Duplicated KAI2 receptors with divergent ligand-binding specificities control distinct developmental traits in Lotus japonicus

Abstract Karrikins (KARs), smoke-derived butenolides, are perceived by the α/β-fold hydrolase KARRIKIN INSENSITIVE2 (KAI2) and thought to mimic endogenous, unidentified plant hormones called KAI2-ligands (KLs). In legumes, KAI2 has duplicated. We addressed sub-functionalization of KAI2a and KAI2b in Lotus japonicus and demonstrate, their binding preferences to synthetic ligands differ in vitro and in a heterologous Arabidopsis background. These differences can be explained by three divergent amino acids near the binding pocket, two of which are conserved across legumes, suggesting legumes produce at least two KLs with different stereochemistry. Unexpectedly, L. japonicus responds organ-specifically to synthetic KAI2-ligands: hypocotyls respond to KAR1, KAR2 and rac-GR24; root systems respond only to KAR1. In hypocotyls, LjKAI2a is required for karrikin responses, while LjKAI2a and LjKAI2b operate redundantly in roots. Our results open novel research avenues into the diversity of butenolide ligand-receptor relationships and the mechanisms controlling diverse developmental responses to endogenous and synthetic KAI2 ligands.


Introduction
repress shoot branching 17,18 . They have also been suggested to affect lateral and adventitious root formation, root-hair elongation, secondary growth and nodulation reviewed in 19,20 .
SLs are perceived by the α/β-fold hydrolase D14/DAD2 that, like KAI2, interacts with the SCF-complex via the same F-box protein MAX2 21, 22 to ubiquitylate repressors and mark them for degradation by the 26S proteasome. The currently most likely repressors of karrikin/KL and SL signalling belong to the SUPPRESSOR OF MAX2 1-LIKE (SMXL) protein family and are closely related to class-I Clp ATPases 23, 24, 25 . They contain a wellconserved ethylene-responsive element binding factor-associated amphiphilic repression (EAR) motif, which interacts with TOPLESS proteins that act as transcriptional corepressors 25, 26, 27 . Thus, it appears that SL and KAR/KL signalling may function in releasing transcriptional repression. In Arabidopsis the SMXL family comprises 8 members and, it is generally assumed that SMAX1 and SMXL2 (SMAX1-LIKE2) repress KAR/KL responses , while SMXL6, SMXL7 and SMXL8 (collectively represented by a single homologue in rice, D53) redundantly repress strigolactone responses 10,23,24,25,26,27 .
Phylogenetic analysis of the α/β-fold hydrolase receptors in extant land plants revealed that an ancestral KAI2 is already present in charophyte algae, while the so-called eu-KAI2 is ubiquitous among the land plants. The strigolactone receptor gene, D14 evolved only in the seed plants likely through duplication of KAI2 and sub-functionalization 28 . An additional duplication in the seed plants gave rise to D14L2 (DLK2), an α/β-fold hydrolase of unknown function, which is transcriptionally induced in response to KAR treatment in a KAI2 and MAX2-dependent manner, and currently represents the best-characterized KAR marker gene in Arabidopsis 4,29 . Despite their similarity, KAI2 and D14 cannot replace each other in Arabidopsis, as shown by promoter swap experiments 30 . This indicates that thei specific expression pattern does not determine their signaling specificity. Instead, specificity must be caused by the tissue-specific presence of their specific ligands, their ability to interact with other proteins or both.
In Arabidopsis and rice, in which KAR/KL signalling has so far been mostly studied, KAI2 is a single copy gene. However, there are other plant species with several KAI2 copies.
For example, the Physcomitrella patens genome contains 11 genes encoding KAI2-like proteins. They have diversified in a loop that determines the rigidity of the ligand-binding pocket, such that some of them preferentially bind KAR1 and others the SL 5-desoxystrigol in vitro 31 . The genomes of parasitic plants of the Orobanchaceae and the genus Striga also contain several KAI2 copies. Some of these have evolved to perceive strigolactones, some can restore KAR-responses in Arabidopis kai2 mutants, and others do not mediate responses to any of these molecules in Arabidopsis 7,32,33 . Together, this indicates that in plant species with an expanded KAI2-family there is scope for a diverse range of ligandbinding specificities (as well as for diverse protein interaction partners). Interestingly, even among plants with only one KAI2 receptor gene, the responsiveness to karrikin molecules can differ significantly. For example, Arabidopsis plants respond to KAR2 with increased expression of the marker gene DLK2, reduced hypocotyl growth and increased root hair length and density and respond more strongly to KAR2 relative to plants treated with KAR1 4,10 . In contrast, rice roots did not display any transcriptional response to KAR2, not even for the marker gene DLK2 11 . It is unclear what determines these differences in KAR2 responsiveness among distinct plant species.
Legumes comprise a number of agronomically important crops and they are unique among plants as they can form nitrogen-fixing root nodule symbiosis with rhizobia in addition to arbuscular mycorrhiza. Given the possible diversity in KAI2-ligand specificities among plant species, we characterized the karrikin receptor machinery in a legume, the commonly used model L. japonicus, for which extensive reverse genetics ressources are available 34,35 . We found that KAI2 has duplicated in the legumes and that L. japonicus KAI2a and KAI2b differ in their binding preferences to synthetic ligands in vitro and in the heterologous Arabidopsis kai2-2 mutant background. We demonstrate that these ligand binding preferences can be explained by three amino acids at the binding pocket, suggesting that the duplicated KAI2 receptors may have sub-functionalized to perceive different versions of endogenous KL-molecules. We found a surprising organ-specific responsiveness to synthetic KAI2-ligands, with hypocotyl development responding to KAR1, KAR2 and rac-GR24, and root system development responding only to KAR1.
These responses depended only on LjKAI2a in hypocotyls, while LjKAI2a and LjKAI2b operated redundantly in roots. Together these findings suggest that a diversity of mechanisms may influence KAR/KL responses including receptor-ligand binding specificity or organ-specific interaction of KAI2 with other proteins.

KAI2 underwent duplication prior to diversification of the legumes
To characterize the karrikin and the strigolactone perception machinery in L. japonicus we retrieved KAI2, D14 and MAX2 by protein BLAST using Arabidopsis KAI2, D14 and MAX2 as templates. A phylogenetic tree revealed that LjD14 (Lj5g3v0310140.4) is a single copy gene whereas LjKAI2 is duplicated in the genome of L. japonicus, in contrast to Arabidopsis and rice (Fig. 1), resulting in two paralogs LjKAI2a (Lj2g3v1931930.1) and LjKAI2b (Lj0g3v0117039.1). The LjKAI2 duplication event must have occurred prior to the diversification of the legumes or at least before the separation of the Millettioids and Robinioids 36 a similar pattern of duplication is also detected in soybean, pea and Medicago truncatula.
The F-box protein-encoding gene LjMAX2 also underwent duplication likely as a result of whole genome duplication, because the two LjMAX2 copies are in two syntenic regions of the genome (Supplementary Fig. S1a). However, only one LjMAX2 copy (Lj3g3v2851180.1) is functional and the other copy ΨMAX2-like (Lj0g3v0059909.1) appears to be a pseudogene as it contains an early stop codon, resulting in a putative truncated protein of 216 instead of 710 amino acids ( Supplementary Fig. S1b). It appears that an insertion of one nucleotide into ΨMAX2-like created a frameshift, because manual deletion of Thymine 453 restores the correct nucleotide sequence, which would allow the synthesis of a full length MAX2-like protein ( Supplementary Fig. S1b).
We wondered why L. japonicus retained two intact copies of KAI2 and hypothesized that they may have functionally diverged, perhaps through changes in their expression pattern and/or amino acid sequence, which may cause differences in the relative spatial distribution of LjKAI2a and LjKAI2b, their ligand binding or their ability to interact with other proteins. We analysed the transcript accumulation of LjKAI2a and LjKAI2b, as well as Overall, both LjKAI2a and LjKAI2b transcripts accumulated to higher levels than those of LjD14 and LjMAX2. LjKAI2a transcripts accumulated approximately 100-fold more in aerial organs than LjKAI2b, whereas LjKAI2b accumulated 10-fold more than LjKAI2a in roots of adult plants, which were grown in a sand-vermiculite mix in pots. However, in 1- week-old seedlings grown on water-agar in Petri dishes in short-day conditions, LjKAI2a transcripts accumulated to more than 10-fold higher levels than LjKAI2b in both roots and hypocotyls ( Supplementary Fig. S2b). This difference between LjKAI2a and LjKAI2b transcript was less pronounced in roots grown in long-day conditions. Together, this indicates that LjKAI2a and LjKAI2b are regulated in an organ-specific, age-and/or environment-dependent manner, which implies that their individual expression pattern is caused by at least partially different transcriptional regulators.
We also examined the sub-cellular localization of the four corresponding proteins in transiently transformed Nicotiana benthamiana leaves, using fusions with TSaphire or mOrange. Similar to observations in Arabidopsis 37, 38 , TSaphire-MAX2 localized specifically to the nucleus, while the /-hydrolases (D14, KAI2a and KAI2b) fused to mOrange localized to the nucleus and cytoplasm ( Supplementary Fig. S3a). Western blot analysis confirmed that the mOrange signal observed in the cytoplasm was due to the fulllength fusion protein and not caused by free mOrange fluorophore, corroborating the dual localization of the two /-hydrolases ( Supplementary Fig. S3b).

L. japonicus KAI2a, KAI2b and D14 can replace their orthologs in Arabidopsis
We examined whether LjKAI2a and LjKAI2b may have evolved ligand binding specificities and/or different functions in plant development. To examine, whether they both function in a canonical manner, we employed a well-established hypocotyl elongation assay in Arabidopsis 7, 30 , after transgenically complementing the Arabidopsis thaliana kai2-2 mutant 4 with LjKAI2a and LjKAI2b driven by the AtKAI2 promoter. Both restored inhibition of hypocotyl elongation in the kai2-2 mutant, however LjKAI2b was not as efficient as LjKAI2a in four independent transgenic lines (Fig. 2a). We also examined the ability of LjD14 to restore hypocotyl growth inhibition, but as expected LjD14 driven by the AtKAI2 promoter did not restore the hypocotyl length in Atkai2-2. However, LjD14 restored repression of shoot branching of the Arabidopsis d14-1 mutant 4 , when driven by the Arabidopsis D14 promoter, which was not the case for LjKAI2a and LjKAI2b ( Fig. 2b and   2c). These results together with the phylogenetic tree ( Fig. 1)  and the paralog LjKAI2b to rescue the Atkai2-2 hypocotyl phenotype might be due to variation in affinity to endogenous karrikin-like ligand(s) or to interacting Arabidopsis proteins caused by variations in amino acids exposed at the protein surface.

Lotus japonicus KAI2a and KAI2b differ in their ligand binding specificity
To explore whether L. japonicus KAI2a and KAI2b can mediate hypocotyl responses to karrikins we quantified hypocotyl length of the Atkai2-2 lines transgenically complemented with LjKAI2a or LjKAI2b after treatment with KAR1 and KAR2 ( Fig. 3a and 3b). Two independent lines complemented with LjKAI2a displayed the same reduction in hypocotyl growth in response to KAR1 and to KAR2, similar to the line complemented with AtKAI2.
However, the two lines expressing LjKAI2b responded more strongly to KAR1 than to KAR2, contrasting with the common observation, that Arabidopsis hypocotyl growth tends to be more responsive to KAR2 4,39 . We examined if the preference towards a specific KAR molecule is also observed with KAI2 from other species. To this end, we used a line resulting from a cross of the kai2 mutant htl-2 with an Arabidopsis line transgenic for the cDNA of the rice D14L/KAI2 11 , and tested its response to the two KAR molecules. In contrast to LjKAI2b, OsD14L/KAI2 mediated a stronger response to KAR2 than to KAR1 (Fig. 3c). This suggests that differential responsiveness of transgenic Arabidopsis lines to different karrikin species is caused by the specific amino acid sequence of the transgenic receptor and does not results from a general incompatibility of a heterologous KAI2 protein with the Arabidopsis background. Together, these results imply that LjKAI2a and LjKAI2b differ in their affinities to KAR1 and KAR2 or their possible breakdown products 30 .
Besides karrikins also the stereoisomers of the synthetic strigolactone rac-GR24, GR24 5DS and GR24 ent-5DS trigger developmental responses via KAI2 as well as in Arabidopsis 10,40 . To confirm the divergent responses mediated by LjKAI2a and LjKAI2b with yet another ligand we complemented the Arabidopsis thaliana d14-1 kai2-2 double mutant with LjKAI2a and LjKAI2b and tested the hypocotyl response to these two GR24 stereoisomers (Fig. 3d). Lines expressing LjKAI2a responded to both with reduced hypocotyl elongation and a much stronger response to GR24 ent-5DS , whereas, interestingly, the lines expressing LjKAI2b did not significantly respond to any of the two stereoisomers. This contrasting sensitivity to GR24 stereoisomers together with the smaller differences in response to the KARs suggests that LjKAI2a and LjKAI2b differ in their binding pocket, resulting in divergent ligand binding specificity.
Three amino acid residues at the binding pocket are decisive for ligand binding specificity To examine differences in binding specificity directly, we analysed the ligand binding of LjKAI2a and LjKAI2b in vitro by differential scanning fluorimetry (DSF) using purified recombinant proteins ( Supplementary Fig. S4). This assay has been successfully used to characterize ligand binding to D14 and KAI2 proteins in vitro 21, 30 . Binding of KAR1 and KAR2 to KAI2 could not be shown with this assay, possibly because karrikins are metabolised in planta and their metabolic products, not the molecules themselves, bind to the receptor in vivo 30 . However, the GR24 stereoisomers GR24 5DS and GR24 ent-5DS are functional and used sucessfully in DSF assays, where GR24 ent-5DS triggers thermal destabilisation of KAI2 proteins from Arabidopsis, Selaginella moellendorfii and Marchantia polymorpha 30 . Here, GR24 ent-5DS induced a thermal destabilization of LjKAI2a at a concentration > 50 µM but it did not cause any significant thermal shift of LjKAI2b ( Fig. 4b).
To determine, which residues could be responsible for differential ligand binding we compared the protein sequences of KAI2a and KAI2b in legumes. This revealed conserved differences between the KAI2a and the KAI2b clade for 16 amino acids ( Supplementary Fig. S5). However, four of these (KAI2a: Y157L, I188T, M223V; and KAI2b: I119V) are not conserved in L. japonicus. We wondered whether the remaining divergent amino-acids could be responsible for the observed differential binding and hypocotyl growth responses mediated by LjKAI2a and LjKAI2b. We used Arabidopsis KAI2 and rice D14L as additional filters because their response pattern to KAR1 and KAR2 in the Arabidopsis background was similar to LjKAi2a. Thus, we focussed on the amino acids conserved within the KAI2b clade, which differed from the KAI2a clade as well as from AtKAI2 and OsD14L/KAI2, namely T103, M161, L191, A226. We modelled LjKAI2a and LjKAI2b on the KAR1-bound AtKAI2 crystal structure (4JYM) 5 , and determined that only M161 (L160 in LjKAI2a) and L191 (S190 in LjKAI2a) are at the entrance or inside the pocket, respectively (Fig. 4a). In addition, we found that inside the pocket a highly conserved phenylalanine is exchanged for tryptophan at position 158 in LjKAI2b. Although this tryptophan is not conserved among other LjKAI2b versions of the investigated legumes, we predicted that this bulky residue may have a strong impact on ligand binding.
To understand whether these three residues are involved in determining the ligand binding specificity we mutated the receptor genes to swap the divergent amino acids (Fig. 4b).
Swapping only the two amino acids that are conserved in legumes already strongly affected the thermal shift in response to GR24 ent-5DS in the DSF assay. LjKAI2a M160,L190 became much less responsive relative to LjKAI2a and displayed a slight thermal shift only with 200 µM GR24 ent-5DS , whereas LjKAI2b L161,S191 gained the ability to respond to GR24 ent-5DS at 200 µM. The changes in ligand-induced thermal shift were even more drastic when all three amino acids were swapped: LjKAI2a M160,L190,W157 did not display any thermal shift in presence of GR24 ent-5DS , whereas LjKAI2b L161,S191,F158 gained a response to GR24 ent-5DS and displayed a thermal shift with ligand concentrations as low as 25 µM. In effect, the thermal shift response of LjKAI2a to GR24 ent-5DS could be recapitulated by changing just three amino acids of LjKAI2b, and vice-versa. We conclude that residues L160/M161, S190/L191 and W158/F159 all contribute to the response of LjKAI2 proteins to GR24 ent-

5DS .
To examine whether these three amino acid residues also determine ligand discrimination in planta, we transformed Arabidopsis d14 kai2 double mutants with the mutated LjKAI2a and LjKAI2b genes driven by the Arabidopsis KAI2 promoter and performed the hypocotyl growth assay in the presence of GR24 ent-5DS . Swapping only the two amino acids conserved in legumes (M160/L161 and S190/L191) was insufficient to exchange the ability between LjKAI2a and LjKAI2b to mediate hypocotyl responses to GR24 ent-5DS .
However, swapping all three amino acids negatively affected the capacity of LjKAI2a M160,L190,W157 to mediate a hypocotyl response to GR24 ent-5DS whereas it reconstituted a response via LjKAI2b L161,S191,F158 in three independent transgenic lines ( Fig. 5a and 5b). Together these results indicate that these three residues determine the difference in ligand binding preference between the two L. japonicus karrikin receptors KAI2a and KAI2b. Although KAR1, KAR2 and GR24 ent-5DS are not the natural ligands of L. japonicus karrikin receptors, the evolution of different residues in the binding pocket of the duplicated KAI2 receptors suggest different functions for LjKAI2a and LjKAI2b.

Identification of L. japonicus karrikin and strigolactone receptor mutants
To explore the roles of LjKAI2a and LjKAI2b in L. japonicus, we searched for mutants in these genes as well as in D14 and MAX2. We identified LORE1 retrotransposon insertions in L. japonicus KAI2a, KAI2b and MAX2 (kai2a-1, kai2b-3, max2-1, max2-2, max2-3, max2-4) 35, 41 and nonsense mutations in D14 and KAI2b (d14-1, kai2b-1, kai2b-2) by TILLING 34 (Fig. 6a, Supplementary Table S1). Since some of the max2 and kai2b mutants had problems with seed germination or production (Supplementary Table S1)  The LORE1 insertion in the kai2a-1 mutant is located close (19 bp) to a splice acceptor site. Since some LjKAI2a transcript accumulated in the mutant, we sequenced this residual transcript to examine the possibility that a functional protein could still be made through loss of LORE1 by splicing. We found that indeed a transcript from ATG to stop accumulates in kai2a-1 but it suffers from mis-splicing leading to a loss of the LORE1 transposon plus 15 bp (from 369 -383), corresponding to five amino acids (YLNDV) at position 124-128 of the protein (Supplementary Fig. S7a and S7b). This amino-acid stretch reaches from a loop at the surface of the protein into the cavity of the binding pocket ( Supplementary Fig. S7c). The artificial splice variant did not rescue the Arabidopsis kai2-2 hypocotyl phenotype, confirming that it is not functional in planta and that the amino acids 124-YLNDV-128 are essential for LjKAI2a function ( Supplementary   Fig. S7d).

LjKAI2a-dependent manner
Phenotypically, d14-1 and all allelic max2 mutants of L. japonicus displayed increased shoot branching, indicating that the L. japonicus strigolactone receptor components D14 and MAX2 are involved in shoot branching inhibition ( Fig. 6b and 6c), similar as in Arabidopsis, pea and rice 4,38,42,43 . In contrast, we could not observe the canonical elongated hypocotyl phenotype, which is observed for Arabidopsis kai2 mutants (and for mesocotyl in rice d14l/kai2 mutants) in white light conditions 4,11 , neither for L. japonicus kai2a and kai2b single mutants nor for kai2a-1 kai2b-1 double mutant or max2 mutants. If anything the kai2a-1 kai2b-1 and max2 mutant hypocotyls were shorter than those of the wild type (Fig. 6d). This indicates that the requirement of KL perception for suppression of hypocotyl elongation under white light is not conserved in L. japonicus or that KL may not be produced under these conditions.
To examine whether L. japonicus hypocotyls are responsive to karrikin treatment, we measured the dose-response of hypocotyl elongation in wild-type to KAR1, KAR2 and also to rac-GR24. Hypocotyl elongation of wild type plants was progressively inhibited with increasing concentrations of all three compounds (Fig. 7a). However, it was not suppressed by KAR1 or KAR2 treatment in the kai2a-1 kai2b-1 double mutant and the max2-4 mutant (Fig. 7b, Supplementary Fig. S8). This demonstrates that similar to Arabidopsis, the hypocotyl response to karrikin of L. japonicus depends on the KAI2-MAX2 receptor complex. We also examined the KAR1 response of kai2a and kai2b single mutant hypocotyls and found that kai2a-1 did not significantly respond to KAR1 and KAR2, while the two allelic kai2b mutants showed reduced hypocotyl growth in response to both karrikins (Fig. 7b). The transcript accumulation pattern of DLK2 (Lj2g3v0765370) -which in Arabidopsis responds to karrikin in a KAI2-dependent fashion 4, 40was consistent with this observation: DLK2 was induced in hypocotyls by KAR1 and KAR2 in a manner dependent on LjKAI2a but not LjKAI2b (Fig. 7c). Furthermore, DLK2 expression was already lower in mock-treated kai2a hypocotyls than in mock-treated wild-type and kai2b- L. japonicus root system architecture is modulated by KAR1 but not by KAR2 treatment It was previously suggested that strigolactone signalling is involved in modulating root development of Arabidopsis and Medicago truncatula and that rac-GR24 treatment can trigger root system architecture changes in both species 44,45,46 . We examined whether L. japonicus root systems would respond to rac-GR24 as well as to KAR1 and KAR2 (Fig.   8a). Surprisingly, in contrast to Arabidopsis and M. truncatula, L. japonicus root systems responded neither to rac-GR24 nor to KAR2. Only KAR1 treatment lead to a dosedependent decrease in primary root length and an increase of post-embryonic root (PER) number and thus, to a higher PER density (Fig. 8a). PERs include lateral and adventitious roots that are difficult to distinguish in L. japonicus seedlings. The instability of rac-GR24 over time in the medium could potentially prevent a developmental response of the root to this compound in our experiments 47 . However, refreshing the medium with new rac-GR24 or karrikins at 5 days post-germination, did not alter the outcome: PER density remained unaffected by KAR2 and by rac-GR24 treatment ( Supplementary Fig. S9).
Consistently, we observed DLK2 induction in roots after KAR1 but not after KAR2 treatment (Fig. 8b).
Together with the L. japonicus hypocotyl responses to KAR1, KAR2 and rac-GR24 this indicates organ-specific sensitivity or responsiveness to treatment with three compounds in L. japonicus and a more stringent uptake, perception and/or response system in the root.
Surprisingly, we found that roots responded to rac-GR24 treatment with increased DLK2 transcript accumulation (Fig. 8c) although no change in root architecture was observed under this condition (Fig. 8a). To confirm the contrasting responses of L. japonicus root systems to KAR1 and rac-GR24, and to test whether they result from divergent molecular signalling outputs that are independent from DLK2 expression, we examined early transcriptional responses after one, two and six hours' treatment of L. japonicus wild-type roots with KAR1 and rac-GR24 using microarrays. Statistical analysis revealed a total number of 629 differentially expressed (DE) genes for KAR1-treated and 232 genes for rac-GR24-treated roots (Supplementary Table S2). In agreement with previous reports from Arabidopsis and tomato 39, 48, 49 the magnitude of differential expression was low.
Most of the DE genes upon KAR1 and rac-GR24 treatment responded solely after 2h ( Supplementary Fig. S10). Interestingly, only a minority of 48 genes responded in the same direction in response to both KAR1 and rac-GR24, while the majority of genes responded specifically to KAR1 (580 DEGs) or rac-GR24 (169 DEGs). In summary, the microarray experiment confirmed that L. japonicus roots respond to KAR1 and rac-GR24 in a mainly distinct manner.

Both LjKAI2a and LjKAI2b mediate root architecture-responses to KAR1
To inspect which α/β-hydrolase receptor mediates the changes in L. japonicus root system architecture in response to KAR1 treatment, we examined PER density in the karrikin receptor mutants. The Ljkai2a-1 kai2b-1 double mutant and the max2-4 mutant did not respond to KAR1 treatment with changes in root system architecture (Fig. 9a,   Supplementary Fig. S11). With 1 µM KAR1 we obtained contradictory results for the single kai2a and kai2b mutants in independent experiments (Supplementary Fig. S11a and S11c). However, kai2a and kai2b single mutants but not the kai2a kai2b double mutant responded to a slightly higher concentration of 3 M KAR1 (Fig. 9a, Supplementary Fig.   S12), indicating that LjKAI2a and LjKAI2b redundantly perceive KAR1 (or a metabolite thereof) in L. japonicus roots. This pattern was mirrored by DLK2 expression in roots: both kai2a and kai2b single mutants responded to KAR1 with increased DLK2 expression, while the kai2a-1 ka2b-1 double mutant and the max2-4 mutant did not respond (Fig. 9b). In summary, we conclude that LjKAI2a and LjKAI2b act redundantly in roots in mediating the responses to KAR1.

Discussion
We found that the karrikin receptor gene KAI2 has duplicated in legumes possibly during duplication of the whole genome that occurred in the Papilionoidaea before the diversification of legumes 59 million years ago 50 . In the model legume L. japonicus, the paralogs KAI2a and KAI2b remained functional since both mediate developmental responses to KARs and each can restore hypocotyl growth inhibition in an Arabidopsis kai2 mutant. We also found two genes encoding the F-box protein MAX2. However, one of them underwent pseudogenization, leaving a single active protein in L. japonicus to deliver its responses to KARs. Gene duplication followed by sub-or neofunctionalization is an important driver in the evolution of complex signalling networks and signalling specificities. We provide evidence that L. japonicus KAI2a and KAI2b diversified in their ligand-binding specificity as well as organ-specific function. KAR1 and KAR2 are highly similar compounds that differ only by one additional methyl group in KAR1. Nevertheless, LjKAI2a and LjKAI2b differ in their sensitivity to these compounds, since in the Arabidopsis hypocotyl assay, LjKAI2a mediates an equal response to KAR1 and KAR2, while LjKAI2b confers a stronger response to KAR1 than to KAR2 (Fig. 10b). Thereby, LjKAI2b changes the response preference of Arabidopsis, which usually responds more strongly to KAR2 2, 39, this work . GR24 ent-5DS , an enantiomer of the synthetic strigolactone analogue rac-GR24 has been shown genetically to act via Arabidopsis KAI2 and to bind to KAI2 in vitro 30,40 . LjKAI2a also mediates strong Arabidopsis hypocotyl growth responses to GR24 ent-5DS but this is not the case for LjKAI2b.
Furthermore, LjKAI2b may be less sensitive to the endogenous ligand of Arabidopsis KAI2, since its ability to restore hypocotyl growth inhibition in untreated Arabidopsis is slightly decreased as compared to LjKAI2a. Together, these results demonstrate that the α/β-fold hydrolase receptor is sufficient to explain ligand sensitivity in the Arabidopsis hypocotyl assay. The differential sensitivity of LjKAI2a and LjKAI2b to GR24 ent-5DS was confirmed in vitro by DSF assay: GR24 ent-5DS induced a thermal shift of LjKAI2a but did not induce thermal destabilization of LjKAI2b.
Identifying the determinants of ligand-binding specificity of D14 and different KAI2 proteins is an area of active research. Although some factors such as geometry and rigidity of the binding pocket have been proposed to determine specificity of KAI2-like proteins for strigolactones vs. karrikins in Physcomitrella patens and in parasitic weeds 31, 32 , it is unclear how differential binding preference for very similar molecules is achieved. We identified three amino acids at the ligand-binding pocket that differ between LjKAI2a and LjKAI2b and explain ligand response to GR24 ent-5DS .Two of these amino acids are conserved across the legume KAI2a and KAI2b clades, namely L160 and S190 in KAI2a and M161 and L191 in LjKAI2b. An exchange of these two amino acids was sufficient to strongly reduce sensitivity of LjKAI2a to GR24 ent-5DS in the DSF assay and to gain a thermal shift of LjKAI2b. Neither amino acid change is predicted to substantially impact the pocket volume or geometry but the amino acids of LjKAI2b are more hydrophobic, which may explain the repulsion of the more hydrophilic GR24 ent-5DS and also the preference for the more hydrophobic KAR1 over KAR2. A similar phenomenon was observed in Brassica tournefortii, a fire-following weed that has three KAI2 genes, of which KAI2c does not seem to be functional 51 . Similar to the situation in L. japonicus, BtKAI2b mediated a greater sensitivity to KAR1 over KAR2 in the Arabidopsis background, while it was the reverse for BtKAI2a. Again, this was explained by two amino acid changes in the binding pocket between LjKAI2a and LjKAI2b towards more hydrophobic amino acids (V98L, V191L). Notably, one of these residues in B. tournefortii (V98L) is in a different position than the specificity-determining residue 160/161 in L. japonicus KAI2a/KAI2b. This suggests that these receptors are highly plastic and that similar binding-specificities may be achieved by changing hydrophobicity in different positions of the pocket.
Furthermore, the position of the change towards stronger hydrophobicity may be involved in conferring sensitivity to the ligand, as B. tournefortii KAI2 proteins respond to lower ligand concentrations than L. japonicus KAI2 proteins in the DSF assay 51, this work .
Exchanging L160/M161 and S190/L191 between L. japonicus KAI2a and KAI2b was sufficient to change their sensitivity to GR24 ent-5DS in the DSF in vitro assay. However, the developmental response of Arabidopsis hypocotyls was hardly changed, possibly because in vivo, suboptimal ligand binding to the receptor can be stabilized by interacting proteins. A third amino acid difference (F157/W158) between the two KAI2 proteins occurs in L. japonicus. This residue strongly determines sensitivity to GR24 ent-5DS likely because in addition to increased hydrophobicity of tryptophan vs. phenylalanine the bulky tryptophan in the pocket of KAI2b may sterically hinder GR24 ent-5DS binding. However, it still allows sensitivity to KAR1, which is larger than KAR2. When all three amino acids are exchanged, LjKAI2a completely loses GR24 ent-5DS -responsiveness in vitro as well as in the Arabidopsis hypocotyl assay whereas LjKAI2b gains full responsiveness.
B. tournefortii is a fire-following plant, whose seeds respond to karrikins by breaking dormancy and germinating 51 . Therefore, it makes adaptive sense for B. tournefortii to maintain two copies of KAI2, one of which is specialized for KAR1, the most abundant KAR in smoke, and the other of which may be specialized for the endogenously produced ligand. However, for L. japonicus, which is not a fire-follower, KAR1, KAR2 and GR24 ent-5DS are likely not natural KAI2 ligands. Nevertheless, the maintenance of two KAI2 genes in the legumes, each with amino acid polymorphisms confering differences in binding preferences to artificial ligands, requires an adaptive basis. One possibility is that L.
japonicus KAI2a and KAI2b have specialized to bind different ligands in planta and that legumes may produce at least two different versions of the as-yet-unknown KL compound.
The distinct expression patterns and developmental roles of LjKAI2a and LjKAI2b might also be consistent with tissue-specific ligands, or even an endogenous ligand versus an exogenous ligand derived from the rhizosphere. From our assays with artificial ligands we extrapolate that KAI2b is likely more stringent regarding the ligand's chemical properties.
The additional amino acid change that has occurred in L. japonicus but not in the other examined legumes may indicate that the KL bouquet of L. japonicus has further diversified. Once the identity of KL and its putative different versions have been identified it will be interesting to investigate the biological significance of this receptor subfunctionalization and the putative diversity of their ligands.
Using kai2a and kai2b mutants of L. japonicus, we determined that LjKAI2a alone mediates developmental and transcriptional responses to exogenously-applied karrikins in hypocotyls, whereas both LjKAI2a and LjKAI2b act in roots (Fig. 10a). In addition, we found that L. japonicus root systems respond to karrikin treatment with slightly increased postembryonic root (PER) density (higher PER number and shorter primary roots).
Surprisingly, this occurred exclusively in response to KAR1, while in contrast, hypocotyl growth inhibition was achieved with KAR1, KAR2 and rac-GR24 (Fig. 10a). To our knowledge such an organ-specific discrimination of different but very similar KAR molecules has not previously been so clearly observed. However, a similar scenario could be at play in rice, in which a transcriptome analysis of KAR2-treated rice roots found no differentially expressed genes, whereas rice mesocotyls responded with growth inhibition to the same treatment 11 . Previous work suggested that KARs are not directly bound by KAI2, but they may be metabolized first to yield the correct KAI2-ligand 30 . It is possible that the enzymes involved in KAR metabolism in hypocotyls and roots differ in their substrate specificities. This would imply that the single methyl group, which distinguishes KAR1 from KAR2, is sufficient to impede or otherwise impact upon specialized metabolism of karrikins. Alternatively, the transport of the KAR2-derived metabolic product could be limited in the root system. Finally, KAR2-derivatives may be specifically catabolised in roots thus limiting the response. Although KAR2 fails to induce increased PER density and DLK2 expression in L. japonicus roots, rac-GR24 is still able to trigger KAI2-dependent DLK2 transcript accumulation albeit being unable to increase PER density. It is possible that DLK2 activation is mediated via D14, which may not be involved in regulating root architecture in L. japonicus. Alternatively, downstream events triggered by KAI2 might differ depending on the specific ligand.
In Arabidopsis, kai2 and max2 mutants display an increased lateral root density 10 . This is somewhat different to our observation that KAR1 treatment triggers increased PER density in L. japonicus. The discrepancy may result from different physiological optima between the two species or from nutrient conditions in the two experimental systems. We observed the KAR1 response of L. japonicus root systems in half-Hoagland solution with low phosphate levels (2.5μM PO4 3-) and without sucrose, whereas the root assay in Arabidopsis was conducted in ATS medium (Arabidopsis thaliana salts) with 1% sucrose.
Phosphate and sucrose levels have previously been described to influence the effect of strigolactone and rac-GR24 on Arabidopsis root architecture 44,52,53 .
In Arabidopsis and rice, KAI2/D14L is required to inhibit hypocotyl and mesocotyl elongation, respectively 3, 4, 11 . Since these two species are evolutionary distant from each other, but have both retained a function of KL signalling in inhibiting the growth of similar organs, it seemed likely that this function would be conserved among a large number of plant species. Surprisingly, in L. japonicus, we observed no elongated hypocotyl phenotype for the kai2a-1 kai2b-1 double and two allelic max2 mutants (Fig. 5). However, we could trigger a reduction of hypocotyl elongation by treatment with KAR1, KAR2 and rac-GR24 in the wild type and in a LjKAI2a and LjMAX2-dependent manner. Thus, it is possible that endogenous KL levels in L. japonicus hypocotyls are insufficient to cause inhibition of hypocotyl elongation, at least under our growth conditions.
In this work, we have demonstrated sub-functionalization of two KAI2 copies in L.
japonicus with regard to their ligand-binding specificity -mediated by three amino acids in the binding-pocket -and organ-specific relevance. Furthermore, we find organ-specific responsiveness of L. japonicus to different artificial KAI2 ligands. Our present work suggests multiple endogenous ligands that can be discriminated by LjKAI2a and LjKAI2b.
It opens novel research avenues towards understanding the diversity in KL ligandreceptor relationships and in developmental responses to both, as yet, unknown and synthetic butenolides that influence diverse aspects of plant development.

Plant material and seed germination
The A. thaliana kai2-2 (Ler background) and d14-1 (Col-0 background) mutants are from 4 , the d14-1 kai2-2 double mutant from 40 , the htl-2 mutant was provided by Min Ni 54 and the cross with K02821 is from 11 . Seeds were surface sterilized with 70% EtOH. For synchronizing the germination, seeds were placed on ½ MS 1% agar medium and maintained at 4°C in the dark for 72 hours.

Protein sequence alignment, phylogenetic tree and synteny
Protein sequences were retrieved using tBLASTn with AtKAI2, AtDLK2 and AtMAX2, against the NCBI database, the plantGDB database and the L. japonicus genome V2.5 (http://www.kazusa.or.jp/lotus). The presence of MAX2-like was identified by tBLASTn in an in-house genome generated by next generation sequencing using CLC Main Workbench 56 . Pea sequences were found by BLASTn on "pisum sativum v2" database with AtKAI2 as query (https://www.coolseasonfoodlegume.org). The MAFFT alignment (https://mafft.cbrc.jp) of the protein sequences was used to generate Maximum-likelihood tree with 1000 bootstrap replicates in MEGA7 57 . For the synteny analysis of MAX2 and MAX2-like, flanking sequences were retrieved from the same in-house genome 56 .

Bacterial protein expression and purification
Full-length L. japonicus coding sequences were cloned into pE-SUMO Amp. Clones were sequence-verified and transformed into Rosetta DE3 pLysS cells (Novagen). Subsequent protein expression and purification were performed as described previously 30 , with the following modifications: the lysis and column wash buffers contained 10 mM imidazole, and a cobalt-charged affinity resin was used (TALON, Takara Bio).

Differential scanning fluorimetry
DSF assays were performed as described previously 30

Plasmid generation
Genes and promoter regions were amplified using Phusion PCR according to standard protocols and using primers indicated in Supplementary Table S3. Plasmids were constructed by Golden Gate cloning 58 as indicated in Supplementary Table S4.

Plant transformation
kai2-2 and d14-1 mutants were transformed by floral dip in Agrobacterium tumefaciens AGL1 suspension. Transgenic seedlings were selected by mCherry fluorescence and resistance to 20 μg/mL hygromycin B in growth medium. Experiments were performed using T2 or T3 generations, with transformed plants validated by mCherry fluorescence.

Shoot branching assay
A. thaliana and L. japonicus were grown for 4 and 7 weeks, respectively in soil in the greenhouse at 16h/8h light/dark cycles. Branches with length superior to 1cm were counted, and the height of each plant was measured.

Treatment for analysis of transcript accumulation
Seedling roots were placed in 1/2 Hoagland solution with 2.5μM PO4 3containing 1 or 3 μM Karrikin1 (www.olchemim.cz for qPCR analysis, synthesized according to 59  Control and rhizobial probesets were removed before statistical analysis. Differential gene expression was analyzed with the Bioconductor package "Linear Models for Microarray Data" (LIMMA version 3.26.8) 63 . The package uses linear models for parameter estimation and an empirical Bayes method for differential gene expression assessment 64 . P-values were adjusted due to multiple comparisons with the Benjamini-Hochberg correction (implemented in the LIMMA package). Probesets were termed as significantly differentially expressed, if their adjusted p-value was smaller than or equal to 0.01 and the fold change for at least one contrast showed a difference of at least 50%. To identify the corresponding gene models, the probeset sequences were used in a BLAST search against L. japonicus version 2.5 CDS and version 3.0 cDNA sequences (http://www.kazusa.or.jp/lotus/). If, based on the bitscore, multiple identical hits were found, we took the top hit in version 2.5 CDS as gene corresponding to the probe. For version 3.0 cDNA search we used the best hit, that was not located on chromosome 0, if possible. For probesets known to target chloroplast genes (probeset ID starting with Lj_), we preferred the best hit located on the chloroplast chromosome, if possible. Probeset descriptions are based on the info file of the L. japonicus Microarray chip provided by the manufacturer (Affymetrix).

qPCR analysis
Tissue harvest, RNA extraction, cDNA synthesis and qPCR were performed as described previously 56 . qPCR reactions were run on an iCycler (Biorad, www.bio-rad.com) or on QuantStudio5 (applied biosystem, www.thermofisher.com). Expression values were calculated according to the ΔΔCt method 65 . Expression values were normalized to the expression level of the housekeeping gene Ubiquitin. For each condition three to four biological replicates were performed. Primers are indicated in Supplementary Table S3.

Statistics
Statistical analyses were performed using Rstudio (www.rstudio.com) after log transformation for qPCR analysis. F-and p-values for all figures are provided in Supplementary Table S5. at LotusBase for the LORE1 insertion lines; and to David Nelson (University of California