To evaluate the effect of UHSMF on genotoxicity, we exposed Arabidopsis dried seeds to three strengths of UHSMF for 1 h. UHSMF was generated by a home-built water-cooled magnet at the National Major Scientific and Technological Infrastructure located in Hefei, China (Gao et al., 2016; Tian et al., 2021). The three treatments were designated as M33, M30.5 and M24.5, in which the number represents magnetic strength in Tesla (T), with a gradient of 0, 95 and 150 T/m, respectively ( Figure 1A ). The seeds without UHSMF treatment, namely geomagnetic filed (GMF), were used as a control (CK). Since the highest strength of SMF was produced in the center of the magnet, the orientation of the magnetic field gradient bellow the center was the same as that of the gravity, which mimics the hypergravity condition, while that above the center mimics the hypogravity condition. Thus, M30.5 and M24.5 treatments included hypergravity (M+30.5 and M+24.4) and hypogravity (M-30.5 and M-24.5) conditions.

The Arabidopsis seeds used in the study were the Col-0 ecotype, whose genome was completely sequenced with high quality in 2000 (Kaul et al., 2000). To identify inheritable mutations possibly induced by UHSMF, we propagated the UHSMF-treated and CK seeds for one generation by self-fertilization, and harvested the seeds as individual plants. In each treatment, a total of 40 independent siblings were randomly selected from the population of the second generation for DNA extraction ( Figure 1B ). Each DNA sample was sequenced by Illumina next-generation technology to reach a coverage depth of 50 folds. Ultimately, we obtained 236 individual genomic sequences with coverage from 33.4 to 74.9 depth ( Figure 2A ). The obtained raw reads were first analyzed and mapped to the reference genome (TAIR10, www.arabidopsis.org) with Bowtie2 program (Langmead and Salzberg, 2012). Sequence variants including single nucleotide variants (SNVs) and InDels (insertions and deletions) ranged from 1 to 28 bp were called by Genome Analysis Toolkit (McKenna et al., 2010) (GATK) ( Figure 1B ). On average, each line contained ~1,600 variants ( Supplementary Figure 1A ). Since this number is much higher than expected (Ossowski et al., 2010), we inferred that it could be due to many variants already present in the seeds before UHSMF treatment, i.e. germline mutations. To solve this problem, we assumed that it is an extremely low possibility that new mutations took place at the same nucleotide position of the genome. Thus, we retained the unique variants (i.e. singletons) in each sample and removed those overlapped with any other ones in all samples. In total, we identified 1,055 singleton variants including 946 SNVs (89.67%) and 109 InDels that consisted of 47 insertions (4.45%) and 62 deletions (5.88%) ( Supplementary Table 1 , Figure 2B ). Among these variants, 178 sites (16.87%) were homozygous while the others (83.13%) were heterozygous ( Figure 2C ). In addition, most (77.06%) of the InDels were short, with the size of 1-3 bp ( Supplementary Figure 2 ). Generally, our data are in agreement with previous reports that SNVs prevailed over InDels (Ossowski et al., 2010; Jiang et al., 2011; Lee et al., 2012). To verify these calls, we randomly selected 56 SNVs and 23 InDels, which were checked by PCR-based Sanger sequencing. Our results showed that 54 (96.43%) SNVs and 22 (95.65%) InDels were confirmed ( Supplementary Figure 3 ), indicating that the accuracy of the variants called by our analytic approaches are compelling.

To quantify the effect of UHSMF on DNA stability, we calculated the average number of mutations per line for each treatment. In general, the orientation of magnetic force, which is generated by magnetic gradient, is ignored in the study of magnetic bioeffects. Here, we first examined the effect of two orientations, namely hypergravity and hypogravity, with the identical strength and gradient on DNA mutations. Our data showed that M+24.5 and M-24.5 or M+30.5 and M-30.5 treatments did not significantly influence the average number of DNA mutations including SNVs and InDels ( Supplementary Figures 1B-D, 4A-C ), indicating that differential orientations with either 95 or 150 T/m have no impact on DNA stability. To focus our analysis on DNA mutation affected by magnetic strength and gradient, we combined the two sets of data from M+24.5 and M-24.5 into one (M24.5), and from M+30.5 and M-30.5 into M30.5. In addition, we confirmed that both the transition rate and the number of InDels larger than 3 bp, which were influenced by UHSMF (discussed latter in this study), were not significantly changed between M+30.5 and M-30.5 or between M+24.5 and M-24.5 ( Supplementary Figures 4D, E ).

We estimated the number of the overall mutations including SNV and InDel per line in each treatment. The highest number of variants per line were found in M24.5 treatment (5.04), followed by M33 (4.38), CK (4.20), and M30.5 (4.08) ( Figure 2D ). However, the differences between any UHSMF treatment and CK were not statistically significant (one-way ANOVA, P > 0.05). We then analyzed the SMF effect on SNV and InDel independently. The number of SNVs per line were the lowest in M30.5 (3.53) and the highest in M24.5 (4.57), while average InDels ranged from 0.35 (M33) to 0.55 (M30.5). M24.5 had the highest SNVs number (4.57) while M33 contained the lowest InDel (0.35). However, no statistically significant differences of both SNVs and InDels were detected between UHSMF-exposed seeds and CK (one-way ANOVA, P > 0.05) ( Figures 2E, F ). To estimate the average mutation rates of SNV and InDel per generation per site per line, we used the sum of homozygous mutations and half of the total heterozygous mutations due to the 50% probability for a heterozygote to be segregated into a homozygote in the offspring. Our data showed that the average mutation rates of SNV and InDel in CK were 9.55 and 0.89 × 10^-9, respectively, which had no statistical difference from the overall UHSMF treated samples (one-way ANOVA, P > 0.05) ( Figure 2G ). Likewise, the average rates of the total mutations (SNV and InDel) in UHSMF-treated samples were similar to those in CK ( Figure 2G ). Taken together, our data suggest that 1 h exposure of UHSMF with the intensity from 24.5 to 33 T has no impact on the rate of heritable DNA mutations.

To analyze possible genotoxic effects of UHSMF in more detail, we examined whether the spectrum of SNVs is altered in the UHSMF-treated groups in comparison with the CK. Consistent with the published data about the bias of SNVs in A. thaliana (Ossowski et al., 2010), E. coli (Lee et al., 2012), D. melanogaster (Keightley et al., 2009) and H. sapiens (Lynch, 2010), the rate of transitions (purine-to-purine or pyrimidine-to-pyrimidine changes) was higher than that of transversions (interchanges between purine and pyrimidine), and the two transitions of C-to-T and G-to-A were most prevalent among all single nucleotide substitutions in CK ( Figures 3A, B ). Although the rate of nucleotide transitions was lower in all UHSMF-treated groups in comparison with CK, a significantly low rate of transitions was detected only in M33 treatment ( Figure 3A ). Further analysis indicated that the most reduced transitions in M33 were the pyrimidine-to-pyrimidine substitution (32.24% to 24.22%), whereas the most increased transversions were A-to-C (5.26% to 9.32%) and T-to-G (5.92% to 11.18%) ( Figure 3B ). Taken together, our data suggest that M33 treatment leads to a significant decrease in the transition rate, particularly in pyrimidine-to-pyrimidine mutations.

Since the C-to-T and G-to-A transitions are predominant in all SNVs, we assumed that the induction or repair of mutations under UHSMF conditions may have a bias on the base type or motif flanking at the mutational site. To test this hypothesis, we analyzed the consensus of the 5 nucleotides flanking the left and right side of all the SNV sites, and visualized by ggseqlogo (version 0.1) (Wagih, 2017). As shown in Figure 3C , the three UHSMF groups displayed the same order as CK with C as the top and G as the second in the mutation frequency. However, the third and fourth mutation frequency were T and A, respectively, in CK and M24.5, and were A and T, respectively, in M33 and M30.5. These results indicate that UHSMF has minor effect on the order of nucleotide mutation frequency. Interestingly, we found that the consensus at the upstream -3 to -1 position flanking the C site was TTA in UHSMF groups but ATT in CK, while flanking the G site was AAT in UHSMF groups but TAA in CK ( Figure 3C ). No consensus of nucleotides was observed at the -5, and -4 positions and at the downstream of the mutation C and G sites in all UHSMF groups. In addition, we analyzed the flanking sequences of the mutated nucleotides, and did not find any typical motifs shared by UHSMF groups, compared to CK ( Supplementary Figure 5 ). These results suggest that the reduced rate of transitions in UHSMF is somewhat related to the characteristic of the upstream flanking sequences of the SNVs.

We then analyzed whether UHSMF has an impact on the distribution of SNVs on the five Arabidopsis chromosomes (Chr). The number of SNVs estimated in each chromosome per line ranged from the lowest of 0.48 (Chr2 in M30.5) to the highest of 1.1 (Chr3 in CK) ( Figure 3D ). We did not detect significant differences in the average number of SNVs distributed on each Chr between UHSMF and CK ( Figure 3D ). These results indicate that the Chr distribution of the spontaneous mutations for SNVs is not altered by UHSMF treatment.

Considering that UHSMF exposure can change the spectrum of SNVs, we assumed that it also affects the spectrum of InDels. Therefore, we analyzed the number of InDels per line independently. As shown in Figure 4A , the number of insertions in M30.5 and M24.5 were 0.23 and 0.25 per line, respectively, which were about two-fold of, but not significantly different from those in CK and M33. In contrast, the number of deletions were similar among all groups ( Figure 4B ). Analysis of the ratio between insertions and deletions clearly showed that M24.5 (1.18) and M30.5 (0.75) were much higher than CK (0.45) and M33 (0.4) ( Figure 4C ). These data suggest that the ratio of insertions to deletions is positively correlated with the gradient in spite of no significant effect of UHSMF on the number of insertions.

We then dissected the effect of UHSMF on the size of InDels. Our data showed that the number of InDels of 1 to 3 bp per line were similar between UHSMF treatments and CK ( Figure 4D ). However, there were obvious differences in InDels larger than 3 bp among the four groups. The number of such large InDels per line was 0.16 in M24.5, significantly higher than that in CK (0.03) ( Figure 4E ), and the percentage of the large InDels gradually increased with an increase in the magnetic gradient ( Figure 4F ). We did not observe any significant differences in distribution of InDels across the five Chrs ( Figure 4G ). In addition, we also analyzed the flanking sequences of InDels, and found that approximately 50% of the InDels were adjacent to a homopolymer or polynucleotide repeat in each group ( Supplementary Figure 6A , Supplementary Table 2 ). InDels smaller than 3 bp showed a similar pattern in CK and UHSMF treatments while homopolymeric and polynucleotide repeat InDels with larger than 3 bp were only present in M30.5 and M24.5, respectively ( Supplementary Figures 6B, C ). Meanwhile, flanking sequence with homopolymers was more frequently detected for insertions but less for deletions in CK and M24.5 ( Supplementary Figures 6D, E ). Taken together, our results suggest that UHSMF gradient is positively correlated with the number of InDels larger than 3 bp.

We mapped all mutations in the genome-wide scale, and found that mutations were randomly distributed across five chromosomes in all groups, except that a relatively greater density of mutations was observed in the proximity of the centromere, particularly on Chr3 ( Supplementary Figure 7 ). In fact, such a distribution pattern of mutations was already reported in Arabidopsis (Ossowski et al., 2010). These results suggest that UHSMF exposure has no bias effect on mutation distribution across chromosomes.

To predict whether mutations affect gene function and expression, we analyzed deleteriousness of all SNVs depending on their sites in intergenic region, untranslated region (UTR), intron and exon based on TAIR10 annotation (Cingolani et al., 2012). In general, most (from 54.77 to 72.57%) of the variants were located in the intergenic region, followed by in the exon (11.46 to 25.88%), intron (8.00 to 13.07%) and UTR (2.29 to 6.28%) in all groups. Interestingly, M24.5 treatment with the largest gradient resulted in the highest mutation (25.88%) in exons while the lowest mutation (54.77%) in intergenic regions ( Figure 5A ). However, no significant difference was detected in the average number of variants in exons and other regions among all the groups ( Figure 5B ).

To further check the potential effects of exon mutations, we analyzed variants deleteriousness via the Sorting Intolerant From Tolerant (SIFT) algorithm, namely SIFT4G (Vaser et al., 2016). SIFT4G predicted deleterious effects ranging from 0 (deleterious) to 1 (tolerated) on amino acid changes. It is generally recognized that the variants with score less than 0.05 are deleterious. Based on this criterion, 7.14% of the variants in CK were deleterious (or intolerance), including the deleterious SNVs (SIFT < 0.05) and InDels in coding region ( Figure 5C ). Compared to that of CK (7.14%), proportion of deleterious variants was higher in M24.5 (8.04%) while lower in M33 (5.14%) and M30.5 (5.10%). On average, the number of deleterious variants were ranged from 0.21 to 0.41 per line in CK and UHSMF groups, and no significant difference was detected between CK and UHSMF samples ( Figure 5D ). Likewise, there were no significant differences in the number of deleterious SNVs and InDels between UHSMF treatments and CK ( Figures 5E, F ). Taken together, our data indicate that UHSMF exposure has no effects on genome-wide distribution and deleteriousness of spontaneous mutations.

To investigate whether UHSMF exposure has a direct effect on seed physiology, we inspected the rate of seed germination in a time course. Our data showed that except for a significant decrease in germination rate of M24.5-expoured seeds at 24 h after seed sowing, other UHSMF treatments had no significant effect on germination rate examined at 24, 48 and 72 h, although a general tendency was observed that germination was faster for M33 seeds but slower for M30.5 and M24.5 seeds, compared to CK ( Figure 6 ). In addition, we found that among the three groups of SMF-exposed seeds, M33 seeds maintained significantly higher germination rate than M24.5 and M30 during the whole germination process, except for M24.5 at 72 h after sowing, while no significant difference in germination rate between M24.5 and M30.5 seeds except for at 24 h after sowing. These results suggest that the gradient but not the strength of SMF has a negative effect on seed germination, particularly at the early stage.

To check the potential impact of genetic variations on seed germination especially in M24.5, genes responsible for the mutation sites were annotated with Gene Ontology (GO) biological process by Metascape (Zhou et al., 2019). Throughout the annotation results, we only found the HEAT SHOCK COGNATE PROTEIN 70-1 (HSC70-1, AT5G02500), which mutated in M24.5 with SIFT score 0, was involved in seed germination process ( Supplementary Table 3 ). However, the function of HSC70-1 was redundant with HSC70-2 and HSC70-3 (Zhao et al., 2021). Thus, the germination rate changes may be not due to variants in lines of HSC70-1. It is not clear whether epigenetic variants occurred in these lines and subsequently affect the germination rates.

UHSMF up to 33.0 T used in this study were produced by a WM5 water-cooled resistive magnet (Gao et al., 2016; Tian et al., 2021) (High Magnetic Field Laboratory of Chinese Academy of Sciences). The temperature inside the device was maintained at 22-24°C by utilizing thermal conduction from temperature-controlled water that circulated through the gap between the inner tube and outer tube. Dried seeds from a single plant (Col-0) were set in five layers with each 50 mm working bore space. The UHSMF parameters for each layer has been summarized in Figure 1A . The UHSMF exposure were lasted for 1 h.

After UHSMF exposure, the seeds were surface-sterilized and stratified at 4°C in dark for 2 days, and then sown on the half-strength Murashige and Skoog (MS) medium (pH 5.7) (Sigma-Aldrich) containing 0.7% (w/v) phyto agar and 1% (w/v) sucrose. Seeds were germinated at 20°C under 100 μmol·m^–2·s^–1 white light with long-day conditions (16 h light/8 h darkness), and at least 50 seedlings for each treatment were transplanted into soil. All the plants were self-fertilized for one generation.

For DNA extraction, rosette leaves were sampled from the second generation plants, and 40 individual plants were used for each treatment. Extracted genome DNA was qualified by checking gel electrophoresis and OD260/280. DNA concentration was accurately quantified by using Qubit. Samples with OD values between 1.8 - 2.0 and DNA content above 1.5 µg were used for library construction. Qualified DNA samples were randomly broken into 350 bp fragments and then constructed by TruSeq Library Construction Kit. Sequencing was done by Illumina platform paired-end 2 × 150 bp sequencing lane. Raw reads in FASTQ format were gotten for the further analysis.

Totally 236 lines were generated for whole-genome sequencing, with five treatment groups and one control group. Raw reads from sequencing were mapped onto the reference genome (TAIR10) using Bowtie2 (Langmead and Salzberg, 2012) software (version 2.3.2) after quality control. Samtools (Li et al., 2009) (version 1.5) was used to convert the mapped sam files to bam files. Then, potential PCR duplicates were removed by using “MarkDuplicates” in picard (version 1.119, https://broadinstitute.github.io/picard/). We employed GenomeAnalysisTK (McKenna et al., 2010) (version 3.4-0) with “RealignerTargetCreator” to identify and generate a list of target intervals. Furthermore, “InDelRealigner” was utilized to realign alignments around small InDels. “UnifiedGenotyper” in GATK was applied to call raw variants. The parameter for the variants calling was “-stand_call_conf 30, -stand_emit_conf 10”.

To reduce the error rate of variant calls, the filtering threshold for the variants are shown below: removal of loci with deletions greater than 20% and heterozygosity greater than 20%. For the loci that are reserved for variation in only one material, we call them unique site. SNPeffect (Cingolani et al., 2012) (version 3.6b) with parameter “-no-intergenic -no-downstream -no-upstream -no-intron -no-utr” to predict the molecular and structural effects of protein-coding variants. SIFT4G (Vaser et al., 2016) was used to predict whether an amino acid change would affect the function of a protein.

The backup of harvested rosette leaves which stored in -80°C were used for variant confirmation. DNA of leaves were extracted by TPS buffer (1M KCl, 0.1M Tris-HCl, 0.01M EDTA-2Na). The candidate sites were randomly selected and specific primers were designed ( Supplementary Table 4 ). For each selected site, the samples containing variant and not (at least 3 lines) were both checked. DNA fragments containing candidate sites in the center with at least 400 bp were cloned from genomic DNA and confirmed by sanger sequencing. The reads were aligned with candidate reference sequences and the sequencing peak diagrams were checked.

The surface-sterilized seeds were stratified at 4°C in dark for 2 days and then sown on the half-strength Murashige and Skoog (MS) medium (pH 5.7) (Sigma-Aldrich) containing 0.7% (w/v) phyto agar and 1% (w/v) sucrose. Seeds were germinated at 20°C under 100 μmol·m^–2·s^–1 white light with long-day conditions (16 h light/8 h darkness). The seeds of SMF-treated and control were sown on the same plate and counted at 24, 48 and 72 h after sowing. The statistical analysis was performed by one-way ANOVA with Tukey`s multiple comparison test and statistically difference was confirmed by P < 0.05.