The genetic framework of shoot regeneration in Arabidopsis comprises master regulators and conditional fine-tuning factors

Abstract

Clonal propagation and genetic engineering of plants requires regeneration, but many species are recalcitrant and there is large variability in explant responses. Here, we perform a genome-wide association study using 190 natural Arabidopsis accessions to dissect the genetics of shoot regeneration from root explants and several related in vitro traits. Strong variation is found in the recorded phenotypes and association mapping pinpoints a myriad of quantitative trait genes, including prior candidates and potential novel regeneration determinants. As most of these genes are trait- and protocol-specific, we propose a model wherein shoot regeneration is governed by many conditional fine-tuning factors and a few universal master regulators such as WUSCHEL, whose transcript levels correlate with natural variation in regenerated shoot numbers. Potentially novel genes in this last category are AT3G09925, SUP, EDA40 and DOF4.4. We urge future research in the field to consider multiple conditions and genetic backgrounds.

Fig. 1. Variation in shoot regeneration among natural Arabidopsis thaliana accessions.

a Geographic distribution of accessions ranked according to the maximum number of regenerated shoots after 21 days with protocol a or b. b Bar chart of shoot numbers after 15 and 21 days with protocol a and b. Colors indicate the batches in which accessions were analysed and labels reflect a compact letter display of two-sided pairwise comparisons (on count data after 21 days) using Dunn’s nonparametric test at a false discovery rate (FDR) of 5% with n = 12 independent biological replicates. Global Kruskal–Wallis tests yielded p-values < 2e−16 for both protocols.

Fig. 2. SNP associations for shoot regeneration.

a, b Manhattan plots of chromosome 2, showing the significance of SNP associations with regenerated shoot numbers after 21 days under protocol a (a) or b (b). The green line marks a Bonferroni threshold of 5%, deduced from an efficient mixed-model association expedited (EMMAX) test with n = 129 and n = 149 independent samples for protocol a and b, respectively. SNPs considered significant in subsequent analyses (raw p-value ≤ 1e−5) are highlighted in red. c Venn diagram showing overlap in genome-wide SNPs that were significantly linked to shoot numbers under protocol a (orange and green parts) or b (blue and yellow parts) after 15d (orange and yellow parts) or 21d (green and blue parts).

Fig. 3. Overlap in candidate genes for various regeneration traits.

a Upset diagram showing overlap in QTGs (i.e., genes closest to SNPs with p ≤ 1e−5) between phenotypes, protocols and time points. Bright blue, dark blue, red, and green colors respectively indicate the most common sets (degree ≥ 6), selected intersections spanning both protocols and the highest order groups within protocol a or b. The same color codes are used for individual SNPs and genes, which are further highlighted in light or dark steel blue if beta (i.e., regression coefficient) ≥ 1.0 or MAF ≥ 10%. b Box plots showing −log10(p-value) (nlog10p) and number of SNPs (avgSNPcount) supporting the genes in each intersection (hinges reflect quartiles). c Scatter plot of SNP significance vs. minor allele frequency (MAF) and histogram of beta values.

Fig. 4. Allelic variation in the top candidate genes.

Distribution of selected SNPs among the ten best and ten of the worst regenerating accessions. Green and red boxes respectively indicate superior and inferior alleles and the three bottom rows depict gradients based on the accumulation of beneficial SNPs and regenerated shoot numbers after 21 days under protocol a or b. The two leftmost columns specify the SNP position along with nearby genes and the two rightmost columns show whether the SNP was significantly linked (p ≤ 1e−5) to data from protocol a and/or b (with brackets reflecting insignificant associations) and the percentage of accessions carrying the minor and usually superior allele in the population (MAF). Blank fields contain missing data.

Fig. 5. Details of the SNPs near WUS and their effect on shoot regeneration.

a SNP associations and linkage (R2) in the region around WUS, showing one upstream SNP (67 bp from the TSS; in pink) and two linked downstream SNPs (in red). b Allele frequencies of the pink SNP. c Box plot of the square root of regenerated shoot numbers after 21 days under protocol b in accessions with either variant of the highlighted SNP (hinges depict quartiles and whiskers extend to 1.5 times the interquartile range). d WUS expression relative to UBC9 and TIP41L in Col-0 and Lp2-2 (respectively a poorly and a strongly regenerating accession) after 3 days on SIM following protocol b. Error bars reflect standard errors and the p-value (corresponding to an estimated effect size of −1.47 within a 95% confidence interval from −3.24 to 0.30) is deduced from a linear model using a two-sided ANOVA with n = 3 independent biological replicates (represented as black dots).

Fig. 6. Shoot regeneration in chromosome substitution lines.

a Bar plot of log-transformed shoot and shoot primordium numbers in CSLs between Col-0 and Ler after 21 days on SIM following protocol b. Per chromosome, the averages of all lines with the Col-0 variant and the Ler variant are compared (irrespective of the other chromosomes) and individual data points are overlaid in jittered strip charts. Error bars reflect standard errors and FDR-adjusted p-values are deduced from negative binomial generalized linear models using two-sided likelihood ratio tests with n = 12 independent biological replicates per CSL and 321 or 318 residual degrees of freedom for shoots or primordia. The two significant terms had estimated sizes of 39.27 (effect of chr2 on shoots; deviance = 7.55) and −2.52 (effect of chr1 on primordia; deviance = 5.90). b Chromosome patterns of the CSLs (with Col-0 alleles in khaki and Ler variants in green), ranked by regenerated shoots, primordia and area.

Fig. 7. Shoot regeneration in mutants for GWAS candidate genes.

a Box plot of the regenerated area in 11 T-DNA insertion lines compared to wild type (WT) plants after 21 days under protocol a, b, or c (respectively shown in bright blue, dark steel blue, and light steel blue). Hinges mark the first and third quartile, horizontal lines reflect the median and whiskers extend to the furthest data point within 1.5 times the interquartile range of the nearest hinge. FDR-adjusted p-values are deduced from a Kruskal–Wallis test and Dunn’s many-to-one tests for two-sided pairwise comparisons to a single control on n = 12 independent biological replicates (quantile estimates are 3.18, −0.47, 2.59, −3.60, 2.59, and 2.42 for significant terms from left to right, i.e., at3g09925 (a), qky (a, b), rlp9 (a, b), and wavh2 (b)). No data was available for at1g20380 under protocol c and wavh2 under protocol a. b Representative images for lines that differ significantly from their wild type counterparts under the respective conditions (protocol a, b or c).

Fig. 8. Visual representation of the four categories of candidate regeneration genes.

The number of associated phenotypes, i.e., shoots, shoot primordia, green area, root-like structures, and undefined structures under protocol a or b (x-axis) was plotted against prior links with shoot regeneration based on literature, where 0 = no link, 1 = possible link, 2 = plausible link, 3 = indirect link, 4 = direct link (y-axis). The size and color of circles reflects the number of genes in each set (n). Blue, green, gray and red outlines respectively define known conditional factors, known master regulators, unknown conditional factors, and unknown master regulators.