Insights into the complex role of GRAS transcription factors in the arbuscular mycorrhiza symbiosis

To improve access to limiting nutrients, the vast majority of land plants forms arbuscular mycorrhizal (AM) symbioses with Glomeromycota fungi. We show here that AM-related GRAS transcription factors from different subgroups are upregulated during a time course of mycorrhization. Based on expression studies in mutants defective in arbuscule branching (ram1-1, with a deleted MtRam1 GRAS transcription factor gene) or in the formation of functional arbuscules (pt4-2, mutated in the phosphate transporter gene MtPt4), we demonstrate that the five AM-related GRAS transcription factor genes MtGras1, MtGras4, MtGras6, MtGras7, and MtRad1 can be differentiated by their dependency on MtRAM1 and MtPT4, indicating that the network of AM-related GRAS transcription factors consists of at least two regulatory modules. One module involves the MtRAM1- and MtPT4-independent transcription factor MtGRAS4 that activates MtGras7. Another module is controlled by the MtRAM1- and MtPT4-dependent transcription factor MtGRAS1. Genome-wide expression profiles of mycorrhized MtGras1 knockdown and ram1-1 roots differ substantially, indicating different targets. Although an MtGras1 knockdown reduces transcription of AM-related GRAS transcription factor genes including MtRam1 and MtGras7, MtGras1 overexpression alone is not sufficient to activate MtGras genes. MtGras1 knockdown roots display normal fungal colonization, with a trend towards the formation of smaller arbuscules.


AM-related GRAs tFs of Medicago truncatula belong to different subgroups. Based on genome-
wide expression profiles recorded by GeneChip hybridizations 47 , a core set of AM-activated GRAS TF genes, namely MtGras1 (designated TF80 in 33 ), MtGras4, MtGras6, MtGras7, and MtRad1 42 was selected. In addition, the MtRam1 gene 31 , encoding a GRAS TF that controls arbuscule branching 33 , was included. The corresponding identifiers from the Medicago truncatula genome 51 , the Medicago Gene Expression Atlas 52 , and the literature are listed in Supplementary Table S1. Analyses of gene expression data from GeneChip hybridizations stored in the Medicago Gene Expression Atlas 52 and from studies on MtRam1 31 revealed two different subgroups among the selected MtGras genes. Whereas MtGras1, MtGras7, MtRad1, and MtRam1 are specifically expressed in mycorrhizal roots and are only activated at background levels in non-mycorrhized controls, MtGras4 and MtGras6 display a low but detectable expression in non-mycorrhizal roots ( Supplementary Fig. S1), yet showing a significant upregulation upon mycorrhization 47 . This corresponds to the data shown in Fig. 1, where MtGras4 and MtGras6 are expressed at significant levels at time point 0. Although being AM-induced, MtGras4 and MtGras6 transcription also occurs in other tissues 52 . We nevertheless selected these genes in addition to the AM-specific MtGras1, MtGras7, MtRad1, and MtRam1 genes, because arbuscule formation responds to a range of nutritional, physiological, and environmental factors 16 that obviously are relevant in non-symbiotic tissues as well.
In a phylogenetic tree of the deduced amino acid sequences of these and selected other symbiotic GRAS TF genes, specific groupings were evident ( Supplementary Fig. S2). Interestingly, MtGRAS4 and MtGRAS7, two GRAS TFs belonging to the MIG1 family defined by 43 share the highest sequence similarities, suggesting www.nature.com/scientificreports www.nature.com/scientificreports/ a functional relationship although only MtGras7 is specifically expressed in AM roots 47 . In addition, the close relation of MtRAM1 and MtRAD1 is of special interest, since these TFs control arbuscule branching as well as arbuscule development and since mutual protein-protein interactions were reported 33,36 . AM-related GRAS TF genes are differentially upregulated in the course of mycorrhization. To reveal the timing of GRAS TF action during the development of an AM symbiosis, a gene expression time course study was performed (Fig. 1). Histological analyses of the harvested mycorrhizal roots indicated the predominant presence of R. irregularis spores and extraradical hyphae until 7 days post inoculation (dpi), while significant intraradical colonization and arbuscule development started from 10 dpi.
Since root material at 0 dpi already contained germinated R. irregularis spores, the AM fungal marker genes GiTefa (encoding a translation elongation factor alpha) and GiaTub (encoding an a-tubulin) are already expressed, whereas transcription of MtPt4, encoding an arbuscule-specific phosphate transporter 50 , is only detected upon arbuscule presence. While the expression of GiTefa and GiaTub showed a linear increase over time, MtPt4 activation rose almost exponentially, mirroring the quick and ongoing process of arbuscule build-up.
During mycorrhization, the transcriptional activation of MtGras4 and MtGras6, being upregulated appr. 16to 17-fold between 0 dpi and 42 dpi (Fig. 1) resembled that of GiaTub and GiTefa, whereas MtGras1, MtGras7, MtRad1, and MtRam1 (appr. 180-to 1200-fold upregulation from the time point of their first expression, Fig. 1) followed the strong rise in MtPt4 transcription. These patterns indicate two different types of activation, with MtGras4 and MtGras6 being already markedly expressed at 0 dpi, probably due to their weak expression in non-mycorrhized roots (Medicago Gene Expression Atlas 52 and 47 ), while the upregulation of MtGras1, MtGras7, MtRad1, and MtRam1 follows the ongoing build-up of functional arbuscules in MtPt4-expressing cells.

AM-related GRAS TF genes differ in their dependency on the GRAS TF MtRAM1.
To study the dependency of AM-related GRAS TF gene expression on MtRAM1, a GRAS TF required for arbuscule branching 33 , comparative gene expression analyses were carried out in R. irregularis-mycorrhized roots of wild type plants and ram1-1 mutants.
Whereas real-time RT-PCR measurements (Fig. 2) of mycorrhizal roots showed a strong decrease in fungal gene expression (GiTefa, GiaTub; down to 1-2%) as well as a complete lack of MtRam1 and MtPt4 transcription in ram1-1 mutants, indicating the absence of highly branched, symbiotically active arbuscules, the reduction of GRAS TF gene transcription appeared diverging. While expression of MtGras1 and MtGras7 strongly decreased   www.nature.com/scientificreports www.nature.com/scientificreports/ to 0.3-0.6% of the level in wild type roots, transcription of MtRad1, MtGras4, and MtGras6 was less affected (reduction to 13-35%). A strongly decreased expression in ram1-1 mutants was also observed for the AM-specific MtMyb1 gene, encoding a key regulator of arbuscule degradation 18 .
Since fungal colonization is impaired in ram1-1 mutants and since expression of AM-induced marker genes as well as MtGras1 and MtRad1 was lower in ram1-1 mutants in response to colonization by Glomus versiforme 33 , a reduced expression of the AM-induced GRAS TF genes studied here was expected. Thus, expression levels of the genes studied ( Fig. 2) were divided by the GiTefa transcription level in order to adjust gene expression to the amount of fungal tissue. This procedure revealed that lower transcription levels of MtGras1 (24.3%), MtGras7 (0.01%), and MtMyb1 (0.03%) were still evident in R. irregularis colonized ram1-1 in comparison to wild type roots, indicating that the expression of these genes is not just reduced in ram1-1 mutants as a consequence of lower fungal colonization.
To achieve a cellular resolution of GRAS TF gene expression, their promoter regions were fused to the gusAint reporter gene, and the resulting transcriptional fusions were expressed in transgenic M. truncatula roots. These studies demonstrated a clearly AM-induced promoter activity for MtGras1, MtGras4, MtGras6, MtGras7, and MtRad1 in wild type roots, with a predominant or exclusive activation in the arbuscule-containing cells (Fig. 3). In ram1-1 mutants, promoter activities of MtGras1 and MtGras7 were completely abolished, even after prolonged staining. In contrast, MtGras4, MtGras6, and MtRad1 promoters are still AM-induced in the ram1-1 mutant background. These findings were in line with our gene expression studies (Fig. 2) and suggested a position of MtGRAS1 and MtGRAS7 downstream of MtRAM1 in the regulatory cascade leading to arbuscule formation, while MtGRAS4, MtGRAS6, and MtRAD1 have to be placed either upstream or parallel to MtRAM1. expression of AM-related GRAs tF genes differs in the dependency on morphologically fully developed arbuscules. To study the dependency of GRAS TF gene expression on the presence of www.nature.com/scientificreports www.nature.com/scientificreports/ functional, phosphate-transporting arbuscules, comparative gene expression analyses were carried out in R. irregularis-mycorrhized roots of wild type plants and pt4-2 mutants.
Real-time RT-PCR analyses of pt4-2 mutants revealed a significant regulation of the GRAS TF genes investigated similar to ram1-1 mutants (Fig. 4). In addition to MtPt4, the MtGras1 and MtGras7 genes are most strongly repressed in the pt4-2 background (to 1.2-2.3% of wild type expression), whereas other GRAS TF genes were less strongly affected. After a division by the GiTefa expression level, only MtGras1 and MtGras7 transcription was reduced to 36.9% and 66.6% in pt4-2 in comparison to wild type roots, respectively, indicating that this downregulation is not just reflecting a lower degree of colonization in pt4-2 roots.
Similar to the results for ram1-1 mutants (Fig. 3), the promoters of MtGras1 and MtGras7 are inactive in pt4-2 roots, while the MtRad1, MtGras4, and MtGras6 promoters are still functional in root areas containing the typical prematurely degrading, stunted arbuscules that were regularly observed for pt4-2 mutants 53 (Fig. 5). On the other hand, since the pt4-2 stunted arbuscule phenotype was not absolutely stable in our growth conditions, some infection units from pt4-2 roots, that in other areas showed typical premature arbuscule degeneration, occasionally developed WT-like arbuscules. Interestingly, wild-type like MtGras1 and MtGras7 promoter activities were now observed ( Fig. 5Ba-h). This local phenomenon suggests that MtGras1 and MtGras7 activation is dependent on a particular stage of arbuscule development that, when modulated by endogenous or exogenous conditions, as e.g. demonstrated for N-starvation 53 ; can alleviate the pt4-2 phenotype of premature arbuscule degeneration. It thus appears that MtGras1 and MtGras7 activation does not simply depend on the pt4-2 genotype, but on the existence of a symbiotic interface beneficial for the plant, enabling the development of mature arbuscules.
A summary of MtGras activity in ram1-1 and pt4-2 mutants is presented in Supplementary Figure S3. In both mutants, MtGras1 and MtGras7 expression is hardly detectable by real-time RT-PCR experiments or histological studies of promoter activity, while MtGras4, MtGras6, and MtRad1 expression comparable to mycorrhizal wild type roots is observed. With respect to the results from our mutant studies, AM-related GRAS TF genes can thus be divided into two groups, being either MtRAM1-and MtPT4-dependent or -independent.
MtGRAS4 and MtGRAS7 form a regulatory module within the network of AM-related GRAS tFs. As one representative of an MtRAM1-and MtPT4-independent GRAS TF gene, MtGras4 was further characterized. To understand its role in AM formation, the Tnt1 transposon insertion line NF4813 was identified in the Medicago truncatula mutant database 54 (Fig. 6A). Plants were inbred to generate a homozygous knockout line, which was tested for the position of the Tnt1 insertion using genomic PCR (Fig. 6B). Real-time RT-PCR measurements revealed strongly reduced levels of the MtGras4 5′ and virtually no remaining MtGras4 3′ transcript region up-or downstream of the Tnt1 insertion site, respectively (Fig. 6C,D).
Among all tested GRAS TF and AM marker genes, the MtGras4 knockout line showed a reduced expression of MtGras7, both on the transcript level (to 22%, Fig. 6E) and the activity of the promoter (Fig. 6I), whereas transcription of all other AM-related GRAS TF genes was unchanged. On the phenotypical level, both the mycorrhization rate and the arbuscule size distribution are unchanged in MtGras4 knockout mutants (Fig. 6F,G). Complementation of the M. truncatula R108-based Tnt1 line with the M. truncatula A17 MtGras4 gene led to a restoration of MtGras7 expression (Fig. 6H), confirming that MtGras7, being strongly upregulated during later stages of mycorrhization ( Fig. 1), is a direct or indirect target of MtGRAS4. Since MtGras4 expression was independent of MtRAM1 and MtPT4, we conclude that the MtGRAS4/MtGRAS7 regulatory module operates parallel to the formation of highly branched, functional arbuscules.
MtGras1 knockdown affects the expression of other AM-related GRAS TF genes. As an example for a GRAS TF gene dependent on MtRAM1 and MtPT4, MtGras1 was functionally studied in transgenic RNAi roots exhibiting an MtGras1 knockdown, since homozygous knockout lines were not available. Due to the fact that RT-PCR measurements indicated effects of an MtGras1 knockdown on AM-related gene expression, a global transcriptomics approach was pursued.
A comparative genome-wide gene expression study of R. irregularis-mycorrhized RNAi:MtGras1 and RNAi:gusAint control roots identified 1020 genes that were at least 2-fold (p < 0.05) downregulated in MtGras1 www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ knockdown roots, indicating the potential of MtGRAS1 to participate in the regulation of gene expression in AM (Supplementary Table S2). A selection of MtGras genes differentially expressed in the MtGras1 knockdown roots is shown in Fig. 7. In line with initial real-time RT-PCR measurements, the RNA interference construct led to a reduction of MtGras1 expression to 29% of the wild type level. This 71% MtGras1 knockdown was able to reduce transcription of the AM-related GRAS TF genes MtGras6, MtGras7, MtRad1 36 , MtRam1 33 , and also MtTF124 33 , while expression of the fungal marker gene GiTefa as well as the M. truncatula MtGras4, MtPt4, and MtMyb1 genes were not significantly affected (Fig. 7).
Especially the only slight, non-significant (p = 0.11; Fig. 7, Supplementary Table S2) reduction of MtPt4 expression in MtGras1 knockdown roots indicates that the formation of active, phosphate-transporting arbuscules is not impaired by a reduced MtGras1 expression. This was also true for the transcription of the MtMyb1 gene, encoding a transcription factor controlling arbuscule degeneration 18 , which is not reduced in MtGras1 knockdown roots (Fig. 7). This indicates that MtGRAS1 does not participate in the initiation of arbuscule degradation, e.g. by activating MtMyb1.
Since MtRAM1 was shown to be required for MtGras1 activation 33 (Figs 2, 3), it is intriguing that MtRam1 expression is downregulated to 24% in mycorrhized MtGras1 knockdown roots (Fig. 7). This finding suggests that similar to the observation for MtRAM1 and MtWRI5a in the regulation of fatty acid biosynthesis 39 Table S2), the question arose, to what extent these patterns of transcriptional changes resemble those in an MtRam1 knockout. To study this, genome-wide expression was recorded in wild type vs. ram1-1 roots. This experiment revealed that the expression of 836 genes were at least 2-fold (p < 0.05) lower in ram1-1 knockout roots, including several marker genes for arbuscule function (e.g. MtPt4, being transcribed at a 689-fold lower level, Supplementary Tables S3 and S4). Although a limited set of 250 genes is at least 2-fold lower expressed (p < 0.05) in RNAi:MtGras1 as well as in ram1-1 roots (Fig. 7B), the major pattern of gene expression regulation is characteristic of either the MtGras1 knockdown or the ram1-1 knockout. Specifically, when looking into the detailed pattern of regulation of 31 well-defined AM marker genes (Supplementary Table S4, Fig. 7C), it is evident that most of these are either only (e.g. MtMyb1) or much stronger (e.g. MtPt4) downregulated in ram1-1 mutants, indicating that the core gene expression program activated in arbuscule-containing cells is strongly affected in ram1-1 mutants, but not or only moderately in the MtGras1 knockdown roots. This finding suggests that although MtGras1 expression depends on MtRAM1 33 (Figs 2, 3), MtGRAS1 is not simply a direct target of MtRAM1 to activate downstream genes related to arbuscule formation and function.
MtGras1 overexpression does not activate other AM-related GRAs tF genes. The massive transcriptional effects resulting from an MtGras1 knockdown prompted us to investigate the effect of MtGras1 overexpression in transgenic M. truncatula roots. In these experiments, the arbuscule-specific MtPt4 and the general ubiquitin3 promoters were used to drive MtGras1 expression in mycorrhized and non-mycorrhized roots, respectively. Although a 7.4-and 1523-fold MtGras1-overexpression was achieved, respectively, leading to comparable amounts of MtGras1 transcripts in mycorrhized and uninoculated roots ( Supplementary Fig. S5), no activation of other GRAS TF or AM marker genes such as MtPt4 was detected ( Supplementary Fig. S5), suggesting that MtGRAS1 is not at the terminal position of a regulatory cascade or requires co-expressed interaction partners to activate transcription of target genes.

MtGras1 knockdown roots show a trend towards the development of smaller arbuscules.
To address the question, whether an MtGras1 knockdown influences the fungal colonization of roots or the maturation of arbuscules, phenotypical studies were performed in comparison to control roots.
While no changes in the mycorrhization rate or the frequencies of arbuscules and vesicles were observed (Fig. 7D) and the arbuscules in MtGras1 knockdown roots did not show symptoms of premature degeneration (Fig. 7E), arbuscule length measurements of mycorrhizal RNAi:MtGras1 in comparison to RNAi:gusAint control roots indicated a shift in the distribution of arbuscule sizes (Fig. 7F). Specifically, MtGras1 knockdown roots tend to contain a higher proportion of smaller and a lower proportion of large arbuscules, suggesting a delayed or less sustained arbuscule development. In line with the fact that the expression of most marker genes for arbuscule formation and function is not markedly affected in mycorrhized MtGras1 knockdown roots (Fig. 7B-C), this effect nevertheless appears subtle and is probably part of a fine-tuning of the arbuscule life-cycle.

Discussion
Colonization of roots by AM fungi ultimately leads to the formation of intracellular arbuscules, functioning as a nutrient exchange interface between plant cells and fungal hyphae. Arbuscule development requires a fundamental transcriptional reprogramming of root cortical cells [44][45][46][47][48][49]55 , being governed by a suite of AM-activated regulators, including several GRAS TFs 29 (Figs 2-5). In our growth conditions, pt4-2 mutants occasionally developed not only prematurely decaying but also apparently vital arbuscules (Fig. 5Ba-h). This phenomenon demonstrates the dependency of MtGras1 and MtGras7 transcription not on the genotype, but on the presence of fully developed arbuscules. Obviously, prematurely decaying arbuscules do not reach the phase of MtGras1 and MtGras7 expression, while fully developed arbuscules do. This indicates that with respect to transcriptional control, MtGRAS1 and MtGRAS7 are placed downstream of MtRAM1 action and might thus be related to stages of arbuscule formation, when arbuscule maturation already occurred. In contrast, MtGras4, MtGras6, and MtRad1 www.nature.com/scientificreports www.nature.com/scientificreports/ expression is not abolished in ram1-1 and pt4-2 mutants (Supplementary Fig. S3), allowing to conclude that the encoded GRAS TFs are not related to arbuscule maturation and are thus likely connected to developmental stages before arbuscule maturation takes place, although a function independent of MtRAM1 and MtPT4 cannot be ruled out. These findings were visualized in a model, where the AM-related GRAS TFs are positioned relative to the establishment of mature arbuscules (Fig. 8).
To study the function of an MtRAM1-and MtPT4-independent GRAS TF gene, an MtGras4 insertion mutant in M. truncatula R108 was analyzed. Although the MtGras4 knockout downregulates MtGras7, which in contrast to MtGRAS4 is MtRAM1-and MtPT4-dependent, it does not affect AM fungal colonization or the arbuscule size distribution (Fig. 6). The MtGras4 knockout, inducing a reduced MtGras7 transcription in the M. truncatula R108 MtGras4 mutants, was successfully complemented by expressing an M. truncatula A17 MtGras4 gene, indicating a conserved regulation in M. truncatula R108 and Jemalong A17 regarding these genes. Interestingly, analyses of MtGras7 and MtGras4 upregulation during the AM time course revealed different patterns (Fig. 1). While MtGras7 appeared to be expressed in an arbuscule-correlated manner that follows MtPt4 activation, MtGras4 expression was already detectable from the onset of fungal colonization. In line with the MtRAM1-and MtPT4-dependency of MtGras7 activation (Figs 2, 3), this further indicates that MtGRAS7 functions in the later stages of arbuscule development, when maturation already took place. In this respect, the dependency of MtGras7 on the MtRAM1-and MtPT4-independent GRAS TF MtGRAS4 is even more remarkable, since it demonstrated a connection between more early and MtRAM1-independent (MtGras4) and more later, MtRAM1-dependent stages (MtGras7) of arbuscule formation. The somewhat outstanding position of the MtGRAS4/MtGRAS7 module within the GRAS TF network is also reflected on the sequence level, where MtGRAS4 and MtGRAS7 share a common clade, being isolated from the rest of the GRAS TFs investigated (Supplementary Fig. S2).
In spite of its dependency on MtRAM1 and MtPT4 ( Supplementary Fig. S3), MtGRAS1 was also found to regulate components placed more upstream of arbuscule development, namely MtGras6, MtRad1, and remarkably also MtRam1 (Fig. 7A). In addition, the AM-related GRAS TF gene and MtRam1-homologue MtTf124 33 appears downregulated in MtGras1 knockdown roots (Supplementary Table S4). Due to the fact that MtGRAS1 regulates MtRam1 expression (Fig. 7A), while MtRAM1 on the other hand is required for MtGras1 activation (Figs 2 and 3), we propose the existence of a regulatory feedback loop, where MtGRAS1 enhances MtRam1 transcription, thus stimulating its own activation as well as the expression of MtRAM1-regulated target genes, ultimately leading to the establishment of a functional, highly branched arbuscule (Fig. 8). Since it was demonstrated that MtRAM1 is not able to directly stimulate MtGras1 expression 33 , the activation of MtGras1 thus has to be dependent on additional, MtRAM1-dependent regulators. It is likely that currently unknown mediators also have to be assumed for most of the relationships shown in the regulatory network visualized in Fig. 8.
Interestingly, MtGras1 expression was downregulated in pt4-2 mutants colonized by Rhizophagus irregularis (Figs 4, 5) and in pt4-1 mutants mycorrhized with Gigaspora gigantea 18 , indicating that the encoded GRAS TF acts downstream of the formation of functional arbuscules. It is thus tempting to hypothesize that MtGRAS1 acts as a checkpoint at a certain point of the later stages of the arbuscule life cycle, where the initial signaling for arbuscule development and branching via MtRAM1 is completed, and the morphological expansion of the functional, phosphate-transporting fungal interface needs to be accelerated or sustained. The slight shift towards the formation of smaller arbuscules observed in MtGras1 RNAi roots (Fig. 7D-F) might thus be caused by the lack of MtGRAS1-activated MtRam1 transcription (Fig. 7A), which would lead to a slower or less sustained, MtRAM1-controled, arbuscule branching. The effect on the arbuscule size distribution is nevertheless small and possibly also suffers from the non-synchronized mycorrhization process, which masks subtle differences in arbuscules sizes.
On the protein level, MtGRAS1 was shown to interact with MtRAM1 in yeast two-hybrid assays 33 , which indicates a joint function of these transcription factors. However, we were able to show that an RNAi-mediated knockdown of MtGras1 leads to a different molecular phenotype than a knockout of MtRam1. Although there are overlaps in the genes regulated by MtGRAS1 and MtRAM1, ram1-1 mutants show a much stronger downregulation of AM-induced genes characteristic of the presence of functional arbuscules and a strong effect on arbuscule branching 33 , which is not the case for MtGras1 RNAi roots (Fig. 7A). The different transcription pattern of mycorrhized MtGras1 RNAi roots and mycorrhized ram1-1 mutants is thus in line with a modifying function of MtGRAS1 during the establishment of a functional arbuscule interface. Although these comparative www.nature.com/scientificreports www.nature.com/scientificreports/ transcription patterns (Fig. 7B) were derived from studies of knockout mutants (MtRam1) and transgenic knockdown roots (MtGras1), the fundamentally different expression pattern observed in ram1-1 and MtGras1 RNAi roots most likely excludes that MtGRAS1 itself acts as a key independent regulator of arbuscule formation, downstream of MtRAM1. An additional hint that MtGRAS1 solely is unable to activate AM-related genes derives from our overexpression experiments, where an upregulated MtGras1 transcription had no effect on selected target genes, probably since MtGRAS1 requires an interaction partner such as MtRAM1 33 .
In addition to the more upstream components regulating arbuscule formation, MtGRAS1 also regulates MtGras7 (Fig. 7A), which is also MtRAM1-and MtPT4-dependent and is thus located at a similar position relative to the formation of mature arbuscules. Incidentally, the downregulation of MtGras7 in MtGras1 knockdown roots might explain why MtGras7 expression is reduced but still detectable in the MtGras4 insertion mutant (that still expresses MtGras1, Fig. 6).
The effect of MtGRAS1 on MtGras7 transcription implies that MtGras7 can be activated by both the AM-related MtGRAS4 and the AM-specific MtGRAS1 TF in an independent manner, suggesting that MtRAM1-/ MtPT4-dependent (MtGRAS1 and its targets including MtGras7) and -independent regulatory modules (MtGRAS4 and its target MtGras7) are connected. The complex network of AM-related and AM-specific GRAS TFs might thus contribute to the balanced expression of AM-related genes at the arbuscule interface, where different physiological and nutritional conditions have to be integrated.

Conclusions
Relying on gene expression and histological studies in symbiotic mutants defective in arbuscule branching or in the formation of phosphate-transporting arbuscules as well as on functional analyses in transgenic knockdown roots or knockout lines, we provide evidence that the network of AM-activated M. truncatula GRAS TFs consists of interconnected modules, including an MtRAM1-MtGRAS1 regulatory feedback loop. MtGras1 knockdown roots displayed normal colonization by AM fungi, but a trend towards the formation of smaller arbuscules was apparent. Although mutants in the AM-specific GRAS TFs RAM1 and RAD1 display more severe AM-related phenotypes in M. truncatula and L. japonicus 33,34,36,42 , our investigation on additional AM-activated MtGras genes provides evidence that a regulatory circuit of multiple GRAS TFs, showing differential dependencies on arbuscule branching and the formation of phosphate-transporting arbuscules, governs and sustains arbuscule development. We propose that this regulatory circuit allows a flexible response of the symbiotic interface towards the external (e. g. nutrient availability 50,53 ; and internal (e. g. plant hormones 20,41 ) stimuli that together influence and adapt the arbuscule life cycle under changing conditions.

Cloning of promoter-gusAint fusions and histological analyses. Promoter sequences of MtGras
genes were amplified from genomic DNA of M. truncatula using oligonucleotides specified in Supplementary  Table S5. PCR-fragments were cloned into pGUS-INT 56 , in front of the gusAint reporter gene cassette. The resulting transcriptional fusions were released using SpeI and subcloned into the SmaI-digested binary vector pRed-Root 57 , after fill-in of 5′ overhangs using the Klenow fragment.
GUS assays were performed by incubating roots in GUS staining buffer 56 for 4 to 8 hours at 37 °C, if not stated differently. Counterstaining of fungal material was performed using Alexa WGA Fluor 488 (Thermo Fisher Scientific, Langenselbold, Germany), as described above.
Cloning of knock-down and overexpression constructs. An RNAi construct for MtGras1 was generated by amplification and recombination of a 379 bp long fragment of the MtGras1 coding region into pDON-R TM 221 (Gateway ® -System, Invitrogen, Karlsruhe, Germany) via the BP-, and subsequently into the binary vector pK7GWIWG2(II)-Q10:DsRED 58 via the LR-reaction. Specificity of the MtGras1 RNAi target sequence was verified via BLAST searches against the current release of the M. truncatula genome 51 and by using the siFi software that predicts potential off-targets 59 .
MtGras1 overexpression constructs were generated by PCR-amplification and cloning of the MtGras1 coding sequence into the vectors 315p9RFP-Pt4-Expr and 917p9RFP-ubi3-Expr 15  Transgenic roots were obtained by Agrobacterium rhizogenes ARqua1 mediated transformation of M. truncatula seedlings according to 62 . Bacteria were grown for two days at 30 °C on selective TY (0.5 g/l tryptone; 0.3 g/l yeast extract; 0.07 g/l CaCl 2 × 2H 2 O) agar plates. Cells were resuspended in 10 ml PS buffer ( www.nature.com/scientificreports www.nature.com/scientificreports/ After four weeks of growth (or four weeks after induction of transgenic roots), each plantlet or composite plant was mycorrhized by inoculation in a small amount of water with 2000 spores of germinating Glomus intraradices Schenck and Smith DAOM197198 spores (Premier Tech Biotechnologies, Rivière-de-Loup, Québec, Canada), having been reassigned to Rhizophagus irregularis (Błaszk., Wubet, Renker, and Buscot) C. Walker & A. Schüßler comb. nov. 63 . After 3-4 hours of inoculation, plantlets were potted into 8 × 7 × 7 cm (height × width × depth) pots filled with sterile Seramis ® (Seramis GmbH, Mogendorf, Germany). Each pot contained two plants, and remaining spore solution was directly pipetted onto the root surface while potting. Mycorrhizal and non-mycorrhizal plants were fertilized with half-strength Hoagland's solution 64 containing 20 µM phosphate. The solution was prepared with deionized water, pH was adjusted to 6.4 with KOH.
To visualize fungal colonization, roots were incubated in 10% (w/v) KOH at 95 °C for 7 min, repeatedly rinsed with water and incubated in 1x PBS buffer (0.14 M NaCl, 2.7 mM KCl, 1 mM Na 2 HPO 4 × 2H 2 O, 1.8 mM KH 2 PO 4 ; pH 7.3) containing 20 µg/ml Alexa WGA Fluor ™ 488 (Thermo Fisher Scientific, Langenselbold, Germany) conjugate overnight. Photo documentation was performed using a Leica MZ 10 F stereomicroscope (Leica Microsystems, Wetzlar, Germany) equipped with an Olympus XC50 camera (Olympus, Hamburg, Germany), a Zeiss Axio Observer Z1 microscope equipped with an AxioCam ICc1 (Carl Zeiss AG, Oberkochen, Germany), and a confocal microscope (Leica TCS SP8 MP, Sohns, Germany). Quantification of fungal colonization was performed using the gridline intersection method 65 . For arbuscule size determination, confocal images of arbuscules were analyzed using the Fiji software 66 . Nine independent roots or four independent pools of two individual roots each were used to determine arbuscule sizes, based on the procedure described by 50,53 and 18 .
To study gene expression in transgenic M. truncatula A17 roots expressing an MtGras1 RNAi construct (RNAi:MtGras1 roots) in comparison to a gusAint gene (RNAi:gusAint control roots), composite plants were mycorrhized with R. irregularis spores as described above and harvested after 54 dpi. To analyse gene expression in M. truncatula ram1-1 roots in comparison to M. truncatula A17 control roots, roots were mycorrhized with R. irregularis spores as described above and harvested after 35 dpi. In all experiments, harvesting time points were selected depending on the mycorrhization rate in the different mutants, in order to obtain sufficient and comparable colonization levels.
Analysis of the Tnt1 insertion line NF4813. The Tnt1 54 insertion line NF4813 (based on M. truncatula R108), harbouring a Tnt1 insertion in the exon of MtGras4 after position +1428, was obtained from the Noble Research Institute (Ardmore, Oklahoma, USA). No stable, homozygous Tnt1 lines could be obtained for other candidate genes. Plants were screened for the Tnt1 insertion via direct PCR from leaf discs using the Phire Plant Direct PCR Kit (Thermo Fisher Scientific, Langenselbold, Germany). The PCR was performed using a Tnt1 binding (Tnt1-F) and two gene-specific primers (NF4813_16_for, NF4813_16_rev). Homozygous plants were selected, selfed, and used for seed propagation. Segregating plants being wild type with respect to the MtGras4 locus were propagated to obtain control plants.
RNA isolation and real-time Rt-pCR. RNA was isolated using the RNeasy Plant Mini Kit (Qiagen, Hilden, Germany). Tissue disruption was carried out via FastPrep ® −24 (MP Biomedicals, Santa Ana, USA). Real-time RT-PCR analyses were performed using the SensiFAST ™ SYBR ® No-ROX One-Step Kit (Qiagen, Hilden, Germany), using primers listed in Supplementary Table S6. Primers were tested for specificity before use. 5 ng of total RNA were used as a template in a 20 µl reaction. RT-PCR reactions followed a three-step cycling program: Reverse transcription at 45 °C for 10 min; polymerase activation at 95 °C for 2 min; PCR amplification with 40 cycles at 95 °C for 5 sec, 55 °C for 10 sec, and 72 °C for 8 sec. The housekeeping gene MtTefa (Medtr6g021805.1 in the M. truncatula genome 51 ) encoding a translation elongation factor was used for normalization. Each biological replicate was measured in three technical replicates. Average values were used to calculate gene expression levels via the 2 −ΔCT method with ΔCT = CT gene − CT MtTefa . Statistical significances were calculated using a two-tailed Student's t test in MS Excel 2016 (Microsoft Corp., Redmond, Washington, USA).
Hybridization and data evaluation of GeneChip ® Medicago transcriptome Assays. Biotinylated aRNA obtained from 100 ng of total RNA for each sample was fragmented as recommended (GeneChip ® Medicago Transcriptome Assay, ThermoFisher Scientific, Schwerte, Germany). The size distribution of the fragmented aRNA was assessed via an Agilent bioanalyzer (Agilent Technologies, Böblingen, Germany) using an RNA 6000 assay. Standard hybridization, post-hybridization wash and double-staining as well as scanning was done as specified for GeneChip ® Medicago Transcriptome Assays (ThermoFisher Scientific, Schwerte, Germany).
Cel files were analyzed using the Expression Console and Transcriptome Analysis Console software (both ThermoFisher Scientific, Schwerte, Germany). Normalization was performed via the Robust Multichip Average algorithm, intensity values for each probe set were log2-transformed and averaged across the three biological replicates using the Tukey's Bi-weight average algorithm, and expression ratios were evaluated statistically via tools of the Transcriptome Analysis Console (ThermoFisher Scientific, Schwerte, Germany).
Original annotations of the genes represented on the GeneChip ® Medicago Transcriptome Assays were updated by annotations from the M. truncatula genome version 4.0 51 , and mapped to probe sets from the GeneChip Medicago genome arrays that were used to construct the Medicago Gene Expression Atlas 52 as well as to UniProt 67 . Venn diagrams were drawn using the VENNY software 68 .

Data Availability
All data generated or analysed during this study are included in this published article and its supplementary information files. In addition, GeneChip ® Medicago Transcriptome Assay data are deposited in the Gene Expression Omnibus repository (https://www.ncbi.nlm.nih.gov/geo/, accession number GSE108867).