GWA mapping reveals loci associated with natural circadian variation
We used the online GWA-portal to perform association mapping on the three circadian traits found to be significantly affected by genetic variation (Seren, 2018). We used an Accelerated Mixed Model (AMM) to account for population structure, although the other association models (linear and non-parametric) gave similar results (See Supplementary Figure 6). Pseudo-heritability estimates for each trait were given as an output of the GWA analysis: period=71%, RAE=37% and phase=13%.
GWA identified multiple genomic regions associated with period, phase and RAE as shown in Figure 3. We used –log10(p ) 6.5 as an arbitrary p-score cut-off to select SNPs for further investigation. This threshold is conservative compared to several other previously published studies in Arabidopsis (Kooke et al., 2016; Proietti et al., 2018; van Rooijen et al., 2015). We investigated known core circadian and flowering time genes to see whether these were significantly associated with any of the traits measured. Other than ELF3 (see below), none of these genes fell within the window of association for significant SNPs. Genes in regions 30kb upstream or downstream of the most significant SNPs were considered as potential gene candidates and were selected for further analysis based on their previously attributed functionality and designated GO term. Genes involved in circadian rhythms, flowering time or chloroplast regulation were given prevalence (details in Supplementary File 2). The most significant associations had three SNPs with a –log10(p ) score of 10.4-11.5, found on chromosome 4 associated with period variation. Within this interval we identified a non-synonymous SNP in the gene COLD-REGULATED GENE 28, a gene that has previously been identified as a negative regulator of several core clock genes (PRR7, TOC1, PRR5 and ELF4 ) and is also implicated in the trade-off between flowering-time and freezing tolerance (X. Li et al., 2016; Wang et al., 2017). The substitution resulted in a tryptophan (W) to serine (S) amino acid change at position 58 within the second exon of COR28 (Figure 4A). This had a SIFT score of 0 indicating a highly probable deleterious effect on protein function. 16 accessions in this study had the minor allele, all found in the South of Sweden. The 58S accessions had a period 1.29h longer than the 58W accessions (t(17.4)=-7.46,p< 0.001, Welch Two Sample t-test) (see Figure 4B) and mostly belonged to the genomic sub-group PC.A. The long-period of cluster PC.C therefore cannot be explained by variation in this SNP. We used the online tool Polymorph 1001 to look for other variants with the serine substitution and found only 5 other variants not assayed in this study (making 21 in total), all of which were also from the South of Sweden (Figure 4C).
We used flowering time data by Li et al. 2010 (Y. Li et al., 2010) and Sasaki et al. 2016 (Sasaki et al., 2015) to look for flowering time differences between the accessions carrying the COR28 -58W major allele and the -58S minor allele (see Table 1). We found that the accessions with COR28 -58S had significantly extended flowering times under simulated seasons for Sweden and Spain (Li) and under a 10°C and 16°C long-day temperature regime (Sasaki) (see Figure 4D). This complements previous findings that COR28 is a flowering promoter and that modifications to this gene increase flowering time as well as lengthening period.
[Table 1]
Other significant SNPs associated with period (chromosome 3) and phase (chromosome 1) identified gene candidates involved in chloroplast function (PARC6, SCO2, DNF, AT1G71015 and FAB1C ). The DF circadian output we used to assay these traits is based on oscillating activation phases of PSII (Goltsev V, Zaharieva I, Chernev P, 2009; Jursinic, 1986) and therefore natural variation in genes regulating chloroplast functionally could also affect the DF output in these candidates.
A RAE associated SNP on chromosome 2 was found 5726bp upstream from a SNP in ELF3 previously characterized as the ELF3-Shaallele. Our associated SNP was found to be under strong linkage disequilibrium with the ELF3-Sha allele (R2=0.86). The alanine-to-valine transition in amino acid position 362 has been associated with naturally occurring alterations to periodicity and robustness in accessions from Central Asia, specifically Tajikistan (Anwer et al., 2014). Here, 13 Swedish accessions were shown to carry the ELF-Sha allele and had mean RAE ratios 0.032 higher on average than for the other accessions. We substantiate evidence that ELF3-Sha accessions have lower rhythmicity and extend the global range of this allele into Northern Sweden. A surprising kinship between Arabidopsis accessions from Northern Sweden and Central Asia has been previously demonstrated through analysis of global population structure and indicates that the presence of this allele has not evolved convergently between the two populations (Nordborg et al., 2005).