Dominance in self-compatibility between subgenomes of allopolyploid Arabidopsis kamchatica shown by transgenic restoration of self-incompatibility

The evolutionary transition to self-compatibility facilitates polyploid speciation. In Arabidopsis relatives, the self-incompatibility system is characterized by epigenetic dominance modifiers, among which small RNAs suppress the expression of a recessive SCR/SP11 haplogroup. Although the contribution of dominance to polyploid self-compatibility is speculated, little functional evidence has been reported. Here we employ transgenic techniques to the allotetraploid plant A. kamchatica. We find that when the dominant SCR-B is repaired by removing a transposable element insertion, self-incompatibility is restored. This suggests that SCR was responsible for the evolution of self-compatibility. By contrast, the reconstruction of recessive SCR-D cannot restore self-incompatibility. These data indicate that the insertion in SCR-B conferred dominant self-compatibility to A. kamchatica. Dominant self-compatibility supports the prediction that dominant mutations increasing selfing rate can pass through Haldane’s sieve against recessive mutations. The dominance regulation between subgenomes inherited from progenitors contrasts with previous studies on novel epigenetic mutations at polyploidization termed genome shock.

inbreeding depression 4,[7][8][9][10][11][12][13][14][15] .However, Stebbins also argued that polyploid species may suffer from a "retarded" phenotypic evolution because of the masking effect 5,6 .Theoretical studies have suggested that additional alleles or duplicated copies may confer a masking effect on recessive and additive mutations because of redundancy, and therefore a mutation in each duplicated gene may need to be accumulated in allopolyploid species for phenotypic evolution such as the evolution of self-compatibility 3,4,16,17 .Dominance relationships can reconcile these two aspects of polyploidy by alleviating the masking effect 2,11,18,19 .Haldane suggested that dominant or partially dominant mutations contribute more to adaptive evolution than recessive mutations, which is referred to as Haldane's sieve 20 .The principle can be valid for both allelic interactions in diploid species and epistatic interactions between homeologous copies in polyploid species because allotetraploid species are effectively fixed heterozygotes of subgenomes derived from different species.Theoretical studies have shown that mutations that increase the selfing rate would be dominant to pass through the Haldane's sieve 21 .
The role of epigenetics in polyploid evolution has long been discussed 3,4,22 .In the synthetic allopolyploid plants of many species, genome-wide instability in gene expression and methylation has been reported and termed as "genome shock" 3,4,22,23 .It has been debated whether novel genetic and epigenetic variants contributed to adaptive evolution of polyploids or were deleterious 3,24,25 .Besides genome-wide studies on small RNA (sRNA) of allopolyploid species 26,27 , the dominance relationship of self-incompatibility in Brassicaceae regulated by sRNA has also been studied in polyploid species 2,11,19 .Selfincompatibility in Brassicaceae including Arabidopsis and Brassica is characterized by the dominance relationship in pollen (hereafter, male dominance) through sRNAs that function as "dominance modifiers" [28][29][30][31][32] .In this self-recognition self-incompatibility system, SRK (S-locus receptor kinase) on the female stigma and SCR/SP11 (S-locus cysteine-rich protein/S-locus protein 11) on the pollen bind together when both of them are derived from the same S-haplogroup and then inhibit the growth of self-pollen tubes through the downstream signaling pathway [33][34][35] .A large number of highly divergent S-haplogroups of the female and male specificity genes (SRK and SCR/SP11, respectively) segregate trans-specifically at the S-locus 11 .The S-haplogroups typically show dominance-recessiveness relationships, i.e., when Sx is dominant to Sy, only Sx-specificity is functional and increases mating success by enabling mating with non-self individuals that possess Sy while preventing self-fertilization caused by Sx [36][37][38] .Both male and female dominance (dominance in stigma) are known in Brassicaceae [36][37][38] , however, male dominance is more frequent, i.e., two S-haplogroups often show dominance in pollen but no dominance (called co-dominance) in stigma 36 .The molecular mechanism of male dominance has been revealed by transgenic experiments in diploid species 33 .sRNAs from a dominant S-haplogroup at the S-locus regulate the tissue-specific methylation of the SCR gene of a recessive S-haplogroup, thereby suppressing its expression and function 28,[30][31][32] .In natural allotetraploid species in Brassicaceae, sequencing and crossing experiments suggested the relevance of dominance in self-compatibility 2,11,19 .In the crop allotetraploid species Brassica napus, loss-of-function mutations were detected in the dominant haplotypes of SRK or SCR 39,40 .In Arabidopsis suecica, a loss-of-function mutation in the SCR-A and mir867 sRNA that has a homology to the recessive SCR02 were inherited from A. thaliana 41 .In Capsella bursa-pastoris, the B subgenome has loss-offunction mutations in SCR as well as mirS3 that has a potential binding site in the A subgenome, although its relevance in SCR dominance regulation is unclear due to the lack of full-length SCR sequence in the A genome 42 .Dominance in a hybrid between different species and in synthetic allopolyploid plants was also studied 43 .Although these data suggest the importance of dominance in the allopolyploid species 2,11,18,19 , functional evidence using the transgenic technique has been awaited to test the relevance of dominance.
Arabidopsis kamchatica (2n = 4x = 32) is a model allopolyploid species with a broad distribution range including locality with heavy metal, and the transgenic technique using floral dip was established [44][45][46][47] .It is a natural self-compatible and allotetraploid species derived from multiple individuals of two progenitor species, A. halleri and A. lyrata 44,48,49 .Both progenitor species are predominantly self-incompatible, although self-compatible individuals of A. lyrata have been reported [49][50][51][52] .A. kamchatica has two S-loci derived from each progenitor species and shows disomic inheritance 18 .S-haplogroups A, B, and C segregate at the S-locus derived from A. halleri, and haplogroups D and E from A. lyrata 18 .The genotyping of 49 individuals covering most of the distribution range identified the following combinations of S-haplogroups: AD (47%, Takashima accession as a representative in this study, abbreviated as Tak), BD (34%, Potter accession, Pot), BE (8%, Okhotsk accession, Okh), individuals with only a single S-haplogroup amplified (6%) and CD (5%) 18 .Despite the self-compatibility of A. kamchatica, full-length sequences with no obvious gene-disruptive mutations were found in SRK-A, SRK-B, SRK-D, and SRK-E 18 .Furthermore, crossing experiments showed functional female selfincompatibility in individuals with full-length SRK-A, SRK-B, and SRK-D 18 .These data showed that these SRK haplogroups as well as downstream female signaling genes retained functionality even though all of these individuals are self-compatible 18 .This suggests that the self-compatibility of A. kamchatica was caused by genedisruptive mutations of the male self-incompatibility gene SCR or other genes necessary for self-incompatibility in pollen, although further transgenic evidence is required to show which gene was responsible for the evolution of self-compatibility 53 .Thus far, no experimental evidence on the existence of dominance either in pollen or in stigma has been reported among the three major Shaplogroup combinations of A. kamchatica (AD, BD, BE), although Shaplogroup C is shown to be most recessive both in male and female self-incompatibility in diploid species 37 .The crossing experiments described above also showed that S-haplogroups B and D exhibit no dominance (that is, co-dominant) in the pistil of A. kamchatica 18 , although it is possible that male dominance exists between B and D haplogroups because many pairs of S-haplogroups showed dominance only in pollen and co-dominance in stigma 36 .
Functional experiments such as transgenic technique and genome editing are important to identify the genes responsible for the evolution of self-compatibility because sequence analysis alone may be confounded by mutations that are difficult to be detected such as those in regulatory regions resulting in expression loss and by secondary degrading mutations in other genes in the selfincompatibility system after the loss of self-incompatibility 11 .In A. thaliana, the standard accession Col-0 had a gene-disruptive mutation both in SCR and SRK 54 but several accessions including Wei-1 have a full-length SRK 55 .By the transgenic introduction of a repaired SCR, A. thaliana Wei-1 restored self-incompatibility 53 .Such transgenic restoration by introducing a single gene shows that the mutation in the gene is responsible for self-compatibility and that all other genes for self-incompatibility are functional.The selfincompatibility system of Brassicaceae involves many genes other than SCR and SRK, and therefore self-compatibility may be attributed to so-called modifier genes such as M-locus protein kinase gene that are not linked to S-locus 35 , or to an unknown locus in the selfcompatible populations of A. lyrata 50,56 .Because the sRNAs regulating male dominance are located at the S-locus 28,30 , the selfcompatibility mutation must be located at the S-locus for dominance to be linked.
In this study, we experimentally examined the male dominance relationships using the self-incompatible diploid species.Based on the sequence analysis of SCR and sRNA sequences in A. kamchatica, we experimentally restored the SCR sequences of S-haplogroups B and D by transgenic experiments.Using them, we tested whether the male self-incompatibility specificity gene SCR was responsible for the evolution of self-compatibility and whether restoration of the dominant but not the recessive S-haplogroup can confer selfincompatibility.

Dominance relationship between S-haplogroups in pollen
By exploiting the trans-specific sharing of the S-haplogroups 18 , we examined the dominance relationship and functional sequences using the diploid self-incompatible A. halleri.Crossing experiments showed that in the pollen, the S-haplogroup B was dominant over D (noted as B > D), that is, the pollen of heterozygote BD was rejected only by the pistil with B specificity, but not by that with D specificity (Fig. 1a, Supplementary Fig. 1a).Similarly, the S-haplogroup A was dominant over D (A > D) (Fig. 1a, and Supplementary Fig. 1a).In addition, S-haplogroups A and B were co-dominant (A = B) (Supplementary Fig. 1a).Together with a previous report showing S-haplogroup D (=Ah12) > E (=Ah02) 37 , our data revealed the dominance relationship in pollen, as A = B > D > E. These experiments suggest that S-haplogroups A and B are dominant, whereas D and E are recessive in A. kamchatica.

Gene-disruptive mutations in SCR genes
The S-locus harbors highly divergent sequences of SRK and SCR and sRNAs 33 .We previously found that female self-incompatibility is functional in some A. kamchatica accessions having full-length alleles of SRK-A, B, and D 18 .Thus, self-compatibility may have evolved through the loss-of-function of SCR.We used the transcriptome sequencing approach to isolate SCR sequences from A. kamchatica and A. halleri because SCR sequences are too short and highly polymorphic to be isolated using polymerase chain reaction (PCR).Then, we confirmed the linkage to S-locus by checking co-segregation with the corresponding SRK haplogroup (Supplementary Table 1).The repressed expression in A. halleri corresponded to the dominance relationships in pollen as described in the previous section (Supplementary Fig. 1b).We searched for loss-of-function mutations in the dominant S-haplogroups of A. kamchatica.AkSCR-B had a 3,960-bp insertion with homology to a transposable element of the Mutator family, AtMU13, at the intron region in all tested A. kamchatica accessions bearing haplogroup B (Fig. 1b).The expression of AkSCR-B was lost in both accessions with the BD and BE haplogroups (Fig. 1c, Supplementary Fig. 1c), strongly suggesting that it is nonfunctional.Among AkSCR-A accessions, Taiwan accessions had a frameshift by G to AC substitution, whereas all the other tested accessions had a frameshift by insertion of a 7-bp repeat (Supplementary Fig. 2).Next, we examined the recessive S-haplogroups.In agreement with the dominance relationship in A. halleri, AkSCR-E was not expressed in the BE haplogroup,  1a).See Supplementary Table 8 for the genotypes of A. halleri individuals.b Gene structure of SCR-B.Blue boxes: exons; blue lines: intron and intergenic regions; red triangle: insertion in AkSCR-B.c The expression level of SCR genes in the anther cDNA of A. halleri and A. kamchatica bearing different S-haplogroups. A. kamchatica accessions from Takashima, Japan (Tak; haplogroups A and D); Murodo, Japan (Mur; haplogroups A and D); Taiwan (Tw; haplogroups A and D); and Potter, Alaska (Pot; haplogroups B and D).ACTIN was amplified as a quantitative control.d Expression of AkAmirS2, AkB-mirS2, and AkBmirS3 precursor genes in the anther cDNA of A. kamchatica from Potter (Pot; haplogroups B and D), Okhotsk (Okh; haplogroups B and E), and Takashima (Tak, haplogroups A and D).+: Reverse transcription with reverse transcriptase; −: reverse transcription without reverse transcriptase as the negative control because primers amplifying the precursor genes do not flank an intron.e Sequence alignment of AkBmirS2 of dominant S-haplogroup B and its target site at the promoter of recessive SCR-D obtained from A. halleri.AkBmirS2 sRNA sequences are highly homologous to their target site (22 of 24 nucleotides were identical).X: mismatches.
despite the absence of a large indel or obvious loss-of-function mutation (Supplementary Fig. 3).Southern blotting analysis and read mapping of high-throughput genomic sequence data of A. kamchatica bearing S-haplogroup D to bacterial artificial chromosome (BAC) sequences of S-haplogroup Ah12 (=AkD) suggested a deletion in the chromosomal region encompassing AkSCR-D (Supplementary Fig. 4), consistent with the lack of AkSCR-D expression (Fig. 1c).Due to the lack of SCR-D sequence in A. kamchatica, that of A. halleri is used in the following analyses.

Isolation of small RNAs and their precursor genes
To examine the involvement of sRNA in A. kamchatica self-compatibility, we searched for S-located sRNAs that show homology to the genomic region near the two recessive SCRs, SCR-D and SCR-E.First, we found the precursor genes of mirS2 and mirS3 sRNAs from each of the two dominant S-haplogroups B and A (Supplementary Figs.5-8, named AkBmirS2, AkBmirS3, AkAmirS2, AkAmirS3 precursor genes).We confirmed that the former three of them co-segregated with SCR sequences at the S-locus (Supplementary Table 1).Second, sRNA highthroughput sequencing was conducted to detect sRNA processed from the precursors.Mapping sRNA reads to the AkBmirS2 precursor gene from the BD (Potter) and BE (Okhotsk) accessions revealed the presence of a 24-nt sRNA in anthers (Supplementary Fig. 6a), and the plant miRNA-target scoring system supported its binding to the promoter of the recessive SCR-D (Fig. 1e and Supplementary Fig. 6c).Similarly, mapping sRNA reads to the AkBmirS3 precursor gene revealed a 24-nt sRNA that targets the intron of SCR-E (Supplementary Figs. 7, 8).These findings support the hypothesis that mirS2 and mirS3 sRNAs are derived from S-haplogroup B precursors and can target SCR-D and SCR-E, respectively.The 24-nt sRNA of AkAmirS3 was found in the AD (Takashima) accession (Supplementary Fig. 8b), but not that of AkAmirS2 (Supplementary Fig. 6b), which was not detected despite the expression of its precursor gene AkAmirS2 (Fig. 1d).This may indicate a secondary decay of the precursor processing of AkAmirS2 and provide a unique control material for subsequent experiments.

Evolutionary restoration of self-incompatibility by repairing a mutation in the dominant S-haplogroup
In natural populations, the S-haplogroup B was combined with both the recessive haplogroups D and E. The sRNA data suggested that the 3,960-bp insertion in AkSCR-B represents a self-compatible mutation that is dominant over the S-haplogroups D and E. These data suggested that self-incompatibility should be restored when the function of the dominant SCR is restored, but not when that of the recessive SCR is restored.Therefore, a restored AkSCR-B of Potter accession without the 3960-bp insertion was generated.The restored AkSCR-B, under the control of its native promoter, was transformed into Potter accessions (Supplementary Fig. 9a).AkSCR-B expression at the anthers was detected in four independent transgenic lines.Self-incompatibility reaction upon self-pollination was detected, except for the line with a low expression (Pot_SCRB_4) (Fig. 2a-f and Supplementary Table 2).Moreover, they produced significantly shorter siliques with fewer seeds per silique than did the wild-type Potter accession (Fig. 2g-i).Further crossing experiments confirmed the viability of the pollen and pistils of these transgenic lines (Supplementary Fig. 9b, c, and Supplementary Tables 3 and 4).These results indicate that selfincompatibility was restored by repairing the single loss-of-function mutation in AkSCR-B.

No restoration of self-incompatibility by repairing a mutation in the recessive S-haplogroup
In contrast, four transgenic Potter lines transformed with AhSCR-D under the control of its native promoter neither expressed SCR-D nor recover self-incompatible reactions upon self-pollination (Fig. 3a, b and Supplementary Table 5).No significant differences in the silique length or number of seeds per silique between the transgenic lines and the wild-type Potter accession were observed (Fig. 3c, d).This result in the recessive S-haplogroup contrasts with that in the dominant one; however, it does not rule out the possibility that the AhSCR-D genomic fragment is nonfunctional because of the lack of some regulatory elements.
To confirm that the AhSCR-D construct is adequate to confer selfincompatibility, it was introduced into the Takashima accession.Consistent with the lack of the 24-nt mirS2 sRNA in this accession, the expression of SCR-D was detected in all seven transgenic lines with a variable expression level.Self-incompatibility reaction was fully or partially observed in all the lines (Supplementary Table 6).Significantly shorter siliques and fewer seeds compared to those in the wild type were observed except for the silique length of the transgenic line Tak_SCRD_9, which showed the weakest SCR-D expression (Fig. 4a-d and Supplementary Table 7).Lines with a higher expression level such as Tak_SCRD_8 and Tak_SCRD_13 produced nearly no seeds (Fig. 4a-d).This experiment showed that the AhSCR-D genomic fragment is adequate to confer self-incompatibility in this accession.Furthermore, when the AhSCR-D promoter was introduced into the self-compatible A. thaliana, it was active in the stamen (Supplementary Fig. 10), indicating that the cloned promoter region is sufficient for expression.

Discussion
In this study, we first found the dominance of S-haplogroups A/B over D in pollen, and identified SCR sequences and sRNA.Transgenic plants with repaired SCR of the dominant S-haplogroup B restored the selfincompatibility.Similar to a previous transgenic study of the diploid natural species A. thaliana 53 , our results indicate that the mutation in the male specificity gene at the S-locus was responsible for selfcompatibility also in A. kamchatica, supporting the adaptive spread of male self-compatible mutations predicted by the theory of sexual asymmetry 12 .Furthermore, our transgenic experiment showed that the self-compatible mutation was located at the S-locus, which encompasses the sRNAs responsible for male dominance.By contrast, the self-incompatibility was not restored when the recessive S-haplogroup D was repaired in the same accession, although this construct turned out functional in another accession that lacked the mirS2 sRNA.Together, the data suggest that a single loss-of-function mutation in the self-incompatibility specificity gene SCR-B conferred dominant self-compatibility.
The high frequency of loss-of-function of SCR or SRK, which can be induced by various mutations, such as frameshift and transposable insertion, can explain the prevalence and rapid evolution of selfcompatibility in both polyploids and diploids.We suggest that the male dominance relationship plays a role in the evolution of selfcompatibility in many natural and domesticated polyploid species 2 .In Brassica napus 39,40 or the BE haplogroup of A. kamchatica, recessive Shaplogroups did not harbor obvious loss-of-function mutations, consistent with the current functional importance of suppression.In A. suecica 41 , Capsella bursa-pastoris 42,43 , and individuals with S-haplogroups D and C of A. kamchatica 18 , recessive S-haplogroups also had loss-of-function mutations.These mutations may represent secondary decay because self-incompatibility genes are not constrained after becoming self-compatible 53,55 .Alternatively, the mutations may have been segregated or fixed in diploid progenitor species, considering that the transition to self-compatibility is among the most frequent evolutionary transitions in angiosperms 7,11,49,51 .Female dominance can also contribute to the self-compatibility of allopolyploid species.In any case, dominance can allow a single mutation to confer selfcompatibility in allopolyploid species, indicating that the existence of duplicated loci did not necessarily "retard" evolution 6 .In addition to duplicated S-loci in allopolyploid species, where experimental study is facilitated by fixed heterozygosity, we propose that dominance played a role in heterozygous alleles in diploid species and in each S-locus of allotetraploid species transiently during the evolution of selfcompatibility until becoming homozygous.The relevance in diploid species is supported by the prevalence of dominant S-haplogroups having self-compatible mutations 11 .These findings support the theoretical prediction of Haldane's sieve regarding the prevalence of dominant mutations increasing selfing rate 11,20,21 .
Our study highlighted the importance of experimental evidence in addition to sequencing for clarifying complex dominance mechanisms in the Brassicaceae self-incompatibility.The co-dominance of Shaplogroups B and D was observed in pistils 18 , and a previous report pointed out that S-haplogroup D belongs to the most dominant "phylogenetic class" 51 based on phylogenetic analysis of SRK sequences 32 .However, our crossing experiment showed the male dominance of B over D. Furthermore, the evolutionary reversal experiment of SCR-B showed that the mutation in SCR was responsible for the evolution of self-compatibility, not that in other known or unknown genes such as an S-unlinked modifier gene recently reported from a self-compatible population of A. lyrata 56 .In this study, we focused on the transgenic experiments of SCR but did not manipulate sRNA.The epigenetic regulation of SCR expression by sRNA was also supported by transgenic experiments in diploid self-incompatible species 28,[30][31][32] , and therefore we also suggest the importance of epigenetic regulation in the dominance of the allopolyploid A. kamchatica.7 for exact Pvalues).Boxplots h and i show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying f, h and i are provided as a Source data file.
The S-locus inherited from A. halleri and another inherited from A. lyrata were combined in A. kamchatica 18 .Our study suggested that the S-haplogroup located on one of the two subgenomes was dominant over another S-haplogroup located on another subgenome.Analogous to the epigenetic regulation by sRNAs in diploid heterozygotes, our data lead to a hypothesis of epigenetic regulation between subgenomes.Many previous studies on the epigenetics of polyploid species focused on the contribution of stochastic epigenetic mutations in synthetic polyploids, which were considered to yield novel variations 3,4,[22][23][24][25] .Our findings suggest that epigenetic interactions between inherited and combined subgenomes provide a mechanism for facilitating the evolution of a typical trait of polyploidy, i.e., self-compatibility.Arabidopsis kamchatica and other polyploid species with transgenic technique 57,58 will be valuable to assess the role of stochastic and inherited epigenetic regulations in other adaptive traits.

Plant materials
Populations of A. halleri and A. kamchatica used in this study are previously reported 18,48 and listed in Supplementary Table 8.Arabidopsis kamchatica (formerly called Arabidopsis lyrata subsp.kamchatica) is distributed across a broad latitudinal range in East Asia and North America, and can grow in soils with a broad range of the concentration of heavy metals 44,46,47,59 .The plants were grown in a plant chamber under long-day conditions (16 h light/8 h dark cycle, 22 °C).After the production of a few adult leaves, they were moved to a 4 °C chamber (8 h light/16 h dark cycle) for 4-6 weeks for vernalization to induce flowering.

Isolation of SCR genes
Total RNAs were extracted from the anthers of 20 flower buds (stage 12) 60 each of A. halleri bearing haplogroups A, B, or D using an RNeasy Plant Mini kit (Qiagen, Hilden, Germany).Complementary DNA (cDNA) was synthesized using the SMARTer kit (TaKaRa Bio USA Inc., San Jose, CA, USA) and submitted for high-throughput sequencing using the Ion Torrent PGM sequencer (Life Technologies, Carlsbad, CA, USA) at the Functional Genomic Center Zurich (Zurich, Switzerland).RNA-seq data were de novo assembled by Trinity 61 (default parameters).A total of 15 known SCR amino acid sequences were used as queries in the TBLASTN homology search against the assembled data as a database (Supplementary Fig. 1d).
AkSCR-E was isolated using the amino acid sequence of AhSCR29 32 , which is phylogenetically close to SCR-E 18 , as a query in the TBLASTN homology searches against the assembled high-throughput genomic sequence data of A. kamchatica from Okhotsk that has S-haplogroups B and E (assembled by SOAPdenovo2 61 with k-mer = 57 as predicted by KmerGenie 62 ).From the TBLASTN results, potential contigs were selected by manually screening for the presence of the eight conserved cysteine residues in SCR.Next, candidates that were detected in more than one A. halleri bearing different haplogroups were eliminated because SCR genes of different S-haplogroups must be highly polymorphic among each other.Primers were designed around the stop codons of the candidates for genotyping by PCR using ExTaq polymerase (TaKaRa Bio Inc., Shiga, Japan) (Supplementary Table 9).SCR co-segregates with SRK of the same haplogroup, as they are tightly linked at the S-locus.Using the specific primers that amplify putative SCR-A, SCR-B, SCR-E, AkSRK-A, AkSRK-B, and AkSRK-E genes, segregation of SCR-A, SCR-B, and SCR-E were checked in an F 2 population (94 individuals) generated by crossing A. kamchatica bearing S-haplogroups A and D and A. kamchatica bearing S-haplogroups B and E (Supplementary Table 1).Additionally, SCR-D isolated from A. halleri could not be amplified in A. kamchatica.Therefore, the segregation pattern of SCR-D was examined in F 2 populations (32 individuals) generated by crossing the A. halleri bearing haplogroup D with A. halleri bearing haplogroup B or A (Supplementary Table 10).

Isolation of full-length SCR genes
Genomic DNA was extracted from the leaf tissues of A. halleri and A. kamchatica using a DNeasy Plant Mini Kit (Qiagen).Primers were designed using SCR genes based on the assembled RNA-seq data to isolate full-length SCR genes of A. halleri and A. kamchatica (Supplementary Tables 8 and 9).Mutations in the AkSCR genes were detected by comparing the SCR sequences of A. halleri and A. kamchatica in Geneious 6.1.7 63 .Insertion in AkSCR-B was isolated following the instructions of the GenomeWalker™ Universal Kit User Manual (TaKaRa Bio USA Inc., San Jose, CA, USA).The promoter regions of SCR genes were also isolated using the same method.DNA samples of A. halleri from various populations were genotyped to identify the A. halleri bearing haplogroup E for the isolation of full-length AhSCR-E.

Expression of SCR genes
Total RNA was extracted from the anthers at stage 12 of A. halleri and wild and transgenic A. kamchatica using an RNeasy Plant Mini kit (Qiagen).cDNAs were synthesized from 500 ng of each of the total RNA using a High-Capacity RNA-to-cDNA™ Kit (Applied Biosystems, Foster City, CA, USA).Primers flanking the intron of SCR genes were used for reverse transcription polymerase chain reaction (RT-PCR) and a reverse transcription quantitative PCR (RT-qPCR) (Supplementary Table 9).The latter was performed using PowerUp SYBR Green Master Mix (A25742) in StepOnePlus Real-Time PCR System of Applied Biosystems.Gel pictures were produced using Gel Doc XR+ Gel Documentation System with Image Lab Software ver.4.1 (Bio-Rad, Hercules, CA, USA) and ImageJ ver.2.15.0.ACTIN or EF1α (ELONGATION FACTOR 1α) was amplified as the quantitative control.

Probe preparation for southern blot
A 317-bp probe designed at the second exon and 3′ UTR of SCR-D was cloned into the pCR™4-TOPO® vector using the TOPO® TA Cloning® kit (Thermo Fisher Scientific, Waltham, MA, USA) (Supplementary Table 9).Using the vector as a template, the probe was labeled with DIG-11-dUTP via PCR according to the manufacturer's instruction of the PCR DIG Probe Synthesis Kit (Roche Applied Science, Indianapolis, IN, USA).Labeling was confirmed using gel electrophoresis, as the labeled probes have higher molecular weight than unlabeled probes.

Southern blotting analysis
Genomic DNA was extracted from the leaf tissues of A. halleri bearing haplogroup D and three A. kamchatica accessions-Murodo, Takashima, and Potter accessions using DNeasy Plant Maxi Kit (Qiagen, Hilden, Germany).A total of 10 µg of each of the accessions was digested with XbaI (New England Biolabs, Beverly, MA, USA) at 37 °C overnight.Digested DNA, together with DIG-labeled DNA Molecular Weight Marker VII (Roche Applied Science), was separated using 1% agarose gel (at 100 V for 1 h).Blotting analysis and hybridization were   performed according to the DIG Application Manual for Filter Hybridization (Roche Applied Science).Capillary transfer of DNA to a positively charged nylon membrane (Roche Applied Science) was performed overnight.DNA was then fixed to the blot via UV crosslinking.The blots were hybridized at 40 °C in 10 ml of hybridization buffer at a concentration of 6 μl of the labeled probes per ml of hybridization buffer.Membranes were washed and chromogenic detections were performed using anti-DIG-alkaline phosphatase and NBT/BCIP (DIG-High Prime DNA Labeling and Detection Starter Kit I, Roche Applied Science) according to the manufacturer's instructions.To test the dominance relationship between A and B, in addition to stigmas, the pollen grains of A. halleri individuals with AB were crossed to the stigmas bearing haplogroup A, B, as well as D, the last of which was performed to show that the pollen grains were viable in this co-dominant case.Pollen tube growth was examined using aniline blue staining 18 .S-specificity of A. halleri used in this experiment was previously confirmed via interspecific crossings with A. kamchatica 18 .The expression of SCR-A, SCR-B, and SCR-D were also examined using these genotypes.

Isolation of sRNA precursors, detection of sRNA, and mapping
The sRNA precursor of A. lyrata bearing haplogroup Sb has been previously isolated 31 .Primers were designed based on this precursor to isolate sRNA precursors in A. kamchatica (Supplementary Table 9).Both the precursor sequences amplified from A. kamchatica bearing haplogroup A and A. kamchatica bearing haplogroup B showed high sequence homology with the mirS2 family, thus designated as AkA-mirS2 and AkBmirS2, respectively.The segregation patterns of AkA-mirS2 and AkBmirS2 were examined in an F 2 population (94 individuals) generated by crossing A. kamchatica bearing haplogroups A and D and A. kamchatica bearing haplogroups B and E. AkAmirS2 and AkBmirS2 were perfectly linked to S-haplogroups A and B, respectively (Supplementary Table 1).The secondary structures of the precursor genes were predicted by Geneious 6.1.7 63 .
Total RNA was extracted from the anthers of 10 flower buds (stages 9-11) from each of these accessions: A. kamchatica from Potter and Okhotsk and Takashima using a mirVana™ miRNA Isolation Kit (Ambion, Austin, TX, USA).To check the expression of sRNA precursors, total RNA was treated with DNase (DNA-free™ DNA Removal Kit, Ambion) and reverse-transcribed to cDNA by High-Capacity RNAto-cDNA™ Kit (Applied Biosystems).Primers amplifying AkAmirS2 and AkBmirS2 were used for RT-PCR (Supplementary Table 9).ACTIN was amplified as the quantitative control.
Total RNA before DNase treatment was submitted for small RNA sequencing by Illumina HiSeq 2500 (Illumina, San Diego, CA, USA) at the Functional Genomics Center Zürich.Adaptors were trimmed, and reads shorter than 20 bp were filtered out before being mapped to sRNA precursors to examine the presence of AkAmirS2 and AkBmirS2.Reads were mapped to sRNA precursors using Bowtie 2.3.2 (command: bowtie2 --no-unal -x [file name of reference fasta] -U [file name of fastq]) 64 and viewed in the Integrative Genomics Viewer (IGV) 2.3.35 65.To isolate the sRNA targeting SCR-E, known sRNA precursor genes (mirS1, mirS2, mirS4, mirS5, mir867, mir1887, and mir4239) 32 were used as query in a homology search against the de novo assembled sequence data of Okhotsk and Takashima, assembled by SOAPdenovo2 66 with k-mer value of 85 as predicted by KmerGenie 62 .High-throughput sRNA reads of Okhotsk anthers were mapped to the potential precursors to detect the presence of 24-nt sRNA using Bowtie 2.3.2 64 and viewed in the IGV 2.3.35 65as described above.A 24-nt sRNA showing high sequence homology, with a target site at the intron region of AkSCR-E, was identified from the mirS3 precursor gene, and thus designated as AkBmirS3.AkBmirS3 co-segregates with SCR-B (Supplementary Tables 1 and 9).The presence of 24-nt mirS3 targeting AkSCR-E was also detected in Takashima, thus designated as AkAmirS3.BWA 0.7.12 (command: bwa mem -t8 [file name of reference fasta] [file name of fastq]) 67 and IGV 2.3.35 65were similarly used for the mapping and the visualization of genomic sequences (SAMD00045705, SAMD00089932, SAMD00089933) 49,68 on the BAC sequences of the Slocus region of S-haplogroup Ah12 (Genbank KJ772374.1) 32.

Cloning and transformation of restored AkSCR-B and AhSCR-D
The full-length sequences, including the promoter and coding sequences of AkSCR-B of Potter accession, were restored via a series of PCR amplification using high-fidelity PrimeSTAR GXL DNA polymerase (TaKaRa Bio Inc.) (Supplementary Table 9).First, two different fragments eliminating the insertion were amplified (AkSCR-B_1°).Then, 100× diluted PCR products (AkSCR-B_1°) were used as the templates for the second round of PCR to amplify the full-length restored AkSCR-B (AkSCR-B_2°).Since SCR-D has been deleted in A. kamchatica, SCR-D of A. halleri was cloned for transformation into A. kamchatica.A 3092-bp region of AhSCR-D, including 648-bp of promoter regions, two exons, an intron, and 85-bp after stop codon, was amplified by high-fidelity PrimeSTAR GXL DNA polymerase (TaKaRa Bio Inc.) (Supplementary Table 9).Then, 3′ A-overhangs were added to the restored sequences and cloned into pCR®8/GW/TOPO® (Thermo Fisher Scientific) according to the manufacturer's instructions.The sequences were inserted into the destination vector pFAST-G01 69 via LR recombination using LR Clonase® II Enzyme mix (Thermo Fisher Scientific).The vectors were then transformed into Agrobacterium tumefaciens GV3101 via electroporation.Constructs generated were transformed into the Potter accession bearing haplogroups B and D and the Takashima accession bearing haplogroups A and D using Agrobacterium-mediated floral dip transformation 45 .

Genotyping and phenotyping of transformed A. kamchatica
Transformed seeds were selected by observation under a fluorescence stereomicroscope with a GFP filter (Olympus SZX12, Japan).The DNA was extracted from the leaf tissues of T 1 plants, and insertion of generated constructs was confirmed via PCR using primers designed from the vectors and respective inserted SCR genes (Supplementary Table 9).Pollination assays were performed to test whether the male components of the transformed A. kamchatica are functional.The anthers of transgenic lines were removed from the flower buds at early stage 13 (before anthesis) and visualized under a stereomicroscope to ensure that there is no pollen on the stigmas.They were then manually pollinated with self-pollen.Stigmas were harvested 24 h after pollination, and pollen tube growth was examined using aniline blue staining 18 .Silique length and the number of seeds per silique of the wild type (WT) and transgenic plants were measured.Statistical significance was assessed by the Bonferroni-corrected Mann-Whitney U test and one-way ANOVA and Tukey HSD test in R. In addition to selfpollination, crosses between transformed pollen and WT pistil and between WT pollen and transformed pistil were checked to ensure the viability of the pollen and pistil, respectively.

Cloning of the AhSCR-D promoter and GUS staining
The 648-bp of the promoter region of AhSCR-D was amplified and cloned into the pCR TM 8/GW/TOPO® vector using a TA Cloning Kit (Thermo Fisher Scientific) according to the manufacturer's instructions.The promoter was inserted upstream of the β-Glucuronidase (GUS) reporter gene in the binary vector pGWB3 by LR recombination using LR Clonase® II Enzyme mix (Thermo Fisher Scientific).The vectors were then transformed into Agrobacterium tumefaciens GV3101 via electroporation.The construct was introduced into A. thaliana using the floral dip transformation method.Transformants were selected on 50 µl/ml kanamycin selection plates.Flower buds were collected and immersed in ice-chilled 90% acetone for 20 min.Samples were vacuum-filtrated for 15 min with GUS staining solution containing 1 mM X-Gluc, 50 mM sodium phosphate buffer, 0.1% Triton X-100, 2 mM potassium ferrocyanide, and 2 mM potassium ferricyanide.Samples were incubated overnight at 37 °C and treated with Herr clearing solution before observation under microscope 70 .

Statistics and reproducibility
Statistical analyses were conducted using R (4.2.2).Exact P-values are shown in Supplementary Table 7.The results of all RT-PCR gel images were verified by two more independent experiments.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Fig. 1 |
Fig.1| S-haplogroup dominance and expression of SCR genes and small RNAs.a Representative pictures of the crossings to investigate the dominance hierarchy of the S-haplogroups A, B, and D. Stigma of Arabidopsis halleri with the haplogroups A (top left), B (bottom left), and D (top and bottom right) crossed with pollen grains derived from heterozygote of AD or BD.Yellow arrows indicate a bundle of pollen tubes.Scale bar = 0.25 mm.At least six observations showed similar results for each crossing (sample numbers in Supplementary Fig.1a).See Supplementary Table8for the genotypes of A. halleri individuals.b Gene structure of SCR-B.Blue boxes: exons; blue lines: intron and intergenic regions; red triangle: insertion in AkSCR-B.c The expression level of SCR genes in the anther cDNA of A. halleri and A. kamchatica bearing different S-haplogroups. A. kamchatica accessions from Takashima, Japan (Tak; haplogroups A and D); Murodo, Japan (Mur; haplogroups A and D); Taiwan (Tw; haplogroups A and D); and Potter, Alaska (Pot; haplogroups B and D).ACTIN was amplified as a quantitative control.d Expression of AkAmirS2, AkB-mirS2, and AkBmirS3 precursor genes in the anther cDNA of A. kamchatica from Potter (Pot; haplogroups B and D), Okhotsk (Okh; haplogroups B and E), and Takashima (Tak, haplogroups A and D).+: Reverse transcription with reverse transcriptase; −: reverse transcription without reverse transcriptase as the negative control because primers amplifying the precursor genes do not flank an intron.e Sequence alignment of AkBmirS2 of dominant S-haplogroup B and its target site at the promoter of recessive SCR-D obtained from A. halleri.AkBmirS2 sRNA sequences are highly homologous to their target site (22 of 24 nucleotides were identical).X: mismatches.

Fig. 2 |
Fig. 2 | Introduction of restored AkSCR-B gene into Arabidopsis kamchatica (Potter accession) promoted recovery of self-incompatibility in A. kamchatica.Self-pollination assays of a wild-type (WT) Potter, b transgenic Potter with restored AkSCR-B (Pot_SCRB_1), c Pot_SCRB_2, d Pot_SCRB_3, and e Pot_SCRB_4.At least five observations showed similar results for each (sample numbers in Supplementary Table2).Yellow arrows indicate a bundle of pollen tubes.Scale bar = 0.25 mm.f Expression of SCR-B in the anther cDNAs of A. halleri bearing haplogroup B (Ahal-B), WT Potter, and four transgenic Potter lines (mean ± S.D., n each line = 3).g Siliques of WT Potter and four transgenic Potter lines.Scale bar = 10 mm.h Comparison of silique length from self-pollination of WT Potter and four transgenic Potter lines.Silique lengths were significantly reduced in the transgenic lines compared to the WT, except for Pot_SCRB_4 (one-way ANOVA and Tukey HSD test, n each line = 5).i Comparison of the number of seeds per silique from self-pollination of WT Potter and four transgenic Potter lines.The number of seeds per silique was significantly fewer in the transgenic Potter lines than in the WT, except for Pot_SCRB_4 (one-way ANOVA and Tukey HSD test, n each line = 5, see Supplementary Table7for exact Pvalues).Boxplots h and i show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying f, h and i are provided as a Source data file.

Fig. 3 |
Fig. 3 | Self-incompatibility was not restored after the introduction of the repaired recessive AhSCR-D.a SCR-D expression in the anther cDNAs of Arabidopsis halleri bearing haplogroup D (Ahal-D), WT Potter, and four transgenic Potter lines (Pot_SCRD).ACTIN was amplified as a quantitative control.b Siliques of WT Potter and transgenic Potter lines.Scale bar = 10 mm.c Comparison of silique length from self-pollination of WT Potter and transgenic Potter lines.The silique length of transgenic lines was not significantly different from that of the WT.d Comparison of the number of seeds per silique from self-pollination of WT Potter and transgenic Potter lines.The length of silique (c) or the number of seeds per silique (d) in the transgenic lines was not significantly different from that in the WT (one-way ANOVA and Tukey HSD test, n each line = 5, see Supplementary Table7for exact P-values).Boxplots c and d show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying c and d are provided as a Source data file.

7
Fig. 3 | Self-incompatibility was not restored after the introduction of the repaired recessive AhSCR-D.a SCR-D expression in the anther cDNAs of Arabidopsis halleri bearing haplogroup D (Ahal-D), WT Potter, and four transgenic Potter lines (Pot_SCRD).ACTIN was amplified as a quantitative control.b Siliques of WT Potter and transgenic Potter lines.Scale bar = 10 mm.c Comparison of silique length from self-pollination of WT Potter and transgenic Potter lines.The silique length of transgenic lines was not significantly different from that of the WT.d Comparison of the number of seeds per silique from self-pollination of WT Potter and transgenic Potter lines.The length of silique (c) or the number of seeds per silique (d) in the transgenic lines was not significantly different from that in the WT (one-way ANOVA and Tukey HSD test, n each line = 5, see Supplementary Table7for exact P-values).Boxplots c and d show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying c and d are provided as a Source data file.
Fig. 3 | Self-incompatibility was not restored after the introduction of the repaired recessive AhSCR-D.a SCR-D expression in the anther cDNAs of Arabidopsis halleri bearing haplogroup D (Ahal-D), WT Potter, and four transgenic Potter lines (Pot_SCRD).ACTIN was amplified as a quantitative control.b Siliques of WT Potter and transgenic Potter lines.Scale bar = 10 mm.c Comparison of silique length from self-pollination of WT Potter and transgenic Potter lines.The silique length of transgenic lines was not significantly different from that of the WT.d Comparison of the number of seeds per silique from self-pollination of WT Potter and transgenic Potter lines.The length of silique (c) or the number of seeds per silique (d) in the transgenic lines was not significantly different from that in the WT (one-way ANOVA and Tukey HSD test, n each line = 5, see Supplementary Table7for exact P-values).Boxplots c and d show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying c and d are provided as a Source data file.

Fig. 4 |
Fig. 4 | The repaired recessive AhSCR-D construct is functional in the Takashima accession, in which mirS2 small RNA expression is not detectable.a Expression of SCR-D in the anther cDNAs of Arabidopsis halleri bearing haplogroup D (Ahal-D), WT Takashima accession (WT), and seven transgenic Takashima lines (Tak_SCRD).ACTIN was amplified as a quantitative control.b Siliques derived from the WT Takashima accession and transgenic lines.Scale bar = 10 mm.c Comparison of silique length from self-pollination of WT and transgenic lines.Silique lengths were significantly reduced in the transgenic lines compared with those in the WT, except for in Tak_SCRD_9 (one-way ANOVA and Tukey HSD test, n each line = 5).d Comparison of the number of seeds per silique from self-pollination of WT and transgenic lines.The number of seeds per silique was significantly fewer in the transgenic lines than in the WT (one-way ANOVA and Tukey HSD test, n each line = 5, see Supplementary Table7for exact P-values).Boxplots c and d show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying c and d are provided as a Source data file.

7
Fig. 4 | The repaired recessive AhSCR-D construct is functional in the Takashima accession, in which mirS2 small RNA expression is not detectable.a Expression of SCR-D in the anther cDNAs of Arabidopsis halleri bearing haplogroup D (Ahal-D), WT Takashima accession (WT), and seven transgenic Takashima lines (Tak_SCRD).ACTIN was amplified as a quantitative control.b Siliques derived from the WT Takashima accession and transgenic lines.Scale bar = 10 mm.c Comparison of silique length from self-pollination of WT and transgenic lines.Silique lengths were significantly reduced in the transgenic lines compared with those in the WT, except for in Tak_SCRD_9 (one-way ANOVA and Tukey HSD test, n each line = 5).d Comparison of the number of seeds per silique from self-pollination of WT and transgenic lines.The number of seeds per silique was significantly fewer in the transgenic lines than in the WT (one-way ANOVA and Tukey HSD test, n each line = 5, see Supplementary Table7for exact P-values).Boxplots c and d show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying c and d are provided as a Source data file.
Fig. 4 | The repaired recessive AhSCR-D construct is functional in the Takashima accession, in which mirS2 small RNA expression is not detectable.a Expression of SCR-D in the anther cDNAs of Arabidopsis halleri bearing haplogroup D (Ahal-D), WT Takashima accession (WT), and seven transgenic Takashima lines (Tak_SCRD).ACTIN was amplified as a quantitative control.b Siliques derived from the WT Takashima accession and transgenic lines.Scale bar = 10 mm.c Comparison of silique length from self-pollination of WT and transgenic lines.Silique lengths were significantly reduced in the transgenic lines compared with those in the WT, except for in Tak_SCRD_9 (one-way ANOVA and Tukey HSD test, n each line = 5).d Comparison of the number of seeds per silique from self-pollination of WT and transgenic lines.The number of seeds per silique was significantly fewer in the transgenic lines than in the WT (one-way ANOVA and Tukey HSD test, n each line = 5, see Supplementary Table7for exact P-values).Boxplots c and d show center line: median; box limits: upper and lower quartiles; whiskers: within 1.5 times the interquartile range; dots: data points.Source data underlying c and d are provided as a Source data file.
Dominance hierarchy of S-haplogroups A, B, and DA.halleri individuals bearing haplogroup A, B, or D were crossed among each other to generate A. halleri bearing haplogroups A and B (AB), A. halleri bearing haplogroups A and D (AD), and A. halleri bearing haplogroups B and D (BD).To test the dominance relationship between A and D in pollen, the pollen grains of A. halleri individuals with AD were applied to the stigmas of A. halleri bearing haplogroup A and that bearing haplogroup D. Similarly, to test the dominance relationship between B and D in pollen, the pollen grains of A. halleri individuals with BD were crossed to stigmas bearing haplogroup B and those bearing haplogroup D.