mapping (MQM) analyses were performed on the resistant parent dhM map. The segregation of the berry color trait worked fine as a positive control for the QTL analyses, identifying the proper position of the locus BeCo [33]. All BR experimental datasets revealed only one single QTL on LG14. The significance always exceeded the most stringent p < 0.01 ( § § § § ) genome-wise (GW)-threshold (Table S4), but with a different peak position for the experiments on shoot (leaf and shoot internode) compared to bunches ( Figure 4, Table 2), and a minimum phenotypic variance explained (PVE) of 15.7%, hence making all of these major QTLs [34].  Table S11.
The QTL associated with shoot resistance was located between 40.584 and 40.626 cM (two-LOD support interval). The associated marker GF14-42 (40.626 cM) explained from a minimum of 20.4% (GL2) to a maximum of 29.2% (GL1) of the phenotypic variability, and it coincided with the marker segregating with Resistance to Guignardia bidwellii (Rgb)1 locus [21]. The second marker C14_26440029 (40.584 cM) segregated with the QTL with the same LOD and PVE as GF14-42 ( Figure 4, Table 2).
The berry resistance QTL was located between 31.241 and 31.533 cM. The most significantly associated marker was C14_20097630ae, explaining from 20.3% (FC1) up to 79.9% (FC2) of the variability of the trait. A second marker co-segregated with the QTL, with the PVE ranging from 18.4% (FC1) to 75.4% (FC2). The QTL was also confirmed by analyzing the phenotypic distribution of the four FC sub-experiments (FC1a, FC1b, FC2a, and FC2b). According to previous nomenclature [21], the locus was named Rgb3. The KW analyses showed the highest significance (p < 0.000005) for all the associated markers to both loci.
To deepen the genetic basis of BR resistance, recombinant progeny individuals at the locus were studied. Eight individuals were recombinant at Rgb3 locus, clearly separating the resistant (blue) and the susceptible (orange) haplotype phase ( Figure 5). Instead, this approach failed to reduce the Rgb1 locus, since recombinant individuals defined a wider region compared to the MQM QTL analysis, comprised between 40.584 and 40.684 cM. Table 2. Description of the significant (p < 0.01) QTLs on LG14 associated with black rot resistance identified in 'Merzling'. When a blank space is given, it means that the value was equal to that of the cell above it. Abbreviations: cM = centimorgan; FC = field cluster; FL = field leaf; FS = field shoot internode; FC = field cluster; GL = greenhouse leaf; LG = linkage group; LOD = logarithm of odds; min = minimum; PVE = phenotypic variance explained.   Table S11, for resistance descriptors see M&M 4.2. Abbreviations: dhM = doubled haploid; FC = field cluster; GL = greenhouse leaves.

Candidate Genes
The QTL region associated with the BR resistance of shoot (FL, FS, and GL) colocalized with the Rgb1 locus and co-segregated with the same GF14-42 marker [21]. To define the physical position of the QTL, the latest version 4 of the genome assembly from the PN40024 accession (PN40024.v4) [35] was used. The alignment of the highly dense map The haplotype phases are shown in orange and blue, and the color code is used to give a putative association with the overall minimum resistance level of GL and FC experiments. Missing genotypic data are shown in white. The black boxes define the minimal recombinant region, and co-segregating markers significantly associated with the QTLs are highlighted in bold. For the official marker names see Table S11, for resistance descriptors see M&M 4.2. Abbreviations: dhM = doubled haploid; FC = field cluster; GL = greenhouse leaves.

Candidate Genes
The QTL region associated with the BR resistance of shoot (FL, FS, and GL) colocalized with the Rgb1 locus and co-segregated with the same GF14-42 marker [21]. To define the physical position of the QTL, the latest version 4 of the genome assembly from the PN40024 accession (PN40024.v4) [35] was used. The alignment of the highly dense map developed in this work to this version of the grapevine reference genome allowed a size reduction of the locus from 2.4 Mb [21] to a region of 0.7 Mb, comprised between 26.6 and 27.3 Mb on chromosome 14. A total of 22 genes have been annotated within the Rgb1 two-LOD support interval. The associated SNP marker C14_26440029 locates within the coding sequence of the Vitvi14g01631 gene predicting for an uncharacterized membrane protein.
The expanded region to the nearest non-associated flanking markers comprehended 103 genes (Table S5). The enriched categories of the Rgb1 locus based on gene ontology (GO) analysis of the 82 genes having GO description, and the corresponding gene lists, are available in Table S6. Three out of seven enriched gene sets were associated with phloem development (Figure 6b) and comprehended a gene cluster encoding for five unknown proteins (Vitvi14g01641, Vitvi14g01642, Vitvi14g01646, Vitvi14g02997, and Vitvi14g01653), with a fold enrichment (FE) from 10.3 to 124.7 ( Figure 6a). For all of them, the best Arabidopsis match was the sieve element occlusion B (AT3G01680) (Table S5), a scaffold protein required to form the phloem filament matrix in sieve elements. Three additional enriched classes ( Figure 6b) (namely 'Proton-transporting ATP synthase complex assembly', 'Protontransporting two-sector ATPase complex assembly', and 'Mitochondrion') were all related to the mitochondrion (FE from 3.6 to 124.7, Figure 6a). The ATP synthase mitochondrial F1 complex assembly factor 2 (Vitvi14g01626) and Ku70-binding (Vitvi14g01636) genes were present in all three classes, together with a Pentatricopeptide (PPR) repeat-containing protein gene (Vitvi14g01632) in the class 'Mitochondrion' (Table S5). Lastly, another enriched category (FE 19.0, Figure 6a) was represented by the 'Nucleoside triphosphate biosynthetic process' (Figure 6b), composed of three genes coding for a dehydration-responsive protein (Vitvi14g01667), an ubiquinol-cytochrome c reductase cytochrome c1 (Vitvi14g01592), and a thymidylate kinase (Vitvi14g01593) ( Table S5).
The Rgb3 locus was located within 19.8 and 19.9 Mb on chromosome 14, defining a region of 0.2 Mb that comprises 8 functionally annotated genes (Table S7). The most significantly associated marker, C14_20097630ae, was detected within an intron of the gene Vitvi14g01119, coding for a predicted phosphatidic acid phosphatase alpha (PAPα). The other highly associated marker, VMC2C3, was instead located within an intron of the gene Vitvi14g01125, coding for a HEAT repeat-containing protein. Two functionally annotated genes, Vitvi14g01120 and Vitvi14g01124, were predicted as an ATP binding protein and a plastid movement-impaired 15 (PMI15) protein, respectively. PMI15 is required for the so-called "avoidance response" that relocates chloroplasts on the anticlinal side of the cells in response to high-intensity blue light. The expanded region to the flanking neighboring markers (19.7 to 22.1 Mb) comprehends a list of 106 genes. No classical resistance (R)-gene was present, but, of the 66 genes having GO description, a significant enrichment (FE from 4.2 to 53.6) of genes predicted as germin-like protein 3 (Glp3) was detected. In fact, a cluster of 10 Glp3 was present in the VCost.v3 (V3) annotation [36], and it resulted in belonging to seven of the nine enriched gene sets (Figure 6d, Table S8). The enriched 'Extracellular region' category, beyond the 10 Glp3 genes, also contains two Non-specific lipid-transfer protein 2 (LTP 2) genes (Vitvi14g02869 and Vitvi14g01128), while the enriched 'Transition metal ion binding' category counted additional four genes that encoded, respectively, a delta 7-sterol-C5-desaturase (Vitvi14g01152), a mannose-6-phosphate isomerase (Vitvi14g01178), a cytokinin hydroxylase (CYP735A1, Vitvi14g01195), and a DNA-directed RNA polymerase subunit beta' (Vitvi14g01068). (Figure 6c, Table S7). Finally, other two enriched pathways were linked to lipid transport and localization (FE 16.8 and 15.1), and comprehended the same four genes that predict a HEAT repeat-containing protein (Vitvi14g01125), two non-specific lipid-transfer protein 2 (Vitvi14g02869, Vitvi14g01128), and a synaptotagmin (Vitvi14g01186).
(Vitvi14g01636) genes were present in all three classes, together with a Pentatricopeptide (PPR) repeat-containing protein gene (Vitvi14g01632) in the class 'Mitochondrion' (Table  S5). Lastly, another enriched category (FE 19.0, Figure 6a) was represented by the 'Nucleoside triphosphate biosynthetic process' (Figure 6b), composed of three genes coding for a dehydration-responsive protein (Vitvi14g01667), an ubiquinol-cytochrome c reductase cytochrome c1 (Vitvi14g01592), and a thymidylate kinase (Vitvi14g01593) ( Table S5). The Rgb3 locus was located within 19.8 and 19.9 Mb on chromosome 14, defining a region of 0.2 Mb that comprises 8 functionally annotated genes (Table S7). The most significantly associated marker, C14_20097630ae, was detected within an intron of the gene Vitvi14g01119, coding for a predicted phosphatidic acid phosphatase alpha (PAPα). The other highly associated marker, VMC2C3, was instead located within an intron of the gene Vitvi14g01125, coding for a HEAT repeat-containing protein. Two functionally annotated genes, Vitvi14g01120 and Vitvi14g01124, were predicted as an ATP binding protein and a plastid movement-impaired 15 (PMI15) protein, respectively. PMI15 is required for the so-called "avoidance response" that relocates chloroplasts on the anticlinal side of the cells in response to high-intensity blue light. The expanded region to the flanking neighboring markers (19.7 to 22.1 Mb) comprehends a list of 106 genes. No classical resistance (R)-gene was present, but, of the 66 genes having GO description, a significant enrichment (FE from 4.2 to 53.6) of genes predicted as germin-like protein 3 (Glp3) was detected. In fact, a cluster of 10 Glp3 was present in the VCost.v3 (V3) annotation [36], and it resulted in belonging to seven of the nine enriched gene sets (Figure 6d, Table S8). The enriched 'Extracellular region' category, beyond the 10 Glp3 genes, also contains two Non-specific lipid-transfer By blasting the sequence of Vitvi14g01075 Glp3 on PN40024.v4, five more genes were identified within the cluster, bringing the total number of genes to 15 ( Figure S4), whose structural annotation, hence, underwent manual curation. In two cases (Vitvi14g01106 and Vitvi14g02865, V3 annotation), the genes erroneously unify two open reading frames (ORFs). These were split, resulting in two new genes annotated in the PN40024 version 4.3 (V4.3) [37] as Vitvi14g04724 and Vitvi14g04720, which corresponded to VIT_14s0006g02400 and VIT_14s0006g02440 in the CRIBI version 1 annotation (V1) [38]. In addition, two new genes were annotated starting at position 19,604,399 (gene name nota available 1, NA1) and 19,612,558 (gene name NA2), which are in progress to be approved by the curators of PN40024.v4. The manual curation was supported only by the non-PN40024 RNAseq evidence, except for one gene (Vitvi14g01096) supported by PN40024 RNA-Seq data. Instead, four ORFs showed no supporting expression data (Table S9). The multiple alignment of the amino acid (aa) sequences revealed a very high mean identity of 94.3%, and the three sequences NA2, Vitvi14g01112, and Vitvi14g04724 and the two sequences Vitvi14g02863 and Vitvi14g02864 were identical within each other (Figures 7 and S4). Conversely, the lowest identity value (79.5%) was obtained by the pairwise alignment between Vitvi14g01108 and Vitvi14g02865 ( Figure S4). The alignment showed a greater identity of the second exon, while the region of increased variability was in the first exon. In particular, 10 genes shared a predicted protein length of 221 aa, while the genes NA1, Vitvi14g01106, and Vitvi14g01108 showed gaps in the first exon, and the genes Vitvi14g01080 presented an extra upstream sequence (13 aa), and the gene Vitvi14g02865 an extra downstream (7 aa) sequence (Figures 7 and S4).
Finally, publicly available data [39] were exploited to investigate the expression pattern of the genes belonging to the Rgb1 and Rgb3 loci on chromosome 14. The produced heat-maps showed that the clustered Glp3 genes on chromosome 14 are highly expressed in roots, seedlings, post fruit set pericarp, and senescent leaves (Figure 8), with the last three developmental stages known to be recalcitrant to BR infection. Contrarywise, the genes belonging to the Rgb1 locus did not show specific patterns ( Figure S5).  Finally, publicly available data [39] were exploited to investigate the expression pattern of the genes belonging to the Rgb1 and Rgb3 loci on chromosome 14. The produced heatmaps showed that the clustered Glp3 genes on chromosome 14 are highly expressed in roots, seedlings, post fruit set pericarp, and senescent leaves (Figure 8), with the last three developmental stages known to be recalcitrant to BR infection. Contrarywise, the genes belonging to the Rgb1 locus did not show specific patterns ( Figure S5).  [39]. Blue and red colors represent low and high expression levels, respectively (logarithmic data transformation).

High-Density SNP Coupled with SSR Genotyping Provides Highly Informative Linkage Maps
The construction of a map combining existing SSR to new SNP information was an Figure 8. Transcriptomic analysis of Rgb3 locus genes in different tissues and developmental stages using the VITVIZ 'Corvina' ATLAS [39]. Blue and red colors represent low and high expression levels, respectively (logarithmic data transformation).

High-Density SNP Coupled with SSR Genotyping Provides Highly Informative Linkage Maps
The construction of a map combining existing SSR to new SNP information was an effective strategy that eased the comparison of the results obtained in previous works. In fact, in this study, we were able to confirm the co-segregation of BR shoot resistance in the grapevine cultivar 'Merzling' (V. aestivalis var. lincecumii × V. rupestris × V. vinifera) with the SSR marker GF14-42, previously described by Rex et al. [21] as associated with the Rgb1 locus mapped on LG14 in the rootstock variety 'Börner' (V. riparia × V. cinerea) (Figure 9). Moreover, the developed highly dense map allowed for narrowing down the size of the QTL to 0.7 Mb. The same region has been identified also by Hausmann et al. [40] in a breeding population with a complex inter-specific background (V. aestivalis var. lincecumii × V. cinerea var. berlandieri × V. labrusca × V. rupestris × V. vinifera). In that work, the locus has been linked to the SSR marker UDV-095 [41], which was included in the current marker set and colocalized with GF14-42 in the M map ( Figure 9). Finally, Dalbó et al. [42] identified the same genomic region associated with both BR and the powdery mildew resistance locus Resistance to Erysiphe necator (Ren)2 in the maternal map of the population 'Illinois 547-1' (V. cinerea B9 × V. rupestris B38) × 'Horizon' (V. aestivalis var. lincecumii × V. labrusca × V. rupestris × V. vinifera). However, the resistance-linked RAPD marker CS25b [42] turned out to be monomorphic in the present study. Anchoring these four studies to the PN40024.v4 genome [35], an overlapping region of about 0.82 Mb emerged ( Figure 9). Based on these findings and on a preliminary genotypic screening of marker allele segregation in different genetic backgrounds-which will be thoroughly investigated in a future work-it can be asserted that this region of the chromosome 14 is crucial for BR resistance, and different haplotypes could co-exist, analogously to the Resistance to Plasmopara viticola (Rpv)3 multi-haplotype locus [43]. Another advantage of including SSR markers in highly dense SNP-based maps is their informativeness and technical ease of validation in different genetic backgrounds, whereas SNP markers need to be converted into haploblocks first and then checked back first to the original mapping population. Nevertheless, a bottleneck of SNPchip genotyping is that some regions are over-or underrepresented, due to the coverage and depth of previous versions of V. vinifera genomes. In addition, the 18K chip sequencing design dates to 2013 [30]; consequently, it lacks some Vitis species that nowadays have been largely employed into grapevine breeding programs, such as the American V. arizonica, V. riparia, and V. rupestris and the Asian V. amurensis, V. piasezkii, and V. romanetii listed in the 'Table of Loci for Traits in Grapevine Relevant for Breeding and Genetics' from VIVC [44]. To address this problem when working with those genetic backgrounds, genotyping-by-sequencing (GBS) is often preferred, e.g., [45,46], although GBS raises the problem of the low transferability of the outcomes (missing primer and sequence information).
Concurrently with more precise QTL mapping, the highly dense maps allowed the evaluation of recombinant individuals ( Figure 5), which is a good strategy to further investigate and narrow down the genetic basis of a particular trait when it determines a wide percentage of the phenotypic variability [29]. Instead, this approach fails to define loci with complex inheritance, e.g., for the powdery mildew stem resistance locus on chromosome 8 [29] and for the Rgb1 locus. In this regard, possible epistatic effects of the V. vinifera parent 'Teroldego' on the resist trait have been hypothesized. The 'Teroldego' genetic background did not influence the Rgb1 trait segregation, but further studies are ongoing to determine its contribution. In fact, a previous work [47] demonstrate that for the same resistance donor, the V. vinifera cultivar with which it is crossed can influence the expression of the trait. grams, such as the American V. arizonica, V. riparia, and V. rupestris and the Asian V. amurensis, V. piasezkii, and V. romanetii listed in the 'Table of Loci for Traits in Grapevine Relevant for Breeding and Genetics' from VIVC [44]. To address this problem when working with those genetic backgrounds, genotyping-by-sequencing (GBS) is often preferred, e.g., [45,46], although GBS raises the problem of the low transferability of the outcomes (missing primer and sequence information).  40,42] studies, based on marker position (Mb) on PN40024.v4 [35]. The box plots represent QTL intervals based on co-segregating markers (larger font) expanded (error bars) to the nearest non-associated flanking markers (smaller font). The Rgb3 associated to bunch resistance detected in the current study is depicted in purple. The Rgb1 locus associated with leaf resistance in the current and previous works [21,40,42] are colored with different shades of green. The same color code is given for the respective markers. The overlap of the analyses is highlighted in grey. Official names are shown for SSR markers, for the original names of SNP markers see Table S11.
Concurrently with more precise QTL mapping, the highly dense maps allowed the evaluation of recombinant individuals ( Figure 5), which is a good strategy to further investigate and narrow down the genetic basis of a particular trait when it determines a wide percentage of the phenotypic variability [29]. Instead, this approach fails to define loci with complex inheritance, e.g., for the powdery mildew stem resistance locus on Figure 9. Overview of the black rot resistance QTLs at chromosome 14. The map is an integration between the current and the previous [21,40,42] studies, based on marker position (Mb) on PN40024.v4 [35]. The box plots represent QTL intervals based on co-segregating markers (larger font) expanded (error bars) to the nearest non-associated flanking markers (smaller font). The Rgb3 associated to bunch resistance detected in the current study is depicted in purple. The Rgb1 locus associated with leaf resistance in the current and previous works [21,40,42] are colored with different shades of green. The same color code is given for the respective markers. The overlap of the analyses is highlighted in grey. Official names are shown for SSR markers, for the original names of SNP markers see Table S11.

Resistance Evaluation of Shoot and Bunch Reveals Organ-Specific QTLs
The present work corroborated the known foliar Rgb1 locus and reported the first genetic dissection of BR bunch resistance (Rgb3). This finding is a breakthrough since-despite the most valuable product of viticulture are of course grapes-in resistance assessments, mainly leaves are evaluated for experimental reasons. Analogously to other fungal diseases, the occurrence of BR infection on different organs during one season is well-established, introducing the concept of "dual epidemics" [48]. To this purpose, the development of new evaluation protocols of downy mildew resistance on detached inflorescences has been addressed to study divergent dual epidemics cases [49]. However, the outcome of ex vivo experiments in our previous study [22] highlighted the impossibility to follow BR disease progression on detached organs. Therefore, resistance assays on grape berries had to be performed in the field necessarily, followed by the development of a new descriptor for bunch resistance (Table 1).
The detection of organ-specific QTLs has been barely described in grapevines for a very few pest and disease resistance cases, i.e., phylloxera root and foliar resistance [27,50], and multi-organ powdery mildew resistance [29]. This phenomenon is well known and has been studied through QTL mapping also, e.g., for potato late blight [51], maize common smut [52], and the two different diseases head blight (or scab) and crown rot that affect both wheat and barley and are caused by the same Fusarium pathogens, recently reviewed by Su et al. [53]. As a consequence, multi-organ evaluation at multi-developmental stages might be of paramount importance for BR disease since there is evidence of a positive effect of senescence on the resistance trait [16], an observation supported by experimental evidence by the expression pattern of Germin-like 3 genes underlying the Rgb3 locus ( Figure 8).

The Genes Underlyng Rgb Loci Suggest non-R-Gene-Mediated Resistance Mechanisms
Following the comprehensive candidate gene analysis, no classical R-gene was present within the two identified genomic intervals. The R-genes are characterized by the highly conserved NLR (nucleotide-binding leucine-rich repeat) domains required for proteinprotein interaction [54] that allow the host to activate the innate plant effector-triggered immunity (ETI) by the direct or indirect recognition of pathogen effector molecules (avirulence proteins, Avr) delivered into the plant cell to overcome the basal immunity (or pathogen-associated molecular patterns PAMP-triggered immunity, PTI) [55]. Therefore, the possible involvement of underlying genes in alternative resistance mechanisms at the shoot and bunch level was investigated.

Shoot Rgb1 Locus
Upon GO enrichment analysis (Figure 6a,b), to our knowledge, this work suggested for the first time a relation between phloem sieve occlusion elements (SOEs) and shoot resistance to a fungal disease as BR. In fact, the Rgb1 locus includes an enriched cluster of five genes belonging to this family. Traditionally, phloem has been associated with sap-feeding insect resistance [56] since it is the vascular tissue where the translocation of soluble organic compounds occurs. Sieve occlusion elements are phloem proteins (called P-proteins) required for the formation of filamentous or tubular structures [56] that associate with organelles and stay firmly attached to the plasma membrane to avoid sieve elements occlusion [57], the latter being conducting cells, where the sugar transport produced during photosynthesis actually occurs [58]. These findings might explain the reason ex vivo experiments on detached leaves do not allow observing disease progression [22], since the separation from the plant impede phloem flow, thus limiting sugar availability to the pathogen. Moreover, they also support the well-known evidence of a preferential disease progression and pycnidia formation nearby the main leaf veins [8] that was also observed in the present study, a region where sugar translocation is more abundant. All five genes belonging to this enriched cluster encode for the first type of SOEs that comprehends scaffold proteins required to form the phloem filament matrix. The second type of SOEs is represented by callose, whose deposition is notoriously implicated in grapevine resistance to both downy and powdery mildew [59,60], as well as in other plant resistance mechanisms [61]. This suggests that the SOE type B detected in this work might act in an analogous manner to callose, limiting the nutrient uptake by the pathogen.
A second enriched GO class was associated with the mitochondrion proton transfer. Like in animals, in plant systems, the mitochondria determine the hypersensitive response to stress stimuli, initiating programmed cell death (PCD) [62]. The synthesis and export of ATP and the generation of higher amounts of phosphate is necessary for the reduction of excessive amounts of reactive oxygen species (ROS) in the cytoplasm; therefore, the impairment of this mechanism causes ROS accumulation, irreversible membrane and DNA damage, and cell death [63]. Accordingly, the assembly factor 2 and Ku70-binding protein underlying the Rgb1 locus might negatively regulate ATP synthase mitochondrial F1 complex formation and proton transport, inducing the oxidative burst and, consequently, the PCD. A similar mechanism has been described through the silencing of a mitochondrial ATP synthase F1 subunit that caused improved resistance to Ralstonia solanacearum in Nicotiana benthamiana [64]. Within this perspective, another mitochondrial resistance mechanism might be mediated by the significantly associated PPR gene. In fact, their disruption has been demonstrated to be involved in the spontaneous cell death response caused by H 2 O 2 accumulation, resulting in enhanced resistance to some fungal and bacterial pathogens in rice [65]. Finally, a possible role of nucleoside triphosphate biosynthetic processes can also be hypothesized, since the Thymidylate kinase homologous gene in cucumber was significantly overexpressed in a resistant line to Fusarium wilt [66].

Bunch Rgb3 Locus
The most strongly associated marker with Rgb3 locus locates within a gene encoding for a PAPα enzyme. The latter is involved in the dephosphorylation of the phosphatidic acid, thus modulating the concentration of this important signal molecule, which is directly participating in the positive regulation of ROS accumulation [67]. A knock-down mutant in a homologous gene of Arabidopsis-a phosphatase with a central role in maintaining jasmonate and salicylic acid hormone homeostasis and defense signaling-was found to show enhanced susceptibility to necrotrophic pathogens, but did not affect the response to biotrophic microorganisms [68]. This suggests that PAPα could be counted among the susceptibility genes, and therefore, it is valuable for gene/base editing approaches.
The GO analysis (Figure 6c,d) revealed a significantly enriched Glp3 gene cluster. Germin-like proteins are secreted N-glycosylated peptides associated with the extracellular matrix [69], which are involved in plant growth and development [70], as well as in disease resistance through oxidative burst [71], hance in PCD induction. Together with the antimicrobial peptides (AMPs), they have been classified within the pathogenesis-related (PR) proteins, and their accumulation is a key component of the plant innate immune system, and especially of systemic acquired resistance (SAR), a mechanism of induced defense that confers long-lasting protection against a broad spectrum of microorganisms [72]. As a relevant gene cluster, BLAST analysis of the Glp3 nucleotide sequences was not only conducted with the PN40024.v4 (V. vinifera subsp. Vinifera) genome [35], but also with the genomes of the putative ancestors (wild Vitis species) provided by Cantu Lab (UC Davis) [73]. This preliminary analysis revealed an expansion or shrinkage of the Glp3 cluster, both "inter" (between), but also "intra" (within) the wild haplophased chromosomes, advising for the necessity to further investigate these genes in the resistance donor ancestors. In grapevine, Glps were first reported in association with the ontogenic resistance of berries to powdery mildew [74]. Then, the expression of the VvGLP3 gene-coding for a protein with superoxide dismutase (SOD) activity-was detected to be highly induced at Erysiphe necator infection sites [75]. These findings suggest focusing on the investigation of senescence in relation to disease resistance, which might uncover new, promising, broadspectrum and basal resistance mechanisms. In rice, endogenous OsGlp gene silencing increased susceptibility to both sheath blight and blast [76,77]. The overexpression of Glp genes has also been demonstrated to be associated with increased resistance to Verticillium dahliae and Fusarium oxysporum in cotton [78] and to Sclerotinia sclerotiorum in tobacco [79]. Multiple disease resistance (MDR) has been inferred based on the detection of non-specific defense mechanisms and genes or QTL clusters associated with different diseases [80]. This phenomenon is well known and referred to as pleiotropy [81]. In maize, two QTLs have been identified as conferring broad-spectrum resistance to three and two diseases, respectively [80]. Finally, a gene family of 14 Glps was studied in rapeseed, demonstrating that the early induction of the SOD activity of some Glp generate H 2 O 2 and increase the resistance to the necrotrophic fungus S. sclerotiorum [71].
Another significantly enriched group of genes belonged to lipid transport and localization. Within the Rgb3 locus, there are two HEAT repeat-containing proteins, and a third one is located 0.3 Mb downstream of the expanded interval (Table S5). The latter gene results in being tightly associated with the recently published Resistance to Plasmopara viticola 29 (Rpv29) locus [82]. The study highlights the probable requirement of HEAT repeat proteins for basal non-host resistance, as it is the case for ILA in Arabidopsis [83]. This implies that the HEAT repeat protein has a pleiotropic effect and corroborates the hypothesis that the Rgb3 locus is a broad-spectrum resistance locus. The lipid transport and localization category also encompassed two non-specific lipid-transfer protein 2 (nsLTPs), considered for all intents the most abundant AMP protein family [84]. Two nsLTPs from maize have also been biotechnologically engineered to obtain peptides able to protect pearl millet from downy mildew [85]. Finally, the 'Transition metal ion binding' enriched gene set includes, besides Glp3 genes, also a mannose-6-phosphate isomerase. Mannose is thought to play an important role in ascomycetes and oomycetes, but also basidiomycetes, and the impairment of its metabolism can cause a reduction in glucose, fructose, and mannose, thus generating chain-inhibitory effects based on the limitation of nutrient uptake that can block initial infection [86].

Segregating Population
An interspecific population was generated in 1998 by crossing 'Merzling' and 'Teroldego' (M × T) at the Fondazione Edmund Mach (FEM, San Michele all'Adige, Italy). Based on pedigree reconstruction [22,87], 'Merzling' is a complex hybrid among V. vinifera, V. rupestris, and V. aestivalis var. lincecumii, which displays high resistance against BR, while 'Teroldego' is a BR susceptible V. vinifera landrace of the Trentino region [22]. The population comprised more than 150 individuals that were planted in an untreated open field at San Michele all'Adige (Trento, Italy) as single plants. For greenhouse experiments, parental and progeny buds were collected for propagation, as described by Vezzulli et al. [88]. Excluding plants derived from self-pollination and outcrossing identified through microsatellite analysis, 147 genetically unique individuals of the M × T progeny were included in this study.

Phenotyping
Two strains of P. ampelicida isolated by FEM in Trentino (Italy), and one isolated by the Julius Kühn Institute (JKI)-Institute for Grapevine Breeding Geilweilerhof (Siebeldingen, Germany), were previously genetically characterized and propagated on oatmeal agar (0.5% w/v) [22]. These strains were combined and used for artificial infection, following the protocol developed by Bettinelli et al. [22]. Briefly, fresh leaf tissues with mature BR lesions were used as the inoculum source. In the greenhouse, the cuttings of the mapping population were cultivated in a climatic chamber at 24 • C, young growing shoots with at least five fully expanded leaves were sprayed with conidia suspension adjusted to 10 4 conidia/mL in the late afternoon and kept at 100% relative humidity overnight. In the field, clusters at susceptible phenological stage [26] BBCH 77 (berries beginning to touch) [89] and growing shoots were sprayed after the sunset to avoid direct UV irradiation. Plastic bags were used to wrap inoculated organs, ensuring a high humidity treatment overnight. Bags were removed the following early morning before sunrise. Overall, two experiments were conducted on clusters in the two growing seasons 2020 (FC1) and 2021 (FC2), while in 2022, disease progression did not occur due to unusually high temperatures during the night (>30 • C). In each field inoculation trial, the entire population was screened simultaneously, and two clusters were evaluated per genotype, generating four sub-experiments (FC1a, FC1b, FC2a, FC2b). Shoots were evaluated once in the field in 2021, distinguishing between leaves (FL) and shoot internodes (FS). A total of eight greenhouse inoculation trials were performed between 2020 and 2021 on subgroups of the population to obtain three experiments (GL1, GL2, GL3). To produce an overall dataset of the resistance trait, the replicated data of GL and FC trials were also combined by (i) retaining only the minimum resistance value per each genotype and experiment type (GL min and FC min), and (ii) calculating the median (GL median, FC median). While the minimum should exclude possible false positive resistance evaluations, the median is the most robust statistical parameter for non-normally distributed data. Resistance evaluation of leaves in the greenhouse was conducted following the 5-step scale proposed by Rex et al. [21], which resembles OIV descriptors [90], i.e., (1) very low, (3) low, (5) medium, (7) high, and (9) very high resistance. Contrariwise, leaves and shoot internodes in the field were evaluated with a binomial scale as susceptible (S that is 1) or resistant (R that is 9), respectively, in the presence or absence of lesions, because the occurrence of other lesions caused by multiple abiotic and biotic stresses did not permit using a more detailed scale. To screen cluster resistance in the field, a 5-step rating scheme was developed based on the percentage of infected berries (Table 3), following the example of OIV 459 descriptor for degree of cluster resistance to Botrytis bunch rot [90]. Berry color (white or black, coded as a 0/1 valued binary variable) was also recorded to be used as a positive control in QTL analyses, thanks to the well-known location of the responsible QTL on chromosome 2 [33]. The normality of trait distribution was determined by means of the Shapiro-Wilk normality test (p < 0.05). All statistical analyses were executed with the software PAST 3.26 [91]. Table 3. Rating scale used for the evaluation of degree of cluster resistance to black rot.

Rate Description
1 very low resistant: more than 60% of infected berries with pycnidia formation 3 low resistant: up to 60% of infected berries with pycnidia formation 5 medium resistant: no more than 20% infected berries with pycnidia formation 7 high resistant: few infected berries with pycnidia formation 9 very high resistant: no infected berries

Genotyping
Genomic DNA extraction from young leaf tissue, quality evaluation and quantification, and array hybridization by Illumina SNPchip Infinium HD Ultra, GrapeReSeq 20K technology were performed at the Sequencing and Genotyping Platform of FEM (San Michele all'Adige, Italy). GenomeStudio 2.0 software (Illumina, USA) was used for SNP calling (GenCall score cutoff 0.15). A call rate of 95% was used to discard poorly performing samples. Quality control parameters as cluster separation score, call frequency, and measures of deviation from expected allele Mendelian inheritance patterns were used to identify SNPs that needed to be manually re-clustered or discarded [92]. The SNP scoring tool ASSiST 1.02 [93] was then implemented, with default thresholds, to filter the dataset and keep only robust SNPs segregating in the population. ASSiST was also used to recode the SNP dataset into JoinMap ® input format [31]. The SNP dataset was integrated with 192 SSR markers of the M × T map published by Vezzulli et al. [88]. Finally, three SSR markers that were previously associated with the Rgb1 locus [40] were amplified in triplex with KAPA2G Fast Multiplex Mix (Sigma-Aldrich/Merck, USA) and included in this study, namely GF14-42 [21], GF14-04 [94] and UDV-095 [41]. The amplification followed a touchdown protocol (60 to 55 • C, −0.5 • C × 10 cycles). Primer sequences and other details are available in (Table S10). The marker UDV-095 was already present in the map from Vezzulli et al. [88], and it served as internal control for SNP and SSR dataset integration. To be run in JoinMap ® , the markers were renamed to have a maximum of 20 characters and to easily distinguish SSR from SNPs by a suffix (_R) (Table S11).

Linkage Mapping
High-density genetic maps were generated with the software JoinMap ® 5.0 [31] by determining LGs through independence LOD. Map quality improvements were achieved following the good practices for datasets obtained by the GrapeReSeq Illumina 20K SNPchip described in Vervalle et al. [95], i.e., chain length and stop criterion parameters optimization to allow convergence of the mapping algorithm. Firstly, an integrated map comprehending all robust SNPs and SSRs was calculated with Cross Pollination (CP) population type (wIN map). Being the number of loci per LG > 100, Maximum Likelihood algorithm was applied for map construction with the required Haldane's mapping function. This analysis also produced parental M and T maps. The same pipeline was followed on a reduced dataset excluding identical segregating markers through the dedicated JoinMap function and producing rIN, rM and rT maps. Afterward, doubled-haploid maternal (dhM) and paternal (dhT) maps were calculated following the two-way pseudo-testcross strategy, [96], to overcome the problem of high number of dominant and missing observations of CP population, i.e., the memory limitations for QTL mapping [31]. To this purpose, the JoinMap function 'create paternal and maternal map' was used to extract loci segregating in each parent and transform them in biallelic markers, lm × ll and nn × np respectively. Loci of segregation type hk × hk were excluded from the dataset to avoid excess of missing data, as there is no possibility to reconstruct from which parent the alleles of heterozygote (hk) offspring derived [31]. Then, markers were converted in doubled-haploid segregation type a × b, with 'a' being either 'lm' or 'np' in the two sub-datasets, and identical loci were excluded. Due to the reduced number of loci, Regression mapping algorithm was also implemented and both Haldane's and Kosambi's mapping functions were tested to obtain the doubled-haploid maps. Since segregation distortion is a common phenomenon in outcrossing species as grapevine, its values were not used as a threshold to exclude markers [97]. Rather, the premature ungrouping-at lower LOD values compared to the other markers belonging to a linkage group-was used to exclude markers prior to grouping calculation. Other marker exclusion criteria applied to map construction were diagnostic values that stand out compared to the rest of the dataset, as high (i) nearest neighbor fit (NN Fit) and stress (NN Stress) that measure how well the markers fit that map order, and (ii) high -Log10(P) Genotype Probabilities Locus Means values, that measure how unlikely genotype observations affect the quality of the map [31,95]. Marker exclusion was checked through the function 'combine maps', to control if it corrupted the order of the map. The overall improvement in the quality of the maps was verified through the increasing (least negative value) of the logE-likelihood [98,99]. Maps were also combined to verify order consistency among complete, reduced and doubled-haploid integrated, maternal and paternal maps.

QTL Analysis
The study of QTLs was performed with MapQTL ® 6.0 [103], and the two-way pseudotestcross strategy [96] was employed to inspect the contribution of the resistant parent 'Merzling' to the trait. The KW non-parametric test (p < 0.005, ****) and Interval Mapping (IM, with Regression mapping algorithm) were implemented to highlight putative QTLs. Automatic cofactor selection (ACS) was then used to confirm significantly associated markers (p < 0.02). To this purpose, a set of cofactors selected at the constant distance of 1 cM on the chromosome where the putative QTL was detected was selected as starting set. This procedure was applied to allow the identification of multiple QTLs on the same chromosome, thus avoiding the occurrence of false positive 'ghost QTL' between adjacent QTLs [103]. The identified cofactors were then included in MQM with Regression mapping algorithm. For IM and MQM analyses, minimum GW LOD thresholds (p < 0.01, § § § § ; p < 0.05, § § § ) and CW (chromosome-wise)-LOD thresholds (p < 0.01, § § ; p < 0.05, § ) were calculated per each trait by three runs of 1000 permutation test, to determine four categories of QTL significance. Multiple rounds of MQM were run until no more loci exceeded CW § thresholds. Adjacent significant loci commonly arose as output of the analysis, but they were not selected as cofactors following the general rule of one marker per locus. Quantitative trait loci were considered major QTLs if explaining more than 10% of the phenotypic variance, otherwise they were considered minor QTLs [34], and a confidence interval of two-LOD was applied to determine QTL limits. Finally, recombinant individuals in the surrounding QTL regions were inspected to deepen the genetic basis of the traits.

Candidate Gene Identification
The latest assembly version 4 of the PN40024 reference genome [35] was used in this work to define the genomic physical positions, since it properly replaced many genes thanks to the increased scaffold size-achieved by 30× PacBio data-assigning to the proper chromosome unknown contigs and genes belonging to heterozygous regions. For this reason, the flanking sequences of all SNPs and SSRs primers-retrieved from [104], and [21,41,88,94], respectively-were blasted on PN40024.v4 [105] to redefine their physical coordinates. Then, the new positions of the markers defining the two-LOD QTL confidence intervals were used to extract the list of genes located within the QTLs and extended to the nearest non-associated marker. This extension was intended to avoid losing associated genes, since it is not possible to determine with which marker the enclosed region co-segregates. Besides the genes annotated according to PN40024.v4.2 (V4.2) [106] two further gene lists were obtained through the Grapevine Expression Atlas web resource (GREAT, account mandatory) [107] to retrieve the V3 functional gene annotation [36], and the correspondence between V4 [106] and previous annotation i.e., CRIBI V1, CRIBI V2, and Vcost V3 [38]. Finally, the lists were merged, and the QTL regions inspected using the genome browser interface Jbrowse [108]. If new genes were predicted in V4, they were retained in the list, while if the V4 lacked genes from other annotation versions, their localization was checked to determine whether to exclude them. The candidate gene lists were studied through GO enrichment executed through ShinyGO 0. 76.3 [109] with false discovery rate (FDR) cutoff of 0.05 and by selecting 'All available gene sets' as pathway database. Basic local alignments were performed at the PN40024.v4 [35] dedicated BLAST sequence server [105]. Multiple sequence alignments were implemented at MUSCLE [110] and the visualization realized through the linked software Mview© 1.63 [111]. Protein sequence conservation was visualized through the NCBI web application Multiple Sequence Alignment Viewer 1.22.2 [112] with MUSCLE alignment as input (Pearson/FASTA format) and 'Column Quality Score' coloring method, that assigns scores to amino acids based on how well they agree with the others at that position. The phylogenetic reconstructions were performed with the web application CLUSTALW [113] using the function 'FastTree' [114] to generate phylograms, that are phylogenetic trees that have branch lengths proportional to the amount of character change [115]. FastTree v2.1.8 function [114] uses a non-parametric branch support based on a Shimodaira-Hasegawa procedure (SH-like) [116] to provide local support SH-like values, that represent the degree of genetic change ranging from 0 to 1. Divergences were verified at structural annotation level through the platform Apollo annotator (training and account mandatory) [117], and manually curated if necessary. In specific cases, functional annotation was further inspected at various databases e.g., In-terPro [118], TAIR [119] or UniProt [120] databases. Gene expression was also explored exploiting published data for the susceptible V. vinifera 'Corvina' at different developmental stages [39] by using the Expression Atlases App (EX-ATLAS) [121], provided by TOMSBio Lab Vitis Visualization (VITVIZ) platform [122].

Conclusions
The novel QTL associated with BR bunch resistance, together with the confirmed QTL linked to shoot resistance, has the potential to become a resource in different frameworks. Firstly, general hypotheses regarding resistance mechanisms were highlighted and set the stage for future basic research on putative susceptibility genes, the Germin-like protein cluster, and metabolites involved in mitochondrial oxidative burst and phloem occlusion.
To unravel this knot of information, the Telomere-to-Telomere (T2T) sequencing of the resistance donor represents the first step and is envisaged. Secondly, this study represents a breakthrough in applied research towards marker-assisted and genomics-aided breeding for BR resistance. In fact, the deployment of organ-specific QTLs is desirable to contain primary infection on leaves and to protect bunches from secondary infection. To reach the goal of routinary adoption, the validation step of the associated markers is essential and already planned in segregating populations derived from different genetic backgrounds. Finally, exploring diverse sources of resistance is crucial to prevent overcoming by distinct pathogen strains and ensure BR resistance durability.