Seeds of 245 A. thaliana accessions were purchased from the Arabidopsis Biological Resource Center. For the mGWAS, plants were grown from the original seeds in a greenhouse at 22°C under fluorescent light (16-h light/8-h dark), and mature seeds were harvested.

Seeds of SALK lines (Alonso et al., 2003) (AT2G32160, SALK_077267 and SALK_118137; AT2G32170, SALK_046329) (Supplementary Figure 1) were also purchased from the Arabidopsis Biological Resource Center. Homozygous mutants were confirmed by genomic PCR using primers obtained from T-DNA Express (http://signal.salk.edu/cgi-bin/tdnaexpress). The transgenic lines named MT160-OE and MT170-OE were established in the Col-0 background by introducing the CDS of AT2G32160.3 or AT2G32170.1 driven by the Cauliflower mosaic virus 35S promoter by the floral dip method (Clough and Bent, 1998) using Agrobacterium tumefaciens strain GV3101 (pMP90) transformed with vectors created as described below (under “Cloning”).

Transgenic plants were selected using GFP fluorescence as an indicator, and T₃ seeds with confirmed fluorescence were used in subsequent experiments. Plants for the expression analysis were grown on half-strength Murashige and Skoog (MS) medium (10 g l⁻¹ sucrose, 8 g l⁻¹ agar, 1 g l⁻¹ MES, and MS vitamin solution, pH 5.7) at 22°C under 16-h light/8-h dark conditions. After 3 weeks, aerial parts were sampled.

AT2G32160-OE plants for πMH analysis were cultured hydroponically as follows. Hydroponic sponges (2.5 × 2.5 × 2.5 cm; Kyowa, Osaka, Japan) were cut horizontally into thirds (2.5 × 2.5 × 0.8 cm), and a slit was introduced into each section. The divided sponges were then soaked in 1× Hyponica hydroponic liquid fertilizer (Kyowa). Water-absorbing sheets (Miki Tokushu Paper Mfg., Ehime, Japan) were soaked in the same medium. Roots of plants grown for 2 weeks under the above-mentioned conditions were placed between a half-folded absorbent sheet, placed in the sponge cutouts, and transferred to 50-ml amber conical tubes (Eppendorf, Hamburg, Germany) filled with the same liquid medium. The plants in each tube were lightly covered with plastic wrap and grown for 1 week. Col-0 plants were grown next to each OE plant as a control. Six grown plants each for different genotypes were used for the metabolomic analysis. A photograph of plants before sampling is shown in Supplementary Figure 2.

πMH and tau-methyl-l-histidine (τMH) for GC-MS/MS structural analyses were purchased from Fujifilm Wako Pure Chemical Corporation (Osaka, Japan) as 3-methyl-l-histidine (product code 139-17851) and 1-methyl-l-histidine (product code 136-17861), respectively.

To obtain metabolomic data for the mGWAS, seed samples were prepared by scooping approximately 50 seeds with a seed spoon (Bio Medical Science, Tokyo, Japan). Except for 10 accessions with insufficient seeds, prepared in triplicate, six replicates were prepared per accession. Metabolite extraction was performed according to previous studies (Sawada et al., 2009; Uchida et al., 2020). A metabolomic analysis for mGWAS was performed on a liquid chromatography–tandem quadruple mass spectrometry system (UPLC-TQS, Waters, Milford, MA, USA) according to previously reported methods (Sawada et al., 2009; Sawada et al., 2012) using the multiple reaction monitoring (MRM) conditions listed in Supplementary Table 2 . The metabolome analysis included blank samples (only the extraction solvent) and the average intensity of the blank samples was used as the basal noise level. We generated six metabolomic datasets: three with data from 245 accessions, and three with data from 235 accessions.

For the metabolomic analysis, plant samples (aerial parts and roots) and seeds were suspended in 80% (v/v) methanol with 0.1% (v/v) formic acid and internal standards (8.4 nM of lidocaine and 210 nM of 10-camphorsulfonic acid) to a concentration of 4 mg/ml and approximately 50 seeds/ml, respectively, and extracted as described above. Quantification of πMH in plant samples and seeds of MT160- and MT170-OE and SALK lines was performed using LC-QqQ-MS (LCMS-8050, Shimadzu, Kyoto, Japan). The analysis was carried out according to our previous study (Uchida et al., 2020) using the MRM conditions detailed in Supplementary Table 3 .

We retrieved a publicly available, genome-wide polymorphism dataset of Arabidopsis accessions genotyped with the Arabidopsis 250k-SNP chip (Atwell et al., 2010) and generated a custom variation dataset comprising 213,925 biallelic SNP loci for these 220 accessions ( Supplementary Table 1 ). To conduct a GWAS based on the six metabolome datasets, we used the multiple loci mixed linear model of GAPIT v3.1.0 (Wang and Zhang, 2021) in R v4.1.3. More specifically, we estimated the genetic relationship using a Multiple Loci Mixed Linear Model. Initially, we listed SNPs with p-value less than 1 × 10 ^{− 5} and from the coding regions of genes overlapping these SNPs, we complied 1,683 genes. We displayed the results as quantile–quantile and Manhattan plots using the R package qqman v0.1.8 (Turner, 2018). The in-house Python and R scripts used to build the custom dataset and perform the GWAS are available at a GitHub repository (https://github.com/junesk9/).

The coding sequence (CDS) of AT2G32160.3 was amplified by nested PCR using two sets of primers (first PCR: 5′-CTTCGTGTATACGAGGAACC-3′ and 5′-CTTAAAAACAAATGCAACAG-3′; second PCR: 5′-CACCATGATTTCATCGTCAGAGAT-3′ and 5′-TTAAGTTGTTGTTATAGCACAC-3′) and PrimeSTAR MAX DNA polymerase (Takara, Shiga, Japan) using cDNA derived from Col-0 rosettes. The amplified CDS was then cloned into a pENTR-D-TOPO vector (Thermo Fisher Scientific, Waltham, MA, USA). The AT2G32170.1 CDS was amplified in the same way using two primer sets (first PCR: 5′-ACCGAAGAGCCACCACC-3′ and 5′-CAGACACAAATAAAGAGAGTC-3′; second PCR: 5′-ATGGTTTCGCCGTCAGAGAGATG-3′ and 5′-TTAAGTTGTTGTTATAGCAC-3′). We attempted to clone the generated amplicon into a pCR8/GW/TOPO vector (Thermo Fisher Scientific); however, most of the resulting insertions were in the reverse direction, probably because the CDS in the desired orientation was toxic to E. coli strain DH5α. In addition, sequences cloned from amplicons inserted in the correct direction were complete aside from the introns. As an alternative, we first amplified the full-length vector sequence without introns using a cloned vector containing one intron in the correct direction as a template, PrimeSTAR MAX DNA polymerase (Takara), and primers for mutagenesis (5′-AATGATACACTGCCATGGGTCATGATT-3′ and 5′- TGGCAGTGTATCATTTTCATCCCAT-3′). Next, the Dpn I-treated amplified product was used to transform E. coli, resulting in the successful cloning of a vector with the CDS in the correct direction. Compared with non-transformed colonies, the colonies of bacteria carrying the vector in the correct direction were very small. The generated AT2G32160.3 (pENTR-D-TOPO) and AT2G32170.1 (pCR8/GW/TOPO) constructs were respectively introduced into overexpression vectors pASG-GW and pAKG-GW (Uchida et al., 2020) using Gateway LR Clonase II Enzyme mix (Thermo Fisher Scientific).

For GC-MS/MS analysis, 100 μl aliquots of 250 μM πMH and τMH solutions were dispensed into separate 1.5 ml tubes. After evaporation of the solution using a centrifuge evaporator (SpeedVac, Thermo Fisher Scientific) and addition of 100 μl Mox regent (2% methoxyamine in pyridine, Thermo) to each tube, the metabolites were methoxylated at 30°C for approximately 6 h with shaking at 1,200 rpm using a thermo shaker (BSR-MSC100, Biomedical Sciences). Next, 50 μl of 1% (v/v) trimethylchlorosilane (TMS, Thermo Fisher Scientific) was added, and TMS derivatization was carried out by incubating the mixture at 37°C for 30 min with shaking at 1,200 rpm. Finally, 50-μl aliquots of the derivatized samples were dispensed into vials for GC-MS/MS analysis (AOC-5000 Plus with GCMS-TQ8040, Shimadzu). πMH-2TMS and τMH-2TMS were annotated using total ion chromatographs obtained by GC-MS/MS in scan mode, and MRM transitions (parent ion > daughter ion) were determined to be 218.00 > 73.10 for πMH-2TMS and 196.00 > 73.10 for τMH-2TMS. Plant extracts were derivatized in the same manner as used for πMH and τMH, and MHs in the extracts were analyzed by GC-MS/MS in MRM mode. Raw data collection was performed using GCMS Solution software (Shimadzu). GC-MS/MS parameters have been detailed previously (Tabeta et al., 2021).

The samples used for expression analysis were aerial parts of 3-week-old plants grown on agar-solidified half-strength MS medium (see the Plants section). Total RNA extraction, cDNA synthesis, and real-time PCR expression analysis were performed according to our previous study (Uchida et al., 2020). The primers used were as follows: 5′-TGATTGGTTGGATTCTTCGTTA-3′ and 5′-TTCCTTATTACCCAACGAACCTT-3′ for AT2G32160, 5′-GGTTGATGTAGATAAGGTTCGTTGT-3′ and 5′-AGGCTTGTAGCATTGATCTCG-3′ for AT2G32170, and 5′-GTTGGGATGAACCAGAAGGA-3′ and 5′-AAGAATACCTCTCTTGGATTGTGC-3′ for actin, the internal control.

We first conducted a preliminary metabolomic analysis based on LC-MS/MS using all A. thaliana seed accessions and selected 147 metabolites that were detected with high signal-to-noise ratios (> 3) and small relative standard deviation (< 10%) for a more detailed metabolomic analysis ( Supplementary Tables 4 – 9 ). Next, we analyzed the metabolomes of seeds of 245 accessions and performed a GWAS of the metabolomic data to find associations (p < 1 × 10 ^{− 5}, Supplementary Figure 3 ) between contents of 140 metabolites and SNPs ( Supplementary Table 10 ). We then examined associations between metabolites and SNPs located in gene regions and found 1,683 candidate genes. As an example, the gene encoding methylthioalkylmalate synthase (MAM) showed associations with a number of glucosinolates with different side chains. In this study, we focused on AT2G32160, which was annotated as a methyltransferase gene and showed an association with 3-methyl-histidine (πMH), as methyltransferase was likely to be directly related to the biosynthesis of πMH, a methylated metabolite. Detected SNPs with low p-values (p < 1 × 10 ^{− 9}) had no apparent effect on the gene function of AT2G32160 because they were located in introns or did not result in amino acid substitutions in exons. In contrast, SNPs with high p-values (p > 1 × 10 ^{− 2} in all batch) gave rise to nonsense mutations (i.e., TGG to TGA) ( Supplementary Table 11 ).

The gene next to AT2G32160, AT2G32170, was annotated as a methyltransferase gene as well ( Supplementary Figure 4 ). The functions of AT2G32160 and AT2G32170 have not been previously determined. In this study, we analyzed AT2G32170 in addition to AT2G32160 associated with πMH. For convenience, AT2G32160 and AT2G32170 are hereafter referred to as MT160 and MT170, respectively.

Expression levels of MT160 and MT170 in aerial parts of 3-week-old plants of the transgenic lines named MT160-OE and MT170-OE (see Materials and Methods) were analyzed by real-time PCR. According to the analysis, expression levels of MT160 were increased by 19–28 fold in MT160-OE lines, whereas those of MT170 were almost identical between MT170-OE lines and wild-type Col-0 ( Figure 1 ). In addition, overexpression and disruption of MT160 had no effect on MT170 expression, and expression levels of MT160 and MT170 in SALK lines were respectively less than 1% and approximately 30% of those in Col-0.

Two types of MH, which differ in the position of the methyl group, have been found in living organisms: πMH and τMH ( Figure 2 ). Because these two isomers could not be easily distinguished by our LC-MS/MS platform, the chemical structure of the metabolite detected in A. thaliana by LC-MS/MS was confirmed to be πMH by GC-MS/MS using TMS-derivatized extracts. The TMS-derivatized standard MH compounds were clearly separated in the GC-MS/MS chromatogram ( Figure 3A ). Comparison of the chromatograms of the seed extracts with those of the two MH standards revealed that the peak corresponding to πMH-2TMS (m/z 218.00 > 73.10), but not that corresponding to τMH-2TMS (m/z 196.00 > 73.10), was present in all samples ( Figure 3B ). This result confirms that the metabolite detected by LC-MS/MS was actually πMH.

πMH contents of seeds of MT160-OE and -knockout lines were respectively approximately 3.7-fold higher and one-tenth lower than those of Col-0 ( Figure 4 ; Supplementary Table 12 ). Although expression levels of MT170 in MT170-OE lines were unchanged relative to those in Col-0, seeds of some lines had increased πMH contents ( Figure 5 ; Supplementary Table 12 ). The amount of πMH in aerial parts and roots of hydroponically grown plants was increased by 20–35-fold and 12–28-fold, respectively, in MT160-OE lines compared with Col-0 ( Figure 5 ; Supplementary Table 13 ). In contrast, πMH contents of MT160- and MT170-knockout lines and MT170-OE lines were unchanged ( Supplementary Figure 5 ; Supplementary Tables 14 , 15 ).