Identification of SNPs associated with magnesium and sodium uptake and the effect of their accumulation on micro and macro nutrient levels in Vitis vinifera

Material and nutrient analyses

Two hundred forty-nine seedlings of a segregating Vitis vinifera F₁ breeding population derived from two heterozygous grape varieties, ‘Verdejo’ × ‘Gewürztraminer’ (V×T) were transplanted in June 2013 into a research plot in Ripperdan, CA (soil type = Cajon loamy sand, Dinuba-El Peco fine sandy loam, Pachappa sandy loam, slightly—moderately saline-alkali; pH = 7.9). All vines were own rooted with no grafting. Row spacing was set at 1.22 m with 2.44 m between rows. Seedlings were trained and managed according to standard grower practices. Plants were fertilized with N, P, and K at rates of 14.5, 18.4, and 12.9 kg/hectare, respectively in 2015 and 18.1, 23.1, and 16.1 kg/hectare in 2016. Fertilization was performed according to industry standard practices; fertigation by applying a liquid fertilizer solution through the drip irrigation every two weeks from the time of fruit set. Lateral shoots were removed from the trunk during establishment and vines were trained to a unilateral cordon and spur-pruned. Plants were visually assessed for foliar symptoms in August (2015) and September (2016) using a 1 (present) or 0 (absence) rating where symptoms were plant stunting and/or leaf chlorosis (Fig. 1). Plant stunting and leaf chlorosis were evaluated separately. For nutrient analysis fully expanded whole leaf (petiole and blade) samples were collected from each vine. Due to variability between genotypes, equivalent leaf volume was collected, typically between 15–25 mature leaves. The leaves were sampled from fertile (fruiting) shoots into brown paper bags and air dried indoors at 22 °C. Once dry, the leaves were submitted to A & L Western Laboratories (Modesto, CA) for nutrient analyses in October 2015 and in October 2016. Nutrient concentrations were measured for nitrogen (N), sulfur (S), phosphorus (P), potassium (K), magnesium (Mg), calcium (Ca), sodium (Na), iron (Fe), aluminum (Al), manganese (Mn), boron (B), copper (Cu), and zinc (Zn). N was measured using automated combustion at 900 °C. S, P, K, Mn, Ca, Na, Fe, Al, Mg, B, Cu and Zn were measured using nitric/hydrochloric acid digestion using a microwave, analysis was by inductively coupled plasma spectrometry (ICP) as detailed by The North American Proficiency Testing Program (Black, 1965; naptprogram.org). N, S, P, K, Mg, Ca, and Na were reported as a percent of dry matter (% dm). Fe, Al, Mn, B, Cu and Zn were reported as parts per million (ppm). At the end of the study, a subset of symptomatic and asymptomatic vines was removed and evaluated for root stunting.

Statistical analysis

Nutrient data were analyzed using JMP v12 statistical software (SAS Institute, Cary, NC) for normality (Shapiro-Wilk W Test), analysis of variance (ANOVA), hierarchical clustering, and correlations for relationships within and between years. Plant symptoms were analyzed as marginal chlorosis only, stunting only, or combined (stunting and/or chlorosis). Data for Zn, Na, P, and Mn were log transformed, S, Mg, Fe, B, and Al were log10 transformed, and Ca and K were square root transformed to fulfill assumptions of normality. Significant differences in nutrient concentrations between years or genotypes were determined using Tukey’s Honest Significant Difference (HSD) (P ≥ 0.05). Correlations were determined using Pearson’s correlation coefficient (r) on the transformed data. Hierarchical clustering was determined using the Ward method on standardized data. Broad sense heritability (H) was calculated based on mean square values using the one location across two years formula modified from Fehr (1987) by Wang, Karle & Iezzoni (2000) with confidence intervals estimated by Knapp, Stroup & Ross (1985). Best Linear Unbiased Predictors (BLUPs) were calculated with the lme4 package in R (v4.0.2 R Core Development Team, 2017) using nutrient concentration, genotype, and year as random effects (Henderson, 1975; Liu, Rong & Liu, 2008; Merk, 2011). Principal component analysis for the population was calculated using the BLUPs for each nutrient concentration within JMP12.0.1.

Genotyping by sequencing and mapping of significant SNP associations

Young grape leaves were collected from each F₁ progeny grapevine in July 2015 and genomic DNA extracted using the Qiagen genomic DNA extraction kit (Qiagen, Inc Valencia, CA). Genomic DNA was sent to the UC Davis Genome Center’s DNA Technologies and Expression Analysis Cores (University of California, Davis) for quality analysis, restriction enzyme digestion (ApeK1), library preparation and Illumina Hi-seq 3000 sequencing. Sequencing coverage was approximately 2.7 million reads per sample. Genotyping by sequencing (GBS) data was analyzed using the Tassel 5.0 GBSv2 pipeline (Bradbury et al., 2007). The table grape/raisin genome of Thompson Seedless was used, in addition to the wine grape-derived inbred genome of PN40024, to capture some of the variability in SNP detection between reference genomes. Quality (≤ 20) and length (≥ 20 bp) filtered and trimmed reads were aligned to the Thompson Seedless genome (Genova et al., 2014; Patel et al., 2018) and the PN40024 12xv2 genome (Canaguier et al., 2017) using BWA (Li & Durbin, 2010). Identified SNPs were further filtered for frequency of minor (0.20) and major (≥ 0.35) allele frequencies, missing data (≤ 10%), and sequencing depth (≥ 5 reads (Thompson Seedless) or ≥ 10 reads (PN40024)) using vcftools 0.1.15 (Danecek et al., 2011). For the PN40024 genome, an additional filtering step to thin SNPs based on physical position to a minimum of 50 bp between sites was completed. A panel of 10,122 and 3,997 filtered SNPs (Thompson Seedless and PN40024, respectively) were used for genome-wide association analyses for each of the ions measured (Table 1). A kinship matrix was estimated in Tassel v 5.2.43 and used in a MLM (mixed linear model) implemented in the software GAPIT v2 for nutrient trait analyses within R statistical analysis software (Lipka et al., 2012; Tang et al., 2015; R Core Development Team, 2017). For binary traits (stunting, chlorosis, and combined symptoms), 2 principal components with P3D were used for analyses implemented within Tassel. Significance of a SNP was based on a P value ≤ 0.05 and a false discovery rate (FDR) ≤ 0.05. Linkage decay was estimated using PopLDdecay for the PN40024 genome (Zhang et al., 2019). In brief, pairwise r² for SNPs within 5 Mb were calculated using PopLDdecay, then the median r² for every 10 kb window were calculated (Zhang et al., 2019). The continuity of the Thompson Seedless genome is of lower quality than that of PN40024. Thus to cross reference significant SNPs detected in each reference genome, PN40024 was used as a coordinate reference. GBS tags with significant SNPs detected in Thompson seedless and their associated flanking sequence were mapped back to PN40024 to unify coordinates. Sequences were mapped using default parameters for short read alignment using Minimap2 (Li, 2018). Uniquely mapped primary alignments with quality higher than 40 were kept in the genome coordinate liftover. Manhattan plots for chromosome 3 were produced using the CMplot package in R (https://github.com/YinLiLin/R-CMplot). This method allowed for the ordering of SNPs in the same coordinate reference and comparison between distributions.

Functional annotation of genes associated with SNPs was determined using Blast2GO v 5.2.5 based on the nonredundant database from NCBI, and protein databases from Uniprot and Swissprot (Gotz et al., 2008; accessed June 2018).

Field symptoms

There were 60 and 52 individual vines exhibiting symptoms, while 189 and 197 did not exhibit symptoms in 2015 and 2016, respectively, roughly following a 3:1 segregation. The parents, not grown at the time of this study, had not previously displayed any symptoms of nutrient imbalances at this location under similar fertilization regimes. Symptoms observed in the F₁ progeny included leaf, internode, and petiole stunting, as well as marginal leaf chlorosis and necrosis (Fig. 1). A subset of symptomatic and asymptomatic plants evaluated for root stunting showed no visible differences (data not shown). Presence of symptoms (stunting or chlorosis) was consistent between 2015 and 2016 for most vines (>90%). Only eighteen vines had symptoms in 2015, but were asymptomatic in 2016. Another two genotypes had no symptoms in 2015, but were symptomatic in 2016. When each symptom was evaluated individually, stunting and marginal chlorosis symptoms were consistent among plants in both years (>80%).

Nutrient compilation

For the population, significant differences were detected in nutrient concentrations between 2015 and 2016 (Table 2). Large increases in Al, Fe, Mg, Zn, Mn and Ca concentrations were observed in leaf tissue between samples collected in 2015 and 2016. Mean values for nutrient concentrations varied for most of the ions evaluated among individuals in the V×T population (Fig. S1; Table S1). A decrease in N, P, K, and B leaf nutrient concentrations was observed from 2015 to 2016. When symptomatic and asymptomatic plants were analyzed separately, differences in nutrient concentration were detected in 2015 and 2016 (Table 3). Higher levels of Mg, Na, Al, and Fe were observed in symptomatic plants in both years, while a decrease in N was observed. Na, P, Cu, Mn, N, S, and Ca concentrations in 2015 or 2016 had no calculable heritability. For Mg (H = 0.34; confidence intervals (CI): 0.18–0.46), B (H = 0.44; CI [0.32–0.55]), K (H = 0.21; CI [0.02–0.36]), Al (H = 0.12; CI [0–0].28), and Fe (H = 0.27; CI [0.10–0.41]) broad sense heritability was moderate to low.

Nutrient concentration and symptom correlations

Symptoms (marginal chlorosis, stunting or both) were positively correlated with Na, Mg, Fe and Al concentrations, and negatively correlated with N across both years (Fig. 2, Table 4, Fig. S2). In 2015 Mg concentrations (r = 0.6146) and in 2016 Al concentrations (r = 0.4805) had the highest correlation with observed vineyard symptoms (Table 4). Na and Fe concentrations were also correlated with symptoms in both years, though at lower r values. In 2015, a significant negative correlation between S concentration (22%) and symptoms was observed, but not in 2016 (Table S2). In 2016, there was a significant negative correlation between P, Cu, and K content and symptoms (Table S2). Other significant correlations among nutrients included positive correlations (r ≥ 0.30) between N, P, and S in 2015 and S, P, and K in 2015 and 2016 (Table S2). A strong positive correlation was also detected for both Mn and Ca with Mg in 2015, but not 2016. Correlation with symptoms were observed for other nutrients, but were not consistent between years. When comparing nutrient concentrations from 2015 to 2016, most nutrients had low to moderate (r = 0.2 − 0.4) correlation, with the exception of copper (r =  − 0.0199) (Table S3). Nutrient ratios were examined between years for potential significant correlations with symptoms. Most ratios did not show consistent differences in values between years for symptomatic and non symptomatic vines (Table S4).

Marker-trait associations

Linkage disequilibrium half-decay distance was estimated to be 100 Kbp (Fig. S3). Genome-wide associations identified several chromosomes associated with differences in the ions evaluated (Fig. S4, Fig. 3). Significant positive associations between SNPs on chromosome 3 and Mg concentration were detected in 2015 for both the Thompson Seedless and PN40024 genomes. SNPs associated with Mg levels explained approximately 6% of trait variation (Table S5). In 2015, 6 SNPs were detected when aligned to the PN40024 genome while 4 SNPs were detected when aligned to Thompson Seedless. No SNPs associated with Mg accumulation were identified in 2016 with either genome at the P = 0.05 FDR level. Only 1 genic SNPs associated with Mg concentration was identified and was co-associated with SNPs identified for marginal leaf chlorosis and stunting (Table S6 and S7). In the Thompson Seedless genome, a small block of SNPs (S3_21825918 , S3_21825925, and S3_21825966), spanning 48 bp, located in a ∼7,000 bp intra genomic region, explained ∼18% of the variation associated with symptoms in 2015. A BLAST search of the region did not identify any significant alignments with any genes (predicted, putative or known) in Vitis or other species.

SNPs associated with Na concentration were detected on chromosomes 11, 12, 13, and 18 had both negative and positive allelic effects ranging from 8 to 13% of the observed variation in the Thompson Seedless genome. Three of the SNPs (S11_16152632, S18_25745134, and S18_25745143) were detected in 2015 and 2016. None of the SNPs associated with Na concentration in 2015 were located in predicted or known genes. Only a single SNP identified in 2016 (S11_11635675) was found in an annotated gene (Table S6). In the PN40024 genome, 11 SNPs with positive allelic effects were detected across chromosomes 3, 11, 15, and 17 in 2016. Individual SNPs explained 6 to 11% of the variation observed. Two of the SNPs on chromosomes 3 and 11 were also associated with Na concentrations in 2015, but were not significant when adjusted for an FDR of 0.05. Using the PN40024 genome annotation, 8 genes were associated with Na accumulation. These included genes putatively involved metabolism and transport. Four of the genes had no functional annotation ascribed (Table S5). One gene, Vitvi11g01139, was associated with Na accumulation and annotated as a Clathrin assembly protein, which is a class of proteins involved in macromolecule transportation. No significant SNPs were identified for any of the remaining ions measured.

For marginal chlorosis, a total of 151 SNPs, 57 in 2015 and 94 in 2016 were detected across 2015 and 2016, respectively in the Thompson Seedless genome. In 2015, 33 of the identified SNPs were located in genes, while in 2016, 20 SNPs were located in genes (Table S5). Sixteen of the genic SNPs were shared between years. Individual SNP effects on the Thompson Seedless genome were both negative and positive and ranged from 16 to 33%. Many of the SNPs identified in only a single year were located in genes with multiple SNPs associated with the trait (Table S6). When aligned to the PN40024 genome, a total of 29 and 55 SNPs in 2015 and 2016, respectively associated with marginal chlorosis were detected. Marginal chlorosis was primarily associated with SNPs on chromosome 3 with other SNPs detected on chromosomes 6, 9, 13, 16, 17 and 19. In 2015, 23 of the identified SNPs were located in genes, while in 2016, 22 SNPs were located in genes, 22 SNPs were shared between years. Similar to the SNPs detected in Thompson Seedless genome, many SNPs were consistent across both years on chromosomes 3 and 16.

A total of 37 and 71 SNPs associated with stunting were identified in 2015 and 2016, respectively when mapped to the Thompson Seedless genome. SNP effects varied from 3 up to 40%. Highest effect SNPs were detected on chromosome 3 with smaller effect SNPs located on chromosomes 18 (2015) and 1, 2, 10, 11, 12, 16 and 18 (2016). Twenty-four and 34 SNPs were associated with genes in 2015 and 2016, respectively. Nineteen of the genic SNPs associated with plant stunting were shared between years. The majority of single-year SNPs were identified in 2016 (Table S5). When mapped to the PN40024 genome, a total of 34 and 44 SNPs were detected in 2015 and 2016, respectively. Only 34 and 35 SNPs associated with stunting were located in genes for 2015 and 2016, respectively of which 29 were shared between 2015 and 2016.

No ion transport pathways were associated with symptom-associated SNPs based on the Thompson Seedless annotation, however approximately 50% of the genes had putative catalytic activity and 50% had binding activity (Fig. S5). None of the genes identified across both years and associated with symptoms were putative transporters, but were instead involved in processes such as oxidation, transcription, development, and stress response (Table 6). When symptom-associated SNPs located in genes based on the PN40024 annotation were evaluated for putative activity, stress response, transcription, growth and development, and metabolic pathways were all represented similar to the Thompson Seedless genome. In addition, SNPs were also detected in several genes related to sugar and nutrient transport. One SNP associated with leaf stunting in both years and symptoms was associated with Calcium ion binding (Vivi03g00243).

The majority of symptom-associated SNPs were found on chromosome 3 in both genomes, but each genome has a unique coordinate system. Therefore, we performed a consolidated genome analysis by mapping significant SNPs in Thompson seedless and their flanking regions to the PN40024 genome in order to order SNP coordinates and look for overlap between references (Fig. 4). When chromosome 3 assemblies were consolidated, a shared cluster of SNPs with significant association with symptoms was observed ∼7.5 Mb and a lesser cluster around ∼15 Mb. When aligned to the Thompson Seedless genome, symptom-associated SNPs located within genes shared across years were predominantly found on chromosome 3 with an additional SNP located on chromosome 10 (Table 6; Table S6). Significant single year SNPs, including those associated in genes, were identified on chromosomes 1, 2, 3, 4, 10, 11, 12, 16, and 18 (Table S5). When symptoms were combined, 28 SNPs were shared across both years, and 29 were only identified in a single year (Table S5). When aligned to the PN40024 genome, significant SNPs were detected on chromosomes 3, 6, 11, 16, and 19. Most of the identified SNPs were located on chromosome 3, and 21 were shared between 2015 and 2016 (Table 5). Twenty SNPs were only detected in a single year, with the majority identified in 2016 (15). Twenty-nine SNPs found on chromosome 3 were shared between years 1 and 2. Sixteen genic SNPs associated with the combined symptoms were identified in both 2015 and 2016 (Table 5).