The root nodule symbiosis regulator NIN exhibits broad DNA binding specificity conferred by an NLP-inherited motif

Abstract

Nitrogen-fixing root nodule symbiosis (RNS) occurs in some eudicots, including legumes, and is regulated by the transcription factor NODULE INCEPTION (NIN), derived from the NIN-LIKE PROTEIN (NLP) family. However, how the NIN protein acquired RNS-specific functions remains unclear. We identify a previously undescribed motif in Lotus japonicus NIN, located downstream of the RWP-RK domain, which we term the FR. This motif broadens NIN’s DNA binding specificity by stabilizing the RWP-RK dimer interface. nin mutants lacking the FR motif show defective nodulation and impaired nitrogen fixation. Arabidopsis NLP2 carries a NIN-type FR and shares key features with NIN. Furthermore, the NIN-type FR had already emerged before the divergence of gymnosperm and angiosperm lineages, suggesting that a specific molecular feature of NIN involved in RNS regulation was inherited from ancestral NLPs prior to the emergence of RNS.

Acquisition of a motif was a prerequisite for NODULE INCEPTION to function in transcriptional control of symbiotic genes.

INTRODUCTION

Root nodule symbiosis (RNS) is a symbiotic relationship between nitrogen-fixing clade plants, including legumes, and nitrogen-fixing bacteria such as rhizobia or Frankia (1, 2). NODULE INCEPTION (NIN) transcription factor (TF) is a master regulator of RNS, controlling rhizobial infection, root nodule development, and symbiotic nitrogen fixation by regulating key RNS-related genes, including NUCLEAR FACTOR-YA (NF-YA), NF-YB, EXOPOLYSACCHARIDE RECEPTOR 3 (EPR3), and RHIZOBIUM-DIRECTED POLAR GROWTH (RPG) (3–6). Previous phylogenomic studies have suggested that the presence or absence of NIN in plants is closely linked to the acquisition and loss of RNS ability (7, 8). In Medicago truncatula, NIN undergoes protein processing during nodule development, with the C-terminal region playing a role in regulating further RNS processes (9). While our understanding of NIN’s function is growing, the fundamental question of which molecular features of NIN enable its regulation of RNS remains unresolved.

NIN is derived from the NIN-LIKE PROTEIN (NLP) TF family, which includes an RWP-RK DNA binding domain and a PB1 domain involved in multimerization (10, 11). A GAF-like domain in the N-terminal region, responsible for nitrate sensing, is conserved among NLPs but absent in NIN (12, 13), which may account for NIN’s loss of nitrate responsiveness. In Lotus japonicus, LjNIN and LjNLP4 form homodimers and bind to similar cis-elements containing two palindromic core sites on the promoters of RNS and/or nitrate-related genes (Fig. 1, A and B) (14). LjNIN can bind to the LjNLP4-binding sites, but LjNLP4 does not necessarily bind to LjNIN-specific sites. The cis-element on ProLjCLE-RS2 is highly palindromic and serves as a common binding site for both LjNIN and LjNLP4. In contrast, some LjNIN-specific binding sites (e.g., ProLjNF-YA and ProLjNF-YB) are less palindromic compared to LjNLP4-binding sites, suggesting that LjNIN has a broader DNA binding specificity than LjNLP4 (14). However, the underlying mechanisms remain unclear.

Fig. 1. An amino acid motif determining the NIN’s broad DNA binding selectivity.

(A) The broad DNA binding selectivity of LjNIN distinct from LjNLP4. Both bind to perfect preferred cis-elements comprising two palindromic core sites (black boxes), whereas only LjNIN can bind partially unpreferred cis-elements containing one unpreferred core site (gray box), enabling the NIN-specific gene regulation. (B) Sequences of the DNA probes for EMSA designed based on the LjNIN/LjNLP4 binding motifs derived from the previously reported genome-wide analyses (14, 65). (C) Domain structures of LjNIN and LjNLP4 with protein sequence alignments of the following RWP-RK (FR) of NINs (FR^NIN) from L. japonicus (Lj), M. truncatula (Mt), Glycine max (Gm), Phaseolus vulgaris (Pv), Parasponia andersonii (Pa), and Casuarina glauca (Cg), and the RWP-RK domains with the FRs from LjNIN/NLPs. White boldfaces and black boldfaces show the completely and partially conserved amino acid residues, respectively. The amino acid sequence identity within the RWP-RK domain among LjNIN and LjNLPs is 52%. Gray circles on alignments indicate the amino acid residues forming a dimer interface via hydrophobic and/or van der Waals interactions in the AlphaFold2-predicted structure of the LjNIN dimer. (D to G) EMSA results of LjNIN (WT) and LjNLP4 (WT) (D) and the LjNIN/LjNLP4 chimeras with the RWP-RK regions swapped [LjNIN (RWP^LjNLP4)/LjNLP4 (RWP^LjNIN)] (E), the FR swapped [LjNIN (FR^LjNLP4)/LjNLP4 (FR^LjNIN)] (F), and both the RWP-RK and the FR swapped [LjNIN (RWP^LjNLP4 FR^LjNLP4)/LjNLP4 (RWP^LjNIN FR^LjNIN)] (G). The constructs of LjNIN/LjNLP proteins are shown at the top panels. Each recombinant protein fused to the maltose-binding protein (MBP) at the N terminus was reacted at 2 μM final concentration with 0.25 μM DNA probe (i to v). The electrophoretic patterns are shown at the middle panels with asterisks and arrowheads indicating the positions of free DNA and the protein-DNA complexes, respectively. Bar graphs at the bottom panels show the fluorescence densitometric profile (means ± SEM; n = 3).

RESULTS

Identification of the motif determining NIN’s broad DNA binding specificity

To identify the key amino acid residues determining the NIN-specific broad DNA binding specificity, we examined amino acid conservation in NINs/NLPs across various plant species. Phylogenetic analysis revealed that the RWP-RK domains derived from NINs showed a pattern similar to that of full-length NINs (fig. S1, A and B). Amino acid sequence alignments further showed that RWP-RK–comprising residues are more conserved in NINs compared to NLPs (Fig. 1C and fig. S1C). To test whether the RWP-RK domain of NIN is crucial for its broad DNA binding specificity, we conducted an electrophoretic mobility shift assay (EMSA) using chimeric proteins (LjNIN/LjNLP4) with swapped RWP-RK domains [LjNIN (RWP^LjNLP4)/LjNLP4 (RWP^LjNIN)], along with wild-type proteins [LjNIN (WT)/LjNLP4 (WT)] (Fig. 1, D and E, and fig. S2). LjNIN (WT) bound strongly to the partially unpreferred cis-elements ProLjNF-YA(iii)/ProLjNF-YB(iv) and the perfectly preferred cis-elements ProLjCLE-RS2(i)/ProLjNIR1(ii), but not to the mProLjNF-YB(v) with only one core site (Fig. 1D). In contrast, LjNLP4 (WT) specifically bound to ProLjCLE-RS2(i)/ProLjNIR1(ii) (Fig. 1D). Under the same conditions, LjNIN (RWP^LjNLP4) still interacted with ProLjNF-YA(iii)/ProLjNF-YB(iv), albeit with reduced binding compared to LjNIN (WT). LjNLP4 (RWP^LjNIN), unlike LjNLP4 (WT), could bind to these two cis-elements, but less effectively than LjNIN (RWP^LjNLP4) (Fig. 1E). These results suggest that the RWP-RK domains contribute partially, but not critically, to the distinctive DNA binding specificities of LjNIN and LjNLP4.

We then focused on a 15–amino acid motif located just following the RWP-RK domains, which we hereby named the following RWP-RK (FR). Notably, this FR motif in NINs (FR^NIN) is evolutionary conserved among nodulating plants of nitrogen-fixing clade, including legumes, Parasponia, and an actinorhizal plant (Fig. 1C) (15, 16). EMSA using chimeric proteins with the FR swapped [LjNIN (FR^LjNLP4)/LjNLP4 (FR^LjNIN)] (Fig. 1F and fig. S2, C and D) showed that LjNIN (FR^LjNLP4) had substantially reduced binding to ProLjNF-YA(iii)/ProLjNF-YB(iv) compared to LjNIN (WT) and LjNIN (RWP^LjNLP4), although it retained substantial interaction with ProCLE-RS2(i)/ProLjNIR1(ii) (Fig. 1F). Conversely, LjNLP4 (FR^LjNIN) bound strongly to ProLjNF-YA(iii)/ProLjNF-YB(iv) (Fig. 1F). A similar binding pattern of the chimeric proteins to cis-elements was also observed in other LjNIN target genes (fig. S3) (5, 14, 17). Chimeric proteins with both the RWP-RK and FR swapped [LjNIN (RWP^LjNLP4 FR^LjNLP4)/LjNLP4 (RWP^LjNIN FR^LjNIN)] exhibited completely reversed DNA binding specificities (Fig. 1G and fig. S2, C and D). Overall, these results indicate that the identified 15-residue FR^NIN is critical for the broad DNA binding specificity of NIN.

We next used LjNIN/LjNLP4 deletion constructs lacking the C-terminal PB1 domain to assess the biochemical role of FR^NIN within a minimal DNA binding module comprising only the RWP-RK and FR. EMSA showed that these modules, including FR-swapped chimeras, retained DNA binding specificity nearly identical to the PB1-containing proteins, each producing a single shifted band regardless of protein concentration (Fig. 2, A to D, and fig. S4). These findings indicate that the DNA binding modules are sufficient to recapitulate the DNA binding properties of LjNIN and LjNLP4, demonstrating the broad DNA binding specificity conferred by FR^NIN is functionally active even in this minimal context.

Fig. 2. Molecular properties of the DNA binding modules, including the RWP-RK and FR.

(A to D) EMSA results of the LjNIN and LjNLP4 proteins with the C terminus region deleted [LjNIN (WT) ΔC (A); LjNLP4 (WT) ΔC (B)] and their FR-swapped chimeras [LjNIN (FR^LjNLP4) ΔC (C); LjNLP4 (FR^LjNIN) ΔC (D)]. The ΔC proteins fused to the MBP at the N terminus were reacted at 0.25, 0.5, 1, and 2 μM final concentration with 0.25 μM DNA probes. The electrophoretic patterns are shown with asterisks indicating the positions of free DNA. These experiments were repeated independently with similar results at least three times. (E) SEC analyses on the MBP-fused LjNIN (WT) ΔC and LjNLP4 (WT) ΔC. Chromatograms show monomeric and dimeric forms indicated by single and double arrowheads, respectively. Asterisk represents protein aggregation. mAU stands for milli-absorbance units as UV absorbance signal. (F) Chemical cross-link analyses of MBP-fused LjNIN (WT) ΔC and LjNLP4 (WT) ΔC using the amine-specific cross-linker DSS. SDS-PAGE results are shown for these proteins under DSS-treated and DSS-free conditions, as well as under DSS-treated conditions in the presence of target DNA including ProLjNIR1 and ProLjNF-YB. The molecular masses of MBP-fused LjNIN (WT) ΔC and LjNLP4 (WT) ΔC are ~56 and 57 kDa, respectively.

The DNA binding modules have an ability to form dimers

Given that the target DNA probes contained two core sites but yielded only a single shifted band in EMSA (Fig. 2, A to D, and fig. S4), we hypothesized that the DNA binding modules, composed of RWP-RK and FR, have an intrinsic potential to form dimers. To assess their oligomerization state, we performed size exclusion chromatography (SEC) and blue native polyacrylamide gel electrophoresis (PAGE), which revealed that the modules of LjNIN and LjNLP4 and their chimeras exist mainly as monomers, but partially dimerize even without DNA (Fig. 2E and fig. S5). Cross-linking experiments using amine-specific cross-linker disuccinimidyl suberate (DSS) indicated that this dimerization was enhanced by their target DNA (Fig. 2F and fig. S6). Collectively, these findings suggest that the DNA binding modules intrinsically dimerize, and that this dimerization is stabilized upon binding to cis-elements with two core sites, potentially providing a basis for the different DNA binding specificities.

FR^NIN reinforces the RWP-RK dimerization for broad DNA binding specificity

Since biochemical approaches left the mechanism by which FR^NIN broadens DNA binding specificity elusive, we next aimed to explore this question from a structural biology perspective. To gain structural insights, we predicted the dimeric structures of the DNA binding modules of LjNIN and LjNLP4 using AlphaFold2. From a total of 100 structures generated by two advanced modes, we selected the one with the highest confidence value (fig. S7, A and B). Both top-ranked predicted structure of LjNIN and LjNLP4 exhibited dimer formation primarily facilitated by the RWP-RK (Fig. 3, A to C). In these models, both RWP-RK domains dimerized primarily through a pair of leucine zipper-like helices, stabilized by hydrophobic and van der Waals interactions via highly conserved residues in the NLP/NIN family (Figs. 1C and 3, A to C, and figs. S1C and S7C). Meanwhile, the fully conserved hydrophobic residues within FR^LjNIN interacted with the dimerization helices of RWP-RK, forming a short helix that further supported the dimer interface (Fig. 3B). However, the FR^LjNLP4 was not predicted to contribute to RWP-RK dimerization (Fig. 3C). This expanded dimer interface was observed in additional predicted structures of the LjNLP4-chimera with FR^LjNIN, but not in those of the LjNIN-chimera with FR^LjNLP4 (fig. S7, D and E).

Fig. 3. FR^NIN-mediated mechanism for NIN’s broad DNA binding specificity.

(A) Top-ranked AlphaFold2-predicted LjNIN dimer structures containing the RWP-RK and FR. Two chains are colored differently. Top: Residues predicted to be involved in dimer formation are shown as stick/sphere models. Bottom: Basic residues concentrated in the putative DNA binding surface are shown as stick/sphere models. Conserved “RWP-RK” residues are in green, and FR^LjNIN regions are in yellow. An ideal B-form DNA structure and LjNIN structure are drawn to scale, with the two core sites of the NIN/NLP-binding motif highlighted in green. (B) Close-up views of the LjNIN dimer structure. The residues forming hydrophobic and/or van der Waals interactions on FR^LjNIN are shown as stick and sphere models. (C) Top-ranked AlphaFold2-predicted LjNLP4 dimer structures containing the RWP-RK and FR, shown in the same manner as in (A). (D and E) The top-ranked AlphaFold3-predicted LjNIN (D) and LjNLP4 (E) dimer structures in complex with DNA (ProLjCLE-RS2). (F) EMSA of LjNIN and LjNLP4 DNA binding modules (ΔC), tested as WT and RWP-RK quadruple point mutants (mutRWP). MBP-fused proteins (2 μM) were reacted with 0.25 μM DNA probes. Schematic models of the LjNIN/LjNLP4 DNA binding mode are described beside the related electrophoretic bands of DNA-protein complexes. (G and H) Results of 100-ns MD simulations based on the AlphaFold2-predicted structures of LjNIN (G) and LjNLP4 (H) in the DNA-free state. (I and J) Proposed model underlying the DNA binding specificity differences between LjNIN (I) and LjNLP4 (J). While both RWP-RK domains can weakly dimerize, LjNLP4 tends to form suboptimal dimers, leading to unstable or inefficient DNA binding (J). In contrast, FR^LjNIN stabilizes an optimal RWP-RK dimer, enhancing binding efficiency and retention even to unpreferred nucleobases (I), thereby conferring broader DNA binding specificity.

In addition, basic residues were concentrated on the same side of the predicted RWP-RK dimer structures (Fig. 3, A and C), a feature typical of DNA binding proteins. Of the five completely conserved residues from which the term “RWP-RK” is derived, three basic residues clustered on the potential DNA binding surface, while the remaining tryptophan and proline residues maintained characteristic structural features (Fig. 3, A and C). The arrangement of two “RWP-RK” residues aligns well with recognition of two core sites ~10 base pairs (bp) apart in the target DNA (Fig. 1B and Fig. 3, A and C). Therefore, the predicted dimerization mode may also reflect the DNA-bound state. The FR regions were positioned opposite the DNA binding surface (Fig. 3, A and C), consistent with our previous results that the FRs do not alter nucleobase-recognition preferences (Fig. 1B) (14). Furthermore, these findings were corroborated by AlphaFold3 predictions of LjNIN/LjNLP4 dimers complexed with DNA, which were structurally reliable with well-supported geometry at both the dimer and protein-DNA interfaces, with the slight distortion observed in the bound DNA (Fig. 3, D and E, and fig. S8). Collectively, these results suggest that the FR^LjNIN contributes an additional dimer interface between the RWP-RKs, stabilizing a dimeric state that is optimal for recognizing the cis-elements harboring two core sites.

To experimentally validate these structural predictions, point mutations were introduced into residues potentially involved in dimer formation within the RWP-RK domains of the LjNIN and LjNLP4 DNA binding modules [Val⁶²¹→Ala (V621A), L625A, L628A, and I632A in LjNIN; V636A, L640A, L643A, and I647A in LjNLP4] (Fig. 3, A to C, and fig. S7C). These quadruple mutants exhibited a reduction in the original DNA-protein complex on both the preferred and unpreferred cis-elements in EMSA (Fig. 3F and fig S9, A to C), indicating that the substituted hydrophobic residues contribute to efficient DNA binding via dimer formation, including the broad DNA binding specificity of LjNIN. Notably, the LjNLP4 quadruple mutant produced a smaller complex, which likely represents a monomeric protein bound to DNA, whereas this type of complex was not detectable in the corresponding LjNIN mutant. These results suggest that LjNIN forms a more stable dimer-DNA complex than LjNLP4, possibly through additional interactions mediated by other amino acid residues, including those at the FR^LjNIN interface. The FR^LjNIN mutant, carrying six substitutions at potential dimer-forming residues in the FR^LjNIN (I642A, I644A, F647A, Y648A, F651A, and L654A; Fig. 3B), lost the broader DNA binding specificity of LjNIN (fig. S9, A and D), indicating that hydrophobic and van der Waals interactions at the FR^LjNIN interface are crucial for its function. Collectively, these mutational results support the structural models predicted by AlphaFold.

To further investigate the DNA-free state, we conducted molecular dynamics (MD) simulations. LjNIN maintained a stable dimeric structure suitable for DNA binding (Fig. 3G and fig. S10A), whereas LjNLP4 lost this conformation despite remaining dimeric (Fig. 3H and fig. S10B). These findings suggest that FR^NIN reinforces the RWP-RK interface both before and after DNA engagement. Consistently, DSS cross-linking of DNA-bound dimeric forms of LjNLP4 partially enhanced binding to unpreferred cis-elements, possibly by mimicking the FR^NIN-stabilized state (fig. S11).

Together, we propose a model explaining the specificity differences between LjNIN and LjNLP4: Although both RWP-RK domains can weakly dimerize (Fig. 3, I and J), LjNLP4 tends to form nonoptimal dimers, resulting in unstable or inefficient binding to the unpreferred cis-elements (Fig. 3J). In contrast, FR^LjNIN stabilizes an optimal dimer interface of RWP-RK, enhancing the DNA binding efficiency and retention, even at unpreferred cis-elements (Fig. 3I). This structural reinforcement underlies the increased versatility and robustness of LjNIN-mediated DNA recognition.

FR^NIN is necessary for NIN to regulate RNS

To test the significance of FR^LjNIN in plants, we overexpressed the full-length chimeric proteins in L. japonicus hairy roots (fig. S12A). In line with the EMSA results, swapping FR^LjNIN and FR^LjNLP4 resulted in a functional conversion of each protein in inducing the expression of LjNF-YA and LjNF-YB (fig. S12, B to E). Moreover, domain swapping including RWP-RK region exerted an even more pronounced effect. In addition, FR-deleted LjNIN [LjNIN (∆FR)] largely lost the ability to bind to ProLjNF-YA(iii)/ProLjNF-YB(iv) and other LjNIN-specific cis-elements, while still interacting with ProCLE-RS2(i)/ProLjNIR1(ii), regardless of the presence or absence of the C-terminal PB1 domain (figs. S3 and S13, A to C). Furthermore, structural prediction combined with MD simulation showed that DNA-free LjNIN (ΔFR) transitioned into a conformation nonoptimal for DNA binding (fig. S13D), like the case of LjNLP4 (Fig. 3H). Overexpression of full-length LjNIN (∆FR) in L. japonicus did not induce the expression of LjNF-YA and LjNF-YB (fig. S14, A to C).

Next, we created an L. japonicus mutant (∆FR nin) using the CRISPR-Cas9 system, which resulted in a 60-bp nucleotide deletion specifically removing the FR containing region, without introducing a frameshift mutation (Fig. 4, A and B). Incidentally, this mutation also caused a single, unintended amino acid substitution (C663R). EMSA revealed that the mutated NIN protein in ∆FR nin [LjNIN (∆FR^CR)] could not bind to NIN target sequences, similar to LjNIN (∆FR), indicating that the accidental C663R substitution had no additional effect on DNA binding activity (fig. S13B). Although most known nin mutants, such as nin-9, completely lack nodulation (18), the ∆FR nin mutants retained nodule formation but developed nonfunctional white nodules with no nitrogen-fixing activity (Fig. 4, C to F, and fig. S15A). Nodule sections from ∆FR nin showed that rhizobia were localized in specific areas rather than uniformly distributed as in WT nodules (Fig. 4, G and H). Transmission electron microscopy revealed that ∆FR nin nodule cells were deformed and had fewer rhizobia colonized compared to WT; some rhizobia were possibly present in intercellular spaces (Fig. 4, I to N). In addition, epidermal infection threads (ITs) were rarely observed in ∆FR nin roots (Fig. 4, O to Q). Cortical IT formation was also severely impaired in ∆FR nin, but rhizobia were observed on the surface of nodule primordia (fig. S15, B and C). Reverse transcription quantitative polymerase chain reaction (RT-qPCR) analysis showed that, unlike in nin-9 mutants, target genes of NIN, such as LjNF-YA, LjNF-YB, LjEPR3, and LjRPG, were induced in ∆FR nin, albeit with delayed and/or reduced expression compared to WT (Fig. 4R and fig. S15D). This delayed gene expression likely contributes to the delayed nodule development observed in ∆FR nin. In M. truncatula, NIN directly regulates Leghemoglobin (Lb) gene expression (19). Consistent with the formation of nonfunctional nodules, LjLb2 was not expressed at any tested stage in ∆FR nin (fig. S15D). We performed transcriptome analysis using roots of WT, the ∆FR nin, and the nin-9 at 0, 7, and 21 days after inoculation (dai). Among the genes up-regulated by rhizobia in WT, 1165 and 1343 genes were down-regulated in the ∆FR nin and the nin-9, respectively, at 7 dai (fig. S16A and data S1 to S3). Among these, 1050 genes were commonly down-regulated in both mutants, suggesting that most function of NIN was attenuated in ∆FR nin. At 21 dai, 1979 genes in the ∆FR nin and 2414 genes in the nin-9 were down-regulated (fig. S16B and data S4 to S6). Expression patterns of NIN target genes were consistent with those detected in RT-qPCR analysis (fig. S16C). In M. truncatula, the nin-16 mutant with nonfunctional nodule formation typically shows up-regulation of senescence- and defense-related genes (20), but such activation was not observed in the ∆FR nin (fig. S16D).

Fig. 4. Phenotypes of ∆FR nin mutants.

(A) Gene model of LjNIN. The black boxes, lines, and arrowhead indicate exons, introns, and the deleted region in ∆FR nin. (B) Protein sequences of LjNIN around FR. (C to E) Nodulation of ∆FR nin at 14 and 28 dai. (C) Nodules. (D) The largest nodule diameter in each plant. (E) Number of total nodules. (F) Acetylene reduction activity (ARA) of WT and ∆FR nin plants at 21 dai. WT plants inoculated with ∆NifH rhizobia were controls for the absence of nitrogen fixation. (G and H) Nodule sections at 28 dai stained with toluidine blue. The white square in (H) is enlarged to the left. (I to N) Transmission electron microscopy images of nodule sections at 28 dai. The white squares in (I), (J), (K), and (L) are enlarged by (K), (L), (M), and (N). WT [(G), (I), (K), and (M)] and ∆FR nin [(H), (J), (L), and (N)]. (O and P) Confocal images of representative ITs of WT at 7 dai (O) and ∆FR nin at 11 dai (P). (Q) Number of ITs at 14 dai. [(O) to (Q)] Plants were inoculated with GFP-labeled rhizobia. (R) RT-qPCR analysis using WT, ∆FR nin, and nin-9 at 0, 1, 3, 5, 7, and 21 dai. Each n contains roots from at least three plants. Data were normalized by LjUBQ expression. In box plots, dots mean individual biological replicates, and different letters or asterisk indicate statistically significant differences: P < 0.05, two-way analysis of variance (ANOVA) followed by multiple comparisons [(D), (E), and (R)]; one-way ANOVA followed by multiple comparisons (F) and Mann-Whitney test (Q). Scale bars, 1 mm (C); 500 μm [(G) and (H)]; 20 μm [(I) and (J)]; 5 μm [(K) and (L)]; 1 μm [(M) and (N)]; and 100 μm [(O) and (P)].

We then examined spontaneous nodule formation by treating with the synthetic cytokinin 6-benzylaminopurine (BAP) or overexpressing a constitutively active form of CCaMK (T265D) (21–23). Spontaneous nodule formation was not induced in the ∆FR nin (fig. S17, A and B). Moreover, complementation analysis in the daphne mutant with defect in cortical NIN function (24) showed that LjNIN (∆FR) induced fewer nodules and impaired nodule development compared to LjNIN (WT) (fig. S17C). These results suggest that the FR of NIN is required not only for rhizobial infection in the epidermis but also for the regulation of nodule organogenesis in the cortex. During the creation of ∆FR nin, we also obtained another nin mutant lacking the entire C-terminal region after RWP-RK, which completely abolished nodule formation (fig. S18).

FR elucidates NIN’s evolutionary trajectory from NLPs

Among the FRs of five LjNLPs, FR^LjNLP1 appears to be the most similar to FR^NIN (fig. S19A). To test functional equivalence of FR^LjNLPs to FR^LjNIN, we performed complementation tests of the nin-9 using chimeric proteins in which FR^LjNIN of LjNIN was swapped with each FR^LjNLPs. Although ITs and immature nodules were occasionally induced by LjNIN with FRs from LjNLP2/3/4, only LjNIN (FR^LjNLP1) formed mature pink nodules with shoots recovery comparable to that of LjNIN (FR^LjNIN, WT) (fig. S19B). We then examined the functional equivalence of full-length LjNLP1 and LjNIN. Overexpression of LjNLP1 in L. japonicus hairy roots could induce the expression of LjNF-YB, but not LjNF-YA or LjEPR3 (fig. S20, A and B). To avoid nitrate-induced suppression of nodule formation by LjNLP1/4 during activation of LjNLP1 by nitrate (14, 25), nin-9 Ljnlp1 Ljnlp4-1 triple mutant was used in the complementation test. LjNLP1 expressed under the LjNIN promoter fragment did not rescue the nodulation defects of the triple mutants (fig. S20C). Thus, these results suggest that functional conservation of the FR alone is not sufficient for NLP to acquire NIN function. Consistent with the FR-swapped complementation tests, predicted structure of FR^LjNLP1 was very similar to that of FR^LjNIN (fig. S21, A and B); however, the DNA binding specificity of LjNLP1 differed from those of LjNIN and LjNLP4, exhibiting a distinct pattern of broad selectivity (Fig. 1D and fig. S21C). In contrast, the FR-swapped chimera LjNIN (FR^LjNLP1) exhibited a DNA binding specificity more similar to LjNIN (Fig. 1D and fig. S21D). These results suggest that the specificity of the FR-independent RWP-RK^LjNLP1 is distinct from that of LjNIN, which may partly explain why LjNLP1 could not rescue the function of LjNIN.

To investigate the functional relevance of NIN-type FR in non-nodulating plants, we next examined AtNLP2, the evolutionarily closest homolog of LjNIN among the nine NLP family members in Arabidopsis thaliana (fig. S1, A and B). AtNLP2, in particular, exhibited broad DNA binding specificity similar to LjNIN, whereas AtNLP2 lacking FR [AtNLP2 (∆FR)] and AtNLP7, an ortholog of LjNLP4 (25, 26), specifically bound to perfect cis-elements (Fig. 5A). The FR-dependent expansion of the dimer interface was predicted in the dimeric structure of AtNLP2, but not in that of AtNLP7 (Fig. 5B and fig. S22A). In addition, the RWP-RK domain of AtNLP2 is more highly conserved with those of NINs than with other NLPs including LjNLP1, LjNLP4, and AtNLP7 (fig. S22B). These findings suggest that AtNLP2, unlike LjNLP1, shares biochemical features with LjNIN not only in the FR but also in the RWP-RK. Based on these observations, we hypothesized that AtNLP2 could mimic the biological function of LjNIN. Overexpression of AtNLP2 in L. japonicus resulted in excessive root deformation, reminiscent of the effects of LjNIN overexpression (fig. S23A). Furthermore, in the absence of rhizobia, AtNLP2 overexpression could induce the expression of LjNIN target genes, including LjNF-YA, LjNF-YB, and LjEPR3 (Fig. 5, C and D). Consistent with the previous report, AtNLP2 activity was dependent on nitrate (fig. S23, B and C) (27). Of note, we found that AtNLP2 expressed under LjNIN promoter could induce ITs and nodule formation in nin-9 Ljnlp1 Ljnlp4-1 mutants in the presence of nitrate in an FR-dependent manner (Fig. 5E), although the nodules induced by AtNLP2 were imperfect in terms of accommodating rhizobia in nodules (Fig. 5F). Together, NLPs from non-nodulating plants that are evolutionarily close to NIN, such as AtNLP2, likely share fundamental functions with NIN. It may be primarily due to similar DNA binding specificity conferred by the combination of RWP-RK and FR, although not entirely identical.

Fig. 5. AtNLP2 shares fundamental features with NIN.

(A) EMSA results of AtNLP2 with or without FR and AtNLP7. In protein sequence alignment of FR^AtNLP2/7, highlighted letters on a gray background and boldfaces on white background show the same amino acids as FR^NIN and amino acids with similar properties to those of FR^NIN, respectively. Each recombinant protein fused to the MBP at the N terminus was reacted at 2 μM final concentration with 0.25 μM DNA probe (i to v; fig. S2A). The electrophoretic patterns are shown with asterisks and arrowheads indicating the positions of free DNA and the protein-DNA complexes, respectively. (B) Top-ranked AlphaFold2-predicted structures of the AtNLP2 and AtNLP7 dimers containing the RWP-RK and the FR. Two different chains are depicted with different colors. The FR^AtNLP2 are highlighted with yellow-based colors. (C and D) RT-qPCR analysis using hairy roots overexpressing LjNIN or AtNLP2 by LjUBQ promoter. Transgenic plants were grown with 0.5 mM KNO₃ in the absence of rhizobia for 1 week. EV, empty vector. Each n contains hairy roots from three plants. Data were normalized by LjUBQ expression. Dots mean individual biological replicates. Different letters indicate statistically significant differences (P < 0.05, one-way ANOVA followed by multiple comparisons). (E and F) Complementation analysis of nin-9 Ljnlp1 Ljnlp4-1 triple mutants. NIN/NLPs were expressed under CE (5 kb)-LjNIN promoter (3.3 kb) in hairy roots. Transgenic plants were inoculated with DsRED-labeled or WT rhizobia in the presence of 5 mM KNO₃ for 30 to 38 days. (E) ITs (top panels) and nodules (bottom panels). Numbers indicate the frequency of plants forming ITs or nodules. (F) Nodule sections stained with toluidine blue. Scale bars, 50 μm [(E), top panels]; 2 mm [(E), bottom panels]; and 500 μm (F).

Last, to gain insights into the origin of NIN-type FR, we collected NLPs that are potentially orthologous to NIN from various land plant taxa (fig. S24, A and B). Full-length LjNIN with these FRs, except for FR^MpNLP, induced ITs and functional mature pink nodule formation, although with varying efficiency (fig. S24C). Consistent with these, AlphaFold3 predictions indicated that functional FRs had the potential to stabilize the DNA-bound RWP-RK dimer through hydrophobic and van der Waals interactions (fig. S25). By contrast, FR^MpNLP was predicted to be deficient in forming these interactions (fig. S25). Therefore, the acquisition of NIN-type FR might have occurred by the time gymnosperms emerged, during the molecular evolution of NLPs.

DISCUSSION

In this study, we identified FR^NIN as a crucial element responsible for NIN’s unique DNA binding specificity. We also propose a fundamental mechanism of DNA recognition within the NLP family, driven by dual dimerization through multiple domains. The finding of FR^NIN exemplifies how a TF can broaden its selectivity without altering its overall preference. To date, NLPs are known to act as dimers through the PB1 domain (28). Here, we demonstrated that the RWP-RK domains also weakly form dimers, where FR^NIN enhances NIN to bind to unpreferred cis-elements by sustaining the dimerization of the RWP-RK domains. Future experimental structural studies will further clarify the mechanism of FR^NIN-mediated broad DNA binding selectivity, as well as how the intrinsic RWP-RK–driven specificity, together with the adjacent region FR, shapes the common and distinct DNA binding features among NIN/NLP family members (Fig. 1 and fig. S21).

Although the ∆FR nin severely impairs rhizobial infection and nodule organogenesis, some RNS processes, including nodule initiation, are still maintained, indicating that NIN’s role is not completely attenuated in ∆FR nin. One possible explanation is that FR^NIN functions primarily to stabilize the NIN-dimer binding to DNA in a context-dependent manner, which may change spatiotemporally following nodule development. Under conditions in which FR^NIN is dysfunctional, there appears to be a mechanism by which NIN can form a dimer through other domains and regulates the expression of its target genes. In particular, considering the expression of LjNF-YA, NIN may have a function that does not require FR^NIN. Furthermore, when NIN function is attenuated, other factors may regulate the expression of NIN target genes through a NIN-independent mechanism.

Previous studies have shown that several prerequisites contributed to the establishment of NIN as the key TF regulating RNS. These include the evolution of the NIN promoter, which enabled its regulation during RNS, and the emergence of NIN-binding site in the promoters of many RNS-related genes (4, 29–31). In this study, we identified an additional prerequisite related to NIN protein function. NIN-type FR, the motif responsible for its broad DNA binding capacity, which is one of NIN’s defining features, had already appeared in certain NLPs before the evolution of RNS. It is likely that NLPs with functional FR similar to FR^NIN had already emerged before the divergence of gymnosperm and angiosperm lineages (Fig. 6). Although the amino acid sequences of NIN-type FRs are unexpectedly variable across species, they are more highly conserved within eudicots. This may imply that eudicot-specific FRs have acquired an additional function. Elucidating the function of NIN-type FR in those NLPs and eudicot-specific FR remains an intriguing subject for future research. Although intact AtNLP2 can substitute for many functions of LjNIN, it does not fully replicate LjNIN’s role. This finding suggests that additional molecular characteristics, such as the loss of nitrate responsiveness and/or other yet unidentified features, are likely critical for the evolutionary transition from NLPs to a fully functional NIN.

Fig. 6. A model for the evolution of NIN and its DNA binding specificity from NLP.

The NLP family underwent multiple duplications during land plant evolution, resulting in functional diversification (11, 66, 67). The closest homologs of NIN in angiosperms are found in two groups: LjNIN and AtNLP1/2/3, and LjNLP1 and AtNLP4/5. These groups are absent in gymnosperms; however, Cryptomeria japonica, a gymnosperm species, has an NLP that includes a functional FR motif resembling the FR motif found in NIN (hereafter referred to as “NIN-type FR”). In contrast, MpNLP, an ancestral NLP from Marchantia polymorpha, a bryophyte species, lacks the NIN-type FR, suggesting that this motif is not an ancestral feature of the NLP family. Rather, it may have originated in the NLPs of gymnosperms. The dimeric conformation of the RWP-RK domain facilitated by the NIN-type FR confers broader DNA binding specificity compared to NLPs lacking this motif. This structural feature may have played a key role in the functional divergence of NIN. The LjNIN and AtNLP1/2/3 group likely emerged in early eudicots, from which NIN was later derived within the nitrogen-fixing clade. Over the course of evolution, NIN appears to have acquired additional molecular features essential for its function, beyond just the NIN-type FR. Orange lines indicate NLPs containing the NIN-type FR.

MATERIALS AND METHODS

Phylogenetic analyses

Phylogenetic analyses in fig. S1 (A and B) were conducted in MEGA X (32, 33) using amino acid sequences of the full-length and DNA binding regions from the 49 and 48 NIN/NLP family proteins, respectively. The evolutionary history was inferred using the neighbor-joining method (34). The optimal trees with the sum of branch length = 12.22349058 and 7.82173710 are shown for the full-length and DNA binding regions, respectively. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches (35). The trees are drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic trees. The evolutionary distances were computed using the Poisson correction method (36) and are in the units of the number of amino acid substitutions per site. All ambiguous positions were removed for each sequence pair (pairwise deletion option). There were a total of 1531 and 156 positions in the final dataset of the full-length and DNA binding regions, respectively. Phylogenetic analysis in fig. S24A was conducted in MEGA11 (37) using amino acid sequences of DNA binding regions from the 42 NIN/NLP family proteins. The tree was constructed by maximum likelihood method using IQ-TREE 2 (38) (JTT + F + R6 substitution model). The optimal tree has the sum of branch length = 42.782. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches. The tree was rooted to the outgroup MpNLP and drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic trees. The evolutionary distances were computed using the Poisson correction method and were in the units of the number of amino acid substitutions per site.

Sequence alignments

CLUSTAL OMEGA (39) was used for multiple sequence alignments among NIN/NLP TFs using default parameters, and the results were displayed by ESPript 3.0 (40). Aligned protein sequences included PaNIN from Parasponia andersonii (accession: PON66248.1), and CgNIN from Casuarina glauca (accession: KF481969.1), in addition to those used in the phylogenetic analyses (fig. S1).

Protein expression and purification

Maltose-binding protein (MBP)-fused recombinant proteins for EMSA, SEC, blue native PAGE, and cross-linking experiments were prepared based on the method previously described (14) with some modifications. All truncated and chimeric constructs were BamHI-introduced into the pMAL-c2X vector (New England Biolabs) and cloned to Rosetta 2 (DE3) (Novagen). Overexpression was induced by 0.25 mM isopropyl-β-d-(-)-thiogalactopyranoside at 18°C for 16 hours. For constructs containing the C-terminal region, cultures were continuously shaken at 110 rpm throughout the 16-hour induction at 18°C. In contrast, for ΔC constructs, cultures were kept stationary for 12 hours and then shaken at 110 rpm for an additional 4 hours to improve the yield of soluble and active protein. After incubation, cells were harvested by centrifugation at 2750g for 15 min and stored at −80°C until use. Harvested cells were resuspended in binding buffer. For constructs containing the C-terminal region, the buffer contained 20 mM tris-HCl (pH 8.0), 1.0 M NaCl, 1 mM dithiothreitol (DTT), 10% glycerol, and 0.5 mM EDTA with the addition of 0.1% (v/v) protease inhibitor cocktail (Nacalai Tesque). For ΔC constructs, the buffer was identical except that DTT was omitted and 1 mM phenylmethylsulfonyl fluoride was added. The resuspended cells were stirred with 0.02% (w/v) lysozyme and then lysed by sonication. The cell debris was removed thoroughly by centrifugation at 40,000g for 30 min at 4°C. Crude protein fractions were applied to fresh (nonregenerated) Amylose Resin (New England Biolabs), washed with the binding buffer, and then eluted with elution buffer (20 mM tris-HCl, pH 8.0, 1.0 M NaCl, 1 mM DTT, 10% glycerol, 0.5 mM EDTA, and 30 mM maltose) with the addition of 0.1% (v/v) protease inhibitor cocktail to obtain MBP-tagged proteins. All eluted fractions were concentrated by VIVASPIN Turbo (30,000 molecular weight cutoff with polyethersulfone) (Sartorius) and further purified by SEC using a Superdex 200 Increase 10/300 GL column (Cytiva) in an ÄKTA go protein purification system (Cytiva). SEC was performed using SEC buffer (20 mM tris-HCl, pH 8.0, 1.0 M NaCl, 1 mM DTT, and 10% glycerol) at 0.4 ml/min flow rate. Purified and unaggregated proteins were reconcentrated and measured for concentration by their absorbance at 280 nm using NanoDrop One (Thermo Fisher Scientific) using their corresponding molecular weight and molar extinction coefficient at 280 nm.

EMSA

Interaction analyses of the NIN/NLP family proteins–DNA were performed based on the method previously described (14, 41) with some modifications. To prepare the probes, DNA fragments were labeled with carboxyfluorescein (FAM). The labeled DNA fragments were annealed in TNE buffer (20 mM tris-HCl, pH 8.0, 200 mM NaCl, and 1 mM EDTA) and then purified on the Superdex 200 Increase 10/300 GL in the ÄKTA go. In a total of 20 μl, each purified DNA fragment (0.25 μM) and poly[deoxyinosine-deoxycytidine (dI-dC)] (50 ng/μl) were mixed with the purified proteins in EMSA reaction buffer (10 mM tris-HCl, pH 7.5, 50 mM KCl, 50 mM NaCl, 1 mM DTT, 2.5% glycerol, and 5 mM MgCl₂) and incubated at 25°C for 20 min. Ten microliters of each mixture was loaded on a 10% polyacrylamide gel in 0.5× TBE buffer (45 mM tris-HCl, pH 8.2, 45 mM boric acid, and 1 mM EDTA). Fluorescence intensity (FI) was detected after 1-s exposure with the preset settings (epi-blue excitation, 466 nm; BPF535 bandpass filter, 535 nm) using LuminoGraph III WSE-6300 (ATTO). The complex ratio of the protein-DNA complex (%) [= 100 × FI〈complex〉/(FI〈free DNA〉 + FI〈complex〉)] was calculated using the Image J (version 1.53a) application.

SEC for oligomeric analysis

For oligomeric analysis, MBP-fused ΔC proteins were preincubated at a final concentration of 500 μM on ice for 1 week. Immediately prior to SEC, each sample was diluted to 60 μM in a total volume of 250 μl, and SEC was performed using the Superdex 200 Increase 10/300 GL column at 4°C, with a flow rate of 0.4 ml/min. The SEC buffer consisted of 20 mM tris-HCl (pH 8.0), 1.0 M NaCl, 1 mM DTT, and 10% glycerol. The elution was monitored at 280 nm using the ÄKTA go system.

Blue native PAGE

Purified proteins were mixed with an EzApply Native (ATTO) and then separated on a u-PAGEL H 4-20% gradient polyacrylamide gel (ATTO) in an EzRun BlueNative buffer (ATTO). The coomassie brilliant blue (CBB)-stained gel was destained with microwave heating and photographed. For comparison, the same protein samples were separated on an e-PAGEL 3-14% gradient polyacrylamide gel (ATTO) by SDS-PAGE and stained with CBB.

Cross-linking experiments

Prior to cross-linking with amine-specific cross-linker DSS, MBP-fused ΔC proteins and MBP alone were buffer-exchanged into 20 mM Hepes-NaOH (pH 7.5), 1.0 M NaCl, 1 mM DTT, and 10% glycerol using the Superdex 200 Increase 10/300 GL column to remove the tris buffer, which interferes with the DSS reaction. For SDS-PAGE–based cross-linking analysis, 10-μl reactions were prepared containing either 2 μM prepared protein with 1 μM nonlabeled DNA fragment, or protein alone (no DNA condition), in a buffer composed of 10 mM Hepes-NaOH (pH 7.5), 100 mM NaCl, 1 mM DTT, 2.5% glycerol, and 5 mM MgCl₂. After incubation at 25°C for 10 min, DSS was added to a final concentration of 200 μM [2% dimethyl sulfoxide (DMSO)], and the reaction was continued at 25°C for 20 min. Reactions were quenched by adding tris-HCl (pH 8.0) to a final concentration of 50 mM. The entire volume of each reaction sample was loaded onto the e-PAGEL 3-14% gradient polyacrylamide gel for SDS-PAGE. The gel was stained with 0.25% CBB R-250 in 45% methanol and 10% acetic acid for 5 min, then replaced with water and heated in a microwave for 10 min, followed by destaining in water for 2 days. The ratio of DSS-mediated cross-linked dimers (in percent) was calculated as follows: 100 × (intensity of the cross-linked dimer band)/(intensity of the cross-linked dimer band + intensity of the non–cross-linked band), based on densitometric analysis of CBB-stained SDS-PAGE gels using ImageJ (version 1.53a). For EMSA-based cross-linking analysis, 10-μl reactions were prepared containing 0, 1, or 2 μM protein with 0.25 μM FAM-labeled DNA fragment and poly(dI-dC) (50 ng/μl), in the same buffer as above. After incubation at 25°C for 10 min, DSS was added to a final concentration of 200 μM (2% DMSO), and reactions were continued at 25°C for 20 min. For DMSO controls, 2% DMSO was added instead of DSS. Reactions were quenched by adding tris-HCl (pH 8.0) to a final concentration of 50 mM, and the entire sample was analyzed under the same conditions described in the EMSA section.

Structure predictions

The protein dimer structures of the DNA binding modules (including both RWP-RK and FR) of LjNIN, LjNLP4, LjNLP1, AtNLP2, and AtNLP7 were predicted under two conditions using ColabFold v.1.0.0 with the AlphaFold v.2.1.0 monomer-ptm model (ptm) or LocalColabFold v.1.5.2 with the AlphaFold v.2.3.0 “multimer_v3” model (mv3) (42–44). A total of 100 protein structures were predicted using 10 seed parameters, with five models generated for each condition. The average-per-residue pLDDT score of the predicted full-length proteins indicated that the monomer-ptm model produced high-confidence predictions (fig. S7, A and B). The protein structures predicted by the monomer-ptm models were used for subsequent MD simulations. These 100 protein structures were classified through principal components analysis based on the xyz coordinates of all Cα atoms. AlphaFold3 predictions were performed by the AlphaFold server (https://alphafoldserver.com) (45). For prediction, two copies of each DNA binding module and ProLjCLE-RS2-derived double-stranded DNA fragment (fig. S2A) were simultaneously input. The predicted structures with pLDDT scores and separate figures showing the corresponding Predicted Aligned Error values were shown as the output from the AlphaFold program, and those of the top-ranked prediction without pLDDT were visualized in PyMOL 2.5.2 (Schrodinger, LLC) (46).

MD simulation

MD simulations were performed using the Maestro interface with the Desmond/GPU program (47). The three predicted protein structures were used as the initial templates for the MD simulations. The initial model of LjNIN with FR deleted was prepared by truncating FR and the post-FR region from LjNIN models. The protonation states and orientations of the residues were determined by the Protein Preparation Wizard in Maestro using PROPKA (48). The solvent box was prepared in a cube, with a TIP3P water model positioned 10 Å away from the protein surface (49). A 150 mM sodium chloride solution was added to the solvent box using the System Builder in Desmond (50). The force field for the protein atoms was assigned to OPLS-AA (all-atom) (51). After a 100-ps energy minimization step, a 100-ns MD simulation was performed under the following conditions: Equilibration was conducted using the isothermal-isobaric ensemble (NPT). Short-range electrostatic interactions between atoms over 9 Å apart were cutoff, and long-range electrostatic interactions were computed using the particle mesh Ewald method. All covalent bonds were constrained using the SHAKE algorithm (52). The thermostat and barostat were Nosé-Hoover chain (53) at 300 K and Martyna-Tobias-Klein (54) at 1 atm, with relaxation times of 1 and 2 ps, respectively. The RESPA integrator (55) was used, with Fourier-space electrostatics computed every 6 fs and all remaining interactions computed every 2 fs. Intermediate structures were recorded every 10 ps, and the Cα root mean square deviation was calculated and analyzed relative to the initial structure.

Plant materials and growth conditions

The Miyakojima MG-20 ecotype of L. japonicus was used as the WT plant (56). A description of nin-9, nlp1, and nlp4-1 plants was published previously (14, 18, 25). Plants were grown with or without Mesorhizobium loti MAFF 303099 in autoclaved vermiculite or on 1% agar plates with Broughton and Dilworth solution (57) under a 16-hour light/8-hour dark cycle at 24°C. ∆NifH strain of M. loti was previously reported (58). For the observation of rhizobial infection, plants were inoculated with M. loti MAFF 303099 constitutively expressing GFP or DsRED.

Stable and hairy root transformation

To create ∆FR nin plants by CRISPR-Cas9 system, guide RNAs (gRNAs) were designed using the CRISPR-P 2.0 program (fig. S18A and data S7) (59). Two gRNAs were used to target a gene, and they were cloned to pMR203_AB and pMR203_BC, gRNA cloning vectors, and then to pMR285_AD, a binary vector (60), by a standard ligation method using a DNA Ligation Kit (Takara) and several restriction enzymes. For stable transformation of L. japonicus, seeds were germinated on germination medium [0.5× Gamborg’s B5 salt mixture (Wako), 1/2× Gamborg’s vitamin solution (Sigma-Aldrich), 1% sucrose, and 1% agar] in a growth cabinet (24°C dark for the first 2 days, 24°C 16-hour light/8-hour dark cycle for the next 2 days). Agrobacterium tumefaciens GV3101 MP90RK strains harboring a construct were streaked on yeast extract peptone (YEP) plate with appropriate antibiotics for 2 days at 28°C. Seedlings were placed in the A. tumefaciens suspension, and then, their hypocotyls were cut into about 3-mm pieces. The hypocotyl pieces were placed onto the top of pilled filter papers saturated with cocultivation medium [1/10× Gamborg’s B5 salt mixture, 1/10× Gamborg’s vitamin solution, BAP (0.2 μg ml⁻¹), naphthaleneacetic acid (NAA; 0.05 μg ml⁻¹), 5 mM MES (pH 5.2), and acetosyringone (20 μg ml⁻¹), pH 5.5] and were incubated in a growth cabinet (21°C dark) for 6 days. After that, the hypocotyl pieces were transferred to callus induction medium [1× Gamborg’s B5 salt mixture, 1× Gamborg’s vitamin solution, 2% sucrose, BAP (0.2 μg ml⁻¹), NAA (0.05 μg ml⁻¹), 10 mM (NH₄)₂SO₄, 0.3% phytagel, meropen (12.5 μg ml⁻¹), and hygromycin B (15 μg ml⁻¹), pH 5.5] and were incubated in a growth cabinet (24°C 16-hour light/8-hour dark cycle) for 2 to 3 weeks. The hypocotyl pieces were transferred to a fresh callus induction medium every 1 week. When calluses became more than 1 mm in size, they were detached from the hypocotyls and transferred onto callus medium without hygromycin B and were incubated for 3 to 7 weeks in a growth cabinet (24°C 16-hour light/8-hour dark cycle) until the leaf primordia became visible. The calluses were transferred onto a new medium every 1 week. The calluses with the leaf primordia were then transferred to shoot elongation medium [1× Gamborg’s B5 salt mixture, 1× Gamborg’s vitamin solution, 2% sucrose, BAP (0.2 μg ml⁻¹), 0.3% phytagel, and meropen (12.5 μg ml⁻¹), pH 5.5] and incubated until their shoot length became about 1 cm. Individual shoots were detached from calluses and transferred to root induction medium [1/2× Gamborg’s B5 salt mixture, 1/2× Gamborg’s vitamin solution, 1% sucrose, NAA (0.5 μg ml⁻¹), 0.4% phytagel, and meropen (12.5 μg ml⁻¹), pH 5.5] and incubated for 8 days. Then, they were transferred to a root induction medium without NAA and cultivated until their root length became about 2 to 3 cm. The resultant transgenic plants were transplanted into soils for further cultivation. Transgenic plants with homozygous mutations were used for analysis.

For LjNIN, LjNLP4, LjNLP1, or AtNLP2 overexpression in L. japonicus hairy roots, the ProLjUBQ-tNOS cassette was amplified by PCR from an original vector (14) and was cloned to pCAMBIA1300-GFP by the In-Fusion (Clontech) reaction. The coding sequence (CDS) of LjNIN, LjNLP4, LjNLP1, or AtNLP2 was amplified by PCR and was cloned downstream of ProLjUBQ by the In-Fusion reaction. Constructs described in protein expression and purification were used as templates to amplify fragments of CDS of the RWP-RK and/or the FR swapped LjNIN/LjNLP4 proteins. For complementation of nin mutants, the 5-kb region around CE (30) and the ProLjNIN 3.3-kb region upstream of the start codon [CE (5 kb)-ProLjNIN (3.3 kb)] (31) and tLjNIN were amplified by PCR from WT genomic DNA and were cloned to pCAMBIA1300-GFP by the In-Fusion reaction. Then, the CDS of LjNIN, LjNLP1, AtNLP2, or AtNLP7 was amplified by PCR and was cloned downstream of CE (5 kb)-ProLjNIN (3.3 kb) by the In-Fusion reaction. To generate LjNIN (FR^swap) and AtNLP2 (∆FR) constructs, two fragments of CDS of LjNIN or AtNLP2 before and after the FR were amplified by PCR using primers shown in data S7 and cloned downstream of CE (5 kb)-ProLjNIN (3.3 kb) by the In-Fusion reaction. For hairy root transformation, seeds were germinated on the germination medium described above in a growth cabinet (24°C dark for the first 2 days, 24°C 16-hour light/8-hour dark cycle for the next 2 days). Agrobacterium rhizogenes AR1193 strains harboring each construct were streaked on YEP plate with appropriate antibiotics for 2 days at 28°C. Seedlings were placed in the A. rhizogenes suspension and then cut at the base of the hypocotyls. The seedlings with cotyledons were transferred onto hairy root medium (1× Gamborg’s B5 salt mixture, 1× Gamborg’s vitamin solution, 2% sucrose, and 1% agar) and were grown in a growth cabinet (24°C dark for the first 1 day, 24°C 16-hour light/8-hour dark cycle for the next 2 days). Then, the plants were transferred onto fresh hairy root medium containing meropen (12.5 μg ml⁻¹) and were grown for 7 to 9 days in a growth cabinet (24°C 16-hour light/8-hour dark cycle) (61). Transgenic roots were identified by GFP fluorescence, and the plants with transgenic hairy roots were used for further experiments.

Acetylene reduction assay

The nitrogenase activity of nodules was indirectly determined by measuring the acetylene reductase activity (25). Nodulated plants were put into 20-ml vials. Subsequently, acetylene was injected into the vials. After incubation for 10 min, the amount of ethylene produced was measured using a GC-2014 (Shimadzu).

Toluidine blue staining

Root segments with nodules were fixed at 4°C overnight with 4% paraformaldehyde and 0.25% glutaraldehyde in 0.05 M phosphate buffer (pH 7.2). The fixed material was dehydrated in an ethanol series and subsequently embedded in Technovit 7100 (Kulzer) according to the manufacturer’s protocol. Longitudinal sections (8 μm) were made by using an RM2265 microtome (Leica Microsystems) and stained for 5 min in 0.05% Toluidine Blue O. Sections were analyzed by using a BX53 microscope equipped with a DP74 camera (Olympus).

Transmission electron microscopy analysis

Nodules were cut from roots and immediately placed in a solution of 2% paraformaldehyde and 2% (v/v) glutaraldehyde in 0.05 M cacodylate buffer (pH 7.4) for fixation and left overnight at 4°C. The fixative was washed out by three successive 30-min washes in 0.05 M cacodylate buffer and postfixed in 2% OsO₄ in 0.05 M cacodylate buffer at 4°C overnight. The fixed material was dehydrated in an ethanol series (50 and 70% each for 30 min at 4°C, 90% and three changes of 100% ethanol each for 30 min at room temperature, and 100% ethanol for 2 days at room temperature). The dehydrated samples were washed out by two successive 30-min propylene oxide (PO) washes. The samples were infiltrated with Quetol-651 resin (Nisshin EM) by successive changes of resin:PO mixes at room temperature (1:1 for 3 hours and 100% resin for 3 hours) and then 100% resin at 60°C for 48 hours to polymerize. The material was sectioned with a diamond knife using a Leica Ultracut UCT ultramicrotome (Leica), and ultrathin sections of ~80 nm were transferred onto copper grids. The sections were stained with 2% uranyl acetate for 15 min at room temperature and lead stain solution (Sigma-Aldrich) for 3 min. The grids were viewed in a JEM-1400Plus (JEOL) at 100 kV and imaged using the charge-coupled device camera EM-14830RUBY2 (JEOL).

Gene expression analysis

The primers used for PCR are listed in data S7. Total RNA was isolated from respective tissues using the PureLink Plant RNA Reagent (Invitrogen) or the Plant Total RNA Mini Kit (Favorgen Biotech). First-strand cDNA was prepared using the ReverTra Ace qPCR RT Master Mix with gDNA Remover (Toyobo). RT-qPCR was performed using a CFX Opus 384 Real-Time PCR system (Bio-Rad) with a THUNDERBIRD SYBR qPCR Mix (Toyobo) or a THUNDERBIRD Next SYBR qPCR Mix (Toyobo) following the manufacturer’s instructions.

RNA sequencing analysis

Total RNA was isolated from roots using the PureLink Plant RNA Reagent (Invitrogen). Libraries were prepared using a NEBNext Ultra II RNA Library Prep Kit from Illumina (New England Biolabs) following the manufacturer’s instructions and sequenced using a NovaSeq 6000 (Illumina) instrument with the 150-bp paired-end sequencing protocol. RNA sequencing (RNA-seq) reads were mapped to the L. japonicus MG-20 genome version 3.0 using HISAT2 (version 2.2.1) with the default parameters (62). Mapped reads were then assembled using StringTie (version 1.3.5) (63). Up- or down-regulated genes were identified with edgeR (version 3.40.2) (64), using a 5% false discovery rate. Only genes with counts per million of ≥0.5 in at least three out of the six samples derived from two different conditions were included in each analysis.

Microscopy

Bright-field images were taken under an S9i stereo microscope (Leica) or a BX53 upright microscope (Olympus). Fluorescence images were obtained using an LSM700 confocal laser-scanning microscope (Carl Zeiss) equipped with ZEN (Carl Zeiss), an M205FA fluorescence stereo microscope (Leica), or a DM6 B upright microscope (Leica) equipped with LAS X software, which processes images following THUNDER algorithms (Leica).

Statistical analysis

Statistical analysis was performed using GraphPad Prism version 9 (GraphPad Software). Normality was checked using the Shapiro-Wilk test, and P > 0.05 was considered as normal distribution. The F test was used to test whether the variances of the two populations were equal. Appropriate methods were chosen according to the nature of the data. The criterion of P < 0.05 means a statistically significant difference in this study.

Supplementary Materials

The PDF file includes:

Figs. S1 to S25

Legends for data S1 to S7

Other Supplementary Material for this manuscript includes the following:

Data S1 to S7