A network of transcriptional repressors modulates auxin responses

The regulation of signalling capacity, combined with the spatiotemporal distribution of developmental signals themselves, is pivotal in setting developmental responses in both plants and animals1. The hormone auxin is a key signal for plant growth and development that acts through the AUXIN RESPONSE FACTOR (ARF) transcription factors2–4. A subset of these, the conserved class A ARFs5, are transcriptional activators of auxin-responsive target genes that are essential for regulating auxin signalling throughout the plant lifecycle2,3. Although class A ARFs have tissue-specific expression patterns, how their expression is regulated is unknown. Here we show, by investigating chromatin modifications and accessibility, that loci encoding these proteins are constitutively open for transcription. Through yeast one-hybrid screening, we identify the transcriptional regulators of the genes encoding class A ARFs from Arabidopsis thaliana and demonstrate that each gene is controlled by specific sets of transcriptional regulators. Transient transformation assays and expression analyses in mutants reveal that, in planta, the majority of these regulators repress the transcription of genes encoding class A ARFs. These observations support a scenario in which the default configuration of open chromatin enables a network of transcriptional repressors to regulate expression levels of class A ARF proteins and modulate auxin signalling output throughout development. Genes encoding the class A auxin-response factor group of plant transcriptional activators reside in constitutively open chromatin, enabling their continual regulation by transcriptional repressors to modulate auxin signalling throughout development.

The regulation of signalling capacity, combined with the spatiotemporal distribution of developmental signals themselves, is pivotal in setting developmental responses in both plants and animals 1 . The hormone auxin is a key signal for plant growth and development that acts through the AUXIN RESPONSE FACTOR (ARF) transcription factors [2][3][4] . A subset of these, the conserved class A ARFs 5 , are transcriptional activators of auxin-responsive target genes that are essential for regulating auxin signalling throughout the plant lifecycle 2,3 . Although class A ARFs have tissue-specific expression patterns, how their expression is regulated is unknown. Here we show, by investigating chromatin modifications and accessibility, that loci encoding these proteins are constitutively open for transcription. Through yeast one-hybrid screening, we identify the transcriptional regulators of the genes encoding class A ARFs from Arabidopsis thaliana and demonstrate that each gene is controlled by specific sets of transcriptional regulators. Transient transformation assays and expression analyses in mutants reveal that, in planta, the majority of these regulators repress the transcription of genes encoding class A ARFs. These observations support a scenario in which the default configuration of open chromatin enables a network of transcriptional repressors to regulate expression levels of class A ARF proteins and modulate auxin signalling output throughout development.
Previous research aimed at understanding how auxin elicits diverse downstream responses in different tissues has focused on asymmetries in the distribution of the hormone 2,4 . However, differences in expression of signalling components could also contribute to the specificity in auxin response. Among the 23 ARFs in Arabidopsis, ARF5, ARF6, ARF7, ARF8 and ARF19 are class A ARF activators of transcription 3 and key regulators of both embryonic and post-embryonic development [6][7][8][9][10][11][12][13] . In the stem cell niches driving post-embryonic plant development, the root and shoot apical meristems 6 , tissue-specific variation in the expression of class A ARF genes (Fig. 1a, b) is thought to be a key determinant of the diversity of auxin responses 14,15 .
(H3K27me3 and H3K4me3) chromatin modifications, which are implicated in repressing and promoting gene expression, respectively 17 . Meta-analysis of published datasets covering a range of tissues and developmental stages showed that H3K27me3 is largely absent, whereas H3K4me3 is present, at all class A ARF loci (Fig. 1c, Extended Data Fig. 2a-c, Supplementary Table 1). These loci are also characterized by accessible regulatory regions in the majority of tissues (Fig. 1c, Extended Data Fig. 2d, Supplementary Table 1). These properties suggest that the chromatin configuration of class A ARF loci allows them to be actively transcribed at different tissues and developmental stages, implying that the spatial expression pattern specific to class A ARF genes does not result primarily from alternate chromatin states with contrasting accessibility.

Repressors as regulators of class A ARF genes
Alternatively, specific spatiotemporal transcription of class A ARF loci could arise from regulatory networks made up of transcription factors (TFs). To identify TFs that could regulate the transcription of class A ARF genes, we used a semiautomated enhanced yeast one-hybrid (eY1H) assay with baits consisting of promoter sequences identical to those from the transcriptional reporter lines described above. The assay yielded 42 previously unrecognized putative transcriptional regulators of class A ARF genes (Fig. 2, Extended Data Fig. 3a, b, Supplementary Table 2). Analysis of this candidate gene-regulatory network indicated that individual class A ARF loci are likely to be regulated by specific sets of TFs, as only four TFs were identified as binding multiple class ARF sequences. Based on the expression of these TFs, the network may contain proteins that mediate either root-or shoot-specific responses (Extended Data Fig. 3c). Most TFs in the network are involved in development, but many putative regulators of ARF8 are associated with biotic and abiotic stress (Extended Data Fig. 3a, d, Supplementary Table 2). ARF8 may therefore act as an environmental hub mediating auxin responsiveness, and indeed it has been shown to be involved in plant responses to both biotic and abiotic stresses 18,19 .
To validate this regulatory network, we searched the class A ARF promoters for the presence of binding sites for the TFs identified by eY1H. We predicted the presence of many of these TF-binding sites within the ARF promoters and found that a small proportion of the inferred bindings have been confirmed experimentally (Extended Data Fig. 3e-g, Supplementary Table 3, refs. 20,21 ). Next, we systematically tested the regulatory activity of each TF through transient expression analysis using the TFs either alone or fused to the VP16 transactivation domain (Extended Data Fig. 4a Table 4). Together, our data reveal a functional regulatory network controlling the transcription of class A ARF genes and demonstrate that this is regulated by TF-mediated repression.

Expression of class A ARF regulators
If the expression of class A ARF proteins is controlled by tissue-specific transcriptional repression, we would expect many of the repressors involved to have expression patterns complementary to those of their target ARF gene. To test for complementarity of expression with a high spatial resolution, we generated transcriptional reporters for six TFs and investigated them in seven combinations with class A ARF reporters in both root and shoot apical meristem (Fig. 3a, b, Extended Data Fig. 5). We observed complementary expression patterns in the root in five of the seven cases (Fig. 3b, Extended Data Fig. 5a, b). In the shoot, we assessed two combinations involving WRKY11 and At2g26940. We detected WRKY11 only in meristem layers L2/3, whereas its target ARF8 is expressed specifically in layer L1 (Fig. 3a). In the shoot apical meristem, At2g26940 is expressed weakly in the centre of the meristem, whereas ARF19 is expressed, also weakly, in flower primordia (Extended Data Fig. 5c). Hence, repressors and their target ARFs have mostly complementary expression patterns in both shoot and root tissues, although repressors and their targets co-localize in some cells, as for other TFs 22,23 .

Mutants of class ARF regulatory genes
To further test the significance of our results in planta, we characterized mutants of 24 TFs from the regulatory network, representing regulators of all five class A ARFs (Supplementary Table 5). We measured the expression of the target class A ARF genes using qRT-PCR in whole root and shoot tissues (Extended Data Fig. 6   sensitivity of expression analysis on whole tissues could also explain our results. This prompted us to determine at higher spatial definition how TF mutations affect class A ARF expression. We first crossed pARF7::mVENUS and pARF19::mVENUS transcriptional reporters into various TF mutants. For the crf10 and wrky38 mutants, in which our qRT-PCR results had not revealed changes in ARF7 mRNA levels, we observed a significant increase in expression and an expansion of the expression pattern for pARF7::mVENUS in the root apical meristem (Extended Data Fig. 8a, b, h). We also observed enhanced expression of pARF7::mVENUS in the root apical meristem of nf-yb13, in this case in agreement with the qRT-PCR results (Extended Data Fig. 8c, h). However, we saw no changes in the expression of pARF19::mVENUS in the root of three mutants we analysed (Extended Data Fig. 8d-f, h). In the shoot apical meristem, pARF7-driven fluorescence in the nf-yb13 mutant was identical to that in the wild type in layer L1 but elevated in layers L2 and L3, indicating a change in the spatial pattern of pARF7 expression (Fig. 3c, d, Extended Data Fig. 8h). We also detected expression pattern changes for pARF7::mVENUS in the shoot apical meristem of the wrky38 mutant (Extended Data Fig. 8g, h). In addition, inducible constitutive overexpression of AL3 or CRF10 in the pARF7::mVENUS background triggered a decrease in mVENUS signal (Extended Data Fig. 8i, j). These results confirm in planta that four TFs are repressors and provide examples of how such repressors shape the expression level or pattern of class A ARF genes.
To investigate the functional role of this network, we scored the 24 TF mutants for defects in auxin-regulated root processes (  Table 7). Although none of these mutants had previously been implicated in auxin-dependent responses, 58% (14/24) showed altered root length in response to auxin and 29% (7/24) showed altered gravitropism. Among mutants with altered root length response, 64% (9/14) showed an enhanced response, and all mutants with changes in gravitropism had a faster response. Thus, for both traits, a majority of the TF mutants with altered auxin response show effects opposite to those observed for mutants in loci known to promote auxin signalling 12,24 , consistent with a repressive role of the TF. We selected two genes with high auxin responsiveness in the root, IAA13 and IAA19, and tested their expression in the TF mutants. Although we mutated only one TF at a time, we found a small but significant increase in the expression of IAA19 in the roots of seven mutants (~28%), two of which also show elevated levels of IAA13. A reduction in either IAA13 or IAA19 was observed in a further three mutants (~12%; Supplementary Table 8). A significant number of the mutants also had altered shoot phenotypes, further demonstrating that these TFs have important roles in development (Extended Data Fig. 10, Supplementary Table 9). Taken together, our results support a negative regulation of auxin responses by the corresponding TFs. That mutation of single genes in the class A ARF regulatory network can significantly affect auxin-dependent developmental responses further demonstrates the functional importance of individual nodes of this network.

Discussion
Although gene repression mediated by polycomb repressive complex 2 (PRC2) proteins plays a broad role in tissue-specific expression 25 , the general absence of H3K27me3, a hallmark of PRC2 activity, at class A ARF loci indicates that their regulation does not rely on this epigenetic mechanism. This may be because such a system would not allow rapid changes in signalling output. Instead, our data suggest a regulatory system based on the use of transcriptional repressors that modulates expression of constitutively active loci and, in combination with post-translational modifications of class A ARF proteins 26,27 , constantly adjust auxin responsiveness during development. Other transcriptional regulation networks defined in eukaryotes involve both transcriptional activators and repressors 28 . Instead, the network we characterize resembles the early scenario proposed by Jacob and Monod 29 for transcriptional regulation by repressors only, indicating that there may be a place for the concept that the expression of key developmental regulators may be controlled via transcriptional repression.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-020-2940-2.

Plant material and growth conditions
All transgenic lines were generated in the Col-0 accession of Arabidopsis thaliana. T-DNA insertion mutants in transcription factor-coding genes and the arf8-1 mutant were obtained from NASC. All T-DNA lines were genotyped to confirm that they were homozygous, and qRT-PCR was used to confirm alterations in transcript levels (Supplementary Table 5). The accession numbers of T-DNA lines and further details are listed in Supplementary Table 5.
For root microscopy and in situ hybridization of ARF transcriptional reporter lines, plants were grown on half-strength Murashige and Skoog (1/2 MS) medium supplemented with 1% sucrose and 1% agar in 24 h light conditions (microscopy) or 12 h light/12 h dark conditions (in situ hybridization). For shoot microscopy, plants were grown in 8 h light/16 h dark conditions for 6 weeks and then transferred to 16 h light/8 h dark conditions for 2 weeks to induce bolting. For the qRT-PCR experiments, the seedlings were grown in 24 h light conditions on 1/2 MS plates containing 1% sucrose and 1% agar for 7 d. For the root imaging of crosses between ARF transcriptional reporter lines and TF mutants and for the co-expression analysis of ARF transcriptional reporter lines with TF transcriptional reporter lines, the plants were grown on 1/2 MS medium supplemented with 0.8% agar in 16 h light/8 h dark light. TF overexpression lines were grown for 12 h light/12 h dark light on 1/2 MS medium supplemented with 1% agar.
All constructs were transformed into Agrobacterium tumefaciens C58pMP90 strain by electroporation and then transformed into Col-0 plants by the floral dip method 30 .
The ARF promoter sequences screened in the eY1H assay were amplified by PCR and sequenced to confirm absence of mutations. The overall ARF promoters screened correspond in length and content to those used in the construction of the transcriptional reporter lines except that the longer promoters were split into two fragments: pARF5 fragment 1: bp -2796 to +134; pARF5 fragment 2: bp -5418 to -2481; pARF6: bp -3255 to +197; pARF7: bp -2973 to +374; pARF8 fragment 1: bp -2899 to +42; pARF8 fragment 2: bp -5091 to -2121; pARF19 fragment 1: bp -2399 to +457; pARF19 fragment 2: bp -4906 to -1992. The amplified fragments were cloned into either pDONR P4P1R or pENTR 5′ TOPO plasmids by the Gateway BP reaction or using the pENTR 5′-TOPO kit, respectively. The resulting plasmids were recombined with the Gateway LR reaction into both pMW2 and pMW3 Gateway destination vectors designed for yeast expression and containing respectively HIS3 or LacZ reporter genes 31 . The resulting plasmids were transformed into the yeast strain YM4271.
Additional transcription factors were cloned and added to the collection of existing root-specific transcription factors (Supplementary Table 10). The transcription factors were amplified by a PCR from cDNA collections obtained by isolating total RNA from various tissues. Each full-length transcription factor cDNA PCR product (without a stop codon) was inserted into a pENTR-Zeo plasmid by the Gateway BP reaction and then recombined into the pDEST-AD-2μ destination vector designed for yeast expression and containing a GAL4 activation domain 31 . The vectors were transformed into the yeast strain Yα1867.
To produce the reporter plasmid for the protoplast assays, the promoter fragment of the respective ARF corresponding to the one used in the eY1H assay, and the ARF transcriptional reporter lines described above, were amplified by PCR and cloned into the plasmid pDONR P4-P1R. For the ARF8 promoter, a short part of the 35S promoter (bp -107 to +1) was inserted at bp -115. Separately, a construct containing an NLS followed by the mVenus coding sequence and an OCS terminator was cloned into the plasmid pDONR 211. Third, a construct containing the promoter of RPS5a (encoding ribosomal protein S5A) driving TagBFP followed by an NLS and a nosT terminator were cloned into the plasmid pDONR P2R-P3. These three plasmids were recombined using a multisite Gateway method to yield the final reporter plasmid pARF-NLS-mVenus-term-pRPS5a-TagBFP-NLS-term. An alternative reporter plasmid contained a shorter ARF promoter fragment that contained sequences upstream and lacked sequences downstream of the start codon (corresponding to the transcriptional reporter lines with shorter promoters described above). To create the effector plasmid for the protoplast assays, the RPS5a promoter was cloned into pDONR P4-P1R; the cDNA of the respective transcription factor without the stop codon was cloned into pDONR 211; and the construct, containing the self-cleaving 2A peptide 32,33 followed by mCherry coding sequence, a NLS and a nosT terminator, was cloned into pDONR P2R-P3. Finally, these three plasmids were recombined with a multisite Gateway reaction to yield pRPS5a-cDNA-2A-mCherry-NLS-term. An alternative effector plasmid included an activator VP16 domain from the herpes simplex virus fused to the TF cDNA.

Microscopy
Roots of ARF transcriptional reporter lines were imaged 5 d after germination. Plant cell walls were visualized by staining with 15 μg ml −1 propidium iodide solution. Roots were examined using a TCS-SP5 confocal microscope (Leica) with excitation at 514 nm and emission at 526-560 nm for mVenus and 605-745 nm for propidium iodide.

eY1H assay
The eY1H assay was conducted according to 31 . The ARF promoters screened correspond in length and content to those used in the construction of the transcriptional reporter lines except that the longer promoters (pARF5, pARF8 and pARF19) were split into two fragments (see Cloning section). With the longer promoters, only 1 out of 39 TFs was identified using the distal fragment of the ARF8 promoter. This suggests that the other 38 TFs bind in a region of the promoter from bp -2480 to +134 for ARF5, bp -2120 to +42 bp for ARF8 and bp -1991 to +457 for ARF19.
We used a TF collection enriched in root-expressed TFs 31 expanded with additional TFs involved either in development of the shoot apical meristem or in hormonal regulation (see Supplementary Table 10).

Transient expression analysis in Arabidopsis protoplasts
For the protoplast assay Col-0 seedlings were grown in short-day conditions ( A Zeiss 710 LSM confocal microscope was used for imaging the protoplasts (Extended Data Fig. 4). Sequential scanning was performed with mVenus (excitation at 514, emission at 520-559), TagBFP (excitation at 405 and emission at 423-491), mCherry (excitation at 561, emission at 598-636) and bright-field channels. z stacks of several protoplasts were taken. The data were analysed using ImageJ software (imageJ.net/Fiji). The image with the best focus for each protoplast was selected from the z stack. The nucleus was selected and the mean fluorescence was measured as illustrated in Extended Data Fig. 4. The number of replicates was between 15 and 54 protoplasts with a majority of experiments including at least 20 protoplasts. For most ARF-TF interactions, 4 or 5 independent experiments were performed (Supplementary Table 4): 2 or 3 experiments with the standard effector plasmid and 2 experiments with alternative effector plasmid containing VP16 domain. For the statistical analysis, we first run a Kruskal-Wallis H-test on all controls for a given set of experiments (TF or TF-VP16). At a significance level of 0.05, all tests rejected the null hypothesis that control populations have the same median, indicating that the data could not be pooled. The results for each experiment was analysed independently using a one-sided Mann-Whitney U-test to test for a significant effect of TF or TF-VP16 and to identify the direction of the change. To take into account the results from several experiments of a given type (TF or TF-VP16), we performed a meta-analysis using the method of Mudholkar and George 34 to combine the P values from the independent experiments. This allows us to obtain 'meta P values' per type of experiment. Note that the meta P value was calculated only if the Mann-Whitney test was significant (with a significance level of 0.05) in at least one of the repetitions.

Expression analysis with qRT-PCR
The whole root and the whole shoot parts of the seedlings were collected separately. For one root sample, roots from 30 seedlings grown on the same plate were pooled together. For one shoot sample, 8 shoots from seedlings grown on the same plate were pooled together. Three independent replicates per genotype were collected. RNA was extracted using Spectrum Plant Total RNA kit (Sigma-Aldrich). The DNA was removed using TURBO DNA-free kit (Invitrogen). The cDNA was produced using SuperScript VILO cDNA Synthesis kit (Thermo Fischer) with 500 ng RNA. The cDNA was diluted 1:100 before use. The qRT-PCR was performed using Applied Biosystems Fast SYBR Green Master Mix. Expression of TUB4 gene was used as standard. The statistical analysis was performed with a one-sided Mann-Whitney test, with P < 0.1 considered as statistically significant. IAA13 and IAA19 were chosen as auxin-responsive genes for qRT-PCR analysis in roots from ref. 35 .

Expression analysis of crosses between ARF transcriptional reporter lines and TF mutants
Mutants of the regulatory transcription factors were crossed with pARF7-mVenus transcriptional reporter line described above. The crosses were selected for the presence of homozygous pARF7-mVenus reporter construct. The F3 generation wild-type and mutant plants were compared.
The roots of 5 d-old plants were stained with 15 μg ml −1 propidium iodide and imaged using the TCS-SP8 (Leica) confocal microscope with excitation at 514 nm and emission at 526-560 nm for mVenus and 605-745 nm for propidium iodide.
For the shoot microscopy the images were taken with a Zeiss 710 LSM confocal microscope. mVenus intensity was measured separately in L1 and in L2/L3 layers in each of the 8 cross-sections with 50 nm distance between each cross-sections. Number of replicates: 7 WT and 7 mutant plants for nf-yb13, 12 WT and 12 mutant plants for wrky38.
Roots of the plants grown for 5-10 d were imaged using the TCS-SP8 (Leica) confocal microscope, with excitation at 514 nm and emission at 526-560 nm for mVenus and excitation and emission at 587 nm and 610-670 nm respectively for mCherry. Total fluorescence was calculated for individual nuclei from two or three individual roots using a 6-px circular selection in ImageJ. These values were then normalized for each channel based on a scale of 0-1 with the brightest nuclei in each root being set to a value of 1. The shoots were examined using the TCS-SP8 (Leica) confocal microscope, with excitation at 514 nm and emission at 526-560 nm for mVenus and excitation and emission at 587 nm and 610-670 nm respectively for mCherry.

Inducible overexpression of TFs
Multisite Gateway cloning technology was used to generate TF inducible overexpression lines. The chimaeric transcription activator p1R4-pG1090:XVE 36 containing XVE followed by the rbs and nos terminators and LexA operon, expressed under UBQ10 promoter was recombined with TF coding sequence (lacking STOP codon) in pDONR211 and the 2A-mCherry-term pDONR P2R-P3 (containing the self-cleaving 2A peptide 32,33 followed by the mCherry coding sequence, a nuclear localization sequence (NLS) and a nosT terminator) and pB7m34GW (the destination vector containing basta resistance gene for in planta selection) to produce pUBQ10-XVE-TF-2A-mCherry oestradiol-inducible constructs. These constructs were transformed in the pARF7-mVenus transcriptional reporter line background by floral dip method 30 .
For the overexpression analysis, roots of the plants grown for 5 d were treated with 10 μM β-oestradiol for 24 h and imaged using the TCS-SP8 (Leica) confocal microscope, with excitation at 514 nm and emission at 526-560 nm for mVenus and excitation and emission at 587 nm and 610-670 nm, respectively, for mCherry.

Shoot phenotype analysis of the TF mutants
24 T-DNA insertion mutants and the wild-type Col-0 were grown in 8 h light/16 h dark conditions on soil for 43 d. Leaf number was counted every 3 d starting from day 24. Rosette diameter was measured at 43 d. After 43 d of growth in the above conditions, the plants were transferred to 16 h light/8 h dark conditions to induce bolting. The following parameters were measured after 21 and 27 d in the 16 h light/8 h dark conditions: length of the main stem, number of cauline branches growing from the main stem, number of axillary branches growing from rosette (the main stem not included). The number of replicates per genotype was 12 plants. For the statistical analysis, an unpaired two-tailed t-test was conducted with P ≤ 0.05 considered as statistically significant.

Root phenotype analysis of the TF mutants
For root length measurement and for gravitropic analysis plants were grown on 1/2 MS medium supplemented with 1% agar in 12 h light/12 h dark conditions. For root length analysis, plants were grown either on medium lacking IAA or supplemented with 10 μM IAA. To reduce plate-to-plate variation wild-type plants and mutants were grown on the same agar plate. Images were taken at 15 d and the root length was measured. The number of replicates per genotype was at least 26 plants without IAA and 15 plants with IAA. For the gravitropic response, plants were grown for 5 d, then turned at a 90° angle and images taken every 1 h for 12 h in the dark using an infrared camera. The number of replicates per genotype was at least 26 plants. Rootnav v.1.8 software (https:// www.nottingham.ac.uk/research/groups/cvl/software/rootnav.aspx) was used for data analysis. Statistical analysis was done with unpaired two-tailed t-test with P ≤ 0.05 considered as statistically significant.

In silico analyses
Analysis of expression and function of regulatory TFs. Expression of TFs in the root and the shoot apical meristems was analysed using cell type-specific expression profiles from refs. [38][39][40] .
Overrepresentation of TF gene families was analysed for families represented by two or more members in the network. The number of gene family members in the network was compared to total number of genes from the same family in the TF library. Statistical analysis was done using a hypergeometric test, with P ≤ 0.05 considered as statistically significant.
Involvement of TFs in specific developmental processes (development, biotic and abiotic stress) was analysed based on literature description.
Chromatin state analysis. Binary data on H3K27me3-and H3K4me3marked genes and chromatin accessibility regions were retrieved from multiple datasets covering a range of tissues and developmental stages. For each dataset, at least two biological replicates were considered, and only the presence of a given ARF in both gene lists was scored as a positive association with a chromatin mark or an accessible region.
Visualization of epigenomic data was carried out using the IGV software 50,51 .
Binding motif search and reanalysis of DAP-seq data. Position weight matrices (PWM) available for TFs identified in the eY1H screen were retrieved from the Jaspar 52 and CisBP 53 databases. Using these PWMs, we computed the best score of the TF binding sites present in each Arabidopsis 2-kb promoter with an R script using the Biostrings library (https://bioconductor.org/packages/release/bioc/html/Biostrings. html) and ranked the class A ARF gene promoter among all Arabidopsis promoters based on this score. As negative control, this operation was repeated identically five times for each class A ARF promoter with 20 randomly selected TFs (excluding specific TF classes and families identified in the eY1H screen). The distributions of class A ARF promoter ranks with eYI1H-selected and randomly selected TFs were compared using a one-sided t-test.
DAP-seq files containing the peak list from ref. 20 were retrieved (GEO accession number GSE60141). Bedtools intersect (bedtools.readthedocs. io/en/latest/index.html) was then used with the -wb option to determine which DAP peak overlap with each promoter.

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

Data availability
The data including the source data that supports the finding of this study are available within the paper, its supplementary information files or publicly available datasets. Publicly available position weight matrices were obtained from the Jaspar and CisBP databases. Publicly available chromatin marking and accessibility datasets were acquired from the Article Extended Data Fig. 1 ARF5 (a, f), ARF6 (b, g), ARF7 (c, h), ARF8 (d, i) and ARF19 (e, j) in the RAM and the SAM using promoters that lack sequences downstream of the start codon but contain the long upstream sequences (pARF −intron ::mVenus) (~3 kb for ARF6 and ARF7; 5 kb for ARF5, ARF8 and ARF19) (see Methods). For SAM images (f-j) an orthogonal projection is shown below to provide information about expression in different layers. k-o, For comparison, the expression of each class A ARF gene in the SAM using the previously published pARF::GFP lines with shorter (~2 kb) promoters containing sequences upstream of the start codon is shown in panels k-o 14 .

| Analysis of class A ARF expression in the RAM and the SAM using transcriptional reporter lines and in situ hybridization. a-j, Confocal images showing expression of
ARF5 (k), ARF6 (l), ARF7 (m), ARF8 (n) and ARF19 (o). (p-r) In situ hybridizations through the RAM for ARF5 (p), ARF6 (q) and ARF8 (r). Note that expression patterns of the class A ARF reporters (a-j) differ from those with shorter (2 kb) promoters (k-o 14 ) and recapitulate the patterns observed with RNA in situ hybridization (p-r; ref. 16 ). This was particularly clear in the shoot for ARF5 and ARF6. Shorter promoters drive GFP expression mostly in flower boundaries for ARF5 and throughout the meristem for ARF6, in contrast with detection of both genes throughout the periphery of the meristem both with longer promoters (k-o; also Fig. 1f-j) or using in situ hybridization 15 . Experiments were done three (a-e) and two times (f-r). Scale bars: 50 μm. Fig. 2 | See next page for caption.

Article
Extended Data Fig. 2 | Distribution of the repressive chromatin marker H3K27me3, the active chromatin marker H3K4me3 and chromatin accessibility at class A ARF loci. a, Chromatin landscape of class A ARF and LEC2 in whole seedlings illustrating the chromatin status of class A ARF loci. Repressive H3K27me3 marker (top row), active H3K4me3 marker (middle row) and FANS-ATAC chromatin accessibility (bottom row; see Supplementary  Table 1). b, c, Chromatin landscape of class A ARF and LEC2 loci showing distribution of the repressive chromatin marker H3K27me3 (a) and the active chromatin marker H3K4me3 (b) in various tissues. Seedling, whole seedlings 17 ; leaf, rosette leaves 42 ; root, whole roots 17 ; seedling 2, whole seedlings 44 ; SAM, shoot apical meristems after 0, 1, 2 or 3 d in long-day conditions 44 . Gene models are shown below with arrowheads indicating direction of transcription. d, The chromatin landscape of class A ARF and LEC2 loci showing chromatin accessibility in various tissues. DNaseI-seq seedling: DNase I hypersensitive sites in whole seedling 46 ; DNaseI-seq root: DNase I hypersensitive sites in root 46 ; FANS-ATAC seedling: FANS-ATAC accessible regions in whole seedling 47 ; FANS-ATAC roots: FANS-ATAC accessible regions in roots 47 ; INTAC-ATAC root tip: INTACT-ATAC transposase hypersensitive sites in root tips 48 . The LEC2 locus is included as a negative control for H3K4me3 marking and chromatin accessibility, and as a positive control for H3K27me3 marking 54 . The y axis scales (at right) show the minimum and maximum number of reads represented in each windows of the same row, except for the data set related to ref. 17  Article Extended Data Fig. 3 | Characterization of the TFs and TF binding sites that regulate class A ARF expression. a, Yeast one-hybrid promoter-transcription factor interaction network for class A ARF genes. Green boxes correspond to the class A ARF; pink boxes are transcription factors binding to the ARF promoters. TF-associated functions and expression analysis are indicated in the upper and lower small boxes and colour-coded as indicated in the key. Note that when two promoter fragments were used for the screen (see Methods), 35 out of 36 regulators bound to the more proximal fragment, supporting previous observations that the majority of transcription factor binding sites reside within a few kb of the transcriptional start site 55 . b, Frequency of TF gene families in the Y1H library collection (black) and in the Y1H network (white). Only families represented by at least two members in the Y1H network were analysed. The network is overrepresented with members of the WRKY and SPL TF families. Statistical analysis: hypergeometric test significant to 5% (*; P = 4e-05 for WRKY family and P = 0.044 for SPL family). Sample sizes for TFs in Y1H library in black/Y1H network in white: n = 29/8 TFs (WRKY); n = 68/6 (ZFP); n = 91/6 (AP2/ERF); n = 44/2 (NAC); n = 7/2 TFs (SPL); n = 52/2 TFs (homeobox); n = 61/2 TFs (bHLH). c, TF expression in the RAM 38 and the SAM 39,40 . 50% of the identified TFs are expressed in both shoots and roots, whereas 24% and 14% are expressed specifically in roots or shoots respectively. d, Known functions of the TFs in the Y1H network based on a literature search (see also Supplementary Table 2). e, Boxplot representation of the distribution of class A ARF promoter ranks. For TFs with established binding models, we ranked class A ARF promoters among all Arabidopsis promoters based on the score of the predicted TF binding sites. We repeated the same operation with a set of randomly chosen TFs from different families (see Methods). The comparison of rank distributions with those of a set of randomly chosen TFs from different families revealed significantly higher ranks for eY1H-identified TFs (see also Supplementary Table 3). Statistical analysis: one-sided t-test. Sample sizes: n = 29 for eY1H-selected TFs and n = 100 for randomly selected TFs. Data are represented as boxplots where the middle line is the median, the lower and upper hinges correspond to the first and third quartiles, the upper whisker extends from the hinge to the largest value no further than 1.5× interquartile range (IQR) from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5× IQR of the hinge. All the individual values are plotted. f, Summary of the DAP-seq analysis for the 17 TFs (see also Supplementary Table 3). g, Example of DAP-seq data, here a DAP-seq peak for WRKY33 in the promoter of ARF8. DAP-seq (f, g) thus confirms experimentally inferred bindings (e) for 4 of the 17 (24%) TFs for which DAP-seq data are available (see also Supplementary Table 3). Note also that chromatin immunoprecipitation sequencing (ChIP-seq) confirms the binding of WUSCHEL to the ARF8 promoter 21 . Fig. 4 | Methodology used for the transient protoplast assay. a, Design of the standard reporter plasmid containing sequences upstream and downstream of the ARF promoter including the first intron (1), the alternative reporter plasmid containing only sequences upstream of the ARF promoter (2), the standard effector plasmid (3), and an alternative effector plasmid containing the VP16 domain fused to the TF coding sequence (4). b, Example of a nucleus of a transformed living protoplast imaged with confocal microscopy with channels for mVenus, TagBFP, mCherry and bright-field. The presence of TagBFP and mCherry specifically in the nucleus is used as a transformation control and as a test of viability of the protoplasts. Quantification: definition of the nucleus as a region of interest using ImageJ to quantify fluorescence (see also Methods). Measurements were conducted in at least 4 independent experiments for each TF (minimum of 2 experiments for TF alone and 2 experiments for TF fused to VP16 domain). Scale bars, 10 μm. c, d, Example of results using the ARF5 reporter plasmid, with (c) and without (d) the VP16 activator domain fused to the TF coding sequence (left and right).  Fig. 3), we saw complementary patterns of expression between transcriptional repressors and their ARFs in the root. b, To further quantify the complementarity of TF versus ARF expression, we quantified the red versus green fluorescence levels in individual nuclei from different cell types (root cap, blue diamond; columella, green triangle; epidermis, red square; vascular cells, purple cross). These values were normalized so that the brightest nucleus of each channel in each line was set to 1, and values were plotted onto scatter plots. Any value falling outside the reference lines shows a >4× bias for expression of either TF or ARF (n = 3 for pAT2G26940::mCherry and pAT2G44730::mCherry in pARF8::mVenus; n = 2 for the remaining genotypes). In some cases there was clear complementarity in some cell types but not others. For example, ZFP6 shows complementary expression patterns in the root cap, epidermis and columella but overlaps with ARF8 in the vascular tissues. c, Analysis of At2g26940 expression in the SAM, where it was found in organ primordia and weakly in the centre of the SAM; no clear expression was observed in roots. As previously observed with other developmental and hormonal regulators 22,23 , co-localization of repressors and their target ARF occurs in some cells as in the case of ZFP6/ARF8 in the root epidermis (a, b) and At2g26940/ARF19 in shoot organ primordia (c), suggesting potential regulatory interactions to modulate transcription levels. Scale bars, 60 μm (a) and 40 μm (c). Experiments were done twice (a, c). Fig. 6 | Expression of class A ARF in mutants for the regulatory transcription factors. Expression of class A ARF in 24 mutants of the regulatory TFs measured with qRT-PCR, in whole root and whole shoot tissue of 7-d-old seedlings. Green boxes indicate statistically significant upregulation of the corresponding ARF in the mutant background compared to wild-type control, and blue boxes indicate statistically significant downregulation. Statistical analysis was performed using a one-sided Mann-Whitney test and a threshold at P ≤ 0.1. For simplicity, only the interactions predicted by the Y1H are shown, with other combinations shaded with a grey box. The full data set is available in Supplementary Table 6. Fig. 8 | Modulating the levels of ARF transcriptional regulators regulates the expression of associated ARFs. a-f, Comparison of ARF expression in wild-type versus mutants in roots. g, Comparison of pARF7::VENUS expression in wild-type versus wrky38 shoot. For quantification (see f), fluorescence was measured in the central zone and primordia 2 (green circles). h, Quantification of fluorescence changes shown as relative changes in mean fluorescence level in mutant compared to wild type (single value). Quantifications are shown for a-g and for Fig. 3c,d. In roots, the total pARF7/19driven fluorescent signal was quantified within a standardized zone covering the stele meristem zone and quantified relative to the wild-type controls. In the shoot, L1 and L2 correspond to quantification in the corresponding layers in the SAM of wild-type and nf-yb13 (see also Fig. 3c, d). Quantification demonstrated a significant change in pattern in wrky38 mutant SAMs (g), with an increase of pARF7 activity in the centre and a loss of the differential expression between the SAM centre and lateral organs. Statistical analysis: unpaired two-sided t-test with P ≤ 0.01 (**). Number of samples observed and quantified: for mutant/wild type roots, 13/13 for crf10, 12/14 for wrky38, 9/9 for nf-yb13, 9/8 for At2g26940, 12/11 for myb65, 12/10 for nlp5; 7 shoots for nf-yb13 and wild-type controls; 7 shoots for wrky38 and 6 wild-type controls. P values from left to right: 0.003, 2e-05, 3e-08, 0.26, 0.57, 0.11, 0.84, 0.007, 0.009. Raw data are provided in Supplementary Table 11. i, Inducible constitutive overexpression of CRF10:mCherry and AL3:mCherry in the pARF7::VENUS line. pARF7::VENUS is shown in yellow and the transcription factors fused to mCherry in red following a 24h induction with β-oestradiol. j, Both lines shown in i show a significant reduction in pARF7::VENUS expression. Unpaired two-sided t-test: P = 4e-10 (CRF10) and 2e-10 (AL3). Number of plants: wild-type control, n = 15; CRF10, n = 21; AL3, n = 20. Error bars: mean ± s.d.. Scale bars: 45 μm for root images; 50 μm for shoot images. For each analysis, the confocal settings were identical in the compared genetic backgrounds. All experiments were done two times.